A New Family of Potent AB₅ Cytotoxins Produced by Shiga Toxigenic Escherichia coli

AB₅ toxins produced by pathogenic bacteria comprise an A subunit with enzymic activity and a B subunit pentamer responsible for interaction with glycolipid receptors on target eukaryotic cells (1). The three AB₅ toxin families recognized to date are the Shiga toxin (Stx), cholera toxin and the related Escherichia coli heat labile enterotoxins, and pertussis toxin (Ptx). In each case, they are key virulence determinants of the bacteria that produce them (Shiga toxigenic E. coli [STEC] and Shigella dysenteriae, Vibrio cholerae, and enterotoxigenic E. coli, and Bordetella pertussis, respectively). Collectively, these pathogens cause massive global morbidity and mortality, accounting for millions of deaths each year, particularly amongst children in developing countries. The AB₅ toxins exert their catastrophic effects by a two-step process involving B subunit–mediated entry of their respective target cells, followed by A subunit–dependent inhibition or corruption of essential host functions. The A subunits of Stx have RNA-N-glycosidase activity and cleave 28S rRNA, thereby inhibiting host protein synthesis. The A subunits of cholera toxin/labile enterotoxin and Ptx are ADP ribosylases that modify distinct host G proteins, resulting in alteration of intracellular cAMP levels and disregulation of ion transport mechanisms (1).

STEC are an important cause of gastrointestinal disease in humans, particularly because these infections may result in life-threatening sequelae such as the hemolytic uremic syndrome (HUS; 2, 3). STEC are a very diverse group comprising >200 E. coli O:H serotypes (2), but epidemiological data indicate that not all of these are highly virulent for humans. Thus, although Stx is generally considered to be a sine qua non of virulence, additional STEC properties undoubtedly contribute to the pathogenic process (2, 3). Moreover, there has been a finding of strains of E. coli O157:H7 and O157:H⁻ that do not produce Stx being associated with cases of human gastrointestinal disease, including HUS (4). Here, we demonstrate that certain STEC strains produce a hitherto unknown yet highly lethal AB₅ toxin. We have characterized the prototype of this new toxin family, which was secreted by a highly virulent O113:H21 STEC strain responsible for an outbreak of HUS.

Bacterial strains, plasmids, and oligonucleotides used in this work are listed in Tables I and II. The O113:H21 STEC strain 98NK2 was isolated from a patient with HUS at the Women's and Children's Hospital, South Australia, as described previously (5). Other clinical STEC strains used in this work were also isolated at the Women's and Children's Hospital. All E. coli strains were routinely grown in Luria-Bertani (LB) medium (6) with or without 1.5% bacto-agar. Where appropriate, ampicillin or kanamycin were added to growth media at a concentration of 50 μg/ml.

E. coli CWG308:pJCP-Gb₃, expressing a modified lipopolysaccharide that mimics the Stx receptor Gb₃ (7) was grown overnight in LB broth supplemented with 20 μg/ml isopropyl-β-d-thiogalactopyranoside (IPTG), and 50 μg/ml kanamycin. Cells were harvested by centrifugation, washed, and resuspended in PBS at a density of 10⁹ CFU/ml. 250 μl of 98NK2 culture supernatant was incubated with 500 μl of this suspension for 1 h at 37°C with gentle agitation. The mixture was centrifuged, filter sterilized, and assayed for cytotoxicity. The same procedure was also used to compare toxin neutralization using derivatives of E. coli CWG308 expressing mimics of Gb₄, lactoneotetraose, and GM₂ (reference 8 and unpublished data). Neutralization of cytotoxicity (%) was calculated as described previously (7).

All tissue culture media and reagents were obtained from Life Technologies. Vero (African green monkey kidney) cells were grown at 37°C in DMEM supplemented with 10% heat inactivated FCS, 50 IU penicillin, and 50 μg/ml streptomycin, unless otherwise indicated. Chinese hamster ovary (CHO) cells were grown in Ham's F12 medium, whereas human colonic epithelial (Hct-8) cells were grown in RPMI 1640 medium. For cytotoxicity assays, cells were seeded into 96-well flat-bottom trays and incubated overnight at 37°C until confluent. Confluent monolayers were washed twice with PBS, treated with 50 μl of filter-sterilized toxin extracts that had been serially diluted in the appropriate tissue culture medium (without FCS), and incubated at 37°C for 30 min. After incubation, 150 μl of medium supplemented with 2% FCS was added per well. Cytotoxicity was assessed microscopically after 3 d of incubation at 37°C. The toxin titer was defined as the reciprocal of the maximum dilution producing a cytopathic effect on at least 50% of the cells in each well (CD₅₀/ml).

Recombinant DNA experiments were approved by the Office of the Gene Technology Regulator (Australia) and were performed under PC2 level containment. Routine DNA manipulations (restriction digestion, agarose gel electrophoresis, ligation, transformation of E. coli, Southern hybridization analysis, etc.) were performed essentially as described previously (6). For DNA sequencing, a plasmid DNA template was purified using a QIAPrep Spin miniprep kit (QIAGEN). The sequence of both strands was determined using dye-terminator chemistry and either universal M13 sequencing primers or custom-made oligonucleotide primers on an automated DNA sequencer (model 3700; Applied Biosystems).

The subA, subB, or both subA and subB (subAB) open reading frames (ORFs) were amplified from 98NK2 genomic DNA by PCR using primer pairs SubAF/SubAR, SubBF/SubBR, and SubAF/SubBR, respectively, using the Expand™ High Fidelity PCR system (Roche Molecular Diagnostics), according to the manufacturer's instructions. The purified PCR products were blunt cloned into SmaI-digested pK184, and transformed into E. coli JM109. Recombinant plasmids were extracted from transformants and confirmed by sequence analysis. In all cases, the inserts were in the same orientation as the vector lac promoter.

To raise specific antisera, we first purified SubA and SubB using a QIAexpress kit (QIAGEN). The subA and subB ORFs, without the 5′ signal peptide-encoding regions, were amplified by high fidelity PCR using 98NK2 genomic DNA template and primer pairs pQEsubAF/pQEsubAR and pQEsubBF/pQEsubBR, respectively. Purified PCR products were digested with BamHI–SacI or SphI–SacI, respectively, ligated with similarly digested pQE30, and transformed into E. coli M15. Correct insertion of the genes into the vector, such that the recombinant plasmids encode derivatives of SubA and SubB with His₆ tags at their NH₂ termini, was confirmed by sequence analysis. For purification of His₆-fusion proteins, transformants were grown in 1 L of LB supplemented with 50 μg/ml ampicillin and, when the culture reached an A₆₀₀ of 0.5, the culture was induced with 2 mM IPTG and incubated for an additional 3 h. Cells were harvested by centrifugation, resuspended in 24 ml buffer A (6 M guanidine-HCl, 0.1 M NaH₂PO₄, 10 mM Tris, pH 8.0), and stirred at room temperature for 1 h. Cell debris was removed by centrifugation at 10,000 g for 25 min at 4°C. The supernatant was loaded (at a rate of 15 ml/h) onto a 2 ml column of nickel nitrilotriacetic acid (Ni-NTA) resin (ProBond; Invitrogen), which had been preequilibrated with 20 ml buffer A supplemented with 0.5 M NaCl and 15 mM imidazole. The column was washed with 40 ml buffer A, followed by 20 ml buffer B (8 M urea, 0.1 M NaH₂PO₄, 10 mM Tris, pH 8.0), and 16 ml buffer C (8 M urea, 0.1 M NaH₂PO₄, 10 mM Tris, pH 6.3) supplemented with 0.25 M NaCl and 5 mM imidazole. The fusion proteins were eluted with a 30-ml gradient of 0–500 mM imidazole in buffer C, and 3-ml fractions were collected and analyzed by SDS-PAGE. Peak fractions were pooled, and the denatured SubA and SubB were refolded by dialysis against 100 ml of buffer B to which 1 L of PBS was added dropwise at a rate of 60 ml/h. This was followed by dialysis against two changes of PBS. The purified SubA and SubB were >95% pure, as judged by SDS-PAGE after staining with Coomassie brilliant blue R250.

BALB/C mice were immunized by intraperitoneal injection of three 10-μg doses of purified SubA or SubB in 0.2 ml PBS containing 5 μg of alum adjuvant (Imjectalum; Pierce Chemical Co.) at 2-wk intervals. Mice were exsanguinated by cardiac puncture 1 wk after the third immunization, and pooled antisera were stored in aliquots at −15°C.

Crude lysates or culture supernatants of E. coli strains, or purified proteins were separated by SDS-PAGE (10), and antigens were electrophoretically transferred onto nitrocellulose filters (11). Filters were probed with polyclonal mouse anti-SubA or anti-SubB sera (used at a dilution of 1:5,000), or monoclonal antibody to His₆ (QIAGEN), followed by goat anti–mouse IgG conjugated to alkaline phosphatase (Bio-Rad Laboratories). Labeled bands were visualized using a chromogenic nitro-blue tetrazolium/X-phosphate substrate system (Roche Molecular Diagnostics).

A derivative of JM109:pK184subAB with a point mutation such that the predicted active site serine residue (S₂₇₁) in SubA was altered to alanine was constructed by overlap extension PCR mutagenesis. This involved high fidelity PCR amplification of pK184subAB DNA using primer pairs SubAF/SubOLR and SubOLF/SubBR. This generates two fragments with the necessary mutation in codon 271 of SubA incorporated into the overlapping region by the SubOLR and SubOLF primers. The two separate PCR products were purified and mixed together, and the complete subAB region was reamplified using primer pair SubAF/SubBR. The resultant PCR product was blunt-cloned into SmaI-digested pK184, and transformed into E. coli JM109. Recombinant plasmids were purified from the resultant transformants and subjected to sequence analysis to confirm that the mutation had been introduced, and that the modified subAB operon was inserted in the vector in the same orientation as in pK184subAB. This construct was designated pK184subA_A271B.

Nonpolar subA and subB deletion mutants of STEC 98NK2 were constructed using the λ red recombinase system (12). This involved high fidelity PCR amplification of the kanamycin resistance cartridge in pKD4 using primer pairs SubAmutF/SubAmutR and SubBmutF/SubBmutR, incorporating the direct repeated FLP recognition target common priming site and sequences derived from the 5′ and 3′ ends of the subA or subB genes, respectively. The resultant linear fragments were electroporated into 98NK2 carrying the temperature-sensitive plasmid pKD46, which encodes the λ recombinase. Allelic replacement mutants were selected on LB-kanamycin plates at 37°C. Replacement of nucleotide (nt) 169–908 of the subA coding sequence or nt 83–352 of subB with the kanamycin resistance cartridge was confirmed by PCR and sequence analysis of the mutants, which were designated 98NK2ΔsubA and 98NK2 ΔsubB, respectively.

RNA was extracted from log-phase LB cultures using TRIzol reagent, according to the manufacturer's instructions (Life Technologies). RNA was precipitated in 1/10 volume of sodium acetate, pH 4.8, and 2 volumes of 100% ethanol at −80°C overnight. RNA was pelleted by centrifugation at 12,000 g for 30 min at 4°C, washed in 70% ethanol, and resuspended in nuclease-free water. RNasein ribonuclease inhibitor (Promega) was added to the samples. Contaminating DNA was digested with RQ1 RNase-free DNase, followed by DNase stop solution, according to the manufacturer's instructions (Promega).

The comparative levels of subA, subB, and subAB transcripts were determined using quantitative real-time RT-PCR, using primer pairs RTsubAF/RTsubAR, RTsubBF/RTsubBR, and RTsubABF/RTsubABR, respectively. These direct amplifications of 220-bp, 238-bp, and 232-bp fragments were within subA, within subB, or spanning subA and subB, respectively. RT-PCR was performed using the one-step access RT-PCR system (Promega) according to the manufacturer's instructions. Each reaction was performed in a final volume of 20 μl, containing 20 nmol of each oligonucleotide, and a 1/20,000 dilution of Sybr green I nucleic acid stain (Molecular Probes). The quantitative RT-PCR was performed on a cycler (Rotorgene model RG-2000; Corbett Research) and included the following steps: 45 min of reverse transcription at 48°C, followed by 2 min denaturation at 94°C, and 40 cycles of amplification using 94°C for 30 s, 56°C for 30 s, and 72°C for 45 s.

To purify the SubAB holotoxin, the complete subAB coding region was amplified by high fidelity PCR using 98NK2 DNA template and the primer pair pETsubAF/pETsubBR. The resultant PCR product was digested with BamHI and XhoI, ligated with similarly digested pET-23(+), and transformed into E. coli Tuner™(DE3). This resulted in IPTG-dependent production of both the SubA and SubB proteins (including their respective signal peptides), but with a His₆ tag fused to the COOH terminus of SubB. Correct insertion of the genes into the vector was confirmed by sequence analysis. Cells were grown in 1 L of LB supplemented with 50 μg/ml ampicillin and, when the culture reached an A₆₀₀ of 0.5, the culture was induced with 5 mM IPTG and incubated for an additional 3 h. Cells were harvested by centrifugation, resuspended in 20 ml of loading buffer (50 mM sodium phosphate, 300 mM NaCl, pH 8.0), and lysed in a French pressure cell. Cell debris was removed by centrifugation at 100,000 g for 1 h at 4°C. The supernatant was loaded onto a 2-ml column of Ni-NTA resin that had been preequilibrated with 20 ml of loading buffer. The column was washed with 40 ml wash buffer (50 mM sodium phosphate, 300 mM NaCl, 10% glycerol, pH 6.0). Bound proteins were eluted with a 30-ml gradient of 0–500 mM imidazole in wash buffer, and 3-ml fractions were collected and analyzed by SDS-PAGE.

Purified SubAB was treated with 0.5% formaldehyde for 60 min at room temperature and heated at 60°C for 10 min before SDS-PAGE analysis to determine the size of the holotoxin. Purified E. coli heat labile enterotoxin (unpublished data), which is known to have AB₅ stoichiometry, was treated and analyzed in parallel.

Vero cells were grown on glass coverslips in 24-well tissue culture plates and treated with or without 1 μg/ml of purified SubAB. After 1 or 48 h, cells were fixed with 4% formaldehyde in PBS for 10 min, and in some cases permeabilized with 0.1% Triton X-100. Coverslips were washed in PBS and blocked with 20% FCS in PBS for 1 h at 37°C. They were treated with either anti-SubA, anti-SubB, or nonimmune mouse serum (diluted 1:800 in PBS/10% FCS) for 2 h at 37°C. After three washes with PBS, the coverslips were reacted with goat anti–mouse IgG–ALX488 conjugate (Molecular Probes), diluted 1:250 in PBS/10% FCS, for 30 min at 37°C. The coverslips were washed three times with PBS, twice with water, dried, and mounted on glass slides using 3 μl of Mowiol solution with antibleach. Slides were examined with a microscope (model IMT-2; Olympus) equipped with epifluorescence optics, using a 60× oil-immersion apochromatic objective.

Crude lysates of STEC strains were subjected to PCR amplification using primer pair RTsubABF/RTsubABR. Alternatively, HindIII digests of genomic DNA purified from the STEC strains were transferred to nylon membranes and probed at high stringency with a digoxigenin-labeled subAB DNA fragment obtained by PCR amplification of pK184subAB using primer pair subAF/SubBR.

Animal experimentation was conducted in accordance with the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes, and was approved by the Animal Ethics Committee of the University of Adelaide. Groups of eight 5–6-wk-old BALB/C mice, each weighing ∼17–19 g, were given oral streptomycin (5 mg/ml in drinking water) for 24 h before oral challenge with ∼10⁸ CFU of a streptomycin-resistant derivative of E. coli DH5α (DH5α^SR) carrying pK184, pK184subAB, or pK184subA_A271B, suspended in 60 μl of 20% sucrose and 10% NaHCO₃. Drinking water was supplemented with 5 mg/ml streptomycin and 100 μg/ml kanamycin. Mice were weighed daily, and numbers of the recombinant bacteria in fecal samples from each group were monitored by plating on LB agar supplemented with 50 μg/ml streptomycin and 50 μg/ml kanamycin. Alternatively, pairs of BALB/C mice were injected intraperitoneally with either 25 μg, 5 μg, 1 μg, or 200 ng purified SubAB in 0.1 ml PBS.

Antibodies to SubAB were measured by ELISA using 96-well Costar PVC plates that were coated overnight at 4°C with 100 μl of 5 μg/ml of purified SubAB in TBS (25 mM Tris-HCl, 132 mM NaCl, pH 7.5). Plates were washed with TBS-0.1% Triton X-100 and blocked with TBS-0.05% Tween-20, 0.02% BSA (TBS-Tween-BSA) for 2 h at 37°C. Plates were washed again and incubated for 4 h at 37°C with 100 μl of serial dilutions of mouse serum in TBS-Tween-BSA, commencing at 1:50. Plates were washed, incubated with goat anti–mouse IgG alkaline phosphatase conjugate (EIA grade; Bio-Rad Laboratories), and diluted 1:15,000 in TBS-Tween-BSA for 2 h at 37°C. Plates were washed and developed with 1 mg/ml p-nitrophenyl phosphate substrate (in 12.5 mM triethanolamine, 135 mM NaCl, 0.02% BSA, 1 mM MgCl₂, 2.5 μM ZnCl₂, pH 9.6) for 2 h at 37°C, after which Absorbance at 450 nm was determined. Absorbance above background was plotted against serum dilution, and the ELISA titer was defined as the reciprocal of the serum dilution resulting in an A₄₅₀ reading of 0.2 above background.

Initially, we discovered the new toxin by testing fresh culture supernatant of the O113:H21 STEC strain 98NK2 for residual cytotoxicity on Vero cells, after absorption with a recombinant E. coli strain that binds and neutralizes all members of the Stx family with high avidity. The latter construct expresses a modified LPS, which mimics the Stx receptor (globotriaosyl ceramide; Gb₃; reference 7). The absorbed 98NK2 supernatant exhibited significant residual cytotoxicity, with a titer of ∼1,280 50% cytotoxic doses (CD₅₀) per milliliter, compared with 10,240 CD₅₀/ml for unabsorbed supernatant. A similar degree of residual cytotoxicity was also observed in supernatant from a derivative of 98NK2 with a deletion mutation in its single Stx-encoding gene (13). The cytopathic effect was maximal after 3 d of incubation and was characterized by rounding of cells, detachment from the substratum, and loss of viability (judged by Trypan blue exclusion).

To isolate the novel cytotoxin genes, we tested culture supernatants from a 98NK2 cosmid gene bank previously constructed in E. coli DH1 (14) for Vero cytotoxicity. Two cosmid clones with partially overlapping inserts were cytotoxic (titers were ∼1,280 CD₅₀/ml). The inserts of these cosmids do not contain stx genes, and are derived from a 36.8-kb portion of the 98NK2 megaplasmid pO113, the sequence of which has been deposited in GenBank/EMBL/DDBJ (accession no. AF399919.3). The organization of genes within the region from nt 5,000 to 17,000 in this sequence is represented in Fig. 1. Within this region, there are two closely linked genes that we have designated subA and subB. The subA gene is located on the complementary strand (nt 13,725–14,768 of AF399919.3) and is preceded by a ribosome binding site (GGAGGAG; nt 14,772–14,778). A putative promoter sequence was identified using the NNPP program (15) with transcription predicted to start at nt 14,831. The subA gene encodes a 347–amino acid (aa) putative secreted protein with a modest degree of similarity to members of the Peptidase_S8 (subtilase) family of serine proteases (pfam00082.8). Its closest bacterial relative is the BA_2875 gene product of Bacillus anthracis (26% identity, 39% similarity over 246 aa). The deduced aa sequence includes a predicted signal peptide cleavage site (determined using the program SignalP V1.1; reference 16) between A₂₁ and E₂₂, which was subsequently confirmed by NH₂-terminal aa sequence analysis of isolated protein. PROSITE analysis also indicated that SubA contains three conserved sequence domains, designated the catalytic triad, characteristic of members of the subtilase family (17). The SubA domain sequences match the consensus sequences for the so-called Asp, His, and Ser subtilase catalytic domains at 11/12, 10/11, and 10/11 positions, respectively, including the known active site residues (Fig. 2).

The subB gene is 16 nt downstream of subA (nt 13,283–13,708 of AF399919.3) and is preceded by a ribosome-binding site (GGAGG; nt 13,714–13,718). An additional putative promoter sequence was identified upstream of subB with transcription predicted to start at nt 13,774. A potential stem-loop element (ΔG = −19.6 kcal/mol) located immediately downstream of subB (nt 13,176–13,206) may function as a transcription terminator for both subA and subB because no such elements were identified downstream of subA. The subB gene encodes a 141-aa protein with significant similarities to putative exported proteins from Yersinia pestis (YPO0337; 56% identity, 79% similarity over 136 aa) and Salmonella typhi (STY1891; 50% identity, 68% similarity over 117 aa). STY1891 has similarity (30% identity over 101 aa) to the S2 subunit of Ptx, but there is negligible similarity between SubB and the latter. Like SubA, the deduced aa sequence of SubB includes a predicted signal peptide cleavage site between A₂₃ and E₂₄, which was also confirmed by NH₂-terminal analysis.

To examine the cytotoxicity of their products, we amplified subA, subB, or both subA and subB (subAB) by PCR, subcloned them into pK184, and transformed them into E. coli JM109. Culture supernatant of JM109:pK184subAB was strongly cytotoxic for Vero cells (>40,960 CD₅₀/ml). As observed with Stx-absorbed 98NK2 culture supernatant, the cytopathic effect was maximal after 3 d of incubation and was characterized by rounding of cells, detachment from the substratum, and loss of viability (Fig. 3). However, culture supernatants of JM109:pK184subA and JM109:pK184subB were not cytotoxic (<10 CD₅₀/ml). Western blot analysis of the supernatants using polyclonal murine antisera raised against purified SubA or SubB confirmed that the appropriate clones produced immunoreactive species of the expected sizes (35 and 13 kD for SubA and SubB, respectively; Fig. 4). Cell lysates of the clones were also tested on Vero cells, and that of JM109:pK184subAB was at least 10 times more cytotoxic than the respective culture supernatant, which is consistent with poor release of secreted proteins from the periplasm of E. coli K-12 strains. CHO and Hct-8 cells were also susceptible to the JM109:pK184subAB culture supernatant, albeit to a lesser extent (toxin titers were 2,000 and 250 CD₅₀/ml, respectively). CHO cells are known to be refractory to Stx (18), whereas Hct-8 cells are sensitive (19).

The requirement for both subA and subB for cytotoxicity was confirmed by constructing nonpolar subA and subB deletion mutants of STEC 98NK2. Replacement of nt 169–908 of the subA coding sequence or nt 83–352 of the subB coding sequence with a 1.6-kb kanamycin resistance cassette was confirmed by PCR and sequence analysis. The resulting mutants were designated 98NK2ΔsubA and 98NK2ΔsubB, respectively. Western blot analysis confirmed that the former produced SubB, but not SubA, whereas the latter mutant produced SubA, but not SubB, as expected (unpublished data). The cytotoxicity of culture supernatants and cell lysates of 98NK2ΔsubA and 98NK2ΔsubB were examined using CHO cells, which as aforementioned, are susceptible to SubAB, but refractory to the effects of Stx. Unlike the wild-type 98NK2 extracts, those from either of the mutants had undetectable cytotoxicity. Thus, cytotoxicity requires the presence of both subA and subB.

To determine whether polyclonal murine anti-SubA or anti-SubB were capable of neutralizing toxin activity, serial dilutions of toxin were preincubated with 10-μl volumes of serum at 37°C for 30 min and assayed for residual cytotoxicity on Vero cells. No neutralization was detected using nonimmune mouse serum, but anti-SubA and anti-SubB neutralized 6,400 and 3,200 CD₅₀ of toxin per ml, respectively (unpublished data).

To determine the extent to which the cytotoxicity of SubAB was dependent on its putative subtilase activity, we constructed a derivative of JM109:pK184subAB with a point mutation such that the predicted active site serine residue (S₂₇₁) in SubA was altered to alanine. Culture supernatant from this derivative (designated JM109:pK184subA_A271B) contained both anti-SubA– and anti-SubB–reactive species of 35 and 13 kD, respectively (Fig. 4). However, supernatant and cell lysate fractions from this clone exhibited markedly reduced cytotoxicity for Vero cells, with titers of 40 and 320 CD₅₀/ml, respectively (Fig. 3). Thus, the point mutation reduced specific cytotoxicity by >99.9%. Therefore, we have named the new toxin “Subtilase cytotoxin.”

We assessed transcription of subA and subB in 98NK2 and JM109:pK184subAB by real-time RT-PCR using primer pairs that direct amplification of ∼230-bp fragments within subA, within subB, or spanning subA and subB. RNA templates from both strains yielded similar quantities of RT-PCR product with all three primer sets (unpublished data). This indicates that the subA and subB ORFs are cotranscribed.

The aforementioned clear requirement for both SubA and SubB for cytotoxicity strongly suggests that the two proteins function together. To examine whether they form an active complex (i.e., an AB_n holotoxin), we subcloned a DNA fragment containing the complete subAB region into the expression vector pET-23(+) such that a His₆ tag was fused to the COOH terminus of the expressed SubB protein. We subjected lysates of E. coli Tuner™(DE3) expressing this construct to Ni-NTA affinity chromatography. Proteins were eluted from the column with a 0–500-mM imidazole gradient, and fractions were analyzed by SDS-PAGE and Coomassie blue staining, as well as by Western blot using polyclonal anti-SubA or monoclonal antibody to the His₆ tag (Fig. 5). The earlier fractions (nos. 3 and 4) contained multiple protein species, including small amounts of anti-SubA–reactive material. However, all the later fractions (references 6–8 and unpublished data) contained only two protein species with sizes of 35 and 14 kD, as predicted for SubA and SubB, respectively (allowing for the extra His₆ at the SubB COOH terminus). These species reacted strongly with anti-SubA and anti-His₆, respectively. Examination of the Coomassie blue–stained SDS-PAGE gel indicated that the SubA and SubB species were present in apparently constant proportions in each of the fractions (∼1:5 on a molar basis, as judged by densitometry). However, the purified SubAB migrated as a single species when subjected to PAGE under nondenaturing conditions, and was not dissociated by treatment with 5% 2-mercaptoethanol (unpublished data). Further confirmation of the stoichiometry of the association between SubA and SubB was obtained by subjecting purified SubAB to mild cross-linking conditions before SDS-PAGE analysis, which indicated that the holotoxin has a molecular size of ∼105 kD (unpublished data). Collectively, these data indicate that SubA and SubB form a stable complex under nondenaturing conditions, at a ratio of 1:5.

Purified SubAB was highly toxic for Vero cells, with a specific activity >10¹⁰ CD₅₀/mg. That is, <0.1 pg of SubAB is sufficient to kill at least 50% of the ∼3 × 10⁴ Vero cells present in a microtiter plate well. For comparison, the specific Vero cell cytotoxicities reported for both major Stx types (Stx1 and Stx2) are in the range of 1–10 pg/CD₅₀ (20–24). Heating SubAB at 56°C for 30 min had no impact on in vitro cytotoxicity, but exposure to 75°C for 30 min resulted in ∼75% inactivation (unpublished data). This degree of heat stability is similar to that reported for Stx2 (20). Like the aforementioned crude extracts tested, SubAB cytotoxicity was maximal after 72 h incubation of toxin-treated Vero monolayers, and there was little evidence of a cytopathic effect at 24 h, even at high toxin doses. Interestingly, however, if Vero cells were treated with ∼1,000 CD₅₀/ml of SubAB for 60 min, followed by removal of the medium, washing of the monolayers three times with fresh medium, and continuation of incubation in fresh medium, significant cytotoxicity was still evident 48–72 h later. This suggests that significant amounts of SubAB were already either tightly bound to the Vero cell surface, or had entered the cells within the first hour.

We examined entry of SubAB into Vero cells directly, by immunofluorescence microscopy (Fig. 6). After 48-h exposure of Vero cells to 1 μg/ml of purified SubAB, both anti-SubA– and anti-SubB–reactive material was clearly evident within the cytoplasm. No significant labeling was seen in toxin-treated cells after staining with nonimmune mouse serum, or in nontoxin-treated cells stained with the specific antisera. Furthermore, if SubAB-treated cells were not permeabilized before staining, very little immunoreactive material was observed. Thus, most of the detectable SubAB appeared to be inside the Vero cells, rather than bound to the outer surface (Fig. 6). Alternately, when toxin-treated cells were examined by immunofluorescence after only 1 h, significant staining of toxin-treated Vero cells was observed using anti-SubA or anti-SubB, regardless of whether the cells were permeabilized or not, suggesting that much of the toxin was bound to the outer surface. No labeling was observed using nonimmune mouse serum, or if Vero cells were incubated without toxin (unpublished data).

The B pentamers of previously characterized AB₅ toxins are known to recognize specific oligosaccharide moieties displayed by host cell glycolipids (1). Differences in receptor specificity of the toxins, as well as in the distribution of the target glycolipids between host species and tissues, has a major impact on host susceptibility and tissue tropism, and the pathology and clinical manifestations of toxin-mediated disease (25). In an attempt to identify candidate glycolipid receptors for SubAB, we absorbed toxin extracts with suspensions of recombinant E. coli strains expressing mimics of the oligosaccharide components of glycolipids Gb₃, Gb₄, lactoneotetraosyl ceramide, and GM2 (references 7, 8 and unpublished data). We tested the absorbed extracts for Vero cytotoxicity. No detectable neutralization of cytotoxicity was observed after absorption with any of the first three constructs, or with the host E. coli strain used to express the oligosaccharides. However, absorption with the GM2 mimic neutralized 93.4% of the SubAB activity. Thus, the oligosaccharide expressed by this strain (GalNAcβ [1→4]{NeuAcα[2→3]}Galβ[1→4]Glcβ−) may be a functional receptor for SubAB. To examine whether Vero cells contain GM2, gangliosides were extracted and analyzed by thin layer chromatography. The extract contained at least eight species, as judged by staining with resorcinol, one of which comigrated with a commercial GM2 standard (unpublished data).

The in vivo toxicity of SubAB was examined by intraperitoneal injection of pairs of mice with 25 μg, 5 μg, 1 μg, or 200 ng of purified toxin. All of the mice died and their survival times were inversely related to dose levels, ranging from 2 d at 25 μg to 8–10 d at 200 ng. Death was preceded by ataxia and hind limb paralysis, suggestive of neurological involvement. Histological examination of organs removed from a moribund toxin-treated mouse revealed extensive microvascular thrombosis and necrosis in the brain, liver, and kidneys (Fig. 7). One out of two mice injected with 1 μg of partially heat-inactivated (75°C for 30 min) SubAB died after 8 d, whereas the other was alive and well at 14 d.

We also challenged groups of streptomycin-treated mice orally with E. coli DH5α derivatives carrying pK184, pK184subAB, or pK184subA_A271B. The drinking water was supplemented with streptomycin and kanamycin to inhibit endogenous gut flora and to select for plasmid maintenance. Each of the constructs was maintained at levels of ∼10⁸–10⁹ CFU/g of feces throughout the experiment. During this period, none of the mice exhibited any obvious diarrhea. The mice challenged with the control strain DH5α:pK184 or DH5α:pK184subA_A271B remained healthy and active, and gained weight steadily. By day 6, mice in these groups had gained 1.71 ± 0.21 and 1.98 ± 0.18 g, respectively. However, the mice colonized with DH5α:pK184subAB appeared ill and lethargic, and steadily lost body weight during the first 6 d after challenge. By day 6, they had lost 2.78 ± 0.41 g of body weight (Fig. 8 A), which is equivalent to 15.7% of their mean starting weight on day 0 (17.7 g). Even on day 3, the difference in weight gain between mice challenged with DH5α:pK184subAB and the other two groups was highly significant (P < 0.01; Student's t test). The severe weight loss experienced by the group challenged with the active SubAB-producing clone indicates that toxin delivered via the gut has significant deleterious effects on the host. Moreover, the fact that the growth of mice challenged with the clone expressing the SubAB protein with the mutation in the active site Ser residue was indistinguishable from that of mice challenged with DH5α:pK184 unequivocally attributes the weight loss to subtilase-mediated cytotoxic activity. Interestingly, the mice challenged with DH5α:pK184subAB started to gain weight from day 7, although they lagged significantly behind the other two groups for the entire duration of the experiment (P < 0.001; Fig. 8). To determine whether seroconversion could account for this apparent recovery, sera collected from each of the mice on day 15 were tested for antibodies to SubAB by ELISA. Only one of the mice challenged with DH5α:pK184 had detectable anti-SubAB levels, and this was at the lower limit of detection (titer = 50). However, seven out of eight mice challenged with DH5α:pK184subAB and five out of eight mice challenged with DH5α:pK184subA_A271B seroconverted; the highest titers for these groups were 3,100 and 720, respectively (Fig. 8 B). One out of two mice injected intraperitoneally with 1 μg of purified SubAB that had been preincubated for 1 h at 37°C with 100 μl of convalescent serum from one of the mice challenged with DH5α:pK184subAB died after 8 d, whereas the other was alive and well at 14 d. Preincubation of toxin with normal mouse serum had no such protective effect, with death occurring at 4 d (unpublished data).

The distribution of subAB-related sequences in other STEC strains isolated from patients with HUS and/or diarrheal disease, or from contaminated food linked to an outbreak of HUS, was investigated by PCR and Southern hybridization analysis (Table III). The genes are not present in the two published O157:H7 STEC genome sequences (26, 27). However, subAB sequences were present in 32 out of 68 other STEC strains tested, including representatives of serogroups O23, O48, O82, O91, O111, O113, O123, O128, O157, OX3, and O nontypable strains. The subAB probe-reactive strains included O111:H⁻ and O157:H⁻ isolates linked to a large outbreak of HUS (28), and lysates of these strains were cytotoxic for (Stx resistant) CHO cells (unpublished data). The subAB hybridization signal obtained for the O111:H⁻ and O157:H⁻ isolates was not as strong as that seen for the other subAB-reactive strains, suggesting the possibility of sequence differences (unpublished data). All of the subAB positive STEC strains were also positive by PCR for the STEC enterohemolysin gene ehxA, which is commonly used as a marker for the presence of an STEC megaplasmid (unpublished data). The presence of subAB in diverse clinical isolates may be a consequence of the fact that at least in 98NK2, the operon is located on a megaplasmid (pO113) that is capable of conjugative transmission (29). However, STEC megaplasmids are heterogeneous and several of the subAB positive strains tested in this work have been previously shown to lack the pilS gene, which facilitates such transfer (29).

In the present work, we have characterized a novel Subtilase cytotoxin at both the gene and protein level. SubAB clearly belongs in a separate family to the other AB₅ toxins characterized to date, as it has distinct A subunit enzymic activity (subtilase rather than RNA-N-glycosidase or ADP-ribosylase). Moreover, we have demonstrated that the potent cytotoxicity and deleterious in vivo effects of SubAB are a consequence of this subtilase activity. The subtilases are a family of serine proteases found in a wide variety of microorganisms (17, 30), but to date, no other members have been shown to have cytotoxic activity. In the present work, we also exposed Vero monolayers to 1 μg/ml purified Subtilisin Carlsberg (a prototype subtilase from Bacillus licheniformis; Sigma-Aldrich) and observed no cytotoxic effect whatsoever (unpublished data). SubA did not appear to have broad-spectrum proteolytic activity and was unable to cleave substrates such as collagen or fibronectin, which might have accounted for the detachment of tissue culture cells from the substratum (unpublished data). Collagen cleavage was assessed colorimetrically using dye-linked hide powder substrate, or by electrophoresis in gelatin-impregnated nondenaturing PAGE gels. Capacity to digest fibronectin was assessed by coincubation of purified human fibronectin with SubAB followed by analysis by SDS-PAGE. Immunofluorescence microscopy demonstrated that SubA and SubB bound to the surface of Vero cells within 1 h and, at 48 h, were detectable inside toxin-treated cells, but the intracellular substrate of SubA is still unknown. Nevertheless, the substrate is clearly essential for cell survival, and elucidation of this target may enable SubAB, like Ptx, to be used as a tool in cell biology.

The presence of SubB was essential for cytotoxicity, and it is likely that it is required for recognition and/or entry of target cells. The neutralization of SubAB activity achieved by treatment with the GM2 mimic probiotic suggests that this ganglioside (or one displaying a closely related oligosaccharide) may be a functional receptor for the toxin. To confirm this specificity, we attempted to neutralize SubAB with micelles of GM2 purified from bovine brain, but without success. However, the conformation of oligosaccharide moieties displayed by glycolipids in biological membranes is also heavily influenced by their own lipid component, as well as by other lipids present. This has a major impact on their capacity to interact with binding sites on AB₅ toxins (31), and purified glycolipids generally exhibit weak binding when presented in micelle form.

The evolutionary origin of Subtilase cytotoxin is unclear, but the data presented here demonstrate the potentially dire consequences that might arise from genetic rearrangements that bring seemingly innocuous genes such as subA and subB into juxtaposition. The closest bacterial homologue of SubA is BA_2875 from B. anthracis, but examination of the genome sequence of the latter did not reveal the presence of a gene encoding a homologue of SubB in the immediate vicinity. To our knowledge, there is also no precedent in the literature for members of the subtilase family of proteases forming stable associations with heterologous polypeptides that impact on biological activity (17, 30). Examination of the Y. pestis and S. typhi genome sequences also did not reveal the presence of subA-like genes in the vicinity of their respective subB homologues, both of which encode products of unknown function. Nevertheless, the degree of similarity between these proteins and SubB is substantial (56% identity and 79% similarity over 136 aa in the case of Y. pestis YPO0337), and this raises the possibility that they are structurally and functionally related. Given the impact of plague and typhoid on human health, it would be of considerable interest to determine whether these SubB homologues can (a) form pentamers capable of binding to eukaryotic cells, and (b) interact with heterologous proteins produced by the respective organism.

The production of two distinct and highly potent AB₅ toxins (Subtilase cytotoxin and Stx) by a bacterium responsible for life-threatening human disease is an important finding that raises the possibility that both contribute (perhaps synergistically) to pathogenesis in some cases of STEC disease. Typically, the relative contributions of virulence factors can be dissected by examination of the behavior of toxin mutants in an animal model. However, existing animal models do not mimic all of the features of STEC disease in humans. Interspecies and age-related differences in receptor distribution have a major impact on host susceptibility, tissue tropism, and the resultant pathology generated by a toxin. Nevertheless, the presence of microvascular thrombi in the brain and other organs, including the renal tubules and glomeruli, of a Subtilase cytotoxin–treated mouse is suggestive of endothelial injury, and is reminiscent of the pathology seen in cases of HUS in humans. This finding is particularly intriguing in the light of the report of strains of E. coli O157:H7 and O157:H⁻ (common STEC serotypes) that do not produce Stx, being associated with HUS (4).

The presence of subAB in diverse STEC isolates from cases of severe human disease demands rigorous investigation of the toxin's biological effects in vitro and in vivo. The fact that subAB is carried on a mobile DNA element and its presence in a diverse range of E. coli O serogroups also raises the possibility of further transmission to other enteric bacteria. If an unequivocal role for Subtilase cytotoxin in disease in humans or animals becomes apparent, the work presented here will provide the foundation for effective diagnostic, therapeutic, and preventative strategies. We have reported PCR primers suitable for use in direct detection of subAB-carrying bacteria in complex clinical and environmental samples. We have demonstrated that a Ser₂₇₁-Ala substitution in SubA virtually abolishes cytotoxicity of SubAB, and that expression of this protein in the GI tract of mice has no adverse consequences, yet elicits a serum antibody response. Thus, we have identified a safe candidate vaccine antigen. Finally, by demonstrating that a harmless strain of E. coli expressing a mimic of the oligosaccharide component of ganglioside GM2 neutralizes SubAB, we have identified a means of absorbing Subtilase cytotoxin in the gut of infected individuals. Previously, we have demonstrated the in vivo efficacy of this receptor-mimic therapeutic strategy using a mimic of the Stx receptor (7).

We wish to thank L. van den Bosch and J. Cook for assistance with immunofluorescence and animal experimentation, respectively, and R. Morona and M. Jennings for helpful discussions.

This research was supported by project grant 207721 from the National Health and Medical Research Council of Australia.