The crystal structure of staphylococcal superantigen-like protein 11 in complex with sialyl Lewis X reveals the mechanism for cell binding and immune inhibition



This article is corrected by:

  1. Errata: The crystal structure of staphylococcal superantigen-like protein 11 in complex with sialyl Lewis X reveals the mechanism for cell binding and immune inhibition Volume 67, Issue 2, 473, Article first published online: 18 December 2007

*E-mail; Tel. (+64) 9 373 7599 86036; Fax (+64) 9 373 8774.


Staphylococcus aureus is a major pathogen that produces a family of 14 staphylococcal superantigen-like (SSL) proteins, which are structurally similar to superantigens but do not stimulate T cells. SSL11 is one member of the family that is found in all staphylococcal strains. Recombinant SSL11 bound to granulocytes and monocytes through a sialic acid-dependent mechanism and was rapidly internalized. SSL11 also bound to sialic acid-containing glycoproteins, such as the Fc receptor for IgA (FcαRI) and P-selectin glycoprotein ligand-1 (PSGL-1), and inhibited neutrophil attachment to a P-selectin-coated surface. Biosensor analysis of two SSL11 alleles binding to sialyl Lewis X [sLex– Neu5Acα2-3Galβ1-4(Fuc1-3)GlcNAc] coupled to bovine serum albumin gave dissociation constants of 0.7 and 7 μm respectively. Binding of SSL11 to a glycan array revealed specificity for glycans containing the trisaccharide sialyllactosamine (sLacNac – Neu5Acα2-3Galβ1-4GlcNAc). A 1.6 Å resolution crystal structure of SSL11 complexed with sLex revealed a discrete binding site in the C-terminal β-grasp domain, with predominant interactions with the sialic acid and galactose residues. A single amino acid mutation in the carbohydrate binding site abolished all SSL11 binding. Thus, SSL11 is a staphylococcal protein that targets myeloid cells by binding sialyllactosamine-containing glycoproteins.


Staphylococcus aureus is a common human pathogen and the most frequent cause of hospital-acquired infection. The bacterium produces an array of virulence proteins designed to subvert the human immune system (Foster, 2005), including the newly described staphylococcal superantigen-like (SSL) proteins (Williams et al., 2000; Kuroda et al., 2001), which are structurally similar to superantigens (Sags). The genes encoding the SSLs are found in all strains of S. aureus and reside in an almost contiguous region within a pathogenicity island (SaPIn2) (Fitzgerald et al., 2003; Aguiar-Alves et al., 2006) that can be transferred through excision and encapsulation into infectious phage-like particles (Lindsay et al., 1998; Ubeda et al., 2005).

Alignments of SSL and Sag amino acid sequences indicate that the two families have evolved from the same ancestral gene, although they have diverged to the extent that they now display minimal global sequence similarity. Nevertheless, their crystal structures reveal a shared architecture that comprises a small N-terminal OB-fold domain (Agrawal and Kishan, 2003) and a slightly larger C-terminal domain of the β-grasp type (Papageorgiou and Acharya, 2000; Arcus et al., 2002).

Despite their obvious similarity to Sags, none of the SSLs bind major histocompatibility complex (MHC) class II or T-cell receptor (TCR) molecules and do not stimulate T cells (Arcus et al., 2002; Al-Shangiti et al., 2004; Langley et al., 2005). SSL7 is one member whose function has been determined. It binds with nanomolar affinity to the Fc region of IgA and blocks interaction with the myeloid IgA receptor FcαRI (CD89) (Langley et al., 2005; Wines et al., 2006; Ramsland et al., 2007). In addition, SSL7 binds and inhibits complement C5 and serum bactericidal activity (Langley et al., 2005). SSL5 has also been studied (Arcus et al., 2002; Bestebroer et al., 2006). It has the highest amino acid sequence identity to SSL11 out of all the SSLs, and binds the cell surface protein PSGL-1 to block neutrophil tethering to a P-selectin-coated surface (Bestebroer et al., 2006).

Less is known about SSL11. It is expressed in all staphylococcal strains examined, and is under control of the SaeRS two-component regulatory system (Rogasch et al., 2006). Loss of SSL11 appears to coincide with reduced staphylococcal adherence to the surface of a surgical implant in a rat surgical infection model (Gan et al., 2002). SSL11 exists in several alleles, with 54% sequence identity between four published S. aureus genomes, MW2 (Baba et al., 2002), Col (Gill et al., 2005), Mu50 and N315 (Kuroda et al., 2001). Most of the amino acid variation is located in the N-terminal region as a result of a relatively recent recombination event with the 5′hsdM (Fitzgerald et al., 2003).

In this paper, SSL11 is shown to bind to cells via a sialic acid-dependent mechanism, and the molecular basis for this activity is confirmed by high-resolution crystal structures of SSL11, alone and in complex with the carbohydrate cell-adhesion ligand sLex. Further analysis shows that SSL11 targets glycoproteins expressing the common trisaccharide sialyllactosamine (sLacNac) and is rapidly internalized by myeloid cells.


SSL11 binds to neutrophils and monocytes

Freshly isolated human peripheral leucocytes were incubated with FITC-labelled SSL11 (allele GL10) and counter-stained with PE-labelled antibodies that define the major leucocyte cell subsets. Stained cells were analysed by fluorescence-activated cell sorting (FACS). Significant binding to monocytes (CD14) and granulocytes (CD10) was observed, with only limited staining of T cells (CD3) and no detectable staining of B cells (CD19) (Fig. 1A).

Figure 1.

Sialic acid dependence of SSL11 binding.
A. FACS analysis of human leucocytes double labelled with SSL11-FITC (x-axis) and CD19 B-lymphocytes, CD3 T-lymphocytes, CD10 granulocytes and CD14 monocytes (y-axes).
B. Recombinant soluble FcαRI protein treated (+) or untreated (–) with neuraminidase and probed with SSL11 (top panel) and MIP8a (bottom panel).
C. FACS analysis of human neutrophils treated (+) or untreated (–) with neuraminidase and stained with SSL11-FITC or (panel D) MIP8a antibody.
E. Human neutrophils stained with 0.037 mM IgA-FITC in the presence of 0.94–30 mM (a–f) SSL11.

SSL11 binding to FcαRI is dependent on sialic acid

During studies with SSL7, which binds IgA Fc, we found that SSL11 also inhibited IgA binding to neutrophils. On further examination, SSL11 was found to bind a recombinant-soluble form of FcαRI protein (CD89) produced from Chinese Hamster Ovary (CHO) cells but not FcαRI produced from Sf9 insect cells (data not shown). This suggested that binding was dependent on complex carbohydrate (Altmann et al., 1999). To determine whether sialic acid was important, CHO cell-derived FcαRI protein was incubated with neuraminidase before SDS-PAGE and transferred to a membrane for detection of SSL11 binding (Fig. 1B). SSL11 bound strongly to untreated FcαRI, but all binding was lost in the neuraminidase-treated sample. The membrane was stripped of SSL11 and blotted with the anti-CD89 antibody MIP8a, which binds to a non-carbohydrate epitope to confirm that equivalent amounts of treated and untreated FcαRI had been loaded.

To determine whether sialic acid was also important for cell binding, freshly isolated neutrophils were treated with neuraminidase prior to staining with SSL11-FITC. SSL11 binding was completely abolished (Fig. 1C). Other non-carbohydrate epitopes remained unaffected by neuraminidase treatment. The MIP8a antibody slightly increased binding to neuraminidase-treated cells (Fig. 1D), presumably because of improved access by the antibody to the less negatively charged surface.

SSL11 binding to FcαRI blocks IgA

Because SSL11 bound to glycosylated, soluble FcαRI protein, SSL11 was tested for its ability to block IgA binding to neutrophils. Cells were incubated with IgA-FITC in the presence of increasing amounts of recombinant SSL11. Dose-dependent inhibition was observed (Fig. 1E), confirming that SSL11 binding to a glycan on FcαRI prevents access of IgA.

Sialyl Lewis X is a ligand for SSL11

Sialyl Lewis X is a tetrasaccharide involved in selectin-mediated leucocyte adhesion and extravasation (Somers et al., 2000), and is associated with several forms of cancer (Kannagi, 2004). SLex is expressed on the myeloid glycoprotein P-selectin glycoprotein ligand-1 (PSGL-1). To determine whether SSL11 bound to PSGL-1, a detergent lysate of fresh human neutrophils was incubated with streptavidin-sepharose beads coated with biotin-SSL11. Biotin-SSL7 was used as a control. Proteins precipitated by either SSL11 or SSL7 were resolved by SDS-PAGE and blotted onto a membrane to be probed with the anti-PSGL-1 monoclonal antibody (mAb) KPL1. PSGL-1 was only detected in SSL11-precipitated proteins (Fig. 2A).

Figure 2.

SSL11 binds PSGL-1 and inhibits neutrophil binding to P-selectin.
A. Detection of PSGL-1 in proteins in the supernatants or precipitates from a neutrophil lysate by treatment with underivatized beads or beads with immobilized SSL11 or SSL7.
B. Relative numbers of neutrophils adhered to a P-selectin coated glass slide after treatment with increasing concentrations of SSL11 compared with inhibition by anti-sLex antibody KM93 or the SSL11 mutant T168P.

SSL11 inhibits neutrophil attachment to P-selectin

The functional consequences of SSL11 binding to neutrophils were examined using an in vitro assay that examined neutrophil adherence to P-selectin-coated glass chambers (Fig. 2B). SSL11 added at 50 nM (∼1.1 μg ml−1) reduced neutrophil attachment by 45%, and 90% inhibition was achieved at concentrations above 75 nM. This degree of inhibition was equivalent to that achieved by 22 nM of the bivalent anti-sLex monoclonal antibody KM93.

Biosensor analysis of SSL11 binding to carbohydrates

To determine the specificity and affinity of SSL11 to carbohydrates, a biosensor assay was employed using several immobilized ligands, including rsFcαRI protein or bovine serum albumin (BSA) conjugated with a set of glycans, including sLex, sLacNac, Lex[Galβ1-4(Fuc1-3)GlcNAc] and LacNac (Galβ1-4GlcNAc) (Fig. 3).

Figure 3.

Surface plasmon resonance studies of SSL11 alleles and T168P mutant binding to sLex and FcαRI.
A. Sequence comparison of two naturally occurring alleles of SSL11, US6610 (used for crystallization) and GL10.
B. Equilibrium binding analysis comparing the affinities of US6610, GL10 and US6610-T168P binding to soluble FcαRI or sLex-BSA immobilized to the biosensor chip.
C. Top panel: A280 values corresponding to the leading, centre and trailing fractions of SSL11 protein eluted from a Superdex G75 column.
Bottom panel: Biosensor response plots of the Superdex G75 fractions (10 μM) binding to sLex-BSA. Solid lines are 1 min injections; dotted lines are 10 min injections. Preformed slower-dissociating oligomers predominate in the leading edge higher-Mr fractions 1 and 2, while the trailing edge low-Mr fractions (3, 4 and 5) depleted of preformed oligomers still show evidence of the slow-dissociating species of SSL11 that increases with surface contact time.

Two alleles of SSL11 were tested to determine whether glycan binding was conserved in naturally occurring variants (Fig. 3A). The SSL11–US6610 allele gave identical equilibrium binding curves to rsFcαRI and sLex-BSA, with calculated dissociation constants (KD) of 0.72 ± 0.05 and 0.72 ± 0.11 μM respectively (Fig. 3B). The SSL11–GL10 also produced identical equilibrium binding curves for rsFcαRI and sLex-BSA, but displayed dissociation constants of 7.3 ± 0.3 and 7.1 ± 0.5 μM that were 10-fold higher than for SSL11–US6610. Identical binding curves were observed for sLacNac-BSA. No binding was detected to either Lex-BSA or LacNac-BSA (not shown).

Interestingly, bound SSL11 exhibited two binding species with different dissociation rates: the first with a rapid dissociation, and the second with a slower rate that accumulated in proportion to the time that SSL11 contacted the sLex-BSA or sLacNac-BSA surfaces. This suggested that some SSL11 formed avidly binding oligomers once bound to the chip surface. To determine whether this was self-assembly of monomeric SSL11 and not merely preformed aggregates in the protein sample, highly pure SSL11 was prepared by high-resolution gel filtration chromatography immediately prior to biosensor analysis. A sharp symmetrical protein peak eluted from the column at the predicted retention time of a 23 kDa protein, indicating the protein was essentially monomeric (Fig. 3C, inset). Five individual fractions representing the leading, centre and trailing edge of the eluted peak were analysed for binding to LacNac-BSA and sLex-BSA with a short (1 min) and a longer (10 min) injection at a saturating concentration of approximately 10 μM (Fig. 3C). The leading edge high Mr fractions (fractions 1 and 2) contained some preformed aggregates as evidenced by the dominant slow-dissociation phase even with the 1 min binding reaction. The leading edge material also had higher total bound response (678 and 547 RU, respectively) when compared with the centre and trailing edge material in fractions 3, 4 and 5, which essentially contained monomer, without significant contamination with aggregate as evidenced by the similar plateau of total bound response (483, 469 and 458 RU). Nonetheless, the monomeric fractions 3, 4, and even the trailing fraction 5, displayed nearly identical biphasic dissociation curves, consisting of both the rapid-dissociating and a slower-dissociating component that increased proportionally with the length of time the ligand surface was exposed to soluble SSL11.

The crystal structure of SSL11

The crystal structure of SSL11–US6610 was solved by molecular replacement and refined at 1.7 Å resolution to an R factor of 19.2% (Rfree = 23.9%) with excellent geometry (Table 1). The four SSL11 molecules in the asymmetric unit are organized as two putative dimers. The final model comprises residues 6–196 for molecules A, B and D, and 5–196 for molecule C, with the exception of residues 73–78 in molecule A and residue 96 in molecule B, which lacked interpretable density; the N-terminal residues are not modelled due to lack of interpretable density.

Table 1.  Data collection and refinement statistics.
 NativesLex complex
  1. The statistics for the collection and refinement for the native SSL11-US6610 (column 1) and SSL11-US6610 in complex with sLex (column 2).

Crystal data
 Space groupP21C2
 Axial lengths (Å)54.35, 99.62, 79.3496.70, 57.80, 43.40
 Angles (°)90.0, 91.5, 90.090.0, 101.9, 90.0
 Data collection
 Resolution rangea (Å)50–1.7 (1.76–1.7)50–1.6 (1.69–1.6)
 Data collection temperature (K)113113
 Unique reflections8602628611
 Completenessa (%)96.4 (71.5)91.5 (80.1)
 Mosaicity (°)0.30.67
 Rmergea (%)2.8 (46.3)5.6 (13.4)
 I/σIa15.0 (3.0)30.4 (9.3)
Resolution range (Å)30.9–1.730.7–1.6
 Protein atoms [mean B-value (Å2)]Mols A–D: 6268 (30.6)Mol A: 1567 (23.2)
 Water molecules [mean B-value (Å2)]737 (39.1)350 (29.4)
 Ligand atoms [mean B-value (Å2)]4 phosphate 20 (40.1)1 sLex 56 (13.3)
1 citrate 13 (23.8)
3 potassium 3 (23.8)
 rms deviations from standard values
 Bond lengths (Å)0.0140.009
 Bond angles (°)1.431.21
 Ramachandran plot
 % residues in most favoured regions92.393.8

SSL11 has the two-domain fold characteristic of both Sags and SSLs (Fig. 4A) (Papageorgiou and Acharya, 2000; Arcus et al., 2002). Residues 22–89 form the N-terminal OB-fold domain, a five-stranded β-barrel comprising β-strands β1–β5. A flexible linker connects the N-terminal domain to the C-terminal β-grasp domain, residues 98–196, in which an amphipathic α-helix, α4, packs against a mixed five-stranded β-sheet with topology β7–β6–β12–β9–β10. An N-terminal helix α2, residues 8–18, sits with its C-terminus in the cleft between the two domains. SSL11 has two notable features that it shares with SSL5 and which differentiate these two proteins from the Sags. First, strands β6 and β7 are longer in the SSLs, forming a prominent β-hairpin that is important for dimerization. Second, the SSLs have a deletion of 9–10 residues relative to most Sags following strand β10; this results in a V-shaped structure that forms the binding site for sLex, described below.

Figure 4.

Crystal structure of the SSL11–sLex complex.
A. Ribbon diagram of the SSL11 US6610 monomer, with secondary structure elements labelled. The C-terminal domain is on the left, and N-terminal domain on the right. Residues 165–182, which form the sLex binding site, are in brown with sLex (yellow) in stick mode.
B. SSL11 dimer in two orthogonal views with sLex in green in space-filled mode.
C. Stereo view of sLex orientation and formation of H-bonds with protein atoms and water molecules as small red spheres.
D. Surface-rendered image of SSL11 with sLex bound in the same orientation as in C. The C1 carboxyl group of sialic acid (*) points down into a positively charged (blue) surface, and the C6 hydroxyl group of galactose (+) points into a deep hydrophilic pocket and H-bonds to E170 and Q176.

A search of all structures in the Protein Data Bank using secondary structure matching (Krissinel and Henrick, 2004) shows that the closest structural homologue of SSL11 is SSL5, with 44% sequence identity and 184 residues matching with an rms difference in Cα atomic positions of 0.91 Å. In contrast, SSL7 (34% sequence identity with SSL11–US6610) is much less similar, with an rms difference of 1.71 Å for 180 Cα positions, similar to the two streptococcal Sags, SPE-C and SPE-J.

Dimerization of SSL11

Examination of the crystal packing reveals that SSL11 forms a dimer through its C-terminal domain. In both native and sLex-complex crystal structures, pairs of SSL11 molecules pack with their β7 strands (residues 112–118) antiparallel and linked by geometrically favourable β-type hydrogen bonds, of length 2.8–2.9 Å. This extends the 5-stranded β-sheet of the two C-terminal domains into a continuous 10-stranded β-sheet encompassing both molecules (Fig. 4B). The buried surface area (295 Å2, or 2.7% of the molecular surface) is smaller than in typical dimers (Jones and Thornton, 1996), but the same association is found in both the native and sLex-bound SSL11 structures, and for SSL5 (Arcus et al., 2002), despite different crystal forms and crystal packing in each case. SSL11–US6610 is monomeric in solution below ∼9 mg ml−1 (400 μM), as determined by dynamic light scattering, but the conserved nature of this interaction suggests that it is a preferred mode of association at high concentrations and may be of functional significance at the cell surface.

Structure of the SSL11–sLex complex

The SSL11–sLex complex was formed by co-crystallization, and its structure was determined and refined at 1.6 Å resolution to values of R = 17.4% and Rfree = 20.7%. The protein structure, comprising residues 5–196, has excellent geometry (Table 1). Importantly, the electron density for the bound sLex molecule is continuous, strong and well defined; the average B-factor for the entire tetrasaccharide is 13.3 Å2 compared with 23.2 Å2 for the protein atoms.

The binding site for sLex is a V-shaped depression in the C-terminal domain formed by residues 165–182. One wall of this site is formed by β-strand 10 (residues 165–171), and the other by an irregular polypeptide structure that includes one 310-helix (residues 176–180). Seven residues hydrogen bond directly to the bound sLex (Phe-166, Thr-168, Glu-170, Gln-176, His-178, Arg-179 and Asp-182; Fig. 4C), and binding is augmented by 15 well-defined water molecules that form water bridges with the protein structure or between sLex atoms. A total of 22 hydrogen bonds are made by sLex, all with lengths 2.7–2.9 Å and excellent geometry. The sialic acid and galactose sugars form the majority of hydrogen bonds, eight with protein atoms and eight with bound water. The glycerol and carboxylate substituents of the terminal sialic acid are most extensively hydrogen bonded; the carboxyl group makes four hydrogen bonds and the glycerol moiety six, including an intramolecular interaction with the carboxyl group. A key feature of sialic acid recognition is that the carboxyl group docks at the centre of a strong patch of positive charge due to Arg-158 and Arg-179 (Fig. 4D). The carboxyl group is not directly hydrogen bonded to either arginine (binding instead to Thr168 at the centre of the patch), but similar electrostatic recognition is important in sialic acid binding to influenza virus neuraminidase (Varghese et al., 1992).

The α2,3 glycosidic bond between sialic acid and galactose appears to be a crucial determinant of binding, ensuring that both hexoses remain on the plane of the SSL11 surface. The third sugar GlcNAc and fourth sugar fucose each have just a single H-bond with protein atoms, through their C6 and C3 hydroxyl groups respectively.

A single-site mutation abolishes sLex binding

A T168P mutation was introduced into the US6610 allele. Substitution of Thr168 to proline was chosen to disrupt two essential hydrogen bonds with the sialic acid carboxyl group (from the main-chain NH and side-chain OH groups of Thr168). The T168P substitution was first modelled into the structure to confirm that it should not disrupt the polypeptide conformation, and that the (ϕ, ψ) angles of Thr168 (−117°, 133°) were appropriate for substitution of proline. The T168P mutant protein expressed well from Escherichia coli and remained soluble and stable in solution. The mutant protein was still recognized by anti-SSL11 antibodies, and circular dichroism spectra for both wild-type and mutant were identical. This confirmed there were no major changes in folding. The SSL11–T168P mutant failed to recognize any of the ligands recognized by wild-type SSL11, failed to bind to any cell types (not shown), and also failed to inhibit neutrophil attachment to P-selectin, indicating this that single mutation completely abolished carbohydrate binding (Fig. 2B).

Fine specificity of SSL11 binding by glycomics array

SSL11 was tested for binding against 285 different glycans using a microarray format provided by the Consortium for Functional Glycomics (San Diego). Glycans were attached through two different spacer groups (SP8 and SP0). SSL11 was tested at 2, 20 and 200 μg ml−1. Binding of SSL11 was observed to 22/285 glycans, and the results for the 10 strongest binding glycans at 2 μg ml−1 are presented in Table 2, with the results for these glycans at 20 and 200 μg ml−1 aligned. SSL11 also bound to five glycoproteins that were included as controls. These were glycans 1-acid glycoprotein (AGP), 2-AGP-A, 3-AGP-β1, 4-ceruloplasmine and 6-transferrrin (see Fig. S1 and Table S1). Notably, sLacNac was present in all of the strongest binding glycans and was itself represented at position 8 in Table 2. Further examination of the glycomics array results confirmed the restricted specificity of SSL11 towards sLacNac. SSL11 did not bind to very similar glycans such as sGal (Neu5Acα2-3Gal) and sialyllactose (Neu5Acα2-3Galβ1-4Glc) to any significant degree (Table 3). Moreover, SSL11 also did not bind sLacNac with a β1-3 linkage instead of a β1-4 linkage between galactose and N-acetyl glucosamine (Table 3). These data confirmed that the principal ligand for SSL11 was sLacNac, a subcomponent of sLex.

Table 2.  Strongest SSL11 binding glycans from the glycomics consortium array at three concentrations of SSL11.
Glycan nameAverage RFU (μg ml−1)
Neu5Acα2-3Galβ1-4(Fucα1-3)GlcNAcβ1-3Galβ1-4GlcNAcβ-Sp8419432 59847 672
331839 28145 413
Neu5Acα2-3Galβ1-4(Fucα1-3)GlcNAcβ1-3Galβ-Sp8264824 80346 794
Neu5Acα2-3Galβ1-4(Fucα1-3)GlcNAcβ-Sp0 (sLex)234823 52944 951
Galβ1-3-(Neu5Aa2-3Galβ1-4GlcNacβ1-6)GalNAc-T185816 13452 876
Neu5Acα2-3Galβ1-4(Fucα1-3)GlcNAcβ-Sp8181419 75848 198
Neu5Acα2-3Galβ1-4GlcNAcβ1-3Galβ1-4GlcNAcβ1-3Galβ1-4GlcNAcβ-Sp0172220 71849 273
Neu5Acα2-3Galβ1-4GlcNAcβ-Sp8 (sLacNac)153719 14249 622
Neu5Acα2-3Galβ1-4GlcNAcβ1-3Galβ1-3GlcNAcβ-Sp082910 62251 814
Neu5Acα2-3Galβ1-4GlcNAcβ1-3Galβ1-4GlcNAcβ-Sp071012 94443 191
Table 3.  Fine specificity of SSL11 binding to sialyllactosamine (sLacNac) and the most similar glycans.
Glycan name Average RFU (μg ml−1)
Neu5Acα2-3Galβ-Sp8sGal411094 119
Neu5Acα2-3Galβ1-3GlcNAcβ-Sp8 101692 286
Neu5Acα2-3Galβ1-4GlcNAcβ-Sp0sLacNac21015 75042 358
Neu5Acα2-3Galβ1-4GlcNAcβ-Sp8sLacNac153719 14249 622
Neu5Acα2-3Galβ1-4Glcβ-Sp0sLac1418455 739
Neu5Acα2-3Galβ1-4Glcβ-Sp8sLac226112 859

Rapid internalization of SSL11 into cells

Cy5-labelled SSL11 bound to human neutrophils and was rapidly internalized at 37°C (Fig. 5). Intracellular staining was evident within 5 min (the earliest time point that could be reliably imaged on the live cell imaging system) after warming the stained cells from 4°C to 37°C. At 4°C, SSL11-Cy5 remained bound to the cell surface for the duration of the incubation, indicating that the internalization was energy dependent. Clear and strong staining within the cytoplasm was evident in all cells with exclusion of SSL11-Cy5 from the nucleus (Fig. 5). Consistent with FACS binding studies, no visible staining of cells was observed with SSL11–T168P-Cy5 at 1 μM (not shown).

Figure 5.

SSL11 is rapidly internalized by neutrophils. Projection image of a Z-series stack of eight confocal images performed on human neutrophils stained with SSL11-Cy5 at 5, 15 and 25 min at 37°C in PBS containing Ca2+ and Mg2+. Panel 4 shows the contrast surface staining of neutrophils after 25 min at 4°C. No staining was visible with cells stained with SSL11–T168P-Cy5 (not shown).


SSL11 binds to 3′-sialyllactosamine-(sLacNac)-containing glycoproteins to gain entry into myeloid cells such as monocytes and granulocytes. The bound carbohydrate fits neatly into a V-shaped depression in the C-terminal β-grasp domain of SSL11, in a region which is a novel binding site among members of the wider Sag/SSL superfamily. A total of 22 hydrogen bonds secure the sLex molecule, 16 of them involving the terminal sialic acid and galactose residues. Extensive hydrogen bonding occurs with the carboxylate and glycerol substituents on sialic acid and the 6-hydroxyl on galactose, which fit into a deep, hydrophilic pocket and hydrogen bonds with Glu170 and Gln176. One striking feature is that sLex adopts an identical conformation to the sLex bound to P- and E-selectin, although the binding mechanisms between SSL11 and the selectins are quite different (Somers et al., 2000). The rms difference for all 56 sLex atoms between the SSL11 and E-selectin-bound structures is only 0.62 Å. SSL11 may have evolved to accept sLex in a restricted conformation that is crucial to immune-related recognition.

Comparison with allelic variants and other SSLs

A 10-fold difference in affinity towards both sLacNac and sLex was observed between two naturally occurring alleles of SSL11. The only differences between these alleles in or around the sLex binding site are at residues 163 and 165. These residues immediately precede the key sialic acid binding residues Phe166 and Thr168. It is not immediately obvious how these substitutions influence binding but could result from a change in solvent structure or a slight displacement of the polypeptide.

Comparison with the known structures of SSL5 and SSL7 is highly informative. Superposition of the SSL5 structure (Arcus et al., 2002) onto that of SSL11 indicates that the two are essentially identical in the crucial residues 165–182 that form the binding site. The rms difference between these regions is only 0.45 Å for all main-chain and side-chain atoms. Notably SSL5 has also been shown to bind to PSGL-1 and block P-selectin-mediated neutrophil adherence (Bestebroer et al., 2006). The crystal structure of sLex bound to SSL5 has been determined, and the glycan is bound in an almost identical position and orientation to SSL11 (Baker et al., 2007).

In contrast, superposition of the SSL7 structure (Al-Shangiti et al., 2004) on SSL11 reveals that SSL7 is quite different in this region, with an rms difference of 1.73 Å for main-chain atoms and several side-chain substitutions. No evidence for SSL7 binding to carbohydrate was obtained.

Biosensor analysis of SSL11 on sLacNac-BSA and sLex showed evidence of a fast- and a slow-dissociating species, with the latter increasing in proportion to the length of time the soluble SSL11 contacted the ligand surface. The slower-dissociating species may represent avid binding homodimers that form at high local surface concentrations. Notably, the crystal dimer of SSL11 places two sLacNac binding sites in an ideal orientation to simultaneously bind ligands extending from a planar surface. Thus, self-assembly at the cell surface might reflect a mechanism by which SSL11 increases its avidity to surface glycoproteins. Alternatively, it may represent the formation of oligomers or aggregates in solution. This is less likely, because the tighter binding species formed from monomer protein isolated from the trailing edge of a size exclusion chromatography peak. Formal proof that the dimer is functionally important must await the deletion of the seven-amino-acid region in the β7 strand that contributes to dimerization. Mutagenesis of individual residues is unlikely to have a significant effect because dimerization involves six hydrogen bonds distributed across main-chain atoms.

Cellular targets of SSL11

SSL11 affinities towards FcαRI and sLex-BSA are identical, even though FcαRI glycans terminate with sLacNac not sLex. The qualitative levels of binding to sLex and sLacNac in the glycomics array were also similar; confirming that sLacNac represents the core ligand recognized by SSL11. Sialic acid and galactose clearly contribute the majority of the interactions and most of the specificity, but the N-acetyl glucosamine residue is likely to stabilize the carbohydrate.

Many glycoproteins and glycolipids express sLacNac and would in principle be targets for SSL11; yet, with leucocytes, binding was selective for granulocytes and monocytes. Myeloid cells constitutively express sLex whereas lymphoid cells do not, so it is possible that sLacNac is similarly predominantly restricted to myeloid cells. Alternatively, there might be an additional ligand required for SSL11 binding that is only expressed on myeloid cells. There is certainly precedence for dual ligand binding for bacterial toxins. Botulinum toxin B binds to sialolactose and synaptogamin to gain entry into cells at the neuromuscular junction in order to cleave synaptobrevin to impair synaptic transport (Jin et al., 2006). Tetanus toxin has a similar mechanism (Louch et al., 2002).

Although SSL11 blocked IgA binding to cell-bound FcαRI, this is unlikely to be a defensive mechanism of S. aureus. The amount of SSL11 required to effectively block this interaction is significantly higher than the concentrations that would be achieved in situ. SSL11 binds to FcαRI glycans with micromolar affinity, which is less than the affinity of the bivalent FcαRI/IgA interaction (FcαRI2/IgA) (Herr et al., 2003). Inhibition is more likely to be a consequence of SSL11 binding to the extensive N-linked glycans present on FcαRI, which has an estimated 20 sialic acid residues (Monteiro et al., 1990). S. aureus produces SSL7, which is a far more effective inhibitor of IgA/FcαRI binding with nanomolar affinity directly to many of the same residues recognized by the receptor (Langley et al., 2005; Wines et al., 2006; Ramsland et al., 2007). Blocking neutrophil attachment to the endothelium is more likely to be important for defence by the bacteria because the P-selectin/PSGL-1 binding is a lower-affinity interaction and thus more readily inhibited by the concentration of SSL11 likely to be achieved during infection. Moreover, sLex is an invariant ligand for P-, E- and L-selectin (Somers et al., 2000) and is essential for leucocyte recruitment. Notably, SSL5 has also been shown to be a potent inhibitor of neutrophil adhesion to P-selectin (Bestebroer et al., 2006). Despite these membrane activities, we still propose that the principal role of carbohydrate binding is to gain entry into cells involved in defence against S. aureus. Thus, we believe that SSL11 targets an as-yet-undetermined intracellular function, and our current investigations are focusing on cellular changes that occur upon SSL11 internalization.

A significant proportion of SSL11 is found bound to the surface of S. aureus, and a recent paper suggests that surface-bound SSL11 could be involved in bacterial adherence. A soluble factor from Lactobacillus reuteri RC-14 significantly attenuated S. aureus adherence and growth in a rat surgical implant model (Gan et al., 2002), and this coincided with selective reduction in SSL11 expression (Laughton et al., 2006). Surface-bound SSL11 thus might act as a bacterial adhesin to sLacNac expressed on epithelial cells. Interestingly, H. pylori adheres to gastric cell via the sialic acid binding adhesin SabA, which binds the minimal epitope Neu5Acα2-3Gal (Roche et al., 2004). Moreover, SabA is solely responsible for the ability of H. pylori to bind to gangliosides (Aspholm et al., 2006).

The emerging roles of SSLs in innate immunity and their homology with bacterial Sags (Papageorgiou and Acharya, 2000; Arcus et al., 2002) indicate that the highly successful superantigen fold has been employed to generate a range of virulence proteins with different specificities and targets. It is intriguing to consider the likely effects on bacterial survival under conditions where multiple SSLs are expressed together. Further studies are likely to reveal a range of different functions for the SSL family that contribute to the virulence of an organism that continues to cause global morbidity and mortality.

Experimental procedures

Mouse anti-human-CD3, -CD10, -CD14 and -CD19 mAbs conjugated to phycoerythrin (MEM-57-PE, SN5c-PE, MEM-18-PE and LT19-PE, respectively), and mouse anti-human CD89 mAb (MIP8a), were purchased from Serotec. Mouse anti-sLex (KM93) mAb was purchased from Calbiochem. Goat anti-rabbit-HRP, anti-mouse-HRP and anti-mouse-FITC were purchased from Dako Cytomation. 3-Sialyl-N-acetyllactosamine BSA (14 atom spacer; sLacNac-BSA), LacNac-BSA, Sialyl-Lewisx BSA (sLex-BSA), and Lex-BSA were from Dextra Laboratories, UK. Neuraminidase was purchased from New England Biolabs.

Production of recombinant SSL11

The primers SSL11–N315-Fw (5′-GTAAAACGACGGCCAGGGATCCAGTACATTAGAGGTTAGATC-3′) and SSL11–N315-Rv (5′-CAGGAAACAGCTATGACGAATTCCGAATAATTTTATAAATTCACTTC-3′) amplify the coding region of the ssl11 gene for the published N315 strain (Kuroda et al., 2001). The resulting PCR product was cut with BamHI (underlined in Fw primer) and EcoRI (underlined in Rev primer) and cloned into the pET-32a-3C vector. Expression and purification of the thioredoxin–SSL11 fusion protein were as described for SSL5 (Arcus et al., 2002; Bendtsen et al., 2004; Langley et al., 2005). Recombinant SSL11 cleaved with 3C protease retains the N-terminal sequence GPGSSTLEVR, where the additional GPGS is the combined remnants of the 3C protease site and the translated BamHI restriction site.

Production of recombinant SSL11–T168P

Site-directed mutagenesis to produce the T168P mutation was performed by overlap PCR (Baker et al., 2004) using the mutant primers T168P-Fw (5′-GTGATTTTTATCCTTTTGAAT-3′) and T168P-Rv (5′-CTTTTTATTCAATTCAAAAGG-3′). Amplified product was cloned into a 3C modified version of pGEX-2T (Amersham Biosciences), and the mutation was confirmed by sequencing. Protein was expressed as a glutathione S-transferase fusion protein in E. coli DH5α, purified on glutathione sepharose. SSL11–T168P protein was released from the column by 3C-protease cleavage and purified by size exclusion chromatography. The mutant protein remained soluble and showed no changes in its circular dichroism spectrum from the wild-type SSL11, confirming normal folding.

Production, purification and biotinylation of recombinant FcαRI

A plasmid encoding the C-terminal His6-tagged ectodomains of FcαRI (Wines et al., 2001) was the template for sequential PCR reactions using the forward primer oBW146 CCGAATTCCACGATGGACCCCAAAC and the overlapping reverse primers oBW214 CTCGAAGATGTCGTTCAGACCGCCACCCGGGTGATGATGGTGATGATG and oBW218 CTTATTCGTGCCATTCGATTTTCTGAGCCTCGAAGATGTCGTTCAGAC. The C-terminus of the expressed receptor following Ile208 was GPGSSSHHHHHHPGGGLNDIFEAQKIEWHE, with linking sequences indicated by italics and the biotinylation target sequence underlined. The PCR product was cloned into pIRESneo (Clontech), transfected into CHO-K1 cells and selected as described (Wines et al., 2001).

CHO-K1 cells expressing rsFcαRI were grown over 10 days in protein-free media (JRH Ex Cell™ 325 PF CHO). Centrifuged, sterile cell supernatant was diluted 1:1 with phosphate-buffered saline (PBS) and loaded onto a 2 ml TALON™ (BD Biosciences) column. Bound protein was eluted in PBS, 100 mM imidazole and stored at 4°C. RsFcαRI was dialysed against 10 mM Tris-HCl, pH 8.0, 7.5 mM MgCl2, 5 mM NaCl and biotinylated with the addition of 5 μm biotin ligase, 5 mM freshly prepared ATP (Amersham Biosciences) and 1 mM biotin (USB Corp.) for 1 h at 37°C. Free biotin was removed by dialysis against PBS at 4°C.

Western blotting of SSL11 to rsFcαRI

RsFcαRI was neuraminidase-treated or untreated (according to manufacturers' instructions), separated on a 12.5% SDS-PAGE, then transferred to nitrocellulose membrane. The membrane was blocked with 5% milk powder in Tris-buffered saline (TBS) for 1 h at room temperature, then incubated for 1 h in 2.5% milk powder in TBS containing 1 μg ml−1 SSL11 at room temperature. The membrane was rinsed twice with TBS with 0.01% Tween 20 (TBS-T), followed by 3 × 5 min incubations in TBS-T with constant shaking. Bound SSL11 was detected using affinity-purified rabbit anti-SSL11 polyclonal antibodies and goat anti-rabbit-HRP mAb at 0.1 μg ml−1 in 2.5% milk powder in TBS. Immobilized protein–antibody complexes were visualized by chemiluminescence using ECL Western Blotting Detection Reagents (Amersham) and exposure to BioMax film (Kodak). The membrane was blocked as above, washed with H2O, stripped by incubating in 0.2 M NaOH for 5 min, then rinsed with H2O. The membrane was then blocked and reprobed as above, except that 1 μg ml−1 MIP8a was used as the primary antibody and goat anti-mouse-HRP mAb was used as the secondary antibody.

Flow cytometry analysis of SSL11 cell binding populations

Human blood (1 ml) was washed twice in 10 ml PBS with 2% v/v fetal calf serum (FCS) and centrifuged at 400 g for 5 min. Cells were resuspended in 1 ml PBS/2% FCS and 50 μl aliquots incubated with 2.5 μg SSL11-FITC, 1 μg of either CD3-PE, CD10-RPE, CD14-PE or CD19-PE, and 10 μg rabbit Ig in the dark for 15 min, after which time 100 μl of 8% formaldehyde was added at room temperature for 2 min. Red blood cells were lysed by addition of 1 ml H2O at 37°C for 2 min. After centrifuging at 400 g for 1 min, the cells were resuspended in 0.5 ml PBS with 2% v/v FCS and analysed by flow cytometry.

Neutrophils were obtained by centrifuging 5 ml fresh human blood over 2.5 ml of Histopaque 1191 layered over 2.5 ml of Histopaque 1077 (Sigma-Aldrich). Cells at the interface were diluted with PBS, washed and resuspended at 1 × 107 cells ml−1 in PBS−1% BSA. Recombinant SSL11 (30–0.94 μm) was incubated with 100 μl cell suspension in the presence of 1 μg (0.037 μm) of IgA-FITC, MIP8a-PE or KM93 for 15 min on ice. Cells stained with KM93 were washed twice with PBS−1% BSA and incubated with 1 μg goat anti-mouse-FITC antibodies for 15 min on ice. All cells were washed twice and resuspended in PBS−1% BSA before flow cytometric analysis.

For neuraminidase treatment, neutrophils were washed once with 150 mM NaCl, 5 mM CaCl2 pH 6.0 and resuspended at 1 × 107 cells ml−1, then incubated with 25 U ml−1 neuraminidase (New England Biolabs) for 1 h at 37°C in 5% CO2. rsFcαRI protein was treated with neuraminidase according to the manufacturer's instructions with supplied buffers (New England Biolabs).

Crystallization of SSL11 and its complex with sLex

The SSL11–US6610 allele was used for crystallization. Crystallization trials were performed in sitting drops (100 nl protein + 100 nl precipitant) using a Cartesian HONEYBEETM nanoliter dispensing robot (Genomic solutionsTM) and a 480-condition crystallization screen (Moreland et al., 2005). After optimization, the best crystals of native SSL11 grew at 18°C in sitting drops comprising 1 μl protein solution (9.2 mg ml−1 in 20 mM Tris-HCl pH 7.4) and 1 μl of 20% PEG 3350, 0.2 M NaH2PO4. For the SSL11–sLex complex, crystals grew at 18°C from sitting-drops prepared by mixing 0.5 μl protein solution (15.8 mg ml−1 in 20 mM Tris-HCl pH 7.4) with 0.5 μl of 10 mM sLex in water (Dextra Laboratories, UK) and 1 μl of 20% PEG 3350, 0.2 M potassium citrate.

Data collection and crystal characterization

Native SSL11 crystals were flash-frozen in 70% paratone oil, 30% mineral oil for data collection. X-ray diffraction data were collected at 100 K on beamline 9-1 at the Stanford Synchrotron Radiation Laboratory (Menlo Park, CA). A complete data set was collected to 1.7 Å resolution and processed with HKL2000 software (Collaborative Computational Project, 1994). The crystals were monoclinic, space group P21, with unit cell dimensions a = 54.35 Å, b = 99.62 Å, c = 79.34 Å, β = 91.5°, and a Matthews coefficient of 2.2 Å3 Da−1 (45% solvent) for four molecules per asymmetric unit. For the SSL11–sLex complex, crystals were flash-frozen in 70% paratone oil, 30% mineral oil as for native SSL11. X-ray diffraction data were collected with an R-Axis IV image plate detector on a Rigaku MicroMax-007HF rotating anode generator at the University of Otago, New Zealand. A full data set was collected to 1.6 Å resolution, and processed with MOSFLM and SCALA from the CCP4 program suite (Collaborative Computational Project, 1994). The crystals were monoclinic, space group C2, with unit cell dimensions a = 96.70, b = 57.80, c = 43.40, β = 101.9° and a Matthews coefficient of 2.47 Å3 Da−1 (50.3% solvent) for one SSL11–sLex complex per asymmetric unit. Data collection and processing statistics are shown in Table 1.

Structure determination and refinement

The native SSL11 structure was solved by molecular replacement with Molrep (Vagin and Teplyakov, 2000), using the structure of SSL5 (Arcus et al., 2002) as search model. The molecular replacement phases were then used with ARP/wARP (Perrakis et al., 2001) to build contiguous molecule (C; residues 7–195) and three incomplete ones (A and B missing loop regions 73–77 and 92–97, and D missing residues 66–82). Subsequent cycles of manual building in COOT (Emsley and Cowtan, 2004) and refinement in Refmac5 (Murshudov et al., 1997) using data to 1.7 Å resolution followed. NCS restraints between the four molecules were used in the initial rounds of refinement, replaced in later rounds by TLS restraints. Water molecules were included in the model if they had spherical density, height greater than 3 σ in Fo–Fc difference maps and favourable hydrogen bonding contacts. The quality of the model was checked periodically with PROCHECK (Laskowski et al., 1993).

The structure of the SSL11–sLex complex was determined by molecular replacement in PHASER (McCoy et al., 2005), using the native SSL11 structure. This was followed by cycles of manual model building in COOT (Emsley and Cowtan, 2004) and refinement in Refmac5 (Murshudov et al., 1997). Clear, well-defined density was found for the complete sLex molecule but was not modelled until the protein structure was complete. A restraint library for sLex was developed through the program sketcher in the CCP4 suite. Water molecules were added as above, and one citrate ion and three potassium ions were also included. Final refinement statistics for both structures are shown in Table 1.

Surface plasmon resonance analysis

Biotinylated rsFcαRI was captured (typically 200 RU) on a streptavidin-coated biosensor chip (GE Healthcare BIAcore, Australia). Binding to BSA–sugar conjugates (Dextra laboratories) utilized the CM5 chip surface coupled using carbodiimide as per the manufacturer's instructions. Coupling was typically 200–400 RU. Control channels consisted of coupled streptavidin, BSA or LacNac-BSA conjugate as appropriate. SSL11 was purified by gel filtration chromatography and the peak used as the analyte (30 nm to 16 μm) in HBS-EP (0.01 M HEPES pH 7.4, 0.15 M NaCl, 3 mM EDTA, 0.005% Surfactant P20) over the immobilized ligands. The binding response plateau at 30 min for n = 3 was taken as a measure of the response at equilibrium (Req). Equilibrium binding data were fitted to a simple binding site model as described previously; Req/Bmax = ([rSSL11])/(KD + [rSSL11]), where Req is the plateau binding response, KD is the equilibrium dissociation constant, Bmax is the maximal bound analyte at calculated saturation, and Req/Bmax is the fraction of immobilized ligand bound. To detect the presence of oligomers or aggregates, 0.5 ml of purified SSL11 (24 μM) was applied to a Superdex G75 column (GE Healthcare) in 20 mM HEPES, 150 mM NaCl, 3 mM ETDA pH 7.4 at 0.4 ml min−1 and the peak collected in 1 ml fractions. Fractions were immediately tested for binding.

Detection of glycan binding specificity by glycomics array

A sample of purified recombinant FITC-labelled SSL11 was submitted to the Centre for Functional Glycomics (, located at Scripps Research Institute, San Diego, California. Binding was analysed at three concentrations of 2, 20 and 200 μg ml−1 to 235 different glycans arrayed on a glass slide. Relative binding was expressed as relative fluorescence units (RFU). The data from the 20 μg ml−1 experiment are available in Supplementary material.

Detection of SSL11 binding of PSGL-1

Detection of PSGL-1 binding by SSL11 used the following protocol. SSL7 and SSL11 were individually labelled with biotin as described (Wines et al., 2006) and incubated (40 μg, 30 min, 25°C) with streptavidin-coupled sepharose beads (50 μl, Sigma-Aldrich) in PBS (pH 7.2) containing 1% BSA. Neutrophils were lysed with 0.5% Brij 96 in 10 mM Tris-HCl pH 7.4, 150 mM NaCl, 1 mM EDTA, 20 mM iodoacetamide (Sigma), and mini-complete protease inhibitors (Roche Applied Science). The lysate was clarified by centrifugation, and the equivalent of 107 lysed neutrophils were incubated (3 h, 4°C) with the streptavidin beads coupled with SSL7–biotin, SSL11–biotin or mock coupled with BSA. The beads were washed five times with lysis buffer, and the bound proteins were analysed by SDS-PAGE (8% acrylamide gel) and semidry transfer to polyvinylidene difluoride membrane. Bound PSGL-1 was detected using the mAb KPL1 (1 μg ml−1, 4 h), antimouse Ig-HRP conjugate (1/2000, 1 h: Chemicon) and Renaissance ECL reagents (PerkinElmer Life Sciences).

Neutrophil rolling assay

Capillary tubes (VD/3530-050, Camlab, UK) were prepared by cleaning in 50% nitric acid before treatment with 4% 3′-aminopropyltriethoxysilane (APES; Sigma-Aldrich) in anhydrous acetone to create a hydrophobic surface for protein binding. The tubes were coated with purified human P-selectin (Calbiochem) at 1 μg ml−1 in PBS for 1 h before blocking with 1% w/v BSA in PBS for 1 h. Human blood was collected in EDTA tubes (BD Biosciences) and neutrophils purified by Histopaque separation as described above. Cells were washed in Dulbecco's PBS without Ca2+ (Sigma-Aldrich) before suspension at 1 × 106 cells ml−1 in Dulbecco's PBS containing Ca2+ and Mg2+ (Sigma-Aldrich). SSL11, KM93 or PBS was incubated with 1 ml of neutrophils for 10 min on ice. The cell suspension was perfused through a capillary tube at a rate of 0.3 ml min−1 (0.8 dyn cm−2) at room temperature by a Harvard Apparatus syringe pump. Cell binding to the microchambers was visualized with an Axiovert S100 microscope over five separate fields after 3 min of perfusion. Digital images were captured with an Axiocam MR-3 camera and analysed with Axiovision 4.5 software.

Internalization of SSL11

Freshly isolated human neutrophils were allowed to adhere onto 3.5 cm plastic plates (BD Biosciences) in Dulbecco's PBS with Ca2+ and Mg2+ for 30 min on ice. The PBS was removed, and the cells were incubated with 1 μm (22 μg ml−1) Cy5-labelled SSL11 (SSL11-Cy5) in Dulbecco's PBS without Ca2+ and Mg2+ (Sigma) for 10 min on ice, before washing three times with ice-cold buffer. Cells were covered in ice-cold Dulbecco's PBS with Ca2+ and Mg2+ (Sigma) transferred to a Solent incubation system at 37°C, and then visualized using an Olympus FV1000 confocal laser scanning microscope. At each time point, a series of Z-plane images was collected. Corresponding transmitted light images were also collected. A Z-series projection was then constructed for each time point overlaying the transmission and fluorescence images.

Crystal structure co-ordinates

The atomic co-ordinates and structure factors for SSL11 bound by sLex (codes 2RDG and 2RDH) are available from the Protein Data Bank, Research Collaboration for Structural Bioinformatics, Rutgers University, New Brunswick, NJ (


We thank Drs C.J. Squire and T. Caradoc-Davies for help with data collection and processing; H. Holloway and J.M. Ross for help with confocal microscopy; and APAF for protein identification. We also acknowledge support from the Functional Glycomics Consortium, San Diego. We especially thank Dr C Kirton (Cambridge University) for assistance with the leucocyte rolling assays. This work is supported by The Maurice Wilkins Centre for Molecular Biodiscovery and grants from the Health Research Council of New Zealand. B.D.W. is an affiliate of the Pathology Department Melbourne University. B.D.W. thanks the NHMRC for support, and P.M. Hogarth, D. Christiansen and M. Sandrin for helpful discussions.