Molecular Identification of a Novel Fibrinogen Binding Site on the First Domain of ICAM-1 Regulating Leukocyte-Endothelium Bridging*

Binding of fibrinogen to intercellular adhesion mole- cule 1 (ICAM-1) enhances leukocyte adhesion to endothelium by acting as a bridging molecule between the two cell types. Here, a panel of four monoclonal antibodies (mAbs) to ICAM-1 was used to dissect the structure- function requirements of this recognition. All four mAbs bound to ICAM-1 transfectants and immunoprecipitated and immunoblotted ICAM-1 from detergent-solubilized JY lymphocyte extracts. Functionally, mAbs 1G12 and 2D5 inhibited binding of 125 I-fibrinogen to ICAM-1-transfectants and abrogated the enhancing effect of fi- brinogen on mononuclear cell adhesion to endothelium and transendothelial migration. In contrast, mAbs 3D6 and 6E6 did not affect

The regulated adhesion of leukocytes to endothelium followed by their extravascular emigration and tissue homing form the basis of host defense mechanisms and immune-inflammatory responses. These processes depend on a stepwise adhesion cascade coordinated by the sequential ligand recog-nition of cell surface receptors expressed on leukocytes and endothelium, including selectins, integrins, and members of the Ig superfamily (1,2). Constitutively detectable on resting endothelial cells and monocytes, and dramatically up-regulated upon cytokine stimulation (3), intercellular adhesion molecule-1 (ICAM-1, 1 CD54) plays a pivotal role in both leukocyteendothelium interaction and leukocyte transendothelial migration through its recognition of ␤ 2 integrin counterreceptors CD11a/CD18 (LFA-1) (4) and CD11b/CD18 (Mac-1) (5). ICAM-1-dependent adherence is also exploited during certain pathogen infections. The major Rhinovirus serogroup and some Coxsackie viruses use ICAM-1 as an adherence and invasion receptor on respiratory epithelium (6), whereas recognition of endothelial ICAM-1 (7) by var gene products expressed on Plasmodium falciparum-infected erythrocytes (8) contributes to the severity of cerebral malaria (9).
In addition to its interaction with cell-associated counterreceptors (2), ICAM-1 recognizes soluble ligands, including hyaluronic acid (10), and fibrinogen (11). Particularly, the functional role of fibrinogen as a novel ICAM-1 ligand has recently received considerable attention. Previous studies demonstrated that through its ICAM-1 recognition, fibrinogen enhanced the adhesion of leukocytes to endothelium (11) and supported transendothelial monocyte migration (12) by acting as a bridging molecule between the two cell types. A similar model was independently proposed for the ability of fibrinogen to increase monocytic cell attachment to mesothelioma cells in a ICAM-1-dependent pathway (13), whereas the ICAM-1-fibrinogen recognition was shown to require intact cytoskeletal organization and topographical receptor distribution on endothelial cells (14). In real-time adhesion experiments in the exposed rabbit mesentery circulation, fibrinogen-ICAM-1 bridging supported firm adhesion of monocytes to endothelium (15), thus suggesting a potential role for this pathway in leukocyte recirculation in vivo (1,2). Finally, consistent with a role of ICAM-1 in vascular cell signal transduction (16,17), binding of fibrinogen to ICAM-1 on saphenous rings modulated vascular tone in a non-nitric oxide-dependent mechanism (18). Because elevated concentrations of fibrinogen constitute a major risk factor for atherosclerosis (19), and increased deposition of fibrinogen (20) and expression of ICAM-1 (21) on atherosclerotic endothelium have been demonstrated, these findings sug-gest that ICAM-1 recognition of fibrinogen may directly contribute to the pathogenesis of vascular injury in vivo.
In order to dissect the structure-function relationship of the ICAM-1-fibrinogen interaction and to begin to elucidate its potential role in vascular cell responses we have generated a panel of monoclonal antibodies (mAbs) to ICAM-1. Using these mAbs to probe the ligand repertoire of ICAM-1, we have identified a discrete region in the first domain (22), which functions as a fibrinogen binding site and is distinct from previously recognized ICAM-1 ligand binding sites (23)(24)(25)(26).
Protein Purification and Labeling-The experimental procedures for the isolation and characterization of fibronectin-depleted (Ͻ0.1 ng/mg of protein) human fibrinogen have been reported (29). Fibrinogen was 125 I labeled by the IODO-GEN method (30) to a specific activity of 0.3 Ci/g of protein, with separation of free from protein-bound radioactivity by chromatography on a Sephadex G-25 column (Pharmacia Biotech, Inc.) preequilibrated with PBS, pH 7.4. Transferrin was purchased from Sigma and dissolved in PBS, pH 7.4.
The immune complexes were precipitated by addition of 0.2 ml of Sepharose CL4B-conjugated protein A (0.2 g/ml) (Pharmacia) for 4 h at 4°C, washed in lysis buffer, boiled for 5 min at 100°C, and electrophoresed on a 7.5% SDS-polyacrylamide gel followed by autoradiography using a Kodak X-Omat AR x-ray film and intensifying screens (DuPont de Nemours, Wilmington, DE). For immunoblotting, detergent-solubilized JY extracts were electrophoresed on a 7.5% SDSpolyacrylamide gel, transferred onto polyvinylidene difluoride membranes (Immobilon, Millipore Corp., Bedford, MA) for 2 h at 450 mAmp, blocked in 5% nonfat dry milk for 1 h at 4°C, and incubated with 25 g/ml aliquots of the various primary mAbs for 1 h at 4°C in 5% dry milk. After the washes, the transfer membranes were incubated with alkaline phosphatase-conjugated goat anti-mouse IgG (Promega, Madison, WI) for 30 min at 22°C and washed, and protein bands were visualized using nitro blue tetrazolium (Sigma) as a substrate.
Binding Studies-The experimental procedures for the binding of 125 I-fibrinogen to ICAM-1-expressing cells have been described (11). Briefly, serum-free suspensions of JY lymphocytes (1.5 ϫ 10 7 /ml) were mixed with 0.44 M 125 I-fibrinogen in the presence of 2.5 mM CaCl 2 for 20 min at 22°C. At the end of the incubation, cell surface-associated radioactivity was separated from unbound material by centrifugation of 300-l aliquots of the JY incubation reaction through a mixture of silicone oil (Dow Corning, New Bedford, MA) at 15,000 ϫ g for 5 min, and radioactivity was determined in a gamma counter. Alternatively, confluent monolayers of ICAM-1 transfectants were incubated with 0.44 M 125 I-fibrinogen for 20 min at 22°C as described above, washed three times in serum-free RPMI 1640, solubilized in 20% SDS, and counted in a gamma counter. Nonspecific binding (10 -30%) was assessed in the presence of a 50-fold molar excess of unlabeled fibrinogen added at the start of the incubation and was subtracted from the total to calculate net specific binding. In mAb inhibition experiments, JY lymphocytes or ICAM-1 transfectants were incubated with 25 g/ml control mAb 6A11 or the various anti-ICAM-1 mAbs for 45 min at 37°C, washed, and incubated with 0.44 M 125 I-fibrinogen as described above before determination of specific binding.
Fibrinogen-dependent Leukocyte-Endothelium Interaction-Serumfree suspensions of JY lymphocytes at 5 ϫ 10 6 /ml were labeled with 0.5 mCi 51 Cr (Na 2 CrO 4 ; specific activity, 487.4 mCi/mg; DuPont NEN) for 2 h at 37°C with incorporation of ϳ2-6 cpm/cell. After being washed in PBS, pH 7.4, cells were equilibrated with 0.44 M fibrinogen or control protein transferrin in the presence of 2.5 mM CaCl 2 and 100 M Dphenylalanyl-L-prolyl-L-arginine chloromethyl ketone for 20 min at 22°C before addition (1.5-3 ϫ 10 5 cells/well) to resting or TNF␣-stimulated HUVEC monolayers. After a 30-min incubation at 22°C, wells were washed and attached cells were solubilized in 20% SDS with determination of radioactivity in a beta counter. For mAb inhibition experiments, resting or TNF␣-stimulated HUVECs were incubated with 25 g/ml control mAb 6A11 or the various anti-ICAM-1 mAbs for 45 min at 37°C, washed, and mixed with 51 Cr-labeled JY lymphocytes preequilibrated with 0.44 M fibrinogen before determination of specific cell adhesion under the various conditions tested, as described above. Fibrinogen-dependent monocyte transendothelial cell migration was assessed as described (12), using HUVEC monolayers grown to confluency on gelatin-coated porous Transwell membranes (diameter 3-8 m, Costar). Cells were incubated with 25 g/ml control mAb 6A11 or the various anti-ICAM-1 mAbs for 30 min at 37°C before addition of vitamin D 3 -differentiated HL-60 cells, stimulated with 10 M formyl-methionyl-leucyl-phenylalanine (Sigma) and preequilibrated with 0.44 M fibrinogen or control protein transferrin, and 2.5 mM CaCl 2 . After a 2-h incubation at 37°C, HL-60 cells transmigrated under the various conditions tested were recovered from the bottom of the well and counted microscopically by vital staining.
Adhesion Assay-The experimental procedures for the preparation and purification of recombinant ICAM-1-Fc from COS7 cells transiently transfected with the ICAM-1 construct were reported previously (23,27). The effect of various anti-ICAM-1 mAbs on T cell adhesion to immobilized recombinant ICAM-1-Fc was investigated as described (27). Briefly, 5 ϫ 10 7 PHA/interleukin 2-activated T cells were labeled with 25 Ci of [ 3 H]thymidine overnight and washed three times in assay medium. Following the LFA-1 activation protocol in 2 mM MgCl 2 and 1 mM EGTA in 20 mM HEPES buffer containing 140 mM NaCl and 2 mg/ml D-glucose (31), T cells were briefly spun at 75 ϫ g for 1 min onto 96-well microtiter plates preparatively coated with recombinant ICAM-1-Fc at 0.24 g/well overnight at 4°C and postcoated with 2.5% bovine serum albumin for 1 h at 22°C (31). In inhibition experiments, ICAM-1-Fc-coated plates were incubated with increasing concentrations (0.5-10 g/ml) of the various anti-ICAM-1 mAbs before addition of [ 3 H]-labeled T cells (2 ϫ 10 5 /well) for a 30-min incubation at 37°C. After the wells were washed, radioactivity associated with the wells under the various conditions tested was determined in a beta counter.
In another series of experiments, ICAM-1-Fc was spotted onto plastic microtiter plates at a concentration of 50 g/ml and allowed to adsorb for 2 h at 37°C. Spots were aspirated to dryness and wells were postcoated with PBS, pH 7.4, plus 1% bovine serum albumin. Plates were incubated with 10 g/ml of control or the various anti-ICAM-1 mAbs for 30 -60 min before addition of aliquots of a suspension of trophozoite-infected erythrocytes at 8% parasitemia, 2% hematocrit, for 1 h at 37°C, resuspending the mixture every 10 min, as described previously (23). Plates were washed with binding medium containing 1% glutaraldehyde to fix cells to the plate, prior to Giemsa staining and drying. The number of parasitized red blood cells (PRBCs) adherent per square millimeter was determined under high power light microscopy.
Epitope Mapping-A panel of domain expression and homolog-scanning mutants of ICAM-1 was utilized as described (23). The panel permits the assignment of mAb epitopes to individual domains or combinations of domains and sublocalization within domain 1 of ICAM-1 (23). cDNAs encoding ICAM-1 mutants into the pCDM8 expression vector were transfected into COS7 cells using the DEAE-dextran method. Transfection was allowed to proceed for 2-4 h in the presence of chloroquine before washes and a 2-min treatment with 10% DMSO (23). After culture in fresh Dulbecco's modified Eagle's medium plus 10% newborn calf serum, 2 mM L-glutamine, and penicillin/streptomycin for 24 h at 37°C, transfected cells were detached by trypsin-EDTA treatment and replated. Forty-eight to 72 h after transfection, cells were detached with 2 mM EDTA, washed, and suspended in PBS, pH 7.4, containing 1% bovine serum albumin, 1% newborn calf serum at a concentration of 5 ϫ 10 5 /ml. Cells were added to 96-well microtiter plates (10 5 cells/well) and stained with control or with the various anti-ICAM-1 mAbs for 30 min on ice before being washed and incubated with fluorescein-conjugated polyvalent goat anti-mouse IgG (Sigma) for 30 min on ice. Cells were washed, fixed in PBS plus 1% FBS and 1% formalin, and stored at 4°C before flow cytofluorometric analysis on a Coulter EPICS cell sorter as described (23). In another series of experiments, aliquots (1 ϫ 10 7 /ml) of resting or TNF␣-stimulated HUVECs or wild-type ICAM-1 transfectants were harvested, washed once in PBS/ EDTA and twice in PBS, pH 7.4, blocked in 20% human serum for 30 min at 0°C, and incubated with 25 g/ml of the various primary mAbs in PBS, pH 7.4, plus 2% bovine serum albumin for 1 h on ice. After the washes, the cells were stained with a 1:20 dilution of fluoresceinconjugated goat F(abЈ) 2 anti-mouse IgG (Tago Inc., Burlingame, CA) for 45 min on ice, washed, and immediately analyzed on a FACScan flow cytometer (Becton Dickinson, Mountain View, CA).

Establishment of Novel
Anti-ICAM-1 mAbs-Four murine mAbs raised against ICAM-1 ϩ Daudi cells were analyzed for reactivity with ICAM-1-expressing cells by flow cytometry. The newly generated mAbs 1G12 and 2D5 bound to resting HU-VECs with a broad and heterogeneous pattern of reactivity, indistinguishable from that observed with anti-ICAM-1 mAb LB-2 (Fig. 1A). HUVEC stimulation with TNF␣ resulted in a coordinated 5-10-fold increased mAb reactivity (Fig. 1B), in agreement with previous observations (3). In parallel experiments, all four newly generated mAbs (1G12, 2D5, 3D6, and 6E6), plus control anti-ICAM-1 mAb LB-2, strongly and homogeneously bound to stable ICAM-1 transfectants (Fig. 1C). In immunoprecipitation and immunoblotting, all four new mAbs recognized a band of ϳ90 kDa, indistinguishable from that resolved by control anti-ICAM-1 mAb LB-2 ( Fig. 2 and data not shown) and consistent with the molecular size and structural organization of ICAM-1 (2). In contrast, no specific bands were immunoprecipitated or immunoblotted by control mAb OKM1 under the same experimental conditions (Fig. 2).
Differential Inhibition of ICAM-1-dependent Adherence by

FIG. 2. Characterization of anti-ICAM-1 mAbs by immunoprecipitation and immunoblotting.
A, 125 I-labeled detergent-solubilized extracts of CD11b/CD18 Ϫ JY lymphocytes were immunoprecipitated with control mAb OKM1 or anti-ICAM-1 mAb 1G12 2D5, 3D6, 6E6, or LB-2. After addition of protein A-conjugated Sepharose CL4B, immunoprecipitated proteins were washed, electrophoresed on a 7.5% SDS-polyacrylamide gel under reducing conditions, and visualized by autoradiography. B, detergent-solubilized JY lymphocyte extracts were electrophoresed on a 7.5% SDS-polyacrylamide gel under reducing conditions, transferred to Immobilon membranes, and incubated with control mAb OKM1 or anti-ICAM-1 mAb 1G12 or 2D5 for 1 h at 4°C. The membrane was washed, incubated with alkaline-phosphatase-conjugated goat anti-mouse IgG, and washed again, and protein bands were visualized by using tetrazolium salts.
The effect of the new mAb panel on ICAM-1 recognition of ␤ 2 integrin LFA-1 (32) or P. falciparum-infected erythrocytes (7) was next investigated. Consistent with the inability of mAbs 1G12 and 2D5 to diminish binding of vitamin D 3 -differentiated HL-60 cells to ICAM-1 transfectants (12), these mAbs did not significantly reduce the binding of PHA-activated human T cells to immobilized recombinant ICAM-1-Fc at any concentration tested (Fig. 5A). In contrast, increasing concentrations of mAb 3D6 or 6E6 completely blocked T cell adhesion to ICAM-1-Fc-coated plates in a dose-dependent manner (Fig. 5A). In control experiments, anti-CD11a mAb 38 and anti-ICAM-1 mAbs 15.2 and 7.5C2 also inhibited T cell attachment to immobilized ICAM-1-Fc in a dose-dependent fashion (Fig. 5A), in agreement with previous observations (23). Finally, the fibrin-  -1 (A) or P. falciparum-infected erythrocytes (B). Ninety-six-well plastic microtiter plates were coated with recombinant ICAM-1-Fc, washed, and postcoated with assay buffer containing 2.5% albumin. A, ICAM-1-Fc-coated plates (0.24 g/well) were incubated with the indicated increasing concentrations of the various anti-ICAM-1 mAbs or control anti-CD11a mAb 38 and further mixed with [ 3 H]-labeled PHA-activated human T cells (2 ϫ 10 5 /well) for 30 min at 37°C before washing and quantitation of specific cell attachment. B, the experimental conditions were essentially as in A, except that immobilized ICAM-1-Fc (50 g/ml) was incubated with the various mAbs (10 g/ml) for 30 -60 min at 22°C before addition of a suspension of trophozoite-infected erythrocytes at 8% parasitemia, 2% hematocrit for 1 h at 37°C, and quantitation of PRBC adhesion by Giemsa staining. For both panels, data are the mean Ϯ S.D. of 2 (A) or 4 (B) determinations. ogen-blocking, LFA-1-nonblocking mAbs 1G12 and 2D5 both completely inhibited adhesion of P. falciparum-infected erythrocytes to ICAM-1-Fc-coated plates (Fig. 5B), whereas the LFA-1-blocking, fibrinogen-nonblocking mAbs 3D6 and 6E6 produced only a partial and variable degree of inhibition of PRBC binding to immobilized ICAM-1-Fc under the same experimental conditions (not shown).
Epitope Mapping and Residue Assignment of Novel Anti-ICAM-1 mAbs-We initially tested the four new mAbs on domain expression constructs. It was apparent from direct mAb binding experiments to recombinant human/mouse ICAM-1 chimeras that mAbs 1G12 and 2D5 both recognized constructs containing human domain 1, as expressed in the wild-type ICAM-1 sequence or in the h123, h1234, or h1 m2345 construct (Fig. 6). In contrast, mAbs 3D6 and 6E6 both recognized ICAM-1 constructs containing human domain 2, as expressed in wild-type ICAM-1, h123, h1234, and m1 h2345 but not in h345. Thus, mAbs 1G12 and 2D5 map to domain 1 of ICAM-1, whereas mAbs 3D6 and 6E6 recognize domain 2 (Fig. 6). Within domain 1, the binding of mAbs 1G12 and 2D5 was completely disrupted by the mutation D26QPKL/KEDLS (Fig.  6), a region predicted to lie at the amino-terminal apex of ICAM-1 (23), and was reduced by 60 -70% as a result of the mutation P70DG/GTV (Fig. 6) in a region predicted to form a neighboring loop in the first ICAM-1 domain (23). Established epitope-mapped anti-ICAM-1 mAbs were also tested for their ability to affect fibrinogen-dependent adhesion of JY lymphocytes to HUVECs. In agreement with the epitope prediction studies reported above, two D26-sensitive mAbs, 7.5C2 and CBR-IC1/4, (23,28) inhibited lymphocyte-endothelium bridging mediated by fibrinogen by 97 and 43%, respectively (Table  I). Consistent with the requirement of the neighboring loop in the formation of the fibrinogen binding pocket on ICAM-1, mAb RR1/1, whose epitope(s) was disrupted by P70 mutation (23), also inhibited fibrinogen-dependent JY lymphocyte adhesion to HUVECs by 85% (Table I). In contrast, mAbs 7F7, which maps to another area of domain 1 (23), or 8.4A6, recognizing domain 2 of ICAM-1, failed to decrease fibrinogen-dependent intercel-lular adhesion under the same experimental conditions (Table I). DISCUSSION In this study, we used a novel mAb panel specific for distinct ICAM-1 epitopes to dissect the ICAM-1 recognition for fibrinogen and its relationship with the binding site(s) for LFA-1 integrin and P. falciparum-infected erythrocytes. Based on inhibition studies and epitope mapping with recombinant ICAM-1 chimeras and receptor mutants, we have identified a novel fibrinogen binding site on the first Ig-like domain of ICAM-1, completely disrupted by mutation of D26 and partially affected by mutation of P70. This region was distinct from the LFA-1 interacting site (26), whereas it more completely overlapped the ICAM-1 recognition of P. falciparum-infected erythrocytes (23). Interestingly, the same residues, D26 and P70, have been previously shown to contribute part of the binding site of the major Rhinovirus serogroup on domain 1 of ICAM-1 (25).
The initial suggestion that the association of fibrinogen with ICAM-1 involved structurally distinct requirements from LFA-1 recognition (26) came from the limited inhibition of ligand binding obtained with mAb LB-2 (11), which maps to the LFA-1 binding site at K40 and L43 on domain 1 (23). Consistent with this prediction, the novel anti-ICAM-1 mAbs 2D5 and 1G12 described here completely suppressed the recognition of fibrinogen without affecting the LFA-1 binding site, and conversely, mAbs 3D6 and 6E6 failed to reduce fibrinogen binding but completely abolished the ICAM-1-LFA-1 interaction. Previous studies demonstrated that the binding sites for PRBCs and LFA-1 are spatially distinct (23) and could be located on essentially opposite sides of domain 1 of ICAM-1 by molecular modeling (23). The fact that mAbs 1G12 and 2D5 inhibit equally well both fibrinogen and PRBC binding to ICAM-1 places the fibrinogen binding site in a structural region separate from the LFA-1 site (23,32). Three of the four described mAbs sensitive to the mutation D26QPKL/KEDS (1G12, 2D5, and 7.5C2) block fibrinogen binding, whereas the fourth mAb (CBR-IC1/4) has a partial inhibitory effect. Intriguingly, mAbs 7.5C2 and CBR-IC1/4 also block LFA-1 access to ICAM-1 but not PRBC binding (23,28). Thus, four mAbs have now been shown to recognize a very limited region of the domain but have quite distinct functional effects. The simplest explanation for this finding, which has previously been demonstrated for CD4 (33), is that the two sets of mAbs bind to a closely related region of the molecule but approach it in quite different directions. This may also mean that the direct binding site(s) for fibrinogen, PRBCs, or LFA-1 may not involve these specific residues, but an adjacent region, dependent on the precise footprint of  the mAb and the positioning of the Fc portion once it is bound.
We were unable to test this possibility using direct binding of 125 I-fibrinogen to human/murine ICAM-1 chimeras because it has been previously demonstrated that murine ICAM-1 recognizes human fibrinogen (11). The main implication of these observations is that the ICAM-1-fibrinogen pathway of intercellular bridging (11)(12)(13)15) operates structurally independently of ␤ 2 integrins LFA-1 and Mac-1, whose binding sites on ICAM-1 have been previously localized to domains 1 (see above), and 3 (34), respectively. This suggests that leukocyte-endothelium interaction mediated by fibrinogen may be also independently regulated from the stepwise adhesion cascade contributed by selectins and ␤ 2 integrins (1,2). In this context, fibrinogen supported firm adhesion of monocytic cells to rabbit mesentery endothelium in vivo, even in the absence of an initial selectin-dependent component of leukocyte tethering and rolling (15). Alternatively, the membrane-distal location of the fibrinogen-recognition site on ICAM-1, at the apex of the protein, makes it ideally positioned to act in an accessory fashion to cooperatively potentiate LFA-1-or Mac-1-dependent leukocyte adherence, bridging the distance between cells while the steric inhibition of the cellular glycocalyx is overcome. Along these lines, it is noteworthy that although the fibrinogen binding site may have an impact on the LFA-1 site, as judged by the effect of cross-blocking mAbs and P70 mutations, the ICAM-1-LFA-1 interaction, conversely, can be entirely abrogated by mAbs to domain 2, which have no effect on fibrinogen recognition, including the novel mAbs 3D6 and 6E6 and mAbs 8.4A6 and R6.5D6, characterized in previous studies (23). This opens the attractive possibility that fibrinogen occupancy, while initiating intercellular bridging per se, may shift the adhesive balance between ␤ 2 integrins and ICAM-1 by primarily cooperating with Mac-1-dependent interactions. The potential pathophysiological implications of these observations are underscored by the prominent role of fibrinogen as a major risk factor for atherosclerosis and vascular diseases, invariably characterized by increased leukocyte adhesion to endothelium and infiltration of the arterial intima (19). Indeed, elevated plasma concentrations of fibrinogen have been shown to correlate with an increased adhesion of leukocytes to endothelium in patients with advanced atherosclerosis (35).
In summary, the identification of a novel fibrinogen binding site on ICAM-1 may help in elucidating the pleiotropic contribution of this adhesive receptor to inflammation and vascular injury (36). The differential inhibitory properties of the mAb panel described here may be beneficial to selectively targeting specific aspects of ICAM-1-dependent adherence and leukocyte recruitment at inflammatory sites in vivo.