Intercellular adhesion molecule-4 binds α₄β₁ and α_V-family integrins through novel integrin-binding mechanisms: Presented in part as abstracts at the 40th annual meeting of the American Society of Hematology, Miami, FL, December 1998 (Blood 1998;92:589a) and at the annual meeting of the British Blood Transfusion Society, Edinburgh, United Kingdom, September 1999 (Transfus Med 1999;9:[suppl 1]9).

Knowledge of cell-cell and cell-extracellular matrix interactions in bone marrow is essential for an understanding of the formation of blood cells and their migration into the peripheral circulation. Such interactions are also relevant to the biology of diseases like sickle cell disease and malaria where adhesive interactions involving red cells and damaged endothelium are crucial features of the pathology.1-3 These adhesive interactions frequently involve molecules belonging to the immunoglobulin superfamily (IgSF) of proteins and the family of proteins known as integrins. Integrins are heterodimers composed of 2 noncovalently associated transmembrane subunits denoted “α” and “β.” Eight different β subunits combine in a restricted manner with 18 α subunits to form some 22 different integrins. Beta subunits complex with several different α subunits, while most α subunits (with the exception of α_V, α₄, and α₆) combine with only one β subunit.4 

The LW blood group glycoprotein, or ICAM-4, is a member of the IgSF subfamily known as intercellular adhesion molecules (ICAMs). It has 2 predicted IgSF I-set domains,5-7 and within the subfamily it is most similar to ICAM-2. ICAM-4 has been found expressed only on erythroid cells and weakly in placenta.8,9 ICAM-1, -2, and -3 function in inflammation and immune responses,10 but no function for ICAM-4 has been defined.

The ICAMs are ligands for β₂ integrins, and all family members, including ICAM-4, are reported to bind α_Lβ₂.10-13 ICAM-1 and ICAM-4 are also ligands for α_Mβ₂,10,11 while the preferred ligand for ICAM-3 is α_Dβ₂.14Unlike ICAM-1, -2, -3, -5, which have the α_Lβ₂-binding motif (L/I)ET(P/S)L (Figure 1), ICAM-4 has the nonconsensus sequence LR⁵²TPL within the C strand of domain 1 and also has an adjacent potential N-glycosylation site (N⁴⁸; Figure 1). Another residue (Q) in the G strand is also important for α_Lβ₂binding in ICAM-1, -2, and -3.15-18 ICAM-4 has T⁹¹ in this equivalent position (Figure 1), as does ICAM-5. Several other IgSF molecules, including vascular cell adhesion molecule (VCAM)-1, mucosal addressin cell adhesion molecule (MAdCAM)-1, CD31, and L1 are also integrin ligands.10,19-22 VCAM-1 and MAdCAM-1 are related to the ICAM family but are bound by α₄ integrins. An aspartic acid residue in the CD loop within the motif QIDSP or GLDTS, respectively, is critical for integrin binding in these molecules.19,23These motifs are all variants of the LDV sequence, which exhibits considerable sequence variation in different ligands.24 

X-ray crystal structures of 2-domain fragments of ICAM-1, ICAM-2, VCAM-1, and MAdCAM-1 provide insights into the structural basis of IgSF-integrin interactions.25 ICAM-1 and -2 are both ligands for the I domain–containing α_Lβ₂integrin. In these ICAMs, a critical glutamic acid residue is located on the relatively flat surface of the CFG face of the molecule.26-28 In contrast, VCAM-1 and MAdCAM-1 are ligands for non–I domain α₄ integrins and have a critical aspartic acid residue on a protruding CD loop.29 30 

We have modeled ICAM-4 based on the published crystal structure of ICAM-2 to identify the positioning of the ICAM-4 nonconsensus LR⁵²TPL motif and to test whether an LD⁷³V motif located at the end of the predicted E strand is solvent-exposed. Using soluble, recombinant ICAM-4 constructs and by mutating several key residues to restore consensus ICAM-2 sequences, we have explored the integrin-binding properties of the molecule in cell-based adhesion assays. We have also targeted the LD⁷³V motif in domain 1 by site-directed mutagenesis to test for involvement in integrin binding. Our results show that ICAM-4 has an unusual integrin-binding profile in that it binds α₄β₁ and α_V-containing integrins and suggest that the reported binding of α_Lβ₂ is of lesser significance. The results also show that ICAM-4 must have novel ligand-binding motif(s).

A homology model for ICAM-4 was constructed by mutation of the ICAM-2 structure26 followed by energy minimization. Using an alignment of the ICAM-2 and ICAM-4 sequences, the coordinates of ICAM-2 were mutated to match the corresponding sequence for ICAM-4. Where insertion or deletion of residues was necessary, these were constructed using the loop-building facility in InsightII (MSI, San Diego, CA) maintaining maximal homology to ICAM-2 and reasonable peptide geometry. After soaking with an 0.8-nm (8-Å) layer of water, the model structure was energy-minimized while tethering the backbone atoms to their initial positions. The tethering force was reduced during this procedure from 4200 kJ/nm (100 kcal/Å) to a final value of 21 kJ/nm (0.5 kcal/Å). The energy-minimized structures (average derivative < 0.84 kJ/nm [0.02 kcal/Å]) were analyzed using Procheck59 and found to be of similar or better quality compared with the original crystal structure. Structural manipulations were carried out using InsightII⁹⁷and energy minimizations using the cvff forcefield implemented in Discover (version 2.97, MSI, San Diego, CA).

Cells lines were mostly from European Culture Collection, Wiltshire, United Kingdom. The β₁ integrin-negative subclone of JY (Prof M. Humphries, University of Manchester, United Kingdom), fibrosarcoma (FLYRD18 [FLY], Dr C. Porter, Hammersmith Hospital, London, United Kingdom), and endothelial (ECV304 and SK-HEP1, Dr P. Evans, Southampton General Hospital, United Kingdom) lines were gifts. Human umbilical vein endothelial cells (HUVECs) were as described.31 K562 cells transfected with α_Lβ₂ (KL/4) or α₄ (KA4C6, which expresses a constitutively active form of α₄β₁) were as described.32 All cells were maintained in Iscoves modified Eagle medium (IMEM), 10% fetal bovine serum.

Function-blocking monoclonal antibodies to integrin subunits were anti-β₁ (clone13, Becton Dickinson, Oxford, United Kingdom); anti-β₂ (MHM23, Dako, Bucks, United Kingdom; YFC118.3, Serotec, Oxford, United Kingdom; 1B4, Alexis, Bingham, United Kingdom; P4H9-A11, Chemicon International, Harrow, United Kingdom); anti-β₃ (PM6/13, Harlan Sera-Lab, Loughborough, United Kingdom; RUU-PL 7F12, Becton Dickinson); anti-β₄ (ASC-3, Chemicon); anti-α_L (38, Dr N Hogg, London, United Kingdom; MHM24, Dako); anti-α₁ (FB12, Chemicon); anti-α₂ (JA218, Prof M. Humphries, University of Manchester); anti-α₃ (C3[VLA3], Immunotech, High Wycombe, United Kingdom); anti-α₄ (HP2/1, Serotec; L25.3, Becton Dickinson; Max68P, Dr T. Shock, Celltech, Slough, United Kingdom); anti-α₅ (SNAKA55, Prof M. Humphries); anti-α₆ (NKI-GoH3, Serotec); anti-α_V(69.9.5, Immunotech; CLB-706, Chemicon); anti-α_Vβ₃ (23C6, Serotec; LM609, Harlan Sera-Lab); and anti-α_Vβ₅ (P1F6, Becton Dickinson). Activating antibodies to integrin subunits were anti-β₁ (TS2/16, American Type Culture Collection, Manassas, VA); anti-β₂ (KIM127, KIM185 as described33 34); and anti-α₄ (44H6, Serotec). Antibodies against the first domain of ICAM-4 (BS46, BS56) were from Dr H. Sonneborn, Biotest, Dreieich, Germany. Linear peptides GRGDSPK and EILDVPST were synthesized in-house.

ICAM-4–Fc fusion proteins (ICAM4Fc) used in the study comprised the 2 extracellular domains of ICAM-4 and the hinge region and Fc domains of human IgG1 as described.35 ICAM-4 complementary DNA (cDNA) encoding leader sequence and the 2 extracellular IgSF domains (ICAM-4 amino acid residues −30 to 196) was amplified by polymerase chain reaction (PCR) using sense primer (TTCCCAAGCTTTGCCATGGGGTCTCTGTTCCCT), antisense primer (ACGGATCCACTTACCTGTGGGGCTCCAAGCGAGCATCAGTGT), and full-length ICAM-4 cDNA template. This DNA was subcloned into pBluescript and used for subsequent steps. Mutant ICAM-4 cDNAs encoding consensus ICAM-2 residues at amino acid residues 50, 52, and 91, IC4S⁵⁰G (removes consensus N⁴⁸ glycosylation site), IC4R⁵²E, and IC4T⁹¹Q were made using inverse PCR.36 Sense primer 5′-ACCCCGCTGCGGCAAGGCAAGACGCTCAGA was used for mutants IC4S⁵⁰G and IC4R⁵²E. Antisense primer for IC4R⁵²E was 5′-TTCGAGGCTGGAATTCTGCGGCTGGGGACA and for IC4S⁵⁰G was 5′-GCGGAGGCCGGAATTCTGCGGCTGGGGACA. Sense primer for IC4T⁹¹Q was 5′-ACGCTGGGCCACCTCCAGGATCACCGCCTA and antisense primer was 5′-TGTTTTCCTGCGCAGGTCACGAGGCAGTGC. Native and mutant ICAM-4 cDNAs were subcloned into pIg vector as described.35 A mutant encoding IC4D⁷³R was generated by overlap extension PCR36 using native ICAM-4 cDNA in pIg as template with complementary, mutational ICAM-4 primers 5′-GCTGCTCCGCGTGAGGGCCTGG and 5′-CCAGGCCCTCACGCGGAGCAGC together with sense and antisense pIg vector primers 5′-AGAACCCACTGCTTACTGGCT and 5′-TGAGCCTGCTTCCAGCAGACA. All clones were verified by sequence analysis. The cDNA clones encoding the extracellular domains of ICAM-1, -2, and -3 and neural cell adhesion molecule (NCAM) in pIg were gifts from Dr D. Simmons, SmithKline Beecham, Harlow, United Kingdom. CAMFc proteins were expressed in COS-7 cells as described35 and extracted from culture supernatant on protein A–Sepharose. Fc fusion proteins containing the first 2 domains of VCAM-1 (VCAMFc) or MAdCAM-1 (MAdCAMFc) were as described.37 ICAM4Fc and mutant ICAM4Fc proteins migrated with M_r about 100 000 kd on sodium dodecyl sulfate–polyacrylamide gel electrophoresis under reducing conditions. IC4S⁵⁰G migrated with somewhat lowerM_r, suggesting an absence of N-glycan at residue N⁴⁸. Western blotting (nonreducing conditions) using antihuman IgG and anti–ICAM-4 antibodies BS46 and BS56 was also performed (not shown). Protein concentrations were determined by the “Nano-orange” technique (Molecular Probes, Leiden, The Netherlands).

Cells were analyzed for antigen expression as described.38 Mean fluorescence intensity was used as a measure of antibody binding.

Immulon-4 96-well plates (Dynex Technologies, Billingshurst, United Kingdom) were coated with 1 μg/well goat-antihuman–Fc (Sigma, Poole, United Kingdom) for 18 hours at 4°C, blocked with phosphate-buffered saline (pH 7.4), 0.4% bovine serum albumin (Fraction V, Sigma) for 2 hours at 22°C, and coated with chimeric proteins in phosphate-buffered saline (1 μg/well unless stated otherwise) for 2 hours at 37°C. Hemopoietic cells were washed once in assay buffer (IMEM, 2 mM EGTA, 5% human group AB serum [National Blood Service, Bristol, United Kingdom]). Nonhemopoietic cells were lifted in phosphate-buffered saline containing 2 mM ethylenediaminetetraacetic acid (EDTA), 0.1% bovine serum albumin and washed once in IMEM containing 0.1% (wt/vol) bovine serum albumin. Cells (10⁷/mL in assay buffer) were labeled with 10 μg/mL 2′,7′-bis(2-carboxyethyl)-5(6)-carboxyfluorescein acetoxymethyl ester (Sigma) for 15 minutes at 37°C and washed thrice in assay buffer. Cells were activated with 80 μM phorbol myristate acetate (PMA, Sigma) in assay buffer for 15 minutes at 37°C and washed twice in assay buffer containing cations. Cells were incubated for 15 minutes at 0°C with 10 μg/mL antibodies (KIM and anti–ICAM-4 antibodies at 25 μg/mL), 500 μM peptides, or 25 μg/mL ICAM4Fc in assay buffer containing 2 mM Mn²⁺ or 10 mM Mg²⁺. Cells were added to CAMFc-coated plates (5 × 10⁴/well in 100 μL) for 30 minutes at 37°C. Plates were read on a fluorescence microplate reader (excitation 485 nm, emission 530 nm, Bio-Tek Instruments, VT), given standardized washes in assay buffer, and read after each wash. Washing was performed by flooding the plates with assay buffer at 37°C and then vigorously decanting the buffer to waste by rapid inversion of the plate. The percentage of input cells bound was calculated. Each data point is the mean of 3 or more replicates, and assays were performed on at least 3 independent occasions.

A homology model of ICAM-4 was constructed (Figure2A) based on the crystal structure of ICAM-2,26 which, in the region modeled, has 29% amino acid sequence identity with ICAM-4. Because this tentative model is derived from sequence homology with ICAM-2, it closely follows the reported fold for ICAM-2 with domain 1 adopting an I-1 fold and domain 2 belonging to the I-2 subset.25 Numbering of residues within the ICAM-4 model follows the numbering reported for the N-terminal amino acid sequence derived from ICAM-4 isolated from red cells.6,7 The N-terminal region of ICAM-4 has an additional 15 residues not present in the ICAM-2 crystal structure, but these have not been included in the model, although it is possible that they form an additional strand adjacent to the existing A/A′ edge strand. The model of ICAM-4 therefore starts at residue 16 (VPF), which is equivalent to residue 1 (KVF) in the structure of ICAM-2.26 After energy minimization, the root mean square deviation of 185 equivalent Cα positions between the ICAM-4 model and the ICAM-2 crystal structure is 0.065 nm (0.65 Å). Like ICAM-2, ICAM-4 also has an additional disulphide bond in domain 1 between the B/C loop and the end of the F strand, a feature commonly associated with IgSF domains that act as integrin ligands.25 There are 2 regions in domain 1 of the model for ICAM-4 where the conformation differs significantly from that of ICAM-2. These are the D/E loop, which forms a prominent protrusion from the top of the molecule, and the E/F loop, which is in close proximity to domain 2, where ICAM-4 has an insertion of one additional residue. In the model, residue R52, which is equivalent to the proposed integrin-binding E37 in the LETSL motif of ICAM-2, adopts a similar location and conformation to the glutamic acid side chain in ICAM-2. Significant differences are also observed in domain 2 at the C-terminal end of the A strand and in the C′/E and F/G loops both at the top of domain 2, where contacts are made with domain 1. These latter changes suggest the contacts between the domains—and hence relative orientations of the domains—are likely to differ between ICAM-2 and ICAM-4. Our model makes no attempt to predict these movements between whole domains, which are beyond the limitations of current modeling techniques. A similar model for ICAM-4 has also recently been described.39 Although it is not possible to directly compare these models because the coordinates are not available, the energy minimization of our model appears to have generated marginally more loop movements in regions of poor sequence homology. For 122 equivalent Cα positions, the model of Hermand et al differs from the ICAM-2 crystal structure coordinates by 0.05 nm (0.85 Å), whereas the difference for our model is 0.10 nm (1.0 Å). Nonetheless, the placement of key residues is essentially the same in both model structures.

The exposure and conformation of residues forming the adhesion surface for domain 1 of ICAM-4 are largely similar to those reported for ICAM-2, although ICAM-4 has a paucity of acidic residues in the expected binding surfaces. Indeed, a distinctive feature of domain 1 in the ICAM-4 model is the concentration of basic residues in these regions, in particular on the CFG domain face. The acidic residue within the LD⁷³V motif at the end of the E strand is exposed at the domain surface in the modeled structure (Figure 2B). The location and interactions of other residues mutated in this study were also examined to ensure, as far as possible, that the substituted side chains would not be deleterious for the protein conformation. S50, R52, and T⁹¹ are all surface residues in the model; none appear to be involved in critical hydrogen or other bonds at the domain surface (Figure 2B); and, from the model, no significant rearrangements of the protein surface would be expected to accompany their mutation to the ICAM-2 concensus residues G, E, and Q, respectively.

A panel of 12 hemopoietic and 10 nonhemopoietic cell lines was tested for adhesion to soluble recombinant ICAM4Fc. To define optimal activating conditions, cells were tested in the presence of cations and after stimulation with phorbol ester. Six hemopoietic (HEL, THP1, KG1a, Raji, Jurkat, and Molt-4) and all nonhemopoietic cell lines bound to ICAM4Fc at levels that were characteristic of each cell line (Figure 3A, Tables 1 and2). Adhesion of these cells was abrogated in the presence of EDTA, suggesting that binding was integrin-mediated (Figure 3B). Binding of hemopoietic HEL cells was markedly inhibited by LDV-containing peptide, whereas binding of nonhemopoietic FLY cells was partially inhibited by LDV peptide but was abrogated in the presence of RGD-containing peptide (Figure 3B). Two anti–ICAM-4 antibodies, BS46 and BS56, did not inhibit HEL or FLY cell adhesion to ICAM-4 (data not shown). A measure of the relative avidity of adhesion to ICAM-4 was obtained by examining binding of several hemopoietic and nonhemopoietic cell lines to titrated ICAM4Fc. For most hemopoietic lines that bound to ICAM-4, a plateau of binding was reached at 10 μg/mL (Figure 3C), while the plateau of binding for the nonhemopoietic cell lines ranged between 2.5 μg/mL (DX3) and 15 μg/mL (HT29) (Figure 3D).

We tested whether the LDV-inhibitable hemopoietic integrin was α_Lβ₂ and whether adhesion to any of the mutant ICAM4Fc fusion proteins containing consensus ICAM-2 residues (IC4S⁵⁰G, IC4R⁵²E, and IC4T⁹¹Q) was different compared with native ICAM4Fc. Each mutant was titrated and tested, in comparison with native ICAM4Fc, for adhesion of HEL cells. Similar results were obtained with wild-type and all mutant ICAM4Fc proteins, demonstrating that each mutant was functionally active (Figure 4A). Hemopoietic cell lines were tested for adhesion to ICAM4Fc or mutant ICAM-4–Fc proteins in comparison with ICAM-1–Fc, -2–Fc, and -3–Fc or NCAMFc under optimal activating conditions for each cell line (Table 1). Every cell line that adhered to ICAM4Fc also bound to all mutant ICAM4Fc proteins, and no cell line that failed to adhere to native ICAM4Fc bound mutant ICA4Fc proteins. The level of α_Lβ₂ integrin expression did not correlate with adhesion to ICAM4Fc. HEL and Molt-4 cells expressed low levels of α_Lβ₂; adhered poorly to ICAM-1–Fc, -2–Fc, or -3–Fc; but adhered well to ICAM4Fc. Conversely, IM9 and EBV-LCL expressed high levels of α_Lβ₂; adhered to ICAM-1–Fc, -2–Fc, or -3–Fc; but did not adhere to ICAM4Fc or to mutant ICAM4Fc proteins (Table 1).

Under conditions where adhesion to ICAM-1–Fc, -2–Fc, or -3–Fc was markedly inhibited, function-blocking antibodies to integrin subunits α_L and β₂ did not abrogate binding of hemopoietic cell lines to ICAM4Fc or to mutant ICAM4Fc proteins (Figure4B; several α_L and β₂ antibodies were tested and representative results are shown). A small reduction of THP1 cell adhesion to ICAM4Fc in the presence of α_L or β₂ antibodies was observed, suggesting there may be some α_Lβ₂-mediated adhesion of THP1 cells. An alternative interpretation is that this apparent blocking was nonspecific because a similar reduction in background THP1 cell adhesion to NCAMFc control protein was also observed (Figure 4B). Treatment with β₂-activating antibodies (KIM127 and KIM185) increased the percentage of IM9 and EBV-LCL cells adhering to ICAM-1–Fc, -2–Fc, or -3–Fc proteins but not to ICAM4Fc or to mutant ICAM4Fc proteins (Figure 4C and not shown). An α_Lβ₂-transfected cell line (KL/4) also adhered to ICAM-1–Fc, -2–Fc, or -3–Fc but failed to adhere to ICAM4Fc or mutant ICAM4Fc proteins under maximally activating conditions (not shown).

All hemopoietic cell lines that adhered to ICAM4Fc were found to express integrins of the β₁ family (Table 1). In adhesion assays, a β₁ blocking antibody (clone 13) abrogated binding while an activating antibody (TS2/16) stimulated binding (β₁^b, β₁^a, Figure5A). Other β subunit antibodies had no effect. Blocking α₄ antibodies (HP2/1, Max68P, and L25.3) totally inhibited binding while an activating antibody (44H6) partially inhibited binding (α₄^a, α₄^b, Figure 5A). Other antibodies against α subunits of β₁-family integrins had no effect.

The integrin α₄β₁ has 2 other well-characterized IgSF ligands, VCAM-1 and MAdCAM-1, both of which are related to the ICAM family. A measure of the relative avidity of α₄β₁-expressing cells for the 3 ligands ICAM-4, VCAM-1, and MAdCAM-1 was obtained by defining the minimum concentration of ligand that promoted adhesion of HEL cells in titrations. Values of 10 μg/mL for ICAM4Fc, 0.05 μg/mL for VCAMFc, and 15 μg/mL for MAdCAMFc were obtained (Figure 5B). These findings are consistent with those of Newham et al,40 who also reported that the relative avidity of α₄β₁-mediated adhesion to VCAM-1 was more than 10 times the avidity for MAdCAM-1. The avidity of α₄β₁/ICAM-4–mediated adhesion was therefore either greater or of a similar order to α₄β₁/MAdCAM-1–mediated adhesion. A β₁-negative, β₇-positive subclone of the JY cell line that expresses α₄β₇ adhered to VCAMFc or MAdCAMFc but did not adhere to ICAM4Fc (Figure 5C). The data suggest that ICAM-4 may not be a ligand for α₄β₇ in this B-cell line.

Every nonhemopoietic cell line tested, whether of human or animal origin, bound to ICAM4Fc (Figure 3A,D; Table 2). Several cell lines did not express the α₄ integrin subunit (FLY, ECV304, HUVEC, and HT29), whereas the integrins α_Vβ₁, α₂β₁, α₃β₁, and α_Vβ₅were consistently expressed (Table 2). Function-blocking integrin antibodies were tested for inhibition of adhesion of these α₄-negative cell lines to ICAM4Fc (Figure 5D and not shown). Of the β subunit antibodies tested, only β₁antibody had an effect, causing partial inhibition of binding. Antibodies to α subunits of the β₁ family were tested, and only α_V antibodies had an effect, causing partial inhibition of binding (irrespective of the antibody or concentration used). The α_V complex antibodies were tested, and an α_Vβ₅ antibody partially inhibited adhesion. Complete inhibition of binding was not observed with any combination of inhibiting antibodies tested. The integrin profiles of the α₄-negative cell lines, together with antibody inhibition data, suggest that ICAM-4 is a ligand for both α_Vβ₁ and α_Vβ₅integrins (the profiles of FLY, HT29, and 293 cells are informative; Table 2). Expression levels of α_V integrins did not correlate with percentage of cells adhering or with the ICAM-4 coating concentration at which this was maximal (Figure 3D, Table 2). The restriction of the α₄β₁/ICAM-4 interaction to a subset of hemopoietic cell lines contrasts with the observation that all nonhemopoietic lines expressing α₄β₁ adhered to ICAM4Fc (DX3, HFFF, SK-HEP-1, 293; Table 2). This adhesion was partially inhibited by β₁, α_V, α_Vβ₅, and α₄ antibodies, but no combination of antibodies completely inhibited binding (data not shown). Expression levels of α₄ and α_V integrins did not correlate with percentage of cells adhering or the ICAM-4 coating concentration at which this was maximal (Figure 3D, Table 2). Untransfected monkey (COS-7) or hamster (CHO) cells also showed integrin-mediated adhesion to ICAM-4 (Figure 3A, Table 2). Titration studies with 3 α₄β₁-negative lines (FLY, HT29, ECV304; Figure 3D) indicated that the avidity of ICAM-4/α_Vintegrin–mediated adhesion was of a similar order to ICAM-4/α₄β₁–mediated adhesion and was equal to or greater than MAdCAM-1/α₄β₁–mediated adhesion, which suggests biological relevance.

Mutation of residues on the CFG face of ICAM-4 (IC4R⁵²E, IC4T⁹¹Q, and IC4S⁵⁰G), homologous to those found to be important for α_Lβ₂ binding in ICAM-1, -2, and -3, did not reduce or increase α₄β₁-mediated adhesion of hemopoietic cell lines (Figure 4A, Table 1) or α_V-mediated binding of nonhemopoietic, FLY cells (not shown). The ICAM-4 model predicts a solvent-accessible, consensus LDV site on domain 1, located on the E strand, toward the bottom of the domain (Figure 2B). α₄β₁-mediated adhesion of HEL cells and α_V integrin–mediated adhesion of FLY cells were unaffected by mutation of this motif (IC4D⁷³R) (Figure 6).

Our results indicate that ICAM-4 is unique in the ICAM family because it binds through novel motifs to α₄β₁ and α_V integrins. We have demonstrated adhesion of hemopoietic (LDV peptide–inhibitable) and nonhemopoietic (RGD peptide–inhibitable) cells to soluble, recombinant ICAM-4. LDV- and RGD-containing peptides have been used extensively as inhibitors of integrin-mediated interactions because these motifs, or their variants, are found in many integrin ligands.24 41 Given the short incubation times in our experiments and our observation that EDTA also inhibited adhesion, we concluded that adhesion was integrin-mediated and hypothesized that there are 2 different integrin ligands for ICAM-4: an LDV-inhibitable hemopoietic integrin and a nonhemopoietic cell integrin that has a preference for RGD- over LDV-containing sequences.

Adhesion of hemopoietic cells to ICAM4Fc was markedly affected in the presence of activating or inhibiting antibodies against α₄ or β₁, which strongly suggests that ICAM-4 is bound by the hemopoietic integrin α₄β₁. The observation that an activating anti-α₄ caused partial inhibition of HEL cell adhesion might indicate that the binding site for ICAM-4 on α₄β₁ is qualitatively different from that occupied by VCAM-1 or MAdCAM-1. There is evidence for overlapping but distinct binding sites for ICAM-1 and ICAM-3 on α_Lβ₂, which might represent an analogous situation.42 The ICAM-4/α₄β₁interaction was seen only with certain α₄β₁-expressing hemopoietic lines (but was seen with all α₄β₁-positive nonhemopoietic lines; see above) and did not correlate with α₄β₁ expression levels. An activating β₁ antibody did not induce ICAM-4 binding in the nonbinding cell lines, suggesting that in hemopoietic cells interaction with ICAM-4 is dependent on additional factors. Masumoto and Hemler43 suggest that the activation state of α₄β₁ is regulated in each cell type by unknown cell-specific constraints and that not all cell types can achieve the highest states of activation and bind the full repertoire of ligands. Our data are consistent with the notion that not all α₄β₁-positive hemopoietic cell lines can achieve the activation state required for binding to ICAM-4. The restriction of the α₄β₁/ICAM-4 interaction to a subset of hemopoietic cell lines contrasts with the observation that all nonhemopoietic lines expressing α₄β₁ adhered to ICAM4Fc. An α₄β₁-expressing, K562, α₄-transfectant (KA4C6, PMA-activated–positive cations) adhered to VCAMFc but not to MAdCAMFc or ICAM4Fc (not shown). This result might reflect the cell-specific activation state of α₄β₁ in this transfectant. We were unable to test transfectants of COS or CHO origin because untransfected cells adhered to ICAM4Fc (Figure 3A, Table 2).

Notably, HEL cells that bound with the highest avidity to ICAM-4 under minimal activating conditions are of erythroid origin. Given that ICAM-4 expression appears to be restricted to erythroid cells and placenta,8 it is tempting to speculate that the molecule has a particular function in erythropoiesis. ICAM-4 is first expressed early in erythropoiesis, at the colony-forming unit–erythroid stage.9,44 In mammalian bone marrow, erythropoiesis occurs in discrete anatomic units known as erythroblastic islands.45,46 The integrity of these structures is maintained at least in part by the interaction of α₄β₁ expressed on erythroblasts, with VCAM-1 expressed on a central macrophage.47 In vivo treatment of mice with VLA-4 antibodies specifically induces anemia.48 Our results show that the apparent avidity of α₄β₁-mediated adhesion of an erythroblast cell line to ICAM-4 is 200-fold lower than that estimated for α₄β₁-mediated adhesion to VCAM-1. We have postulated elsewhere that the interaction of ICAM-4 with its ligand α₄β₁ on adjacent erythroblasts may help to stabilize erythroblastic islands.49 It is likely that there are multiple interactions between adjacent cells in these structures, and further studies will be necessary to assess the physiologic relevance of those involving ICAM-4.

Our finding that the adhesion of various cell lines to ICAM4Fc does not correlate with the level of α_Lβ₂expression; that adhesion is not abrogated by function-blocking, α_Lβ₂ antibodies; that adhesion is not enhanced by activating α_Lβ₂ antibodies; and that an α_Lβ₂ transfectant does not adhere all suggests that adhesion to ICAM-4 or ICAM-4 mutants is not mediated by α_Lβ₂ in the hemopoietic cell lines examined. In contrast to these findings, others have reported that ICAM-4 is a ligand for both α_Lβ₂ and α_Mβ₂ on peripheral blood lymphocytes and transfected cell lines.11,39 In these other studies, α_Lβ₂-mediated adhesion was observed only in the presence of 2 mM CaCl₂, and the maximum adhesion observed was only about 12% bound cells. In adhesion assays using ICAM4Fc and α₄β₁-expressing hemopoietic cell lines, we found that adhesion was inhibited in the presence of 2 mM CaCl₂ and was maximal in the presence of Mg²⁺ or Mn²⁺ ions and EGTA (see “Materials and methods”). It is also reported that Ca²⁺ ions are negative regulators of α_Lβ₂-mediated adhesion to ICAM-1.50 We suggest that the adhesive interactions of ICAM-4 with β₂ integrins are of lower avidity than those between ICAM-4 and α₄β₁- or α_V-family integrins described here and reflect the rather promiscuous nature of ICAM-4–ligand interactions. Clearly, further studies are needed to establish whether any of these interactions observed in vitro are physiologically significant.

Antibody inhibition studies suggested that the adhesion of nonhemopoietic cells is mediated by α_V integrin(s) (probably α_Vβ₁ and α_Vβ₅), though complete inhibition of binding was not observed with any combination of inhibiting antibodies tested. A possible explanation for this observation is that the ICAM-4 binding site on α_V integrins may be qualitatively different from that occupied by its other ligands, as suggested for α₄β₁, above. It is also possible that ICAM-4 is bound by other α_V integrins, α_Vβ₆ and α_Vβ₈, that might be expressed by these cell lines but may not be inhibited by these α_V antibodies. We did not find α_Vβ₃-mediated adhesion of nonhemopoietic cells to ICAM-4. Adhesion to ICAM-4 of the α_V-positive, hemopoietic cell line HEL (which expresses only α_Vβ₃, data not shown) was not blocked by inhibiting α_V antibodies. These findings were not unexpected because the assays were performed in the presence of EGTA, which is reported to inhibit α_Vβ₃-mediated binding to CD31.21 Monkey and hamster cells also showed integrin-mediated adhesion to ICAM-4. CHO cells express α_V integrins, and CHO transfectants expressing hamster α_V/human β₅ hybrid integrins are active in functional assays with human-derived ligands.51 52 It is therefore likely that adhesion of both COS-7 and CHO cells is also α_V integrin–mediated. Indeed, our data are consistent with the hypothesis that ICAM-4 is a truly promiscuous α_Vintegrin ligand and is bound by all α_V-family integrins.

These results raise the possibility that interactions between ICAM-4 on erythroblasts and α_V integrins on the central macrophage of erythroblastic island structures in bone marrow might also contribute to the integrity of the island structures. It is also possible that ICAM-4 has a wider role in erythrocyte recognition by α_V integrin-expressing cells of the reticuloendothelial system and may thereby serve as an “erythrocyte addressin.” In sickle cell disease, episodes of acute pain are linked to vascular occlusion caused by the abnormal adhesion of sickle red cells to vascular endothelial cells.1,2,53,54 There are some data suggesting that ICAM-4 expression is abnormally elevated on the erythrocytes of patients with sickle cell disease and that antibodies to ICAM-4 partially inhibit adhesion of sickle red blood cells to tumor necrosis factor-α–activated endothelial cells.49 These observations raise the possibility that ICAM-4/α_Vintegrin interactions may contribute to the abnormal adhesion of sickle red cells to activated endothelium during sickle cell disease crisis.

Our results strongly suggest that ICAM-4 is a ligand for α₄β₁ and α_V integrins, although we cannot rule out the possibility of crosstalk between these integrins and other integrins or unidentified molecules not targeted by antibody inhibition studies and that these molecules, rather than α₄β₁ or α_V integrins, are ligands for ICAM-4. In hemopoietic cell lines, α₄β₁ antibodies, an LDV peptide, or EDTA completely blocked adhesion, strongly suggesting that ICAM-4 was bound by α₄β₁ and the absence of crosstalk. In nonhemopoietic cell lines, both the RGD peptide and EDTA completely blocked adhesion, but no combination of inhibiting antibodies completely blocked binding. In these cells it is possible that crosstalk between α_V integrins and other molecules underlies this observation. In this context α_Vβ₃ is reported to regulate both α₄β₁-mediated lymphocyte migration and α₅β₁-mediated adhesion and endocytosis of phagocytic cells.55 56 

ICAM-4 is clearly an unusual molecule because it can be a ligand for α₄β₁, which has a preference for LDV-based sequences, and also for α_V integrins that usually bind ligands containing the RGD sequence.24,41 The α_V integrins generally bind extracellular matrix molecules, and α_Vβ₃ has the largest repertoire of ligands.24 α_Vβ₃is the only α_V integrin previously shown to have IgSF ligands and binds CD31 and L1.21,22 The integrin-binding site for L1 has been identified as an RGD sequence in domain 6,22 while that of CD31 has been shown to reside largely on domain 1.21 This domain has no RGD motif, suggesting that α_Vβ₃ can use other sequences.57 This is clearly also the case for the α_V integrin–binding site of ICAM-4 which, like CD31, has no RGD sequences. Most integrin-binding sites on IgSF molecules contain an acidic residue, and they are commonly located on the N-terminal domain.21,25,58 Unlike other ICAMs, ICAM-4 domain 1 does not have the consensus α_Lβ₂-binding motif in the C strand. Unlike VCAM-1 and MAdCAM-1, ICAM-4 also lacks a consensus α₄β₁-binding motif in the C-D loop of this domain. IgSF molecules are highly versatile and can use almost any surface for ligand recognition.25 It is therefore possible that integrin-binding motif(s) may be located in domain 2 of ICAM-4, where there are 7 acidic residues. With the exception of D¹⁷³ the model shows that all of these residues are exposed at the surface and are potentially available for interaction. Further mutational analysis of domain 2 is clearly necessary to define the integrin-binding site of this novel ICAM.

We thank Dr D. Simmons and Dr M. K. Robinson for helpful discussions, P. Martin for DNA sequence analysis, and Dr W. Mawby for peptide synthesis.