(Received for publication, March 4 1997, and in revised form, May 5, 1997)
From the Imperial Cancer Research Fund Cell Adhesion Laboratory,
Imperial Cancer Research Fund Laboratories, University of Oxford,
Institute of Molecular Medicine, John Radcliffe Hospital, Headington,
Oxford OX3 9DU, United Kingdom and the Laboratory of
Molecular Biophysics and the Oxford Centre for Molecular Sciences,
Department of Biochemistry, University of Oxford,
Oxford OX1 3QU, United Kingdom
CD31 (PECAM-1) is a member of the immunoglobulin superfamily whose extracellular domain is comprised of six immunoglobulin-like domains. It is widely expressed on endothelium, platelets, around 50% of lymphocytes, and cells of myeloid lineage. CD31 has been shown to be involved in interendothelial adhesion and leukocyte-endothelial interactions, particularly during transmigration. CD31-mediated adhesion is complex, because CD31 is capable of mediating both homophilic and multiple heterophilic adhesive interactions. Here we show that the NH2-terminal (membrane-distal) immunoglobulin domain of CD31 is necessary but not sufficient to support stable homophilic adhesion. Key residues forming the binding site within this domain have been identified by analysis of 26 single point mutations, representing the most systematic analysis of a fully homophilic interaction between immunoglobulin superfamily family members to date. This revealed five mutations that affect homophilic binding. Uniquely, the residues involved are exposed on both faces of the immunoglobulin fold, leading us to propose a novel mechanism for CD31 homophilic adhesion.
CD31 (also known as PECAM-1, platelet-endothelial cell adhesion molecule) is a heavily glycosylated transmembrane protein of approximately 130 kDa (1, 2). It is widely expressed on circulating platelets, vascular endothelium, myeloid cells, around 50% of circulating lymphocytes (primarily CD8+, CD45RA naive lymphocytes), and CD34+ hemopoietic progenitor cells in human bone marrow. Molecular cloning identified CD31 as a member of the immunoglobulin superfamily (3-5), whose extracellular domain of 574 amino acids is comprised of six immunoglobulin-like domains, each encoded by a single exon (6), a membrane-spanning hydrophobic region and a cytoplasmic tail of 118 amino acids. Unusually among adhesion molecules, the cytoplasmic tail is comprised of eight exons from which it is possible to derive a number of splice variants (6). These splice variants have been identified as being expressed in a developmentally specific manner (7) and also in regulating CD31-mediated adhesion (8).
There is now significant evidence from both in vivo and in vitro studies to show that CD31 is involved in the extravasation of monocytes and neutrophils at inflammatory sites (9-13). These studies indicate that CD31 is a potential target for therapeutic intervention in both acute and chronic inflammatory conditions.
CD31-mediated adhesion is complex, because in common with other members
of the immunoglobulin superfamily, such as neural cell adhesion
molecule (14) and L1 (15, 16), it is capable of binding both to itself
(homophilic adhesion) and to non-CD31 ligands (heterophilic adhesion).
A number of potential heterophilic ligands have been identified,
including the integrin v
3 (17, 18), a
120-kDa ligand on T-cells involved in down-regulation of T-cell
responses (19) and an as yet uncharacterized glycosaminoglycan decorated ligand on L-cells (20). In addition, it has also been shown
that CD31 is capable of up-regulating both
1- and
2-mediated adhesion following homophilic engagement
(22-24).
Given the complexity of CD31 interactions and the wide distribution of potential ligands, it has proved necessary to study CD31-mediated adhesion in the context of heterologous systems. To define the domain or domains responsible for mediating homophilic binding, we have previously used chimeric fusion proteins comprised of the NH2-terminal 1, 1-2, 1-3, 1-4, 1-5, and 1-6 Ig domains of CD31 fused to the Fc portion of human IgG1 to form a nested series of deletions (24). COS cells expressing full-length CD31 allowed to adhere to surfaces coated with this domain deletion series of CD31-Fc proteins showed that domains 5 and 6 were necessary to support homophilic adhesion but did not exclude the possibility that other more NH2-terminal domains are also required. To address the roles of domains 1 to 4 in homophilic adhesion, we assessed the ability of antibodies to block homophilic adhesion. It was found that antibodies mapping to domain 1-2 of CD31 were able to inhibit the binding of CD31(D1-D6)+COS to CD31(D1-D6)Fc. We subsequently proposed a model in which the NH2-terminal domains of CD31 on the surface of one cell bind to the membrane proximal domains of CD31 expressed on the surface of an apposing cell and vice versa, in a fully interdigitating anti-parallel mode of adhesion. This double reciprocal model of adhesion has also been proposed for homophilic interactions mediated by other members of the immunoglobulin superfamily, for example carcinoembryonic antigen (25), neural cell adhesion molecule (26), and L1 (27). In the present study we set out to test this model and to identify specific residues involved in mediating CD31 homophilic adhesion.
Unless specified otherwise, all reagents and chemicals were purchased from Sigma. Protein A-Sepharose was purchased from Pharmacia Biotech Inc., and Immulon-3 microtitre plates were from Dynatech Laboratories Inc. (Chantilly, VA). COS-1 cells were provided by the Imperial Cancer Research Fund Cell Bank (Clare Hall, UK) and grown in Dulbecco's modified Eagle's medium, supplemented with 10% fetal calf serum, 2 mM glutamine, and 100 units/ml penicillin-streptomycin.
AntibodiesThe following monoclonal antibodies (mAbs)1 were used. Anti-CD31 mAbs were: 9G11 (3); L133.1 (Becton Dickinson, Oxford, UK); hec 7.2 (Endogen, TCS Biologicals Ltd., Bucks, UK); 5.6E (Immunotech, Marseilles, France); 10B8 (R & D Systems, Oxford, UK); CLB-486 and CLB/CD31 (Monosan, Bucks, UK); VM64 (Monosan, Bucks, UK), WM59 (TCS Biologicals Ltd., Bucks, UK); IGi18 (Bender Medsystems, Vienna, Austria), P2B1 (Chemicon International Ltd., Harrow, UK). Anti-ICAM-3 mAbs were CH3.2, CAL 3.10, CAL 3.24 (28), and KS128 (30). Anti-CD14 mAb UCHM1 was obtained from the Imperial Cancer Research Fund hybridoma unit (Clare Hall, UK).
Polyclonal anti-sera specific for the cytoplasmic tail of CD31 were raised in rabbits against the peptide sequence ETVYSEVRKAVPDAVESR (Imperial Cancer Research Fund Peptide Synthesis Laboratory) coupled to keyhole limpet hemocyanin as immunogen according to the method of Hancock and Evan (29), and the IgG fraction from sera isolated by protein A-Sepharose chromatography. Immunoflourescence of COS cells expressing CD31 with either pre- or post-immune sera showed a specific anti-CD31 response following immunization. Monospecificity was demonstrated by blocking staining with the addition of a 200-fold molar excess of the immunizing peptide, but a second peptide (sequence NHAMKPINDNKE) also derived from the cytoplasmic tail of CD31 had no effect (data not shown).
Generation of Chimeras and MutantsRecombinant soluble adhesins, comprised of the extracellular domain of adhesion molecules fused to the constant (Fc) portion of human IgG1 have been described previously: CD31 (D1-D6)Fc (24); ICAM-3 (D1-D5)Fc; and MUC-18 (D1-D5)Fc (31).
Chimeras between CD31 and ICAM-3 were generated as follows. ICAM-3
(D1)/CD31 (D2-D6)TM, in which the NH2-terminal Ig domain of
human full-length CD31 is replaced by the equivalent domain from human
ICAM-3, and ICAM-3 (D1-D2)/CD31 (D3-D6)TM, in which the two
NH2-terminal Ig domains of human full-length CD31 are replaced by the equivalent domains from human ICAM-3, were generated by
a two-step recombinant PCR method (31, 32). All PCR amplifications were
performed using Pwo DNA Polymerase (Boehringer Mannheim), using CD31 in pCDM8 (3) or ICAM-3 in pCDM8 (30) as templates, and using
a maximum of 10 cycles to reduce the rate of adventitious mutation. For
ICAM-3 (D1)/CD31 (D2-D6)TM, primers used were: ICAM-3 ATG forward
amplification primer (5-ATATAAGCTTATGGTACCATCCGTGTTGTGGCCC-3
), ICAM-3
D1 reverse amplification primer with CD31 overhanging sequence (5
-CACCCTGGGACTGGGCACTCCACGCTCCGGGAGCCCGTACAC-3
), CD31 D2
forward amplification primer with ICAM-3 overhanging sequence
(5
-GTGTACGGGCTCCCGGAGCGTGGAGTGCCCAGTCCCAGGGTG-3
) and CD31 reverse
amplification primer (5
-TATCTGATGCGGCCGCCTAAGTTCCATCAAGGGAGCC-3
). For
ICAM-3 (D1-D2)/CD31 (D3-D6)TM, primers were: ICAM-3 forward amplification primer, ICAM-3 D2 reverse primer with CD31 overhanging sequence (5
-GTGGAACTTGGGTGTAGAGAAGGGCAGGACAAAGGTTCGGAG-3
), CD31 D3
forward amplification primer with ICAM-3 overhanging sequence (5
-CTCCGAACCTTTGTCCTGCCCTTCTCTACACCCAAGTTCCAC-3
) and CD31 reverse amplification primer. PCR products were digested with
HindIII and NotI and cloned into pIg vector
digested HindIII/NotI. ICAM-3 (D1-D3)/CD31
(D4-D6)TM, ICAM-3 (D1-D4)/CD31 (D5-D6)TM, and ICAM-3 (D1-D5)/CD31
(D6)TM were generated by PCR amplification of the required fragments of
CD31 and ICAM-3. Primers used were as follows: ICAM-3 forward
amplification primer, ICAM-3 D3 reverse amplification primer
(5
-ATATGAATCCAAAGACCGTCAAGTTCTCCCG-3
), ICAM-3 D4 reverse amplification primer (5
-ATATGAATTCTTTATCTTTCCATTTCAAGTGCTG-3
), ICAM-3
D5 reverse amplification primer
(5
-ATATGAATTCATTTCTATCCTGTCAAGTAAGGTG-3
), CD31 D4 forward
amplification primer (5
-ATATGAATTCGGGCTTGGAAAATAGTTCTGTTAT-3
), CD31 D5 forward amplification primer
(5
-ATATGGATTCCCCAGGATTTCTTATGATGCCCAG3
), CD31 D6 forward
amplification primer (5
-ATATGAATTCAATTCTATCCTGTCAAGTAAGGTG-3
), and
CD31 reverse amplification primer. ICAM-3 PCR products digested with
HindIII and EcoRI and CD31 PCR products digested
with EcoRI and NotI were ligated into pCDM8
digested with HindIII and NotI.
ICAM-3 (D1)/CD31 (D2-D6)Fc, ICAM-3 (D1-D2)/CD31 (D3-D6)Fc, CD31
(D1-D2)/ICAM-3 (D2-D5)Fc, and CD31 (D1-D3)/ICAM-3 (D3-D5)Fc constructs
were produced using the same two-step PCR strategy previously
described. ICAM-3 (D1)/CD31 (D2-D6)Fc and ICAM-3 (D1-D2)/CD31 (D3-D6)Fc
were made using the same primers as their membrane bound equivalents,
substituting CD31 D6 reverse amplification primer (5-GATCAGATCTACTTACCTGTTTTCTTCCATGGGGCAAGAAT-3
) for the CD31 STOP
reverse amplification primer. For CD31 (D1-D2)/ICAM-3 (D2-D5)Fc the
following primers were used: CD31 ATG forward amplification primer
(5
-TCTGAAGCTTCCTGCAGTCTTCACTCTCAGGATG-3
), CD31 D2 reverse amplification primer with ICAM-3 overhang
(5
-CGACCCACCCTCCACTACAGAGAAGGATTCCGTCAC-3
), ICAM-3 D2 forward
amplification primer with CD31 overhang
(5
-GTGACGGAATCCTTCTCTGTAGTGGAGGGTGGGTCG-3
) and ICAM-3 D5 reverse
amplification primer (5
-GATCAGATCTACTTACCTGTCGACGGGGGGGTCACGGG-3
). For CD31 (D1-D3)/ICAM-3 (D3-D5)Fc the following primers were
used: CD31 ATG forward amplification primer, CD31 D3 reverse
amplification primer with ICAM-3 overhang
(5
-GGGGGTCACGGGCAGGACCTTGGAAAATAGTTCTGT-3
), ICAM-3 D3 forward
amplification primer with CD31 overhang
(5
-ACAGAACTATTTTCCAAGGTCCTGCCCGTGACCCCC-3
), and ICAM-3 D5 reverse
primer. PCR products were digested HindIII/BglII and cloned into pIG vector digested
HindIII/BamHI. Constructs were verified by
restriction digests and sequencing.
pGPI-CD31, a construct that gives rise to a
glycosylphosphatidylinositol (GPI)-linked form of CD31 lacking the
transmembrane and cytoplasmic tail of CD31, was generated in stages. To
generate pGPI the GPI-anchor signal peptide was amplified from an LFA-3 template in pCDM8 (isolated by expression cloning, sequence identical to that reported in Ref. 35). Primers used were: LFA-3 forward amplification primer (5-ATACGGATTCAAGCAGCGGTCATTCAAGA-3
) and LFA-3
reverse primer (5
-ATCTATGCGGCCGCAAATGAGAAATCAGATGGCTT-3
). The PCR
product was digested with BamHI and NotI and
ligated into pCDNA3 to produce pGPI. The extracellular domain of
CD31 was amplified using CD31 ATG forward amplification primer and CD31
GPI D6 reverse primer (5
-TATAGATCTTTCTTCCATGGGGCAAGAAT-3
), the
PCR product digested with HindIII and BglII and
cloned into pGPI digested with HindIII and
BamHI.
Point mutations were introduced into CD31 (D1-D6)Fc chimera or
full-length CD31 using a two-step PCR strategy using common forward
amplification (5-TCTGAAGCTTCCTGCAGTCTTCACTCTCAGGATG-3
) and reverse
amplification primers
(5
-GATCAGATCTACTTACCTGTAGAGAAGGATTCCGTCACGGT-3
), encompassing the
first two NH2-terminal domains of CD31, in addition to
sequence-specific mutagenic primers (list available on request). PCR
products were digested with HindIII and BglII and
subcloned into pIG vector digested with HindIII and
BamHI. Mutants were verified by sequencing of the amplified
region, excised with HindIII and an endogenous
BamHI site at the 3
end of domain two in CD31, and
subcloned into CD31(D1-6)Fc chimera in pIG or full-length CD31 in
pCDNA3 digested with HindIII and BamHI.
Subcloning was verified by sequencing of the mutated region.
Recombinant chimeric-Fc plasmids were transiently expressed in COS cells, and secreted proteins were purified from tissue culture supernatants after 7-10 days by protein A-Sepharose chromatography as described previously (33). Eluted proteins were buffer exchanged into 20 mM Tris, pH 8.0, by centrifugal dialysis (Centricon-10 columns, Amicon, Berverly, MA) and checked by SDS-polyacrylamide gel electrophoresis.
Transfection and Cytoflourometric AnalysisIndividual constructs (10-20 µg/107 cells) were transiently expressed in COS-1 cells using DEAE-dextran as a facilitator (33). To assess the efficiency of transfection, COS cells were harvested 48 h post-transfection in PBS/2 mM EDTA, washed in PBS/0.25% bovine serum albumin at 4 °C, stained with primary mAbs (10 µg/ml), washed, stained with a 1:200 dilution of fluorescein isothiocyanate-conjugated goat anti-mouse IgG, washed, fixed in PBS/2% formaldehyde, and analyzed on a Becton Dickinson FACScan. COS cell transfectants were routinely 40-60% positive by FACScan analysis.
Adhesion AssaysImmulon-3 96-well plates were precoated
overnight at 4 °C with 1 µg/well goat-anti-human-Fc Ig in
bicarbonate buffer, pH 9.6, blocked with PBS/0.25% bovine serum
albumin (Fraction V) for 2 h at room temperature, and then coated
with Fc chimeric proteins in PBS for at least 2 h at room
temperature. At 48 h post-transfection, COS cells were harvested
in 2 mM EDTA/PBS, labeled with the fluorescent dye
2,7
-bis-(2-carboxyethyl)-5-(and-6)-carboxyfluorescein acetoxymethyl ester (Molecular Probes) for 30 min at 37 °C, and washed twice in
assay buffer (Dulbecco's modified Eagle's medium), 10 mM
HEPES buffer, 0.25% bovine serum albumin). Labeled cells were added to
plates at 5 × 104 cells/well in a volume of 50 µl
and allowed to adhere for 1 h at 37 °C. Plates were washed with
prewarmed assay buffer until the cells in the negative control wells
were sufficiently removed by visual inspection. Typically this was two
or three times. Adhesion was quantitated using a Cytofluor II
fluorescent plate reader (Millipore, Watford, UK), comparing the total
input fluorescence and the bound fluorescence, to calculate the
percentage of input cells bound. Where COS cells expressing different
chimeric constructs were used in the same assay, the percentage input
cells bound was corrected to take into account the transfection
efficiency.
Transfected COS cells were harvested with 2 mM EDTA/PBS, resuspended at 5 × 106 cells/ml in PBS/0.2% bovine serum albumin, and treated with 5 units/ml phosphatidylinositol phosholipase C (Boehringer Mannheim) for 30 min at 37 °C. Cells were washed in PBS/0.2% bovine serum albumin and stained for FACScan analysis as described above. CD31 staining was with mAb 9G11, and CD14 staining was with mAb UCHM1.
ImmunoflourescenceTwo color immunoflourescence of COS cells expressing full-length CD31 and GPI-CD31 was performed as follows. Transfected COS cells were plated at a density of 2.5 × 104 cells/well in 8-well chamber slides (Nunc and Life Technologies Inc.) and grown overnight. Cells were washed in serum-free Dulbecco's modified Eagle's medium, fixed with ice-cold methanol, and blocked with 20% goat serum in PBS. The extracellular domain of CD31 was detected using 9G11, a mouse anti-CD31 monoclonal, at 2 µg/ml and visualized with goat anti-mouse IgG tetramethylrhodamine isothiocyanate conjugated secondary antibody (Sigma). Cells were washed and blocked for a second time with 20% goat serum, and the cytoplasmic tail of CD31 was detected with the rabbit polyclonal serum described above at 20 µg/ml. Bound antibody was visualized with goat anti-rabbit IgG (whole molecule) (Fab)2 fragment fluorescein isothiocyanate-conjugated secondary antibody. The cells were observed using a Zeiss Axiovert photomicroscope with 40× objective and photographed on Kodak Ektachrome 200. A combination of automatic and manual exposure times were used to ensure equal exposure times between different CD31 constructs.
Enzyme-linked Immunosorbant Assay of Fc ProteinsImmulon-3 96-well plates were precoated with 1 µg/well goat-anti human-Fc Ig overnight at 4 °C, blocked with PBS/2% bovine serum albumin for 2 h at room temperature, and then coated with chimeric proteins (0.5 µg/well) for at least 2 h at room temperature. Primary antibody was added in saturating amounts, followed by peroxidase-conjugated goat anti-mouse (1:1000 dilution, Amersham Life Sciences, Bucks, UK). Each layer was incubated for 30 min at room temperature and followed by four washes. The assay was visualized with O-phenylenediamine dihydrochloride, and absorbance was read at 450 nm.
Studies by Sun et al. (34) have
suggested that exons 9 and 10 (the transmembrane domain and the first
15 amino acids of the cytoplasmic tail) are required for CD31-mediated
homophilic adhesion. Clearly, if this were the case, the use of soluble
chimeras comprised of the extracellular domain of CD31 fused to human
Fc would not be an appropriate approach to dissecting CD31 homophilic adhesion. To test this assertion, we constructed a GPI anchored form of
CD31 (CD31-GPI), in which the transmembrane and cytoplasmic domains of
CD31 are replaced by the 31 carboxyl-terminal amino acids of LFA-3 to
provide a GPI-anchor signal sequence (35). Transient expression of both
CD31-GPI and full-length CD31 in COS cells supported comparable binding
in a homophilic adhesion assay to CD31 (D1-D6)Fc immobilized on plastic
compared with negative control protein MUC18 (D1-D5)Fc (Fig.
1A). The expression of a GPI-anchored form of CD31, was verified by two independent methods. Firstly, dual immunoflourescence with 9G11 (a monoclonal antibody raised against the extracellular domain of CD31) and a rabbit polyclonal antibody raised against a cytoplasmic tail peptide of CD31
(Fig. 1B) demonstrated expression of CD31 extracellular domain but no detectable cytoplasmic tail. Secondly, the presence of a
GPI linkage was confirmed by treating COS transfectants with phosphatidylinositol phosholipase C and following the loss of cell
surface expression by FACScan. Full-length CD31 expression was not
significantly affected by phosphatidylinositol phosholipase C treatment
(increase of 5.5%), whereas CD31-GPI expression is reduced by 47.5%,
comparable with the 37.8% reduction observed for CD14, which is known
to be GPI-linked. These results demonstrate that the extracellular
domain of CD31 alone is sufficient to support homophilic adhesion and
that adhesion is not dependent on either transmembrane or cytoplasmic
sequences.
The NH2-terminal Domain of CD31 Is Necessary but Not Sufficient to Support Homophilic Adhesion
Previous studies have
led us to propose a model for CD31 homophilic adhesion in which the
NH2-terminal immunoglobulin domains bind to membrane
proximal domains of CD31 expressed on the surface of an apposing cell
and vice versa, in a fully interdigitating anti-parallel
mode of adhesion (24). Evidence for the role of the
NH2-terminal domains in homophilic adhesion was obtained by antibody blockade and by use of truncated soluble recombinant forms of
CD31 as inhibitors of adhesion. One limitation of this approach is the
paucity of domain-specific blocking antibodies. To directly investigate
the role of the NH2-terminal domains in homophilic
adhesion, we made a series of chimeric ICAM-3/CD31 constructs in which
immunoglobulin domains from CD31 are replaced by the equivalent domains
from ICAM-3 (CD50) as summarized in Fig.
2A. COS cells expressing each
construct were analyzed by FACScan and immunoprecipitation to show
expression of chimeras containing the appropriate domains (Table
I). Transfectants were then allowed to
adhere to plastic coated with CD31 (D1-D6)Fc (Fig. 2B).
Replacement of the first NH2-terminal domain of CD31
results in complete ablation of adhesion, indicating that domain 1 is necessary for homophilic adhesion.
|
A significant prediction of a fully interdigitating mode of adhesion for CD31 is that two binding sites exist on each molecule, and as such, removal of one site would be expected to reduce rather than ablate binding. Clearly the results obtained with the ICAM-3/CD31 chimeras binding to wild type CD31 described above are not fully consistent with this model. The identification of domains 5 and 6 as a component of the homophilic binding site was based upon the observation that in a series of nested deletion constructs, only those containing 5 or 6 domains supported homophilic adhesion. One possibility is that the identification of domains 5 and 6 as a component of the homophilic binding site arose because steric hindrance blocked the binding of cell-expressed CD31 to truncated CD31Fc constructs when in proximity to plastic. Previous experience with similar constructs for CD66 (36), ICAM-3 (37), and sialoadhesin (38) would suggest that this is not the case. However, to completely eliminate this possibility we have produced two additional constructs in which NH2-terminal domains of CD31 are fused to an irrelevant ICAM-3 stalk, in all cases preserving the overall length of the extracellular portion as six Ig type domains (Fig. 2A and Table II). In an adhesion assay where COS cells expressing full-length CD31 were allowed to adhere to each of these constructs (Fig. 2C), the three NH2-terminal domains of CD31 did not support adhesion, even though the CD31 NH2-terminal domains were presented in the context of a construct the same length as wild type CD31. The combined results from these constructs indicate that although the NH2-terminal domains are necessary for CD31 homophilic adhesion, they are not sufficient. Significant homophilic binding occurs only in the context of a full-length extracellular domain.
|
Having established that the NH2-terminal domain of CD31 plays a genetically dominant role in mediating homophilic binding, a mutagenesis screen of domain 1 was undertaken. We aligned the primary amino acid sequences of CD31 and VCAM-1, and using the co-ordinates for the crystal structure of VCAM-1 domains 1 and 2 (39), produced a homology based model of the first two domains of CD31. As targets for mutagenesis, we selected all of the predicted solvent exposed aspartate, glutamate, arginine, and lysine residues in domain 1 and a subset of charged residues in domain 2 encompassing the putative heparin binding motif LKREKN (22). For the most part, charged residues were substituted with alanine, using a two-step recombinant PCR approach to generate 26 single point mutants in a CD31(D1-D6)Fc backbone. Mutant proteins were produced in COS cells and purified from culture supernatants with protein A-Sepharose.
The structural integrity of the mutants was assessed by enzyme-linked immunosorbant assay profile (Table III) using a panel of 12 anti-CD31 monoclonal antibodies predominantly against epitopes in domains 1 and 2 (including blocking antibodies L133.1, hec 7.2, and 5.6E) but also including antibodies mapping throughout the extracellular domain. Of the mutants examined, only one (K50E) was judged to be grossly misfolded and excluded from further analysis.
|
To address the effect of mutations of CD31 on homophilic adhesion,
assays were performed to determine the adhesion of COS cells
transiently expressing full-length CD31 to plastic coated with mutant
and wild type Fc proteins at 0.5 µg/well (a concentration determined
to be saturating). When compared with wild type CD31(D1-D6)Fc, mutations D11A and K89A were found to abolish binding, whereas mutations D33A, K50A, and D51R reduced binding (Fig.
3A). To verify these findings,
mutant proteins D11A, D33A, D51R, and K89A were titrated in the range
0.1-2 µg/well. There was no increase in cell binding with increasing
concentration of protein (Fig. 3B), showing that the assay
was indeed saturating.
This was further confirmed by subcloning each mutant that exhibited a functional effect from the CD31(D1-D6)Fc background into full-length transmembrane expressed CD31. In adhesion assays with COS cells expressing mutant CD31 binding to wild type CD31(D1-D6)Fc, significant inhibition of homophilic adhesion was observed (Fig. 3C). The mutant profile revealed using Fc chimeras was fully confirmed in the context of cell expressed transmembrane bound forms.
Based on the model of CD31, examination of the relative position of the residues identified by site-directed mutagenesis shows both faces of immunoglobulin domain 1 to be involved. Residues D11 and D33 form a discrete acidic cluster on the A and B strands, respectively, located on the ABDE face. K89 is located at the top of the interstrand FG-loop, and K50 and D51 are located in the CD loop, collectively indicating an extensive contact area on the CFG face.
In the present study we have used a well established and defined system of recombinant IgG chimeric constructs and heterologous COS cell adhesion assays to dissect CD31 homophilic adhesion. The major findings may be summarized as follows. First, the extracellular domain of CD31 is capable of supporting homophilic adhesion, independent of any contribution from either the transmembrane or the cytoplasmic tail sequences. Second, CD31 domain 1 is necessary to mediate homophilic adhesion, but together with domains 2 and 3 it is not sufficient to support adhesion when expressed out of the context of the whole extracellular domain. Thus domain 1 is necessary but not sufficient for homophilic binding. Third, we have identified five residues in domain 1 that are essential to mediate homophilic adhesion.
Others have suggested that the transmembrane and cytoplasmic tail of CD31 are required to mediate homophilic adhesion (34); however, in the present study we show that the extracellular domain alone expressed on a GPI anchor is sufficient to support adhesion. Goldberger et al. (40) have identified a soluble form of CD31 present human plasma at levels of 10-25 ng/ml, resulting from alternative splicing of the transmembrane domain. This form of CD31 may therefore function as a homophilic ligand for cell expressed CD31 and serve as an endogenous regulator of leukocyte transmigration either by restricting leukocyte transmigration under normal circumstances or as a mechanism of down-regulating transmigration following inflammatory stimuli.
The experiments described above were designed to test our previous model of CD31-mediated homophilic adhesion, which envisaged that the NH2-terminal, membrane distal domains of CD31 binding to the membrane proximal domains on an apposing cell surface and vice versa. To further explore the role of the NH2-terminal domains, we made a series of chimeric ICAM-3/CD31 constructs, in which a nested series of replacements of CD31 domains by the equivalent domains from ICAM-3 were allowed to adhere to full-length CD31, demonstrating that domain 1, in addition to domains 2 and 3 previously implicated (24), is involved in homophilic adhesion (Fig. 2, B and C). An important prediction of a fully interdigitating anti-parallel model for homophilic binding is that removal of one of the two binding sites on each molecule would be expected to reduce rather than eliminate binding. In our present study we find that replacement of the NH2-terminal domain 1 alone completely ablates binding (Fig. 2, B and C). In principle this could arise because removing one binding site in a two-site interaction reduces the affinity of the interaction below the detectable threshold of the assay employed. The obvious alternative interpretation is that homophilic binding results from a direct interaction between NH2-terminal domains, excluding a role for the membrane proximal portion of the molecule. This latter interpretation is consistent with the results obtained by Sun et al. (34), who used an elegant series of loss-of-function and gain-of-function mutants to demonstrate that homophilic adhesion required the presense of domains 1 and 2 and that it was possible to confer the ability to bind human CD31 on mouse CD31 by replacing its first two domains with the equivalent domains from human CD31. To determine whether the NH2-terminal domains alone are homophilically competent in a non-CD31 backbone, we produced constructs in which the NH2-terminal domains of CD31 were expressed on a stalk consisting of ICAM-3 Ig domains. These chimeras preserved the overall length of the extracellular domain as compared with wild type CD31, thus removing any steric constraints arising from the use of truncated CD31-Fc chimeric proteins.
Importantly, these reagents revealed that although the NH2-terminal domains are necessary, they are not sufficient to support stable homophilic adhesion outside of the context of an accessory function provided by the membrane proximal CD31 domains. We propose a model in which stable homophilic contact between the NH2-terminal domains of CD31 requires that these binding domains are held in the correct spacial orientation. We propose that this function is provided by the membrane proximal domains. This would explain the positive gain-of-function results obtained by Sun et al. (34) and described above. In this construct the binding functions would be provided by the human NH2-terminal domains, whereas the murine domains perform an accessory role in maintaining the human domains in the correct orientation. Interestingly a number of studies have suggested a positive role for domain 6 in CD31-mediated interactions. For example, Fab fragments of the antibody 4G6, which recognizes the epitope CAVNEG in domain 6 of CD31 (41), has been shown to enhance CD31 homophilic adhesion (42). Similarly, in a mixed lymphocyte reaction proliferation assay, an antibody recognizing domain 6 and a peptide derived from this domain were found to exert an inhibitory effect (43).
The mechanism by which the membrane proximal domains modulate the the binding domains is currently under investigation. One attractive possibility is that these domains mediate cis-interactions between CD31 on the same cell to dimerize or cluster multiple low affinity binding domains, resulting in an increase in overall avidity, as has been shown for ICAM-1 (44) and cadherins (45, 46). In addition it has also been demonstrated that immunoglobulin domains distinct from the ligand binding domains are capable of mediating homodimerization, for example, the fourth immunoglobulin domain of the stem cell factor receptor (47).
We therefore envisage a model in which CD31 homophilic engagement
involves both cis- and trans-interactions. In our
model, low affinity trans-interactions mediated by
NH2-terminal domains are clustered to achieve higher
avidity by cis-interactions mediated by the membrane
proximal domains, forming a zipper-like interaction (Fig.
4B). This model provides
interesting regulatory possibilities. A puzzling aspect of CD31 biology
is the fact that although CD31 is widely expressed on mature,
circulating hemopoietic cells, CD31+ cells do not undergo
spontaneous CD31-mediated homotypic aggregation. This could be
explained in terms of regulation of the cis-interactions of
CD31, for example by cytoskeletal association and cell surface distribution. The recent finding that cytokine treatment of endothelial monolayers induces a redistribution of CD31 from sites of cell-cell contact to the apical surface are particularly suggestive and could
also provide a mechanism for switching between homophilic and
heterophilic interactions (48). Alternatively, conformational changes
in the membrane proximal domains may be directly translated into
conformational shifts in the binding domains and consequent changes in
affinity in a fashion analogous to integrins.
The immunoglobulin fold has been identified as mediating a variety of
interactions, for example antigen-antibody complexes (50), heterophilic
interactions with integrins such as ICAM-1,2,3/LFA-1, VCAM/VLA-4,
MadCAM/4
7, and
CD31/
v
3, (reviewed in Ref. 49) pseudo-homophilic interactions, e.g. CD2/CD48 (51), and
fully homophilic interactions such as CD31-CD31. The key feature of the
immunoglobulin fold is a core structure formed by two
-sheets packed
face to face. A comparison of structures solved at atomic resolution
shows that the regions peripheral to the core, such as the edge strands
of the
-sheets, and the loops that link the strands can have quite
different conformations. Mirroring this structural diversity, different
immunoglobulin-folds utilize different regions to interact with their
ligands.
In antigen-antibody complexes, antigen recognition and binding is
mediated by loops at the NH2-terminal end of the fold (50). A recurring theme among integrin-Ig superfamily interactions is the
utilization of acidic residues presented in the
NH2-terminal portion of the loop joining the CD -strands
on the CFG face of the Ig fold, for example, the LETS motif identified
in ICAM-3 (37) and the QIDS motif in VCAM (51). In the case of
CD2/CD48, a pseudo-homophilic interaction, mutagenesis data implicate
an adhesive surface comprised of the equivalent GFCC
C" faces of the
immunoglobulin fold in both CD2 and CD48 (52). To date, no systematic
site-directed mutagenesis of any fully homophilic Ig superfamily
interaction has been undertaken. There is some limited data on the
interaction of neural cell adhesion molecule based on the ability of a
decapeptide derived from its third immunoglobulin domain to block
homophilic adhesion (53). Likewise spontaneous mutations in L1 give
rise to neurological disease such as X-linked hydrocephalus (54);
however, many of these mutations result in frameshifts, premature
truncation, etc., and only 26 result in missense mutations. Of these,
13 are predicted to affect key structural residues or domain boundaries
and destabilize the protein, whereas others introduce cysteines and may
promote intermolecular disulfides. Hence spontaneous mutations have so
far yielded few insights into the nature of homophilic interactions.
Our present study therefore represents the most extensive directed
analysis of the structures mediating Ig superfamily homophilic
adhesion.
Significantly, site-directed mutagenesis of CD31 has identified residues on both faces of the immunoglobulin fold as being involved in the homophilic contact sites. Specifically a cluster of two acidic residues, D11 and D33, which lie on the predicted A and B strands, K89, which lies at the top of the FG loop, and two residues K50 and D51 on the CD-loop. This finding represents a novel mode of interaction between immunoglobulin folds and is consistent with the zipper model proposed above, where each CD31 molecule interacts with two others on an apposing cell surface, thus requiring two distinct binding faces.
CD31-mediated adhesion is complex both in terms of the number of potential ligands, both homophilic and heterophilic, and in terms of the domains involved in each interaction. This study has identified a role for both NH2-terminal and membrane proximal domains of CD31 in homophilic adhesion and proposes a zipper model analogous to that seen among cadherins to explain the available data. This model is consistent with mutagenesis data, which indicates that both faces of immunoglobulin domain 1 are involved in the homophilic interaction and provides a framework for biophysical analysis of homophilic adhesion.
We thank Dr. Ian Bird and Dr. Julia Spragg of the Yamanouchi Research Institute, Oxford, for gifts of antibodies and reagents, for bench space, and for invaluable advice and discussions. We are grateful to Steve Lee for assistance with Fig. 4A. We also thank past and present members of the Cell Adhesion and Molecular Hemopoiesis Laboratories for advice and support and Dr. Elaine Ferguson for critical reading of the manuscript.