HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

J. Biol. Chem., Vol. 275, Issue 28, 21514-21524, July 14, 2000

This Article Abstract Full Text (PDF) All Versions of this Article: 275/28/21514    most recent M002286200v1 Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Huang, C. Articles by Springer, T. A. Search for Related Content PubMed PubMed Citation Articles by Huang, C. Articles by Springer, T. A. Social Bookmarking                      What's this?

Structural and Functional Studies with Antibodies to the Integrin 2 Subunit

A MODEL FOR THE I-LIKE DOMAIN^*

Chichi Huang, Qun Zang^§, Junichi Takagi^¶, and Timothy A. Springer^¶

From the ^¶ Center for Blood Research and Department of Pathology, Harvard Medical School, Boston, Massachusetts 02115,  Pfizer Central Research, Groton, Connecticut 06340, and ^§ Biogen, Cambridge, Massachusetts 02142

Received for publication, March 17, 2000

	ABSTRACT

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

To establish a structure and function map of the 2 integrin subunit, we mapped the epitopes of a panel of 2 monoclonal antibodies including function-blocking, nonblocking, and activating antibodies using human/mouse 2 subunit chimeras. Activating antibodies recognize the C-terminal half of the cysteine-rich region, residues 522-612. Antibodies that do not affect ligand binding map to residues 1-98 and residues 344-521. Monoclonal antibodies to epitopes within a predicted I-like domain (residues 104-341) strongly inhibit LFA-1-dependent adhesion. These function-blocking monoclonal antibodies were mapped to specific residues with human  mouse knock-out or mouse  human knock-in mutations. Combinatorial epitopes involving residues distant in the sequence provide support for a specific alignment between the -subunit and I domains that was used to construct a three-dimensional model. Antigenic residues 133, 332, and 339 are on the first and last predicted -helices of the I-like domain, which are adjacent on its "front." Other antigenic residues in 2 and in other integrin  subunits are present on the front. No antigenic residues are present on the "back" of the domain, which is predicted to be in an interface with other domains, such as the  subunit -propeller domain. Most mutations in the 2 subunit in leukocyte adhesion deficiency are predicted to be buried in the 2 subunit I-like domain. Two long insertions are present relative to -subunit I-domains. One is tied down to the back of the I-like domain by a disulfide bond. The other corresponds to the "specificity-determining loop" defined in 1 and 3 integrins and contains the antigenic residue Glu¹⁷⁵ in a disulfide-bonded loop located near the "top" of the domain.

	INTRODUCTION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

Integrins are heterodimers that bind both cell surface and extracellular matrix ligands (1). The leukocyte or 2 integrins act as traffic signal molecules to regulate leukocyte emigration from the bloodstream in inflammation and lymphocyte homing (2). Additionally, the leukocyte integrin LFA-1 (CD11a/CD18) is important in lymphocyte/antigen-presenting cell and killer cell/target cell interactions in immune responses (3). LFA-1 binds to the immunoglobulin superfamily members intercellular adhesion molecules 1, 2, and 3 (ICAM-1, -2, and -3).¹ Mutations in the 2 integrin subunit cause leukocyte adhesion deficiency (LAD), a disease characterized by life-threatening bacterial infections, granulocytosis, and a lack of neutrophil diapedesis at inflammatory sites (4).

Ligand binding by integrins is dynamically regulated by cellular signals (5, 6). The adhesiveness of 2 integrins can be transiently stimulated after cellular activation through receptor tyrosine kinases and G protein-coupled receptors and can be stably activated with phorbol esters, Mn²⁺, or activating mAb. The mechanisms of activation of 2 integrins appear to include both conformational changes that regulate the affinity or multivalent affinity (avidity) for ligand and association with the cytoskeleton.

Regions in the N-terminal segments of integrin  and  subunits are important for ligand binding. All integrin  subunits contain seven repeats of ~60 residues each that are predicted to fold into a -propeller domain containing seven -sheets (7). The -propeller is toroidal in shape, with the -sheets arranged around a central pseudosymmetry axis like blades of a propeller. Some integrins, including the leukocyte integrins, contain a functionally important inserted (I) domain between -propeller sheets 2 and 3. The I domain has a doubly twisted fold that is found in nucleotide-binding enzymes and G proteins, with a central hydrophobic -sheet surrounded by amphipathic -helices (8-11). Five residues in a central pocket coordinate with Mg²⁺ either directly or via water molecules to form a metal ion-dependent adhesion site (MIDAS). During ligand binding, the sixth coordination position of the Mg²⁺ is hypothesized to be occupied by an acidic residue in the ligand (8). This residue in ICAM-1 is likely to correspond to Glu-34, which is by far the most critical residue for binding to LFA-1 (12). Mutation in I domains of the residues that form the primary or secondary coordination shell of the Mg²⁺ abolishes ligand binding (reviewed in Ref. 8). Specificity for ICAM-1 maps to amino acid residues on either side of the Mg²⁺ (13). Furthermore, mAbs that map to the I domain, but not mAbs that map to surrounding regions in the -propeller domain, block ligand binding by LFA-1 (13). These results show that the I domain in the  subunit of LFA-1 constitutes a ligand binding interface for ICAM-1.

The domain structure for integrin  subunits is less well defined than for  subunits. A region that is highly conserved among integrin  subunits is located between residues 104 and 341 of the 2 subunit. This region is 94% identical between mouse and human 2, and amino acid substitutions that cause LAD map to this region (reviewed in Ref. 14). Ligand cross-linking, mutational, and mAb mapping studies suggest that this conserved region is functionally important (15, 16). Blocking and stimulatory mAb have been mapped to this region in 1 and 3 integrins (16, 17). A MIDAS-like DXSXS motif is present in the conserved region, and the conserved region has been proposed to have an I domain-like fold (8, 18, 19). Mutation of the Asp and two Ser residues of the DXSXS motif as well as other residues in the conserved region with oxygenated side chains abolishes ligand binding in several  integrins. Alignment of hydropathy plots shows good similarity between the first half of the  subunit I domain and the  subunit conserved region (8); however, the second half diverges, suggesting that there are important structural differences in this region. The end of the putative I-like domain is therefore difficult to define. The only previously published three-dimensional model of an I-like domain, for that of the 3 subunit (18), proposed that it terminated at the equivalent of residue 284 in 2. A different structure-sequence alignment has been proposed that places the domain terminus at residue 341 (19); however, no specific three-dimensional model was proposed.

Cysteine-rich regions are present N-terminal and C-terminal to the -subunit I-like domain and are linked by a long range disulfide bond. The N-terminal cysteine-rich region is homologous to a repeat found in plexins, semaphorins, and integrins (PSI domain) (20). mAbs that activate integrins have been mapped to the C-terminal cysteine-rich region in 1 and 3 integrins (21-26), and the KIM127 mAb has been mapped to residues 413-575 in 2, in the middle third of the region C-terminal to the I-like domain (27). Epitopes for several other activating mAbs defined for 2 integrins, CBR LFA-1/2 (28) and KIM185 (29), have not been well localized. While there is no evidence that it participates in ligand binding, the C-terminal region may be important in transducing signals from inside the cell that regulate ligand binding.

To explore the structure and function of the 2 subunit, we have mapped epitopes of a panel of 17 mouse anti-human 2 mAbs and one rat anti-mouse 2 mAb, including function-blocking, nonblocking, and activating mAbs. Nonblocking mAbs map outside the I-like domain, and activating mAbs map to the C-terminal region. The function-blocking antibodies map to the I-like domain, and we have been able to define individual human/mouse amino acid substitutions responsible for five different antibody epitopes in this region. Several epitopes involve residues that are distant in the primary structure and provide constraints on the folding of this region that are used to predict its structure, in conjunction with secondary structure and threading prediction algorithms, sequence-structure alignments, and molecular modeling.

	MATERIALS AND METHODS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

Cell Lines and Monoclonal Antibodies-- COS-7, JY, and SKW3 cells were grown in RPMI 1640 medium supplemented with 10% fetal bovine serum. The mouse anti-human CD18 IgG1 monoclonal antibodies TS1/18 (30), CBR LFA-1/1, CBR LFA-1/2, CBR LFA-1/7 (28), and anti-human CD11a mAb TS1/22 (30) were previously described. Antibody 1C11 was obtained through the Fourth International Leukocyte Workshop. Antibodies 6.7 (31), CLB LFA-1/1 (32), and L130 were obtained through the Fifth International Leukocyte Workshop. mAb MHM23 (33) was a gift from Dr. A. McMichael. CLB-54 (34) was a gift from Dr. R. Van Lier. GRF1 (35) was a gift from Dr. F. Garrido. MEM-48 (36) was a gift from Dr. V. Horejsí. 11H6 was a gift from Dr. H. J. Bühring. May.017 (37) was a gift of Dr. Y. Ohashi. KIM185 (29) and 6.5e were gifts of Dr. M. Robinson. Rat anti-human antibodies YFC51.1 and YFC118.3 (38) were gifts from Dr. G. Hale. Addresses for the above antibody contributors are listed in Refs. 39 and 40. The rat anti-mouse CD18 mAb C71/16 (41) was from Dr. I. Trowbridge (Salk Institute, San Diego, CA).

Human and Mouse Chimeric CD18 Constructs-- Both human and murine 2 cDNA were in Ap^rM8 (13). Chimeras were named according to the species origin of their segments. For example, h98m indicates that residues 1-98 are from the human  subunit and residues 99 to the C terminus are from the mouse  subunit. Amino acid sequence numbering was according to the mature human sequence (42). The primers used for generating chimeras and substitution mutants are listed in Table I. In most of the cases, silent mutations were introduced into the primers to create a restriction site for testing of incorporation of the mutations. Polymerase chain reaction (PCR) was used to construct chimeras h612m, m612h, m521h, m122h, m163h, and h254m. To construct h612m, a DNA fragment from residue 612 to the C terminus of murine 2 subunit (m2) was generated by PCR, digested with BstBI and NotI, and ligated to a 7.7-kilobase pair DNA fragment generated from human 2 (h2) with the same enzymes. Chimeras m612h and m521h were constructed by insertion of PCR fragments (HindIII-612 or HindIII-521) amplified from m2 into HindIII and BstBI or HindIII and Eco47III sites of h2. The HindIII site is located on the 5' end of the cloning site of the vector Ap^rM8. Constructs m122h and m163h were made by insertion of the PCR fragments (HindIII-122 or HindIII-163) from m2 into HindIII sites created in the human to mouse mutants R122N or D163E. h254m was made by ligation of a PCR fragment (KasI-MluI) from m2 into the KasI and MluI sites of h344m.

Other chimeras were constructed by PCR overlap extension (43). In brief, two successive PCRs were used to generate a chimeric fragment, which was then digested with restriction enzymes and ligated into the h2, m2, or certain chimeric constructs. In the first PCR, two separate PCRs were used to generate one fragment from h2 and a neighboring fragment from m2. Two oligonucleotide primers at the overlap region were complementary for at least 24 bases. A silent substitution in the sequence of the overlap oligonucleotide was included to introduce a restriction site for screening of clones and for construction of subsequent chimeras. Chimeras h344m and m344h were constructed by inserting chimeric PCR fragments into HindIII and BstBI sites on h612m or h2, respectively. A silent mutation was included to introduce a MluI site on residue 344. The constructs, h302m, h163m, and h98m, or m302h and m254h, were made by insertion of the chimeric PCR fragments into restriction sites HindIII and MluI in h344m or m344h, respectively.

Construction of Human  Mouse Point Mutations-- To facilitate the construction of point mutants in the conserved region, a silent mutation was introduced into h2 to create a unique MluI restriction site at residue 344 by PCR overlap extension as described above. The PCR fragment containing the MluI site was ligated into the HindIII and BstBI sites of h2. The mutant, HmH, was verified by sequencing and expression on the COS cell surface and binding to ICAM-1 (not shown). The mutant N339Y was generated by PCR amplification with primers that overlapped the Bsu36I and MluI sites and encoded mutations near the MluI site. The PCR fragment was then transferred into the Bsu36I and MluI sites of HmH. PCR overlap extension was used to produce the other point mutants. To create mutants L270M, A290S/N292S, S302K/R303K, E325D, and H332Q, 5' upstream and 3' downstream primers containing the Bsu36I and MluI sites, respectively, were used to generate PCR fragments that were then ligated into the HmH. Similarly, mutants R122N, R133Q, D163E, E175A, N190D, and M218I were generated by using primers encompassing the HindIII and KasI sites. The PCR fragments were digested with these enzymes and ligated into h2. The double point mutants R133Q/H332Q and R133Q/N339Y were constructed by replacing the HindIII-KasI fragment of H332Q or N339Y with the HindIII-KasI fragment of R133Q. The triple point mutant R133Q/H332Q/N339Y was made by PCR amplification with R133Q/H332Q as template and primers that contained the Bsu36I and MluI site and that encoded the N339Y mutation near the MluI site. All point mutants were verified by sequencing about 100 base pairs around the mutation sites.

Construction of Mouse  Human Point Mutations-- Individual human residues were introduced into the murine  subunit sequence by PCR overlap extension (43). Briefly, the outer 5' and 3' primers were selected to include unique restriction sites near the mutation in the  subunit cDNA. Mutations were introduced by a pair of inner complementary primers. After PCR, the products were restriction-digested and ligated into the  subunit cDNA vector cut with the same enzymes. All constructs were verified by DNA sequencing of the region subjected to PCR.

Aggregation Assay-- JY or SKW3 cells used for aggregation assays were harvested near confluence (about 5 × 10⁵ cells/ml). After 20 min of preincubation with ascites (1:200) or purified mAbs (10 µg/ml), cells were stimulated with phorbol 12-myristate 13-acetate (PMA) at a final concentration of 50 ng/ml in 100 µl of L-15 medium (Sigma) supplemented with 5% fetal bovine serum. The reactions were performed in microtiter plates gently shaken for 30 min for JY cells or 2 h for SKW cells at 37 °C. The amount of aggregation was scored as described (13).

Cell Transfection and Immunofluorescence Flow Cytometry-- cDNAs in Ap^rM8 expression vector were purified by Wizard Midiprep kits (Promega, Madison, WI) and ethanol-precipitated. COS cells were transiently co-transfected with wild-type, mutant, or chimeric 2 subunits and human L subunit cDNA constructs using DEAE-dextran (44). Transfected COS cells were treated with trypsin-EDTA on day 2 and replated. On day 3, cells were harvested in 5 mM EDTA/phosphate-buffered saline and washed with L-15 medium supplemented with 2.5% fetal bovine serum.

Results with human  mouse mutations were repeated, and mouse  human knock-in mutations were tested, using transfection of 293T cells with calcium phosphate precipitates (45, 46). Medium was changed after 7-11 h. Cells were harvested for flow cytometry analysis 48 h after transfection.

Immunofluorescence flow cytometry was performed as described (47). Cells were incubated with 10 µg/ml purified mAb or a 1:200 dilution of ascites on ice for 30 min and then stained with a 1:20 dilution of fluorescein isothiocyanate-conjugated secondary antibody. After washing, the cells were resuspended and fixed in PBS containing 1% formaldehyde.

Adhesion Assay-- Transfected COS cells were labeled with 2',7'-bis-(2-carboxyethyl)-5-(and -6)-carboxyfluorescein acetoxymethyl ester (Molecular Probes, Inc., Eugene, OR) and assayed for binding to ICAM-1 as described (13). Binding of mutant transfected cells to ICAM-1 was expressed as a percentage of wild-type transfectant binding, i.e. 100 × (mutant  mock binding)/(wild-type  mock binding). Triplicates in each experiment were averaged and considered as a single data point for calculation of S.D. among at least three different experiments.

Secondary Structure Predictions-- Thirty-six integrin  subunits from six different phyla, Swiss-Prot or GenBank^TM accession codes P05107, P11835, S32659, P32592, P53714, P53713, P055563, P09055, P49134, P07228, P12606, P26010, P26011, S43534, O54890, P05106, I51530, P18084, AF043257, P18564, U77584, AF078802, AF059607, A57283, P11584, AF060203, X98852, AF005356, P26012, P26013, P16144, JN0786, Q64632, AF005357, and L13305 were aligned with PRRP with human 2 as the top sequence (48). The I-like domain and 10 adjoining residues on either end were excised. All columns with gaps in the human 2 sequence were removed from the alignment, and it was submitted as an SAF file to PHD (49).

Threading-- Threading was with THREADER V2.1a (50) (available on the World Wide Web). The data base of 1908 representative chains and files was updated as described previously (51). To this was added the 351 domains and chains with a three-layer sandwich architecture, i.e. each homology family representative from the CATH Database H-level Representatives List with the 3.40 architecture (52) (available on the World Wide Web). Domains were cut out of pdb files using the CATH domain definition list. Additionally, one representative for each conformationally distinct integrin I-domain and von Willebrand's factor A domain structure was added. Domains and chains were converted to THREADER data base format using STRSUM. A total of 2321 structures were present in the data base.

Sequence segments corresponding to residues 102-341 of integrin 2 from all 35 integrin  subunits described above were subjected to threading. Because of the insertions expected in the  subunit I-like domain compared with currently known structures, different loop gap penalty weights were tested (0.5, 0.45, 0.4, 0.35, and 0.30). The energies (pairwise plus solvation) for each structure were averaged for the 35 sequences, and Z-scores for the average energies were calculated with the histogram tool of Excel. Weights of 0.45-0.35 grouped I and A domains together in terms of their scores and thus gave the most consistent results, although the highest Z-score for any structure was obtained with the weight of 0.5 (4.72 with the M I domain).

Molecular Modeling-- The I domain structures of integrin M, 1jlm (9); L, 1zon (53); 2, 1aox (11); and the A1 and A3 domains of vWF, respectively, 1auq (54) and 1atz (55), were superimposed with 3DMALIGN of MODELLER (56) (available on the World Wide Web), using a gap penalty of 4 Å and five iterations of alignment with superposition of C, C, C, C, and finally C atoms. The superimposed structures and the resulting structure-sequence PIR alignment were opened in LOOK (Molecular Applications Group, Palo Alto, CA), and multiple gaps within each loop were condensed into a single gap, leaving insertions/deletions near turns or midpoints of loops. The 2 sequence was aligned using both secondary structure and sequence similarity, as described under "Results." In early models, the specificity-determining loop (57) of 2 was omitted. A LOOK model with 1atz as template was used as the .ini file in MODELLER, to obtain better starting positions for untemplated loops. All five I and A domain structures were used as templates in MODELLER with the alignment of Fig. 4, and an extra -ribbon in the Cheb methylesterase structure 1chd, residues 233-248, was used as a template for 2 residues 253-268. Because the 1chd template did not overlap any of the other templates, its position relative to the rest of the domain was only restrained by the disulfide between Cys²²⁴ and Cys²⁶⁴. Of 100 models, that with the best score with the QUACHK module of WHATIF (58) was chosen. Both this model and the I domain structures were used as templates in a further round of modeling in which the specificity-determining loop, 2 residues 149-176, was inserted. For the loop bonded by the Cys¹⁶⁹-Cys¹⁷⁶ disulfide, a disulfide-bonded loop of the same length was used, residues 5-13 of the CD59 complement-regulating protein structure 1cdq. 400 models were made, and one without knotted loops and with good QUACHK and NQACHK scores was chosen.

	RESULTS

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

Mapping of Epitopes on the 2 Integrin Subunit-- 13 human/mouse  subunit chimeras were constructed to map mAb epitopes and correlate localization in the structure with the effect of mAb on function (Fig. 1A). Proper association with the  subunit was shown by immunofluorescence staining with L mAb of COS cells cotransfected with the chimeric  and the human L subunits. Expression of L on the cell surface requires association with the 2 subunit (47, 59). All 13 chimeric 2 subunits associated with the human L subunit and were expressed on the cell surface as efficiently as wild-type human and murine 2 subunits as shown with TS 1/22 mAb to L (Fig. 1B). Furthermore, the function of all 13 chimeric 2 subunits associated with L in transfected COS cells was demonstrated by binding to purified human ICAM-1 and was equivalent to human L2 (data not shown).

View larger version (42K): [in this window] [in a new window]   Fig. 1.   Structure of chimeric 2 integrin subunits and mAb epitope localization. A, the human 2 subunit (h2) is shown with domains and restriction sites; sites with asterisks were introduced with silent mutations to facilitate the construction of chimeras. Numbers correspond to amino acid residues in the mature human 2 sequence (42). In chimeras, open and hatched bars correspond to human and mouse sequence, respectively. B, mapping of epitopes on the 2 subunit. mAb reactivity was determined by immunofluorescent flow cytometry on COS cells cotransfected with the indicated wild-type or chimeric 2 subunits and the human L subunit. Scoring was as follows: +, percentage of positive cells comparable with the wild-type transfectant; , staining not significantly different from the negative control (mock transfectant).

The chimeras were used to map the epitopes of a panel of 17 mouse mAbs to the human 2 subunit and one rat mAb to the mouse 2 subunit using immunofluorescence flow cytometry of transfected COS cells (Fig. 1B). The rat mAb to mouse CD18, C71/16, bound to an epitope contained within residues 1-122, as shown with m122h, and at least a portion of the epitope localized N-terminal to residue 98, as shown with h98m. This mAb thus maps N-terminal to the conserved domain.

Three mAbs, 6.7, MEM-48, and CBR LFA-1/7, were shown with reciprocal chimeras to bind to an epitope localized between residue 345 and 612, and the absence of staining of m521h further showed that the epitope included amino acids between residues 345 and 521. These mAbs thus map C-terminal to the conserved domain.

Two of the antibodies studied, CBR LFA-1/2 and KIM185, activate binding of 2 integrins to their ligands (28, 29). These mAbs mapped to the C-terminal half of the cysteine-rich region, from residue 522 to 612.

Further mAb mapped to discrete regions within the conserved region (Fig. 1B). The epitope recognized by mAb 11H6 included residues 99-122. The mAb GRF1 bound to an epitope between residues 123 and 163, as shown with chimeras h163m, m163h, h254m, and m254h. mAbs L130, MHM23, and 6.5e bound to the region between residues 164 and 254. mAb May.017 bound to both chimeras m163h and h163m. Lack of reactivity with h98m and m254h showed that this mAb recognizes residues in at least two segments, from 98 to 163 and from 164 to 254. mAb CLB LFA-1/1 and CLB 54 mapped to residues 303-344.

Two pairs of antibodies exhibited complex patterns of reactivity, suggesting recognition of epitopes in noncontiguous conserved subregions. mAbs YFC51.1 and YFC118.3 bound to an epitope between residues 122 and 344, as shown with chimeras m122h, h344m, and m344h. However, these mAbs failed to stain three pairs of reciprocal chimeras, h163m and m163h, h254m and m254h, and h302m and m302h. Thus, replacement of either the N-terminal portion (residues 122-163) or C-terminal portion (residues 302-344) of the conserved domain destroyed the epitope. This suggested that mAb YFC51.1 and YFC118.3 recognize an epitope requiring the presence of amino acid residues from both regions 123-163 and 302-344.

The staining pattern of mAbs TS1/18 and 1C11 was converse to that of mAb YFC51.1 and YFC118.3. These two mAbs could also be localized to an epitope between residues 123 and 344. However, mAb TS1/18 and 1C11 bound to all three pairs of the reciprocal chimeras h163m and m163h, h254m and m254h, and h302m and m302h. Even when these mAbs were titered and used at concentrations just sufficient to give good staining of h2, no significant difference in staining intensity between these reciprocal chimeras and h2 was detected (not shown). The binding to all three reciprocal chimeras suggests that recognition of one of two regions, which are split in all three chimeras, is sufficient for binding. Thus, the epitope includes residues from both regions 123-163 and 303-344, and the presence of human residues from either of these regions is sufficient for binding.

Mapping Epitopes to Individual Residues in the I-like Domain-- Only 14 amino acid residues in the I-like domain differ between the mouse and human, simplifying mapping of epitopes to specific amino acid residues. Site-directed mutagenesis was used to introduce individual mouse amino acid residues into the human 2 integrin sequence (Fig. 2A). Residues found to be important were further studied by introducing mouse  human substitutions into the mouse 2 subunit, co-expression with the human L subunit, and testing for gain of mAb reactivity (Fig. 2B). The mAb mapping results are described in light of a model for the I-like domain, which is described below (Figs. 4 and 5).

View larger version (40K): [in this window] [in a new window]   Fig. 2.   Localization of epitopes with individual amino acid substitutions. A, introduction of mouse residues into human 2 (knock-out). B, introduction of human residues into mouse 2 (knock-in). mAb reactivity was determined by immunofluorescent flow cytometry on COS cells and in independent experiments on 293T cells. Cells were cotransfected with the indicated mutant  subunits and the human L subunit or vector control. Scoring was as follows. +, percentage of positive cells comparable with the wild-type transfectant; ±, percentage of positive cells markedly less than wild-type transfectant but significantly higher than the negative control; , staining not significantly different from the mock transfectant negative control (mock).

mAb CLB LFA-1/1 and CLB 54 map to the predicted C-terminal -helix of the I-like domain. In agreement with mapping of these mAbs to residues 303-344, the human  mouse substitutions H332Q and N339Y reduced binding partially and completely, respectively (Fig. 2A). Expression of a mouse 2 subunit containing three mouse  human substitutions (Q133R, Q332H, and Y339N) yielded gain of expression of the CLB LFA-1/1 and CLB 54 epitopes, whereas the double mouse  human mutant Q133R/Q332H was negative (Fig. 2B). Together, the knock-out and knock-in data demonstrate that both His³³² and Asn³³⁹ contribute to this epitope. These residues are seven sequence positions apart in the predicted C-terminal -helix of the 2 subunit, the 6-helix (see below; Fig. 4). Since an -helix makes one turn per 3.5 residues, these residues are exactly two turns away from one another, on the same face of the predicted -helix, as appropriate for their presence in the same epitope (see below; Fig. 5).

Five mAb (GRF1, YFC51.1, YFC118.3, TS1/18, and 1C11) map to both the first and last predicted -helices of the I-like domain. For GRF1, the knock-out R133Q mutation demonstrated the importance of Arg¹³³ (Fig. 2A). The knock-in Q133R mutation confirmed the importance of Arg¹³³, but the knock-in mutation Q332H also demonstrated a role for His³³² (Fig. 2B). Furthermore, the double knock-in Q133R/Q332H bound the GRF1 mAb better than either single mutation, confirming the presence of both Arg¹³³ and His³³² in the epitope (Fig. 2B). The TS1/18 mAb was unaffected by knock-out of any single human residue, but binding was abolished by the double mutation R133Q/H332Q (Fig. 2A). Binding of TS1/18 was partially restored by knock-in of either Arg¹³³ or His³³² and completely restored by knock-in of both residues in the Q133R/Q332H mutant (Fig. 2B). The binding of YFC51.1 and YFC118.1 mAb was abolished by both the R133Q mutation and by the H332Q mutation (Fig. 2A). In agreement, single knock-ins of either Arg¹³³ or His³³² had no effect, but knocking the double Q133R/Q332H mutation into the mouse -subunit completely reconstituted binding of YFC51.1 and YFC118.1 (Fig. 2B). Binding of the 1C11 mAb was not affected by the single R133Q or N339Y mutations but was eliminated by the double mutant R133Q/N339Y. Introduction of Arg¹³³ into the murine sequence was not sufficient to restore binding, nor was the double knock-in Q133R/Q332H; however, the triple knock-in Q133R/Q332H/Y339N reconstituted binding. Thus, Arg¹³³ and Tyr³³⁹ are important in the 1C11 epitope.

The epitope mapping of the above five mAbs is in excellent agreement with the model described below, which shows that residues Arg¹³³, His³³², and Asn³³⁹ are in adjacent -helices (see below; Fig. 5). Indeed, these residues form a triangle, and each pair of residues that forms a side of the triangle is recognized by a different set of antibodies: Arg¹³³ and His³³² by GRF1, TS1/18, YFC51.1, and YFC118.1; Arg¹³³ and Asn³³⁹ by 1C11; and His³³² and Asn³³⁹ by CLB LFA-1/1 and CLB 54.

Four antibodies map within a short segment closed by a disulfide between Cys¹⁶⁹ and Cys¹⁷⁶, which has been shown in the 1 and 3 integrin subunits to determine ligand specificity (57). The E175A knock-out mutation abolishes recognition by mAb L130, MHM23, and 6.5e (Fig. 2A). The A175E knock-in mutation is sufficient for binding of these three mAbs and additionally May.017 (Fig. 2B). The ability of A175E to knock in binding of May.017 and the inability of E175A to knock out binding are consistent with the finding that human residues 1-162 are sufficient for May.017 binding (Fig. 1B). Residues 1-98 are not sufficient. All of the human residues in the 99-162 interval have been knocked in individually and do not restore May.017 binding; therefore, it appears possible that two or more residues in the 99-162 interval, or one in this interval and another in the 1-98 interval, contribute to the epitope.

Inhibition of Lymphoid Cell Homotypic Aggregation with mAbs to 2-- Two lymphoid cell lines, JY and SKW3, were used to test the ability of CD18 mAbs to block the interactions of LFA-1 with ICAM-1 and ICAM-3. JY cells express ICAM-1, less ICAM-2, and no ICAM-3. On the other hand, SKW3 cells express ICAM-3, less ICAM-2, and no ICAM-1. In agreement with this, the homotypic aggregation of PMA-stimulated JY and SKW3 cells is mediated predominantly by LFA-1 interaction with ICAM-1 and ICAM-3, respectively (60, 61). The inhibition of homotypic aggregation of JY and SKW3 cells by 2 subunit mAbs was concordant (Table II). Furthermore, there was an excellent correlation between the epitopes to which mAb bound and their effect on function. Antibodies that bound C-terminal to the I-like domain were not inhibitory. Of these mAb, those that bound more C-terminally were the ones that have previously been shown to activate LFA-1 adhesiveness: CBR LFA-1/2, KIM185, MEM48 (28, 29), and, according to previous mapping results, KIM127 (27). All 12 mAbs that mapped to the I-like domain were inhibitory (Table II), despite binding to at least five distinct classes of epitopes within this domain.

View this table: [in this window] [in a new window]   Table II Inhibition of PMA-stimulated SKW3 and JY cell aggregation with mAb to the 2 subunit

A Prediction of a Fold for the Conserved Domain-- The secondary structure and solvent accessibility of the 2 subunit were predicted by PHD (49), using as input a multiple alignment containing 35 integrin -subunits (see "Materials and Methods") including all eight  subunits known in Vertebrata and subunits from five other phyla: Arthropoda, Echinodermata, Nematoda, Cnidaria, and Porifera. The predicted secondary structure for the entire extracellular domain is shown to scale (Fig. 3). The PSI-like domain in the first 50 residues of 2 contains two predicted -helices. The only other predicted -helices are in the region corresponding to the I-like domain; these  helices alternate with predicted -strands, as previously noted (19, 62), and thus the I-like domain is predicted to have an / fold. Between the PSI domain and the I-like domain are four predicted -strands. The region C-terminal to the I-like domain contains approximately 20 predicted -strands. The predicted -strands in the I-like domain are hydrophobic, as appropriate for a three-layer sandwich containing a central hydrophobic -sheet and surrounding amphipathic helices. By contrast, the -strands N- and C-terminal to the I-like domain have alternating hydrophobic and hydrophilic amino acids, as appropriate to two-layer sandwiches. Because of the above considerations, and also the I-like domain's greater conservation in evolution, it is justifiable to segment out the I-like domain as an / domain between all- domains and to consider it separately for structure prediction.

View larger version (21K): [in this window] [in a new window]   Fig. 3.   Predicted secondary structure, domain organization, and epitope localization of the 2 integrin extracellular segment. Secondary structure was predicted with PHD using a multiple sequence alignment of 35 integrin  subunits as described under "Materials and Methods." The lengths of predicted secondary structure segments are shown to scale, along with cysteines and confidently defined disulfide bonds in 3 (75).

Threading with THREADER 2.1 and an up-dated fold data base was used to identify a fold for the I-like domain. In threading, a sequence is aligned with or "threaded through" each structure in a data base. The sequence-structure alignments are completely analogous to sequence-sequence alignments, including provision for gaps or insertions, but what is calculated is the pseudoenergy of the test sequence in each three-dimensional structure in the data base. The sequence corresponding to residues 102-341 of 2 for each of the 35 different integrin  subunits was threaded, and scores for each structure in the fold data base were averaged for all 35 sequences. Previous results on threading the I-like domain gave inconsistent results for the two different  subunits that were studied, and the fold data base did not include any integrin I or vWF A domains (19). Averaging results for multiple homologous sequences yields more reliable results (51, 63). Six integrin I domain and vWF A domain structures were in the data base, and all six were among the top seven structural hits (Table III). Furthermore, the remaining structure among the top seven hits, 1chd, is a member of the same fold family, known as the Rossmann, or nucleotide-binding, fold and has some features that may be significant for the  subunit I-like domain that will be discussed below. The Z-scores for I domains and 1chd are in a moderately high range that suggests that the I-like domain has a Rossmann fold and has structurally significant similarities to integrin I domains. Since the scores are only moderately high and a domain with a different type of Rossmann fold is also present in the high scoring group, significant differences with I domains are also likely to exist. This justifies the "I-like" designation, to emphasize both similarities and differences with I domains.

View this table: [in this window] [in a new window]   Table III Averaged threading results for 35 -subunit I-like domains Threading was as described under "Materials and Methods" with a gap penalty of 0.45. Energies for each of 2321 structures were averaged for I-like domains from 35 different integrin -subunits. Z-scores were calculated, and the highest scoring structures are shown. According to the Threader 2 User Guide (50), "Experience has shown the following interpretations to be useful: Z > 3.5, very significant, probably a correct prediction; Z > 2.9, significant, good chance of being correct; 2.7 < Z < 2.9, borderline significant, possibly correct; 2.0 < Z < 2.7, poor score, could be right, but needs other confirmation."

Consistent with the threading results, sequence alignment with I domains reveals significant similarities and differences (Fig. 4). Integrin I domains and vWF A domains, referred to generically as I domains, were structurally superimposed to yield a structure-sequence alignment, and the 2 I-like domain was then aligned using both sequence and secondary structure similarities. The beginning of the -subunit I-like domain can easily be aligned with I domains, using sequence similarities in the 1-strand, the DXSXS motif of the MIDAS, and hydrophobic residues in the 1-helix (Fig. 4). The predicted 2-strand of the I-like domain can also be aligned with the 2-strand of the I domain by sequence. The last 60 residues of the I and I-like domains can be equivalenced based on their alternating 4, 5, 5, 6, and 6 secondary structure units and because the 6-helix is the last predicted -helix in the entire  subunit (Fig. 3). Sequences in this region can be aligned based on secondary structure prediction and hydrophobic amino acids (Fig. 4).

View larger version (82K): [in this window] [in a new window]   Fig. 4.   Alignment of the human 2 I-like domain with integrin I domains and vWF A domains. Five I and A domain high resolution structures were structurally superimposed as described under "Materials and Methods" to yield the shown structure-sequence alignment. The 2 I-like domain was then aligned according to its sequence and predicted secondary structure. -Strand and -helix segments are highlighted in pink and yellow, respectively, as predicted for the human 2 I-like domain with PHD (49) using a multiple alignment of 35  subunit sequences or determined for structures of I and A domains with DSSP (76). Above the human 2 sequence are shown residues that differ in mouse (black) or that are mutated in LAD (green) (14, 68). Dots above the human 2 sequence show every 10th residue. Residues in the primary or secondary coordination shell of the I domain structures are red. Conserved residues with oxygenated side chains in 2 are colored red if mutations of a particular residue in all tested  subunits (1, 2, 3, and 5) abolished ligand binding and not subunit expression (18, 62, 71, 77, 78). The confidence in the alignment between the -subunit and I domains is coded in the bottom bar as follows: correct (white bar); aligned with correct secondary structural unit but perhaps offset in sequence (light gray); uncertain topology (dark gray). Consensus sequences are as follows: aliphatic hydrophobic (@); hydrophobic (o); charged (j); small (dot); aromatic hydrophobic ($); positively charged (+).

In Rossmann folds, the central -strands are the longest and most hydrophobic. In I domains, -strands 4 and 1 are central. The 4-strand of the I-like domain is readily identified by its hydrophobicity and sequence similarity to the 4-strand of I domains (Fig. 4). Alignment of the region with the 3-strand and 2- and 3-helices is difficult, but it is aided by the position of the neighboring 4-strand. Because alignment of 3, 2, and 3 is difficult and there is a weakly predicted -strand in the I-like domain between the 2 and 3 helices, it is possible that the I and I-like domains could differ in topology in this region. The alignment in Fig. 4 indicates regions predicted to be correctly aligned (white bars), aligned with the correct secondary structural unit but perhaps offset in sequence (light gray bars) and of uncertain topology (dark gray bar). The alignment is similar to that of Ref. 19 and very different from that of Ref. 18.

Two long insertions in the I-like domain account for its greater length compared with I domains. A long loop in I-like domains, the specificity-determining loop, is inserted between -strands 2 and 3, and a long sequence with two weakly predicted -strands is inserted between 4 and 4 (Fig. 4).

In I domains, the order of -strands is 6-5-4-1-2-3 (the same numbering is used for the I-like domain model; Fig. 5A). The C termini of -strands 6, 5, 4, 1, and 2 are at the "top," as is the MIDAS. The -helices 6, 1, and 2 form the "front" face, and -helices 3, 4, and 5 form the "back" face, with a counterclockwise order viewed from the top of 1-2-3-4-5-6 surrounding the hydrophobic -sheet (see I-like domain model; Fig. 5A).

View larger version (70K): [in this window] [in a new window]   Fig. 5.   Stereodiagram of the 2 I-like domain model. A, view of the top of the domain centered on the Mg2+ ion. B, view of the front of the domain centered on -helices 1 and 6. B, rotated about an axis through the center of each -strand 90° relative to A. The ribbon diagram (79) backbone is in blue and gray for areas of confident and uncertain topology, respectively. Oxygen and nitrogen atoms are red and blue, respectively. Antigenic residues in human 2 have yellow side chains. Positions corresponding to antigenic residues in human 1 (17) and chicken 1 (80) are shown as yellow lollipops with C-C bonds and large C atoms. Residues with oxygen-containing side chains important in ligand binding (see Fig. 4) have silver side chains. Mg2+ is represented as a large silver sphere. The two disulfide bonds are shown in yellow. In A, mutations in LAD and N-linked sites are shown. Residues mutated in LAD, except those with oxygen-containing side chains shown with silver side chains, are represented with purple lollipops. Positions corresponding to N-linked sites in at least 2 of 35 -subunits appear as green lollipops. Note that antigenic residues and N-linked residues are exposed, while residues mutated in LAD are often buried. Note the absence of antigenic residues and N-linked sites on the 3- and 4-helices and the long loops on the "right" and "back" sides of the domain.

The major differences between the I and I-like domains (i.e. the regions of uncertain topology and the two long insertions) are localized to its "right" and "back" sides (gray main chain trace in Fig. 5, A and B). The 3-strand and 2-helix of uncertain alignment topology are located in this region, as are the long insertions between the 2- and 3-strands and between the 4 and 4 elements. Notably, I domains differ from their close structural homologues, the small Ras-like G proteins, in this same region on the right side of the domain; furthermore, the differences between I domains and 1chd, the other top hit in threading, occur in this region. The 3-helix in the I-like domain is more hydrophobic than other I-like domain -helices (Fig. 4) and resembles a hydrophobic -helix in the same structural location in 1chd, which neighbors and is partly buried by two "extra" -strands. The long insertion between 4 and 4 in the I-like domain probably has a similar structural location (but not topology) to the extra -strands in 1chd, because this loop is disulfide-bonded to the 3-helix. Appropriately, the long loop buries the hydrophobic 3-helix in the I-like domain model (Fig. 5A). Nearby, the specificity-determining loop (57) between -strands 2 and 3 is predicted to project from the top of the I-like domain.

The molecular model of the 2 subunit I-like domain (Fig. 5, A and B) was constructed with LOOK and MODELLER, as described under "Materials and Methods." All five I domains were used as templates, using the alignment shown in Fig. 4. Frameworks were provided for modeling portions of the two long insertions, but the only restraint on orientation relative to the rest of the I-like domain was the disulfide bond between Cys²²⁴ and Cys²⁶⁴. The quality of the models was evaluated with programs developed to check the quality of x-ray and NMR structures and models, using structural features that differ from those used in refinement (58, 64) (Table IV). The quality of the 2 model is in a range that suggests that it is correctly threaded. The quality is better than for a previously described 3 model with a markedly different alignment and C-terminal domain boundary. Furthermore, the 2 model includes two long untemplated loops, which cannot be accurately modeled and decrease model quality. When these loops are deleted to give a model with the same number of residues as in the 3 model, the quality is markedly better than for the previous model (Table IV).

	DISCUSSION

TOP ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS DISCUSSION REFERENCES

Our studies with mAb to the 2 integrin subunit provide a structure-function map and support a specific structure-sequence alignment and model for the I-like domain. The 12 mAb we mapped to the I-like domain recognized five sets of different epitopes, yet all inhibited binding of LFA-1 to ICAM-1 and ICAM-3 as measured in homotypic adhesion assays. This confirms the importance in 2 function of this domain, as previously suggested by its evolutionary conservation, the high incidence of mutations in this domain in LAD, and mutation of MIDAS residues. Of three mAb that mapped to a region between the C terminus of the I-like domain and the midpoint of the cysteine-rich region, none inhibited LFA-1 function. By contrast, two activating antibodies (28, 29), CBR LFA-1/2 and KIM185, mapped to the C-terminal half of the cysteine-rich domain (Fig. 3).

The epitopes of several 1, 2, and 3 activation antibodies have been mapped to regions close to residues 522-612, to which CBR LFA-1/2 and KIM185 map. mAb KIM185 has independently been mapped to an overlapping region of residues 406-570 (27). Similarly, KIM127, another 2 activation antibody, has been mapped to the region of residues 406-570 (27). LIBS2, which can increase the affinity of the platelet integrin IIb3 to fibrinogen by 20-fold, was localized within an 89-amino acid residue region immediately next to the transmembrane domain (21). TASC, an 1 integrin-activating antibody that promotes adhesion to laminin, binds to the region from residue 493 to 602 (16). Since the large number of disulfides in the C-terminal half of the cysteine-rich region should keep it rigid, it is hard to imagine that antibodies that bind to this region could induce a conformational change in the  subunit that would be propagated to the N-terminal region where the ligand binding sites are located. On the other hand, it seems likely that antibodies binding to this region could act like a "wedge" to keep the region to which they bind in the  subunit further apart from the  subunit and thus alter the relative orientation of the  and  subunits in the more N-terminal ligand binding region. The epitope of a mAb "G" to 1 (16) that has been shown to disrupt subunit association was mapped to the same region. Therefore, this region may have an important function in transducing inside-out signals from the cytoplasm to the ligand-binding head piece of integrins.

Although it has been proposed that the evolutionarily conserved region of integrin  subunits contains a MIDAS and adopts an I domain fold (8), this proposal has been clouded by disagreement on the C-terminal boundary of the domain (18, 19), a lack of consistent threading predictions, and lack of any experimental evidence in support of a particular topology. Our alignment is similar overall in the secondary structure elements that are equivalenced to that of Tuckwell and Humphries (19), although different in many alignment positions, and differs dramatically from that of Tozer et al. (18). We provide computational and experimental evidence for a particular structure-sequence alignment and qualify the confidence in different parts of this alignment. Secondary structure predictions on the entire 2 extracellular domain defined a specific segment with alternating -helices and -strands. This alternating pattern is characteristic of / folds, and we identified it with the I-like domain. Threading predictions on this segment averaged from 35 different integrin  subunits consistently identified I domains and another domain with a Rossmann fold as the top hits. The beginning and end of the I-like domain could be clearly aligned with the I domain, as could the central 4-strand. Model building provided support for the structure-sequence alignment, because the structural quality was better than for a model derived with a different alignment. Finally, mAb mapping showed that residues 133, 332, and 339 are adjacent in the three-dimensional structure. This evidence is independent of, and hence reinforces, the secondary structure prediction and sequence alignment that suggest these residues are in adjacent -helices. The fold is clearly similar overall to that of I domains but also different in some respects. We explicitly acknowledge uncertainty in the threading of 3 and 2 and two inserted loops that are untemplated by different color codes in Figs. 4 and 5. Furthermore, we cannot rule out a contribution by -strands N-terminal to residue 104 or C-terminal to residue 342 to the I-like domain.

Mapping with six different mAbs, which recognize three distinct epitopes, demonstrate that residues 133, 332, and 339 are adjacent in the structure. Each pair of these residues is recognized by a different group of antibodies. The proximity of these residues confirms our structure-sequence alignment and three-dimensional model, which places these residues in structurally adjacent -helices 1 and 6. Residues 332 and 339 are exactly two helix turns apart in predicted -helix 6, as appropriate for presence on the same face of this helix. In the model, residues 133, 332, and 339 form an almost equilateral triangle, with their C carbons 12.8-14.9 Å distant (Fig. 5B). Each pair of residues can readily be encompassed in an antibody footprint, which is approximately 30 Å in diameter.

Of overall interest for integrin domain organization, we can predict which faces of the I-like domain are solvent-exposed and, conversely, which are not exposed and thus may interact with other domains in the integrin  and  subunits. The predicted helices 1 and 6 are highly exposed in the intact integrin, as shown by antibody recognition of residues Arg¹²², Arg¹³³, His³³², and Asn³³⁹ (Fig. 5, A and B). Residues recognized by inhibitory mouse mAb to chicken 1 map nearby to the end of the 1-helix and the loop connecting it to the 2-strand (yellow lollipops in Fig. 5, A and B). Activating and inhibitory mAb to human 1 map to the neighboring 2-helix (yellow lollipops in Fig. 5, A and B). Therefore, the "front" face of the I-like domain that bears the 6, 1, and 2 helices is well exposed in the intact integrin, and this includes residues that extend from the top of this face to the bottom of the domain. The location of the antigenic residue Glu¹⁷⁵ is much less certain, but it appears to be not far from the other antigenic residues, near the "top ri