Identification of a Binding Site for Integrin αEβ7 in the N-terminal Domain of E-cadherin

The integrin αEβ7, which is predominantly expressed on mucosal T lymphocytes, has recently been shown to recognize the cell adhesion molecule, E-cadherin, on epithelial cells. We have carried out mutations on E-cadherin, involving domain deletions as well as substitutions of specific amino acids, in order to identify the sites recognized by the integrin. Binding of αEβ7 required the presence of the first two N-terminal domains of E-cadherin. Deletion of extracellular domains 3 and 4 or truncation of the cytoplasmic domain of E-cadherin had no consequence on integrin binding. Substitution of a glutamic acid in the BC loop of the Ig structure of the fist, N-terminal, domain of E-cadherin abrogated binding of αEβ7. This mutation did not appear to affect the conformation of the domain nor the pattern of expression of E-cadherin on the cell surface. Synthetic peptides encompassing the first domain of E-cadherin had very little inhibitory effect on the interaction with αEβ7. Our results highlight structural dissimilarities between recognition of E-cadherin by αEβ7 and recognition of other members of the IgSF by integrins and show that the heterophilic (integrin binding) and homophilic sites in the N-terminal domain of E-cadherin are distinct.

The integrin ␣E␤7, which is predominantly expressed on mucosal T lymphocytes, has recently been shown to recognize the cell adhesion molecule, E-cadherin, on epithelial cells. We have carried out mutations on E-cadherin, involving domain deletions as well as substitutions of specific amino acids, in order to identify the sites recognized by the integrin. Binding of ␣E␤7 required the presence of the first two N-terminal domains of E-cadherin. Deletion of extracellular domains 3

and 4 or truncation of the cytoplasmic domain of E-cadherin had no consequence on integrin binding. Substitution of a glutamic acid in the BC loop of the Ig structure of the fist, N-terminal, domain of E-cadherin abrogated binding of ␣E␤7. This mutation did not appear to affect the conformation of the domain nor the pattern of expression of E-cadherin on the cell surface. Synthetic peptides encompassing the first domain of E-cadherin had very little inhibitory effect on the interaction with ␣E␤7. Our results highlight structural dissimilarities between recognition of E-cadherin by ␣E␤7 and recognition of other members of the IgSF by integrins and show that the heterophilic (integrin binding) and homophilic sites in the N-terminal domain of E-cadherin are distinct.
The cadherins are a superfamily of Ca 2ϩ -dependent homophilic cell adhesion molecules that regulate recognition and interaction between cells (1,2). E-cadherin, which is expressed by epithelial cells, plays an important role in embryonic development (1) and in tissue morphogenesis (1,3). It is pivotal in the establishment and maintenance of cell polarity (4) and in tumor suppression (5,6). In polarized epithelial cells, E-cadherin is localized at the basolateral membrane and is concentrated at the adherens junctions (7).
The extracellular region of E-cadherin is made up of five tandem repeats of approximately 110 amino acids each (1,2). The adhesion specificity resides in the N-terminal extracellular domain and is dependent on the integrity of the amino acid sequence HAV (8). The cytoplasmic domain of E-cadherin associates with the catenins, which provide linkage to the actin cytoskeleton (9,10). The presence of the cytoplasmic domain is essential for the function of E-cadherin in homophilic cell-cell interactions and cellular organization (10,11). E-cadherin can associate with various membrane and cytoplasmic proteins (1,12) and can transduce information from extracellular contacts, as well as initiate intracellular signaling pathways.
The first, N-terminal, extracellular domain of E-cadherin has an Ig-like structure composed of seven anti-parallel ␤-strands arranged as two opposing sheets (13). A calcium binding pocket is present at the interface between successive domains (13). The protein is most likely to be expressed as a dimer (14), with the junction between domains 1 and 2 providing the interface for dimerization. Three calcium ions bound at the junction between domains 1 and 2 coordinate and stabilize the formation of the dimer (14). A general model for cadherinmediated interactions has been proposed in which cadherin dimers on opposing cells form a zipper-like superstructure compounding weak adhesive interactions to form strong intercellular bonds (15).
In addition to its function in homophilic cell-cell adhesion, E-cadherin has been shown to be a ligand for integrin ␣E␤7 (16,17). This integrin shows a very narrow pattern of expression, being restricted to T lymphocytes in the vicinity of mucosal epithelia (18 -21). The level of expression is particularly high on intraepithelial lymphocytes in the gut, a population that represents a large proportion of the total T cell complement of the body.
In the present report, we describe mutagenesis studies on E-cadherin, involving both domain deletions and point mutations, that aim to identify the binding sites for ␣E␤7. Our results show that the N-terminal domains 1 and 2 of E-cadherin are both required for integrin binding and that a glutamic acid in the first domain is crucial for the interaction. The location of this residue in the loop connecting ␤-strands B and C and the identity of its neighboring amino acids highlight striking differences between the interaction of ␣E␤7 with Ecadherin and the interaction of other leukocyte integrins with the IgSF.
Construction of E-Cadherin Domain Deletion Mutants-cDNA for mouse E-cadherin (24) in pBluescript SK ϩ was obtained from R. Kemler (Max-Planck Institute, Freiburg, Germany). Regions coding for extracellular domains EC1, EC2, EC3, and EC4 were separately deleted from * The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. the cDNA using the overlap extension PCR method (25). The decision on domain boundaries was guided by the assignment of repeats in the E-cadherin sequence obtained from SwissProt Data Base (CADEmouse, AC P09803). EC1 is defined as amino acids 1-108, EC2 as amino acids 109 -221, EC3 as amino acids 222-332, and EC4 as amino acids 333-439. To generate a full-length expression construct, the PCR, utilizing Pfu polymerase, was used with the primers: 5Ј-AGTGGATCCCCCAG-ATCTGAGCTT-3Ј (forward) and 5Ј-GACTTTCTAGACCCCTAGTCGT-CCTCGCCAC-3Ј (reverse). The following oligonucleotide primer pairs were used for generating the domain deletion mutant DNA by PCR: deletion of EC1, 5Ј-CTCAGAAGACAGAAACGAACCCAGGAGGTGTT-TGAG-3Ј (forward) and 5Ј-TCGTTTCTGTCTTCTGAGACC-3Ј (reverse); deletion of EC2, 5Ј-GACAACAGGCCAGAGTTTAACCCGAGCACGTA-TCAG-3Ј (forward) and 5Ј-AAACTCTGGCCTGTTGTCATT-3Ј (reverse); deletion of EC3, 5Ј-AATGACAACGCTCCTGTCTTCATGCCTGCGGA-GAGG-3Ј (forward) and 5Ј-GACAGGAGCGTTGTCATTAAT-3Ј (reverse); deletion of EC4, 5Ј-AATGAAGCCCCCATCTTTATCCCAGAAC-CTCGAAAC-3Ј (forward) and 5Ј-AAAGATGGGGGCTTCATT-3Ј (reverse). Each mutant DNA was prepared using PCR reactions to encompass the domains on each side of the deletion and a third reaction to join the two products. These mutants were constructed in pBluescript SK ϩ , excised by digesting with BglII/XbaI, and ligated into pcDNAI/Neo (Invitrogen) at BamHI/XbaI sites. Another mutant was constructed in which the DNA sequence coding for the last 72 amino acids of the cytoplasmic domain of E-cadherin (amino acids 657-728) were deleted. For this mutation, the primers 5Ј-AGTGGATCCCCCAG-ATCTGAGCTT-3Ј (forward) and 5Ј-GCGGCAGTCGACTAGAAGTTTC-CAATTTCATCAGGATT-3Ј (reverse) were used. The latter incorporates a stop codon. This mutant DNA was excised from pBluescipt SK ϩ using enzymes BglII/SalI and ligated into pcDNAI/Neo at BamHI/XhoI sites. The fidelity of each domain deletion was confirmed by DNA sequencing.
Site-directed Mutagenesis in Domain 1 of E-cadherin-Mutations of single amino acids in domain 1 of E-cadherin were introduced by the site-specific mutagenesis method of Kunkel et al. (26) using oligonucleotides. Wild-type E-cadherin was subcloned into pcDNAI/Amp (Invitrogen) and a Muta-Gene phagemid in vitro mutagenesis kit (Bio-Rad Laboratories Ltd, U. K.) was used to prepare the mutations. Six amino acids (Glu-31, Asp-44, Glu-54, Glu-56, Glu-86, and Glu-93) were mutated to alanine. The oligonucleotides used consisted of 28 -36 DNA bases spanning the mutation sites and incorporating a single base change (A to C) in codons for aspartate or glutamate. DNA sequencing identified plasmids that contained the desired mutation, and these were transfected into mouse L cells.
Transfection of L Cells-Wild-type and mutated E-cadherin cDNAs were stably transfected by the calcium phosphate method into mouse L cells, which do not normally express E-cadherin. Wild-type cDNA and domain deletion constructs were in the vector pcDNAI/Neo. Constructs carrying the point mutations were in pcDNAI/Amp, and these were cotransfected with pSV2*Neo (27) at a 10:1 molar ratio to permit selection. The transfected cells were cultured in Dulbecco's minimum essential medium, 10% FCS, 0.9 mg/ml geneticin (G418). After two weeks of G418 selection, transfectants expressing high levels of Ecadherin were isolated by cell sorting on a Facstar (Becton Dickinson), following immunofluorescence staining for E-cadherin. Alternatively, they were treated with DECMA-1 and positively selected by panning on Petri dishes coated with affinity-purified sheep anti-rat Ig (Seralab, U. K.) at 10 g/ml.
Immunofluorescent Staining and FACS analysis-L cell transfectants were lifted with 0.05% trypsin in Hanks' balanced salt solution in the presence of 2.0 mM Ca 2ϩ to maintain conformational integrity of E-cadherin (28). The cells were stained with mAbs to E-cadherin followed by FITC-conjugated anti-rat IgG. Immunofluorescence was analyzed with a Facscan (Becton Dickinson) using Consort 30 or Lysys software.
Confocal Microscopy-Transfected L cells expressing wild-type or mutated E-cadherin were seeded onto glass coverslips overnight so that they formed a confluent layer. For immunofluorescent staining, live unfixed cells on coverslips were incubated with mAb DECMA-1 supernatant for 12 min at room temperature, washed with medium (6 min), incubated with secondary FITC-labeled antibody for 12 min at room temperature and washed with medium (6 min). Stained cells were observed under an Odyssey XL laser scanning confocal microscope (Noran Instruments Inc.) using standard FITC filters.
Computer Models of E-cadherin Domain 1-The atomic coordinates of E-cadherin domain 1 were obtained from Dr. M. Ikura (University of Toronto, Canada), and the structure was modelled on an Evans & Sutherland computer using the program FRODO.
Synthetic Peptides-Two linear peptides, NRDKETKV and RENKK-TDV, were synthesized in a Milligan Biosearch 9500 synthesizer using N-(9-fluorenyl)methoxycarbonyl-protected amino acids and a N,N-dimethylformamide solvent system. Peptide sequences were confirmed by Edman degradation using an ABI 470A protein sequencer with a 120APTH analyzer. Cyclic peptides, CDKETKC and CDKKTEC, and overlapping peptides of 15 amino acids each, spanning the first domain of E-cadherin, were made by Alta Biochemistry, University of Birmingham, U. K.
Adhesion Assays-Adhesion of MTC-1 cells to a monolayer of CMT93 or L cells was assayed as described previously (17). Blocking antibody, M290 IgG, was used at 10 g/ml. Peptides were used at the indicated concentrations. In some experiments, a variation of the assay was used: cells were plated on the 96-well plate for only six hours, instead of overnight, at a low density (1.0 ϫ 10 4 cells/well instead of 3.5 ϫ 10 4 cells/well). These conditions resulted in a sparse cell monolayer.

Construction, Expression, and Assay of Domain Deletion
Mutants-The domain deletion mutants are schematically outlined in Fig. 1. In four of these (⌬1, ⌬2, ⌬3, and ⌬4) E-cadherin domains EC1, EC2, EC3, and EC4 were separately removed. Mutant ⌬cyt encoded E-cadherin lacking the last 72 amino acids in the C terminus of the cytoplasmic domain, which is the region required for binding to the cytoskeleton (10,11). Expression of mutant E-cadherin by L cell transfectants was assessed by indirect immunofluorescent staining with monoclonal antibodies DECMA-1, ECCD-1, and ECCD-2 (Fig. 2). mAb MAC123 against an irrelevant antigen was used as a negative control. Previous studies have established that DECMA-1, which blocks the homophilic adhesion function of E-cadherin, recognizes an epitope located at the membrane proximal part of the extracellular region of E-cadherin (22) and that ECCD-1, which also blocks E-cadherin homophilic function, binds to an epitope in EC1 (8). The location of the epitope for ECCD-2 (23) has not been reported. L cells transfected with the domain deletion mutant ⌬1 showed no staining with either ECCD-1 or ECCD-2 but good staining with DECMA-1 (Fig. 2). Mutant ⌬4 showed staining with ECCD-1 and ECCD-2 but not with DECMA-1. Transfectants expressing mutants ⌬2, ⌬3, and ⌬cyt showed staining with all three antibodies. The results confirm the location of epitopes for ECCD-1 and DECMA-1 and show that ECCD-2, which does not block the homophilic function of Ecadherin but does inhibit binding to ␣E␤7 (17), detects an epitope in domain EC1.
The transfectants carrying the five E-cadherin domain deletion mutants were tested for adhesion to MTC-1 cells, which express ␣E␤7 at high levels (Fig. 3). Transfectants carrying wild-type E-cadherin or empty vector were used as positive and negative controls, respectively. Mutants ⌬1 and ⌬2 showed no adhesion activity above background, whereas mutants ⌬3, ⌬4, and ⌬cyt supported the same level of adhesion as wild-type E-cadherin (Fig. 3). The adhesion activity of mutants ⌬3, ⌬4, and ⌬cyt was strongly inhibited by mAb M290, which blocks ␣E␤7 function.
Generation of Amino Acid Substitution Mutants in Domain 1 of E-cadherin and the Effect on ␣E␤7 Binding-Studies on interactions between integrins and cell adhesion molecules of the IgSF have shown that certain conserved motives, which invariably include a highly exposed charged residue located within loop regions of the Ig structure, are important for recognition by integrins (29). Therefore, we targeted for mutagenesis six exposed glutamic or aspartic acid residues in domain EC1 of E-cadherin; Glu-31, Asp-44, Glu-54, Glu-56, Glu-86, and Glu-93 were mutated to alanine. A computer-generated model of EC1 showing the location of the six amino acids is seen in Fig. 4. The level of expression of mutant E-cadherin on the surface of L cell transfectants was assessed by indirect immunofluorescence with E-cadherin antibodies (Fig. 5). All transfectants were able to bind antibodies DECMA-1, ECCD-1, and ECCD-2 at comparable levels. It is, therefore, unlikely that the mutations were sited within the epitopes detected by the EC1 antibodies, ECCD-1 and ECCD-2. The mutants were tested for adhesion to MTC-1 cells (Fig. 6A). The mutant E31A showed no adhesion above background levels, whereas all the other mutants supported adhesion of MTC1 cells to the same extent as wild-type E-cadherin. In all cases where adhesion activity was observed, it was inhibitable by the ␣E␤7 blocking antibody M290.
Adhesion assays were also conducted in which a minimum period of time (six h) was allowed for the L cell transfectants to attach to the assay plate before the test. The purpose was to minimize potential redistribution of mutant E-cadherin to the basolateral surfaces of adjacent cells. Furthermore, the cells were plated at lower density to achieve a sparse distribution and thus enhance accessibility of mutant E-cadherin to the integrin. Under these assay conditions, the mutant E31A still failed to support adhesion to MTC-1 cells (Fig. 6B).
From these results, we conclude that Glu-31 in the BC loop of E-cadherin domain EC1 is critical for binding ␣E␤7 and that, unlike the situation with other IgSF integrin ligands, the CD loop of E-cadherin, where Asp-44 is located, does not play a role.
The E31A Substitution Does Not Affect the Cell Surface Distribution of E-cadherin-We investigated the cell surface distribution of mutant E-cadherin carrying the E31A substitution in a monolayer of L cell transfectants, similar to that used in the adhesion assay, to be certain that the mutant protein was exposed on the upper surface of the cells. L cell transfectants expressing wild-type E-cadherin or E31A mutant were grown on coverslips for 24 h to a confluent monolayer and then stained by immunofluorescence, without fixation, for E-cadherin. The labeled cells were observed by confocal microscopy to determine the cell surface localization of E-cadherin. In both transfectants, there was a strong staining pattern on the cell boundaries and weaker but consistent staining on the upper surface of the cells (Fig. 7).

Attempt to Inhibit the Interaction Between E-cadherin and
Integrin ␣E␤7 with Synthetic Peptides-A linear peptide, NRD-KETKV, which represents the sequence of amino acids 27-34 of E-cadherin domain EC1 and spans the entire BC loop, was tested for inhibitory activity in the adhesion assay (Fig. 8). A scrambled sequence, RENKKTDV, served as a control. The epithelial cell line CMT93 was used in these adhesion tests instead of E-cadherin transfectants because these cells show lower background levels of adhesion to MTC-1 cells (17). At 2.0 mg/ml, the NRDKETKV peptide had a small but significant effect, causing approximately 35% inhibition, whereas the control peptide had no activity. We also tested the shorter, cyclic peptide CDKETKC together with CDKKTEC as a control. It had no effect on adhesion (not shown). In an attempt to identify other sites in domain EC1 that may be involved in the interaction with the integrin, a panel of 22 overlapping peptides, each 15 amino acids long and spanning the entire domain EC1 of E-cadherin, was screened for inhibitory activity. None had any significant effect (not shown).

DISCUSSION
The present report describes mutagenesis studies involving domain deletions and mutation of specific amino acids in Ecadherin with the objective of identifying sites recognized by the mucosal integrin ␣E␤7. Similar structure/function analyses on other members of the IgSF, ICAMs 1, 2, and 3 (31-34), VCAM-1 (35)(36)(37)(38), and MAdCAM-1 (39,40), which function as ligands for integrins, have highlighted the importance of domain 1 in integrin recognition. Furthermore, the structural motif, (G/Q)-(I/L)-(E/D)-(T/S)-(P/S)-(L/X), on the CD loop of the Ig structure has emerged as a common theme in the IgSF/ integrin interactions, the glutamic or aspartic acid playing a particularly important role. In ICAMs 1 and 3, a distant residue in FG loop also contributes to integrin binding (32)(33)(34), whereas in VCAM-1, distant residues in FG and EF loops contribute to the interaction (35)(36)(37)(38). Therefore, the CFG face of the Ig structure appears to form the surface for interaction with integrins (29). Our results emphasize the importance of domain 1 of E-cadherin in recognition by ␣E␤7 since deletion of this domain prevented binding of the blocking antibody ECCD-2 and abrogated adhesion mediated by the integrin. We have identified one amino acid in domain 1, Glu-31, as being critical for integrin recognition. In contrast to results with other IgSF members, this important residue resides in the highly flexible BC quasi-helix, spanned by the sequence RD-KETKV, not in the CD loop. We consider the E31A substitution unlikely to have had any distant effects on the conformation of the molecule. Mutation of five other Glu and Asp residues with exposed side chains in the loops or quasi-helices connecting strands CD, DE, and FG, or exposed on the surface of strands D and G, had no effect on integrin binding. We chose not to mutate glutamic or aspartic acids in the loops joining strands AB and EF or in the C-terminal region of domain 1 because their carboxylate side chains are required for the coordination of Ca 2ϩ in the interdomain junction (14). Disruption of divalent cation binding here would be expected to destabilize the junction between domains 1 and 2.
Linear and cyclic peptides mimicking the sequence of the BC loop of E-cadherin had little or no effect on integrin binding. In contrast, synthetic peptides have successfully been used to inhibit binding of ␣4␤1 to VCAM-1 (35,41), LFA-1 to ICAM-2 (33), and ␣4␤7 to MAdCAM-1 (40). The poor inhibitory effects of our peptides may have been due to their failure, even after cyclization, to mimic the helical conformation of the BC loop. Alternatively, there may be other, more complex, structural requirements for integrin binding that cannot easily be simulated by peptides.
The involvement of the BC quasi-helix in integrin binding must be considered in relation to the structural requirements for homophilic E-cadherin interactions. Recent structural studies on this domain (13) have shown that the motif HAV, which is crucial for the homophilic interaction (8), is located in the F-strand, and studies on N-cadherin show that residues in the C strand and in loops CD, DE, and FG are also required (15). Thus, the homophilic adhesive interaction engages the CFG face of the molecule. Our data suggest that this is not the region binding to the integrin. It is also notable that amino acids in the BC quasi-helix are not directly involved in dimerization of domain 1, which is thought to be a prerequisite for the homophilic interaction (14,15). Although homophilic adhe-sion and integrin binding are not likely to occur simultaneously due to steric considerations, it may be important that the two types of interaction engage separate sites on the molecule; this would be especially significant if homophilic and heterophilic adhesion initiate distinct signaling events in the cell.
We have shown that deletion of domain 2 of E-cadherin also abolishes integrin recognition. Recent crystallographic studies (14) have shown that the junction between domains 1 and 2 of E-cadherin participates in the dimerization of the molecule. The three Ca 2ϩ ions located in this interdomain junction play a crucial role in the dimerization process and are coordinated by amino acid residues from both domains 1 and 2. Residues Asp-134, Asp-136, Asn-143, and Asp-195 in domain 2 are particularly important. It is likely that our ⌬2 mutant, in which domain 1 was joined to domain 3, failed to coordinate the interdomain Ca 2ϩ ions correctly and, as a result, the alignment of domain 1 was abnormal. If so, dimerization of E-cadherin in the manner predicted by Nagar et al. (14) would not occur, and the molecule could be expected to be inactive. Calcium binding has only small effects on the conformation of domain 1 itself (13). Whether misalignment at the domain junction alone explains the failure of the ⌬2 mutant to support integrin binding is not known. Our data does, however, parallel observations with VCAM-1 (38) and MAdCAM-1 (39) in that the presence of both domains 1 and 2 of the molecules was shown to be required for integrin binding although critical residues were found to reside in domain 1 only. In these studies, domain 2 was considered to play a secondary role either by contributing additional contact sites to stabilize the interaction or by conferring structural stability to domain 1.
We have shown that deletion of the cytoplasmic region of E-cadherin, which is involved in binding the actin cytoskeleton, does not diminish its ability to bind integrin. Similar studies with ICAM-1 (32) and VCAM-1 (38) have shown that molecules with truncated cytoplasmic domains can still mediate integrin binding. In contrast, the formation of adherens junctions through the homophilic interaction of E-cadherin is dependent on the cytoplasmic domain being intact (11).
A comparison of E-cadherin sequences among vertebrate species (Table I) shows that Glu-31 is highly conserved. The ␣E␤7 integrin is also likely to be widely expressed, as a similar molecule has been described in birds (42). The E-cadherin/ ␣E␤7 interaction may, therefore, serve a fundamental function in vertebrate mucosal immunity. A comparison of sequences from different members of the cadherin family in the mouse/rat (Table 1) shows that Glu-31 is unique to E-cadherin. Among the various mouse cadherins, P-cadherin is the most homologous to E-cadherin, sharing 58% homology in their amino acid sequences (43). We have, in fact, tested L cell transfectants ex-  pressing mouse P-cadherin in our adhesion assay and found them unable to support ␣E␤7 binding (data not shown). The ability to bind ␣E␤7, therefore, appears to be restricted to E-cadherin and is not shared by other members of the cadherin family.
The biological significance of many of the interactions between integrins on leukocytes and members of the IgSF is now understood. The two members of the ␤7 family of integrins, ␣4␤7 and ␣E␤7, have particular significance for the mucosal immune system. Binding of ␣4␤7 to MAdCAM-1 expressed on HEV of gut-associated lymphoid organs is important for lymphocyte homing to mucosal tissue (44). The purpose of the interaction between ␣E␤7 on mucosal T cells and E-cadherin on mucosal epithelial cells has been a matter of speculation for some time, but it is now clear that this adhesion system is not involved in lymphocyte homing to mucosal sites (45) and is not required for entry of T cells into mucosal epithelia. Engagement of ␣E␤7 expressed by intraepithelial T cells can, however, impart costimulatory signals to the lymphocytes, and it is also necessary for their cytotoxic effector function against mucosal epithelial cells (46,47). Since the E-cadherin/catenin complex can associate with a number of cytoplasmic proteins, some of which take part in signal transduction pathways (12,48), Ecadherin is now thought to be involved in a two-way dialogue between extracellular adhesive interactions and the cytoskeleton/intracellular signaling pathways. Engagement of E-cadherin by ␣E␤7 may, therefore, initiate signaling pathways in epithelial cells that are significant in the immunological surveillance of mucosal epithelia.