Abnormal expression of integrin alpha 6 beta 4 in cervical intraepithelial neoplasia.

We have used subunit-specific monoclonal antibodies (MAbs) and immunohistochemistry to examine the distribution of integrin alpha 6 beta 4 in normal ectocervical epithelium and various grades of cervical intraepithelial neoplasia (CIN). Antibodies were first characterised by immunoprecipitation from two surface-labelled tumour cell lines. Monoclonal antibody G71 was found to precipitate integrin beta 4 from BeWo but not T47D cells, while other anti-beta 4 antibodies precipitated beta 4 from both cell lines. Both G71 and an antiserum to the C-terminal peptide of beta 4 precipitated free beta 4 from surface-iodinated BeWo cells. Neither antibody recognised truncated beta 4 chains observed at approximately 160 kDa. These data suggest that different isoforms of beta 4 are expressed in different tumour cell lines, and that there may be a pool of beta 4 at the cell surface that is not complexed to alpha 6. In normal cervix, both the alpha 6 and beta 4 subunits occur at the basal surface of the basal cell layer. In CIN, the distribution is markedly altered, with strong expression of alpha 6 and beta 4 in the upper cell layers of the ectocervical epithelium. All 40 cases of CIN that were studied exhibited this alteration. Furthermore, the extent of extrabasal staining appeared to correspond with the grade of CIN. The form of integrin beta 4 recognised by antibody G71 also appears in the upper cell layers in CIN, but it shows a more restricted distribution than the normal isoform.

Integrins are a family of heterodimeric (ax,B) cell-surface receptors involved in cell-matrix and cell-cell interactions (Hynes, 1992). Integrin a6,B4 has been shown to be expressed by many epithelial cells, usually at the basal cell surface at the site of adhesion to the basement membrane (Sonnenberg and Linders, 1990). Direct evidence for the involvement of oc6,B4 in cell-basement membrane interaction has come from tissues containing hemidesmosomes, in which it has been demonstrated that a6,B4 is specifically localised within these anchorage structures (Stepp et al., 1990;Sonnenberg et al., 1991;Behzad et al., 1995). There is evidence that laminin functions as a ligand for a6,B4 (De Luca et al., 1990;Lotz et al., 1990;Lee et al., 1992;Niessen et al., 1994;Aplin and Church, 1995). Mutations of the (34 subunit have been observed in junctional epidermolysis bullosa, where there is loss of dermal-epidermal adhesion (Phillips et al., 1994;Vidal et al., 1995).
Cell surface molecules that mediate cell-cell or cellmatrix adhesive interactions may be altered qualitatively or quantitatively in carcinoma. Such alterations during neoplastic transformation (Giancotti and Ruoslahti, 1990;Symington, 1990;Plantefaber and Hynes, 1989;Risinger et al., 1994;Tidman et al., 1990) are coupled with the disruption of basement membrane integrity and may occur as a prerequisite of invasion into the underlying stroma (Liotta et al., 1991;Frixen et al., 1991). Some of these changes are likely to be detectable in preinvasive phases. Knowledge of the altered adhesive properties of transformed cells also offers increased insight into the natural history of the disease (Liotta et al., 1991;Giancotti and Ruoslahti, 1990;Symington, 1990;Plantefaber and Hynes, 1989).
Invasive cervical cancer is preceded by a variable period of intraepithelial neoplasia (CIN) thus providing an opportunity to study neoplastic transformation in the preinvasive phase. Here we describe the results of our investigation into the behaviour of integrin a6(34 in CIN, where altered adhesive properties are likely to be important in the development of invasive cells.

Materials and methods
Cervical biopsies Cervical biopsies (n = 40) were selected from areas showing abnormalities according to the standard colposcopic criteria of the presence of acetowhite, iodine-negative lesions and vascular abnormalities. Appropriate local ethical permission was obtained. The biopsies were first washed in phosphatebuffered saline (PBS), oriented correctly, then snap frozen onto cryostat stubs in liquid nitrogen, using Optimal Cutting Temperature Compound (OCT; Orme Scientific). They were sectioned at right angles to the epithelial/stromal junction at 7 gm onto precleaned microscope slides (Taab) on a Reichart Jung E cryostat and stored at -80°C until required. Normal cervical tissues were obtained at hysterectomy from women reported as having recent normal cytology. Initial histopathological assessment indicating normality or the presence of CIN and its grade was made on serial sections. The diagnosis was later confirmed independently by another histopathologist who screened the entire series.

Antibodies
Mouse monoclonal antibody (MAb) 5B5 to integrin #4 was raised against amnion epithelial cells and had been previously characterised by immunoprecipitation from ,B4-positive and -negative cell lines (Sonnenberg et al., 1991;Aplin et al., 1992). Mouse MAb G71 to integrin (4 was raised against epithelial cells obtained from endometrial tissue (Aplin and Seif, 1985;Aplin et al., 1992). Rat MAb GoH3 to integrin a6 (Sonnenberg and Linders, 1990)  Rabbit antiserum to integrin (34 was raised to the synthetic peptide TLSTHMDQQFFQTC based on the cytoplasmic carboxy terminal sequence of the molecule. This was conjugated to rabbit serum albumin and used in repeated subcutaneous injections. The serum was characterised by ELISA on the peptide conjugate and Western blotting on the intact subunit. It was then immunoaffinity purified for use in immunoprecipitation. Correspondence: JD  Immunoprecipitation analysis Human choriocarcinoma (BeWo) and breast carcinoma (T47D) cells have been shown to express integrin subunits on their cell surface and were used as controls. They were grown in a 1: 1 mixture of Ham's F12 and Dulbecco's modified Eagle medium with 10% fetal calf serum (FCS), Hepes, 2 mM L-glutamine, gentamycin and streptamycin. BeWo human choriocarcinoma cells were grown to confluency in Dulbecco's modified Eagle medium supplemented with 10% FCS, glutamine and antibiotics. Cells were surface labelled with 1251 (1 mCi per 10-7 cells) using lactoperoxidase/ hydrogen peroxide (Aplin et al., 1992) and lysed with 2% Triton X100, 5 mg ml-' bovine serum albumin (BSA), 10 mg ml-' leupeptin and 2 mM phenylmethylsulphonyl fluoride (PMSF) in PBS ABC (1 ml 10-7 cells). The resulting lysate was precleared with protein A-Sepharose (Sigma), preloaded with a similar Triton extract made using nonradioactive cells. The supernatant was divided into aliquots for immunoprecipitation; to 25 ml of antibody (GoH3, 5B5 and G71), 150 ml of iodinated cell extract was added in a total volume of 200 ml in PBS ABC and the reaction mixture incubated on ice for 1 h with periodic mixing. Immune complexes were collected with protein A-Sepharose preblocked with cold cell extract and preloaded with antimouse immunoglobulin (Dako) at a concentration of 1 mg IgG per ml of packed beads. This bead preparation, which is loaded with anti-mouse IgG to only a fraction of its total binding capacity, was used for both mouse and rabbit primary antibodies; the latter, used as a polyclonal preparation, binds avidly to unoccupied protein A. The beads were washed six times with 1% Triton XI00 in PBS ABC followed by a final wash in PBS ABC and boiled in gel loading buffer for 10 min. Immunoprecipitates were analysed by SDS-PAGE on 5% gels followed by autoradiography.

Immunohistochemistry
Stored, frozen sections of cervical tissue were brought to room temperature and fixed in cold acetone for 10 min, followed by washing in PBS. Endogenous peroxidase activity was blocked for 60 min at 37°C (Andrew and Jasani, 1987). Sections were then washed in PBS. All antibody incubations were carried out for 60 min at room temperature, with three 5 min washes in PBS between stages. Primary antibody was used at 1/25 (v/v) in PBS. Secondary antibody was biotinylated rabbit anti-mouse (Dako) used at 1/300 (v/v). A drop of avidin in complex with biotinylated peroxidase (ABComplex/HRP; Dako) was then placed on the sections and left for 30 min, rinsed off and washed as before. Bound antibody was visualised by incubation in 3,3'-diaminobenzidine (DAB) for 5-10 min (200 mg of DAB dissolved in 400 ml PBS, filtered, with the addition of 60 ml hydrogen peroxide). After rinsing in running tap water, sections were counterstained in Harris's haematoxylin (Sigma), dehydrated and mounted in Hystomount (Taab). Appropriate controls were included in each run.
Sections were graded for the extent of extrabasal staining. H & E-stained consecutive sections were analysed and graded independently for the extent of CIN by a histopathologist.

Results
Characterisation of antibodies by immunoprecipitation Two carcinoma cell lines which express integrin a6/4 -BeWo (Aplin et al., 1992) and T47D (Sonnenberg and Linders, 1990) -were selected for the characterisation of subunitspecific monoclonal antibodies to be used in the study. Cell surface iodination was carried out before immunoprecipitation from detergent extracts under conditions expected to preserve the association of integrins into heterodimeric complexes. The anti-integrin a6 monoclonal antibody GoH3 precipitated from T47D cells the a6 subunit (120 kDa under Integrin a6,B4 in CIN JD Aplin et a! a 241 reducing conditions) along with the 200 kDa ,B4 chain with which it is associated (Figure 1, lane 1). In addition, a pair of closely spaced bands was visible at approximately 160 kDa; these are probably truncated forms of integrin ,B4 also found in complex with ax6 (Falcioni et al., 1988;Kennel et al., 1989; (Figure 1, lane 2), though somewhat lower a6 subunit signal intensity was detected. This suggested that, as previously noted (Sonnenberg and Linders, 1990), a fraction of the pool of integrin /4 at the surface of T47D cells may be unassociated with a6, the anti-#4 precipitate thus containing a lower abundance of a6. Antibody 5B5 to integrin /34 gave a result identical to that obtained with 439-9B (Figure 1, lane 4) as previously reported (Sonnenberg et al., 1991). We were unable to precipitate any polypeptide from T47D cells with antibody G71 (Figure 1, lane 6).
BeWo cells were used to investigate further the properties of G71. Very similar results were obtained when antibodies GoH3, 439-9B and 5B5 were used in immunoprecipitation from BeWo ( Figure 2) and T47D (Figure 1) cells. Like T47D, BeWo cells expressed an a6,B4 complex that could be immunoprecipitated with either anti-a6 (GoH3: Figure 2 (Figure 2, lane 1). To confirm that G71 was indeed recognising the full length /4 chain, we compared its behaviour in immunoprecipitation from BeWo cells with that of a polyclonal antibody raised to an oligopeptide based on the published C-terminal sequence (Hogervorst et al., 1990;Suzuki and Naitoh, 1990;Tamura et al., 1990). The anti-peptide serum immunoprecipitated the 200 kDa /34 chain (Figure 3, lane 5), which comigrated precisely with the chain precipitated by G71 (Figure 3, lane 1). The anti-peptide serum also precipitated a doublet of bands at approximately 160 kDa (Figure 3, lane 5). These must be assumed to be Nterminally truncated forms of /4, also reported by Giancotti et al. (1992). The bands are substantially weaker than those seen in this molecular weight range when 439-9B (Figure 3 Giancotti et al., 1992); these would not be recognised by the peptide antiserum. Neither G71 nor the peptide antiserum precipitated detectable quantities of the integrin a6 chain. This suggests that free ,B4 is present at the BeWo cell surface, and that both these antibodies preferentially recognise it.
The distribution of integrin oc6,B4 in cervical tissue Monoclonal antibodies GoH3, G71 and 5B5 were used to examine the distribution of integrin a6,B4 in normal and neoplastic cervix. The expression of the ,B4 integrin subunit in normal stratified squamous epithelium was monitored by antibody 5B5. Expression was strongest at the basal aspect of the basal cells (Figure 4a and b), with an accompanying weaker pericellular and diffuse staining pattern in the basal and parabasal layers (Figure 4b). The a6 subunit gave a similar distribution (Figure 4c) as monitored by staining with GoH3.
In cases of CIN (Figure 4d and e), staining with 5B5 and GoH3 was seen on the surface of neoplastic cells which displayed nuclear abnormalities, including an increased nucleocytoplasmic ratio, nuclear hyperchromatism and the presence of abnormal mitotic figures. In CIN I, in which abnormal cells occupy the lower one-third of the epithelium, staining with 5B5 extended beyond the basal region with strong pericellular staining of the abnormal cells. In CIN II, as the abnormal cells extend further through the epithelium to occupy up to two-thirds of its thickness, staining with 5B5 and GoH3 followed suit. In CIN III, pericellular staining was observed throughout the full thickness of epithelium though there was some variation in its intensity (Figure 4d and e). Altogether 40 cases of CIN were studied, and all of them conformed to this pattern of staining.
Antibody G71 showed similar behaviour, with predominantly basal staining in the normal ectocervical epithelium (Figure 5a). This antibody did not stain the lateral plasma membranes or cytoplasmic regions of basal or parabasal cells. The epitope was also present in the walls of small blood vessels in the upper dermis (Figure 5a and b). G71 staining was increasingly visible in the upper layers of the epithelium with increasing grade of CIN ( Figure Sb). Although G71positive cells could be observed throughout the epithelium in CIN III, extrabasal staining was more heterogeneous and less extensive than observed with 5B5 or GoH3, with G71negative cells always present in the lesion.

Discussion
Cervical cancer is one of the major health care issues affecting women in the UK despite the proven success of screening programmes for the preinvasive lesions (CIN) (Anderson et al., 1988). The mechanisms that are involved in the progression of intraepithelial neoplasia from in situ to invasive phenotype remain poorly understood, but include changes in cell adhesion, motility and proteolytic activity leading to altered stability of the basement membrane associated with penetration into the sublaminal matrix (Liotta et al., 1991).
All available evidence indicates that the integrin /4 subunit forms heterodimeric complexes uniquely with integrin a6 (Carter et al., 1990;De Luca et al., 1990;Lotz et al., 1990;Sonnenberg and Linders, 1990;Sonnenberg et al., 1991;Lee et al., 1992). However, in addition to its heterodimeric form, integrin /34 has been suggested to exist at the cell surface in a form uncomplexed to a chain (Sonnenberg and Linders, 1990). Our data demonstrating that anti-/34 antibodies 5B5 and 439-9B precipitate relatively more labelled /34 and less a6 than does antibody GoH3 to the x6 chain can be adduced in support of this suggestion, as can the data of Hodivala et al. (1994) in cultured keratinocytes.
In normal cervix we have detected the a6 and #4 subunits strongly at the basal cell surface, but also more weakly in the cytoplasm and lateral plasma membrane domain of both basal and parabasal cells. This distribution was also reported by Carico et al. (1993) but it differs slightly from other reports in which a basal-only pattern has been described (Sonnenberg and Linders, 1990;Lee et al., 1992;Hughes et al., 1994). The distinction is probably a result of the combination of a high-affinity antibody with a highly sensitive staining protocol, allowing the detection of smaller quantities of antigen. All authors agree that the major location of the a6,B4 complex is at the basal cell surface. In comea (Stepp et al., 1990), skin (Sonnenberg et al., 1991 and amnion (Behzad et al., 1995) it is known to be concentrated specifically in hemidesmosomal junctions, where it is assumed to have a role in anchoring the cell surface to the extracellular matrix. Other normal epithelial cells that lack hemidesmosomes also express a6,B4 basally (Sonnenberg and Linders, 1990;Aplin, 1993 (1994) reported similar findings in the context of a larger panel of integrins. Our data are also consistent with previous work indicating elevated levels of the ,B4 subunit in murine carcinoma (Falcioni et al., 1988;Kennel et al., 1989;Van Waes et al., 1991), and with our previous observation of increased expression in extrabasal locations in squamous cell carcinoma of the skin (Tidman et al., 1990). In contrast, Hodivala et al. (1994) reported a reduced level of integrin a6#4 expression in HPV16/Ha-ras-transformed keratinocytes and in a small group of HPV-positive CIN lesions. In this context it will be of interest to correlate further the behaviour of integrin a6#4 with HPV status in CIN.
The onset of proliferative intraepithelial change implies altered tissue kinetics with more rapid escape of neoplastic cells from the basal layers of the tissue as well as loss of Integrin a6#4 in CIN JD Aplin et al Figure 5 Immunoperoxidase localisation of the G71 epitope in normal cervix (a) and CIN III (b). The basal localisation observed in the normal tissue is largely consistent with that observed using other anti-,B4 antibodies, although no lateral staining is observed. Note also that the epitope is detected in small blood vessels in the superficial dermis. In CIN III, staining in the basal layer is somewhat weaker, but pericellular reactivity appears in upper cell layers within the lesion with considerable variation in reactivity between individual cells. Magnification: x 495.

differentiated cells from intermediate and superficial layers.
This is likely to be accompanied by altered cell polarisation and architecture, intercellular and cell-basement membrane adhesive interactions (Liotta et al., 1991). The presence of /4containing cells in extrabasal layers of the neoplastic epithelium could imply a more rapid escape from the basal layer of cells bearing the residual hallmarks of the basal cell phenotype, or alternatively it may be that integrin ,B4 is capable of a function in intercellular organisation other than adhesion to basement membrane. Loss of hemidesmosomes is a prerequisite of cell migration, either in wound healing (Kurpakus, 1991) or tumour invasion (McNutt, 1976).
During trophoblast invasion loss of a6#4 occurs from the migrating cell population soon after the loss of adhesion to the villous basement membrane (Aplin, 1993); this contrasts with the persistence of a6#4 in cervical neoplastic cells in situ as well as in invasive foci (Carico et al., 1993;Hughes et al., 1994).
The properties of monoclonal antibodies 5B5 and 439-9B show clearly that they recognise the extracellular domain of integrin /4 (Sonnenberg et al., 1991) and immunoprecipitate a6/34 complexes. In addition to the full length polypeptide, these antibodies capture truncated forms of the ,B4 chain present in both cell lines. The properties of these isoforms are similar to those described previously for C-terminally truncated /34 chains (Falcioni et al., 1988;Kennel et al., 1989;Van Waes et al., 1991;Giancotti et al., 1992). These can also be captured using antibody to a6, confirming their ability to complex stably with a /4 chain. We cannot exclude that some proteolysis of /34 occurred during our experiments since calcium ions were present, and Giancotti et al. (1992) have shown that the cytoplasmic domain of the subunit is sensitive to a calcium-dependent protease, calpain. Their evidence suggests that truncated forms of ,B4 are present along with the full length chain in vivo. The immunoprecipitation analysis reported here, in which we have demonstrated that G71 recognises a 200 kDa chain in BeWo cells with precisely the same electrophoretic properties as one recognised by a polyclonal antibody to the C-terminus of human integrin /34, confirms previous evidence (Aplin et al., 1992) that G71 recognises the integrin ,B4 subunit. In contrast, no subunit was precipitated by G71 from T47D cells in several different experiments. This suggests that different /4 molecular isoforms may be present in different cell types, not all of which are recognised by G71. The G71 epitope is in the extracellular domain of /4 as monitored by light and electron microscopic immunolocalisation in normal amnion epithelial cells (Aplin and Seif, 1985; JD Aplin and DR Garrod, unpublished results). G71 recognises the full length /34 subunit in preference to shorter variants suggesting that its binding site may be near the amino terminus of the molecule and therefore lost during N-terminal proteolytic truncation.
It was therefore of interest to compare the distribution of the G71 and 5B5 epitopes immunohistochemically. Differential expression of /4 isoforms may reflect the heterogeneity of neoplastic cell behaviour and could be relevant to pathophysiology of different phenotypes. The greater heterogeneity of binding of G71 to neoplastic cells suggests that a varying degree of proteolytic or other modification may occur after escape from the basement membrane.
We have shown that a6#4 integrin displays an abnormal distribution with remarkable consistency and in a fashion that reflects the extent and grade of CIN. The appearance of a6#4 in association with neoplastic cells suggests the possibility that it may also occur in detectable quantities in cervical smears. Our data support the view that the analysis of molecular changes occurring in preinvasive conditions offers the hope of improved basic understanding of the disease process as well as the possibility of novel, and perhaps more convenient, approaches to diagnosis.