Epithelial tight junction proteins as potential antibody targets for pancarcinoma therapy

Abstract

Recombinant monoclonal antibodies are beginning to revolutionize cancer therapy. In combination with standard chemotherapy, high response rates have been reported with antibodies of the human IgG1 isotype for treatment of non-Hodgkin’s lymphoma and breast cancer. It is becoming apparent that targets for antibody-based therapies do not necessarily need to be absent from normal tissues but can be present there either in low copy numbers or with binding epitopes shielded from the therapeutic antibody. Here, we studied whether claudin proteins that form tight junctions in normal epithelia are still expressed on carcinoma cells and whether their extracellular domains can be recognized by antibodies. We show that mRNAs of claudins 1, 3, 4, and 7 are all expressed in different human carcinoma cell lines, while claudin 8 was selectively expressed in breast and pancreas cancer lines. Chicken polyclonal antibodies were raised against peptides contained within predicted extracellular domains of claudins 1, 3, and 4. Affinity-purified IgG fractions for claudins 3 and 4 were monospecific and bound to human breast and colon carcinoma lines, but not to a line of monocytic origin. Claudin 3 antibodies also homogeneously stained human renal cell carcinoma tissue and micrometastatic tumor cells as identified by cytokeratin staining in bone marrow biopsies of breast cancer patients. Fluorescence-activated cell sorting and immunocytochemistry indicated that claudin antibodies bound to the surface of tumor cells. By analogy to other tumor-associated antigens that are differentially accessible to antibodies on tumor vs normal tissue, we propose that certain claudin proteins have potential as targets for novel antibody-based therapies of carcinomas.

Introduction

Antibody-based therapies for cancer hold the promise of higher specificity and lower side effect profile than conventional anti-proliferative and genotoxic small-molecule drugs. These possible advantages rely on an improved distinction between normal and malignant cells by antibodies and the employment of less toxic immunological anti-tumor mechanisms, such as recruitment of cytotoxic immune cells and activation of complement.

Targets for antibody-based therapies need to have particular qualities that do not solely rely on a numeric overexpression of target molecules on tumor cells. While a high-level overexpression of the human epidermal growth factor receptor type 2 (HER-2) as a result of gene amplification is the basis for the therapeutic window and low side effects of the immunoglobulin G1 (IgG1) antibody trastuzumab (Herceptin) [17], other targets for antibodies on the market and in clinical development for tumor therapy have distinct qualities. In the case of human IgG1 antibodies to CD20 (rituximab), CD22 (epratuzumab), and CD52 (Campath-1H), antibody targets are equally well expressed on tumor cells and normal lymphocytes [8, 16, 20, 51]. Here, the ablation of normal cells by the antibody is tolerable and normal lymphocyte levels are eventually restored by new cells derived from target-negative stem cells. In the case of antibodies to a peptide repeat epitope in the backbone of the proteoglycan mucin-1, the target is selectively exposed on tumor cells as a result of chronic underglycosylation of mucin-1 [2].

Other examples of differential accessibility of antibody targets are carcinoembryonal antigen (CEA/CEACAM5/CD66e) and carboanhydrase IX (CA IX/CA9). Although both antigens are expressed on normal epithelia of colon and kidney, respectively, radioactively labeled antibodies for imaging do distinguish well between tumor and normal tissue, and cytotoxic IgG1 antibodies are well tolerated [7, 37, 57]. This is most likely due to a restricted expression of CA IX (and CEA) on the luminal side of normal epithelial tissue where antibodies of the IgG format do not normally have access. Also the well-established tumor-associated antigen epithelial cell adhesion molecule (Ep-CAM) (TACSTD1/GA733-2) may be less well accessible on normal tissue compared with tumor tissue [39]. Ep-CAM is thought to serve as a calcium-independent homotypic cell adhesion molecule for epithelial cells and, consequently, is confined to the intercellular space of epithelia [30, 31]. In addition, Ep-CAM expression is up-regulated on tumor cells which may contribute to a therapeutic window [15, 26, 47, 48, 56, 65]. In breast cancer, the expression of Ep-CAM on tumors correlates with a decreased survival [15, 56]. Intriguingly, high-affinity anti-Ep-CAM antibodies are very toxic [25, 32, 52], whereas intermediate-affinity antibodies are well tolerated [53]. This shows general accessibility of the Ep-CAM target on normal cells but also indicates that kinetics of antibody binding may open a therapeutic window.

Based on the above insights, we studied other epithelial cell–specific proteins for their utility as antibody targets. We focused on proteins involved in organizing epithelial tissue architecture, which is typically disrupted in carcinoma tissue. By analogy to Ep-CAM, we assumed that also other epithelial surface proteins may still be advantageous for tumor cells and therefore their expression may be maintained or even up-regulated. Likewise, such proteins may be barely accessible in well-structured epithelia to antibodies but become exposed on tumor cells. One class of proteins that may fulfil such criteria are claudins. Claudins are integral membrane proteins located within the tight junctions (TJs) of epithelia and endothelia. TJs organize a network of interconnected strands of intramembranous particles between adjacent cells. In TJs, occludin and claudins are the most prominent transmembrane protein components (for review see [36, 41, 59, 60]). In a fence/gate function, they create a primary barrier to prevent and control the paracellular transport of solutes and restrict the lateral diffusion of membrane lipids and proteins to maintain cellular polarity [35].

Antibodies against different claudin family members were described by several authors. Morita et al. [43] produced polyclonal rabbit antibodies against the C-terminal cytoplasmic domains of claudins 5 and 6. Synthetic polypeptides derived from the C-terminal cytoplasmic domain of claudin 1 were used for immunization of guinea pigs [14]. Wilcox described a rabbit polyclonal antibody directed against a C-terminal peptide of claudin 14 [64]. A variety of monoclonal and polyclonal claudin-specific antibodies are available from Zymed Laboratories (South San Francisco, CA, USA) all recognizing the intracellular C-terminal part of claudins 1, 2, 3, 4, 5, 15, and 16. Several antibodies were generated in rodents against claudins 1, 2, and 3 [29, 42]. These claudin-specific antibodies were useful for detection of claudin expression in fixed or denatured tumor tissue for diagnostic purposes but not on intact cells in situ or in vivo. Up to now, no claudin-specific antibodies were described that selectively bind to the cell surface of intact tumor cells and tissue as would be required for antibody-based therapeutic approaches.

In this study, we show that certain claudin family members are widely expressed on tumor cell lines of various origin. The level and quality of expression was not unlike that of the validated pancarcinoma target Ep-CAM. Polyclonal antibodies are described that specifically detect claudin family members 3 and 4 on the surface of epithelial tumor cells. In this context, selected claudins will be discussed as potential antibody targets due to their wide-spread expression, their overexpression on tumor cells and presumed differential accessibility to antibodies.

Material and methods

Cell lines

The following human cell lines were investigated for expression of claudins. Sources were the American Type Culture Collection (ATCC, Rockville, MD, USA), the Deutsche Sammlung für Mikroorganismen und Zelllinien (DSMZ, Braunschweig, Germany), and the European Collection of Cell Cultures (ECACC, Salisbury, UK). Cell lines were cultivated as recommended by ATCC, DSMZ, and ECACC, respectively.

Colon carcinoma: Caco-2 (ATCC no. HTB-37), SW480 (CCL-228), SW948 (ATCC no. CCL-237), HT-29 (ATCC no. HTB-38). Mammary carcinoma: BT20 (ATCC no. HTB-19), CAMA-1 (ATCC no. HTB-21), ZR-75-1 (ECACC no. 87012601), T47D (ATCC no. HTB-133), MDA-MB-231 (ATCC no. HTB-26), SK-BR-3 (ATCC no. HTB-30). Ovarian carcinoma: EFO-21 (DSMZ no. ACC 235), EFO-27 (DSMZ no. ACC 191). Prostate carcinoma: LNCaP (ATCC no. CRL-1740), DU145 (ATCC no. HTB-81), PC-3 (ATCC no. CRL-1435). Kidney carcinoma: CAL-54 (DSMZ no. ACC 365), Caki-2 (ATCC no. HTB-47). Lung carcinoma: A-427 (DSMZ no. ACC 234), SCLC-21H (DSMZ no. ACC 372), LCLC-97TM1 (DSMZ no. ACC388). Pancreatic carcinoma: PancTu [3]. Gastric carcinoma: KATO III (ATCC no. HTB-103). Cervix carcinoma: HeLa (DSMZ no. ACC 57). Epidermal tumor: A431 (ATCC no. CRL-1555). Hepatocellular carcinoma: Hep-G2 (ATCC no. HB-8065). Melanoma: MelJuso (DSMZ no. ACC 74). Erythroblast: TF1 (ATCC no. CRL-2003). Fibrosarcoma: HT-1080 (ATCC no. CCL-121). Leukemia: Jurkat (DSMZ no. ACC 282). Lymphoma: Daudi (ATCC no. CCL-213), U937 (ATCC no. CRL-1593.2).

Gene expression analysis

RNA was prepared from different cell lines using the RNeasy MiniKit (Qiagen, Hilden, Germany). For preparation of cDNA, 2.5 μl p(dT)15-primer (concentration 0.5 mg/ml; Roche Applied Science, Mannheim, Germany), 10 μg RNA, and H₂O (total volume 22.5 μl) were incubated for 10 min at 70°C. The mixture was incubated with 10 μl 5x First Strand Buffer (Invitrogen/Gibco, Karlsruhe, Germany), 5 μl 100 mM DTT (BD Biosciences/Clontech, Palo Alto, CA, USA), and 2.5 μl deoxynucleoside-triphosphates (dNTPs) (each 10 mM, MBI Fermentas, St Leon-Rot, Germany) for 2 min at 42°C. cDNA synthesis was completed by adding 1 μl Superscript II (reverse transcriptase, 200 U; Invitrogen/Gibco, Karlsruhe, Germany) for 50 min at 42°C.

Sequences coding for claudins were amplified using the following primer pairs: 5′ primer CLDN1back1: 5′-TAATAAAGCTTCCCGAGCGAGTCATGGC-3′ and 3′ primer CLDN1for1: 5′-TAATATCTAGACACGTAGTCTTTCCCGCTG-3′; 5′ primer CLDN3back1: 5′-TAATAAAGCTTGCAGCCATGTCCATGGGC-3′ and 3′ primer CLDN3for1: 5′-TAATATCTAGAGACGTAGTCCTTGCGGTC-3′; 5′ primer CLDN4back1: 5′-TAATAAAGCTTGGACGCTGAACAATGGCC-3′ and 3′ primer CLDN4for1: 5′-TAATATCTAGACACGTAGTTGCTGGCAGC-3′; 5′ primer CLDN7back1: 5′-TAATAAAGCTTCTGAGGGCGGAAATGGCC-3′ and 3′ primer CLDN7for1: 5′-TAATATCTAGACACATACTCCTTGGAAGAG-3′; 5′ primer CLDN8back1: 5′-ATAATAAGCTTTCCCAGAGGATAATGGCAAC-3′ and 3′ primer CLDN8for1: 5′-TATTATCTAGACACATACTGACTTCTGGAGT-3′. As control, the Ep-CAM–encoding sequence was amplified by polymerase chain reaction (PCR) using the following primers: 3′ primer Ep-CAM1for: 5′-GCGGCCGCTTATGCATTGAGTTCCCTATGC-3′ and 5′ primer EpCAM1back: 5′-CCAGTGTGGTGGCGCCACCATGGCGCCCCCGCAGG-3′.

Polymerase chain reaction amplification of the cDNAs was performed with 10 mM Tris-HCl (pH 8.3), 50 mM KCl, 1.5 mM MgCl2, 0.001% gelatin, 0.5 μm of each primer (Metabion, Planegg, Germany), ±1 M betaine, 0.2 mM dNTP Mix (MBI Fermentas, St Leon-Rot, Germany), and 0.75 U Taq DNA Polymerase (Sigma-Aldrich, Munich, Germany). Forty cycles (95°C for 30 s; 50°C for 30 s; and 72°C for 30 s) were used for amplification. PCR products were analyzed by electrophoresis on 1.5% agarose gels, and DNA was stained by ethidium bromide.

To confirm specificity of the PCR reactions for certain claudin species, PCR fragments obtained with primers specific for CLDN1, 3, 4, and 7 were excised from the agarose gel and analyzed by nucleotide sequencing (TopLab, Martinsried, and Sequiserve Vaterstetten, Germany).

Sequence alignment and domain prediction

Nucleotide and amino acid sequences specific for human claudins were available in databases. The accession numbers for the aligned claudin proteins are as follows: CLDN1 XM_003151, CLDN7 NM_001307, CLDN2 XM_010309, CLDN14 XM_009753, CLDN3 XM_057967, CLDN4 XM_057966, CLDN6 XM_012518, CLDN9 NM_020982, CLDN5 XM_009839, CLDN8 XM_009721, CLDN17 NM_012131, CLDN10 XM_007076, CLDN15 NM_014343, CLDN11 NM_005602, CLDN18 NM_016369, CLDN12 XM_004591, and CLDN16 NM_006580. The alignments were performed with the BLASTP programs for proteins at the NCBI (http://www.ncbi.nlm.nih.gov/BLAST/) [1]. For the prediction of transmembrane helices, the TMHMM program (http://ca.expasy.org) was used.

Generation of chicken antibodies

For analysis of sequence homologies between the different claudin proteins and identification of claudin-specific sequences, amino acid sequences corresponding to the two predicted extracellular loops were aligned.

Two peptides derived from the first extracellular loop and one peptide derived from the second extracellular loop of claudin 1, 3, and 4 were selected and used for production of antibodies in chickens (see Table 1). In vitro synthesis of peptides and immunization of chickens with keyhole limpet hemocyanine (KLH)-conjugated claudin peptides was performed at Davids Biotechnologie (Regensburg, Germany). Briefly, in vitro synthesis of peptides was performed in an ABI 433A Peptide Synthesizer (Applied Biosystems, Foster City, CA, USA). Peptides were purified via high-performance liquid chromatoraphy (HPLC) to a purity of >70%, as confirmed by mass spectrometry analysis, and conjugated with KLH protein at the N terminus using glutaraldehyde, or at tyrosine residues using diazobisbenzimide (Davids Biotechnologie, Regensburg, Germany).

Table 1.

Peptides derived from the first (1) and second (2) extracellular loops of claudins 1, 3, and 4, as used for coupling to KLH and generation of antibodies in chicken eggs

Name of peptide	Sequence	Extracellular loop
CLDN1.1	RIYSYAGDNIVTAQA	1
CLDN1.2	WMSCVSQSTGQVQCKVFDSLLNLSSTLQATR	1
CLDN1.3	RIVQEFYDPMTPVNARYE	2
CLDN3.1	WRVSAFIGSNIITSQNIWEGLWMNCVVQ	1
CLDN3.2	QSTGQMQCKVYDSL	1
CLDN3.3	ANTIIRDFYNPVVPEAQK	2
CLDN4.1	RVTAFIGSNIVTSQT	1
CLDN4.2	GLWMNCVVQSTGQMQCKVYDSLLALPQD	1
CLDN4.3	HNIIQDFYNPLVASGQKR	2

Three hens were immunized with 100 μg of a mix of all three peptides derived from claudins 1, 3, or 4, respectively. Injections were given in up to three sequential immunization cycles performed intradermally and intramuscularly (Davids Biotechnologie, Regensburg, Germany). Five days after immunization, IgGs were purified from the egg yolk by ammonium sulfate precipitation, and IgG affinity was chromatographed on Fractogel EMD copolymer (Merck, Darmstadt, Germany) coupled with the individual claudin-derived peptides. The resulting monospecific IgG fractions were more than 98% pure. As control, unspecific IgG was purified from nonimmunized chicken egg yolks by ammonium sulfate precipitation.

To test the binding specificity of the generated monospecific IgG fractions from chicken egg yolks, an ELISA analysis was performed. The claudin peptides were coated to an ELISA plate (Maxi Sorb; Nunc, Neerijse, Belgium) with a concentration of 50 ng/50 μl in phosphate-buffered saline (PBS), pH 7.0. The detection was performed with rabbit anti-chicken IgG conjugates with alkaline phosphatase (paranitrophenylphosphate [PNPP]) using diethanolamine as substrate.

FACS analysis and immunocytochemistry

Nonpermeablized SW480 and CAMA-1 cells were used for fluorescent-activated cell sorting (FACS) analysis. U937 cells served as negative control. Flow cytometric analyses were performed according to the instruction manual provided by Becton Dickinson (FACSCalibur User’s System; Becton Dickinson, Heidelberg, Germany). FACS data were recorded and analyzed with Becton Dickinson Cellquest software. Dilutions of affinity-purified antibodies in RPMI, 1% FCS (fetal calf serum), 10% human AB-serum (Biotest, Dreieich, Germany) were as follows: chicken anti-CLDN1.1, 1:3.3; anti-CLDN3.2, 1:2.5; anti-CLDN3.3, 1:3.3; anti-CLDN4.3, 1:3.2; and antibody IgG fraction from egg yolk of a nonimmunized chicken, 1:3. For standard binding analysis, 200,000 cells were incubated with chicken IgG for 60 min at 4°C. Thereafter, cells were washed twice with 100 μl RPMI, 1% FCS buffer. Subsequently, cells were incubated for 60 min at 4°C with a secondary mouse anti-chicken-FITC antibody (Sigma-Aldrich, Munich, Germany) diluted 1:100 in RPMI, 1% FCS, 10% AB-serum. After two washing steps in FACS buffer (PBS, 1% FCS, 0.05% NaN₃), cells were fixed in a 1:20 dilution of 37% formaldehyde. For FACS analysis, 10,000 events were analyzed.

A FACS-based inhibition assay on A431 cells using claudin peptides CLDN1.1, 1.3, and 4.3 in a concentration of 13 μM was performed as described above.

For immunofluorescence staining, SW480 and CAMA-1 cells were incubated for 60 min at 4°C with anti-claudin IgG, washed as described above, and incubated for 60 min at 4°C with Cy3-conjugated rabbit anti-chicken IgG (Jackson ImmunoResearch Laboratories, West Grove, PA, USA). Cell staining was analyzed by immunofluorescence microscopy using a Nikon TE200 inverse microscope, and micrographs were taken.

Immunohistochemistry of tumor samples

Tissue slides were prepared from normal and malignant kidney tissue. The clear cell renal carcinoma was T3b N0 M0G3 stage (kindly provided by the Department of Urology, University of Munich). Tissue samples were frozen in liquid nitrogen, stored at −80°C, and embedded in Tissue Tek O.C.T. Compound (IMEB, San Marcos, CA, USA) before sectioning. Tissue slices of 5 μm were prepared using a Reichert Jung Frigocut 2800 Cryostat (IMEB, San Marcos, CA, USA). Before staining, the slides were air-dried for 24 h and then stored at −20°C.

Before staining, tissue slides were blocked for 20 min with 10% human AB-serum (Biotest, Dreieich, Germany) in PBS (Invitrogen/Gibco, Karlsruhe, Germany). Subsequently, chicken anti-claudin IgG was incubated for 45 min at the dilutions described above. Mouse anti-Ep-CAM antibody (3B10, 1 μg/μl; Micromet, Munich, Germany) was used at a dilution of 1:400 in RPMI 1640, 1% FCS, 10% AB-serum. Control mouse antibody MOPC21 (Sigma-Aldrich, Munich, Germany) was diluted 1:400 in RPMI 1640, 1% FCS, 10% AB-serum. After incubation with the first antibody, slides were washed three times for 3 min in PBS. Subsequently, slides were incubated for 45 min with anti-chicken IgG AP conjugate or anti-mouse IgG AP conjugate (Jackson ImmunoResearch Laboratories, West Grove, PA, USA). For detection of the rabbit anti-claudin 1 antibody (Zymed, South San Francisco, CA, USA), anti-rabbit IgG AP conjugate (Sigma-Aldrich, Munich, Germany) was used. Staining with rabbit anti-CLDN1 antibody specific for the C-terminal part of CLDN1 was performed on air-dried slides from serial sections of frozen tissue samples derived from human ovarian carcinoma. Slides were washed three times for 3 min. Then freshly prepared Fast Blue substrate (Sigma-Aldrich, Munich, Germany) was added to the slides. Fast Blue substrate was freshly prepared as follows: 20 mg naphtol AS-MX-phosphate was dissolved in 2 ml dimethylformamide (solution A). Levamisol (24 mg) and Fast Blue BB salt (100 mg) were solved in 98 ml 0.1 M Tris/HCl, pH 8.2 (solution B). Solutions A and B were mixed in a ratio 1:2, filtered through a 0.2-μm filter, and incubated on the tissue slides for 10–30 min.

Immunohistochemistry of bone marrow aspirates

Bone marrow samples were aspirated from the pelvis of patients with mamma carcinoma. Mononucleated cells were isolated from aspirates by density gradient centrifugation and air-fixed on slides as described [46]. Areas of dried cells were encircled with grease using a DAKO pen (DAKO, Glostrup, Denmark) and preincubated for 10 min with 10% AB-serum (Biotest, Dreieich, Germany) in PBS. Cells were incubated with chicken anti-claudin IgGs CLDN1.1, 3.2, 3.3, and 4.3 in a dilution of 1:6 in 10% AB-serum in PBS for 45 min. As controls, nonimmunized chicken serum, mouse anti-cytokeratin antibody A45/B/B3 (Micromet, Munich, Germany), or mouse MOPC21 antibody (Sigma-Aldrich, Munich, Germany) was used. After washing three times with PBS, cells were incubated in the dark with anti-chicken antibody labeled with Cy3 (Dianova, Freiburg, Germany) in a dilution of 1:80 or anti-mouse IgG FITC (Sigma-Aldrich, Munich, Germany) diluted 1:32 diluted in 10% AB-serum in PBS. All incubation steps were performed in a humidified chamber. After washing three times with PBS, the slides were mounted with Kaiser’s gelatine (Merck, Darmstadt, Germany), and staining was analyzed by immunofluoresence microscopy using a phase contrast Nikon TE200 inverse microscope equipped with an XYZ axis scanning stage (Maerzhaeuser, Wetzlar, Germany). Pictures were taken by a Vosskuehler CCD-1300B camera and processed by the Lucia G 4.70 program (Laboratory Imaging, Prague, Czech Republic).

Results

Claudin 1, 3, 4, 7, and 8 expression in tumor cell lines of various origins

The expression of mRNAs encoding five selected claudin proteins was investigated in a total of 31 different human cell lines derived from tumors of various origins (Fig. 1). Expression data from a coupled reverse transcription RT-PCR are shown for claudins 1, 3, 4, 7, and 8. PCR signals of the expected sizes were observed for all five claudins. The expression of the Ep-CAM was analyzed as control. The quality of RNA preparations was investigated with PCR-detecting β-actin expression (bottom panel). Very comparable β-actin signals were obtained with all samples.

Fig. 1.

Expression of claudin 1, 3, 4, 7, and 8 mRNAs in various cancer cell lines. Products from RT-PCR reactions were separated on agrose gels, and DNA stained by ethidium bromide. Ep-CAM expression was included as epithelial marker, and β-actin expression to document comparability and quality of samples. The arrow heads mark the 500-bp position within the DNA size markers added on left and right most lanes. A reagent control was done in the penultimate lane.

Claudins 1, 3, 4, and 7 all showed a high-level and relatively broad expression in tumor cell lines derived from epithelia (Fig. 1, lanes 1–25). DNA sequences obtained from PCR amplificates were compared with claudin sequences accessible in the NCBI Entrez database and verified as claudins 1, 3, 4, and 7. Claudin 1 showed an extra expression in a melanoma cell line (MelJuso; lane 26), and claudin 7, an extra expression in two cell lines derived from blood-born tumors (TF-1 and Daudi; Fig. 1, lanes 27 and 30). Claudin 1 was significantly expressed in 19/25 cell lines derived from carcinomas of colorectal, breast, ovary, prostate, kidney, lung, pancreas, stomach, cervix, epidermis, and liver origin. Claudin 3 was expressed in 22/25 carcinoma cell lines, the only exception being the breast cancer line MDA-MB-231 (Fig. 1, lane 9). Claudin 4 expression was detectable in all 25 carcinoma lines except for the lung cancer line A-427 and Hep-G2 (Fig. 1, lanes 18 and 25). Claudin 7 was found expressed in virtually all 25 carcinoma lines analyzed. By contrast, expression of claudin 8 was confined to the six breast cancer cell lines (Fig. 1, lanes 5–10) and one pancreas cancer cell line (lane 21).

The expression patterns in Fig. 1 revealed that some of the cell lines tested were negative for certain claudin species. This suggests that the primers used for PCR reactions selectively amplified claudin-specific cDNA sequences rather than contaminating intronless genomic DNA sequences of claudins. This notion is supported when the expression analysis was repeated with newly prepared RNA from the various cell lines. Identical patterns of claudin-specific fragments were observed in both experiments (data not shown) demonstrating highly reproducible mRNA signals for claudins 1, 3, 4, 7, and 8.

Epithelial cell adhesion molecule expression was not only detected in cell lines of carcinoma origin but also found weakly with TF-1 erythroleukemia cells (lane 27). The high frequency of claudin 1, 3, 4, and 7 expression in carcinoma lines would qualify them as pancarcinoma targets. Because of their more limited expression in noncarcinoma lines, we focused in the following on claudins 1, 3, and 4 for the generation of antibodies against putative extracellular domains.

Prediction of extracellular claudin protein domains

To assess the potential of claudins as targets for antibody-based therapies we needed to identify domains of claudins that are extracellular and lend themselves for production of antibodies that discriminate well between the 22 members of the claudin protein family. Each claudin protein was predicted to have four transmembrane segments with two extracellular loops, and N and C termini located in the cytoplasm [27, 49]. Membrane-spanning domains and putative extracellular domains of claudins were reassessed by the program TMHMM for prediction of transmembrane helices in proteins [28]. Figure 2a depicts the presumed arrangement of claudin proteins in the plasma membrane of epithelial cells. The analysis shown in Fig. 2b revealed that the amino acid sequences of the first and fourth transmembrane segments (shown in italics) and the first and second extracellular loops (shown in bold) are highly conserved, whereas the sequences of the second and third transmembrane segments are more divergent. The first extracellular loop is larger (about 50 amino acids) and more hydrophobic than the second extracellular loop (about 20 amino acids), and was proposed to serve as an intercellular zipper structure [12]. The cytoplasmic C termini displayed the greatest variability in primary structure within the claudin protein family (Fig. 2b).

Fig. 2.

a Schematic model for the membrane arrangement and orientation of claudin proteins according to Tsukita [60]. b Primary sequence alignment of claudins 1–18. Italics putative transmembrane domains; bold putative extracellular domains. The numbers in brackets give the amino acid positions of the first amino acid. Underlined amino acids mark highly conserved residues. c Alignment of protein sequences of human and chicken (gallus) claudins 3 and 4. Sequences for human and chicken claudins were obtained from the Protein Database of the National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov). Sequence of chicken claudin 4 is only known in part (amino acid 1–56) which was aligned to human claudin 4. Differing amino acids in human and chicken sequences are boxed, changes to similar amino acids are underlined. The amino acid identities between claudins 3 and 4 of human and chicken origin were 73.6% and 78.6%, respectively. Sequences for chicken claudins 1, 7, and 8 were not available in any database.

Generation of claudin-specific antibodies

Except for claudins 1 and 2 [61], the domain arrangements of other claudin family members are not well characterized. Potentially immunogenic, claudin-specific peptide sequences for production of antibodies were selected by the following criteria: (1) Peptides should be within presumed first or second extracellular domain of claudins (see Fig. 2b). (2) Selected peptides should have highest possible sequence specificity for a given claudin protein. (3) Peptides should fulfil minimal requirements for being soluble and immunogenic. (4) At least two peptides of two different lengths should be used for antibody generation against presumed extracellular loop 1, and one peptide for the smaller loop 2. The peptides selected for immunization are shown in Table 1.

The sequence alignment in Fig. 2b shows that the amino acid sequences of claudins 1–16 of one species are relatively conserved. Likewise, there is a high degree of conservation among the same claudin in various species [42]. This may explain why conventional immunization approaches using peptide antigens in mice or rabbits have so far proven unsuccessful [21]. Only a combination of genetic and peptide immunization in mice recently resulted in antibodies recognizing the intracellular C-terminal portion of claudin 1 [21]. In our approach, the evolutionarily more distant chicken was therefore chosen as species for immunization with KLH-coupled claudin-specific peptides. Sequence alignments of human and chicken CLDN3 and 4 are shown in Fig. 2c. The sequence of chicken claudin 4 is only partly known. Human claudin 3 showed 73.6% and human claudin 4 showed 78.6% identity to the corresponding chicken sequences. Sequences of chicken claudins 1, 7, and 8 were not available in any database.

Enzyme-linked immunosorbent assay analysis used the various peptide antigens for testing purified IgG fractions from chicken immunized with the individual peptides. ELISA titers between 1:100 and up to 1:65,000 were obtained for all peptide antigens, whereby peptides CLDN1.1, 1.3, 3.2, 3.3, and 4.3 reached titers >1:10,000. Monospecific reactivity with peptides was observed for antibodies against peptides CLDN3.2, 3.3, and 4.3, which were further used for analysis of claudin surface expression.

Specificity of purified IgG fractions from chicken anti-claudin antisera was also shown in a FACS-based inhibition assay using claudin peptides CLDN1.1, 1.3, and 4.3. It was observed that claudin peptides reduced the binding of chicken anti-claudin 1 and anti-claudin 4 antibodies to A431 cells (data not shown).

Expression of claudin1-GST fusion protein was not detectable in Western blot either with the chicken anti-CLDN1 antibody or with a commercially available anti-claudin 1 antibody (Zymed, South San Francisco, CA, USA) suggesting that claudin 1 was not properly expressed as a GST-fusion protein (data not shown).

Tumor cell surface staining with polyclonal anti-claudin 3 and 4 antibodies

To test whether predicted extracellular loops of claudins are accessible to antibodies on the surface of tumor cells, the various peptide affinity–purified chicken IgGs were used for FACS analysis of nonpermeabilized cells. Polyclonal chicken antibodies raised against peptides of presumed extracellular loops of claudins 3 (CLDN3.2 and 3.3) and 4 (CLDN4.3) bound well to the two cancer cell lines SW480 (Fig. 3a) and CAMA-1 (Fig. 3b), which express claudins 3 and 4 (see Fig. 1). The reactivity of claudin 3- and 4-specific antibodies with CAMA-1 was slightly stronger than with SW480 cells, consistent with the difference in RT-PCR signal intensity. U937 cells which did not detectably express claudins 3 and 4 (see Fig. 1) served as negative control and did not exhibit significant shifts in FACS histograms compared with controls using the detection antibody alone (Fig. 3c). The antibody derived from immunization with peptide CLDN1.1 seemed to be nonspecific, as it also reacted with U937 cells. The Ep-CAM-specific positive control antibody reacted strongly with SW480 and CAMA-1, but no staining was detectable on U937 cells (Fig. 3, top panels).

Fig. 3.

Cell surface expression claudins 3 and 4 on cancer cell lines. FACS histograms using the colon cancer line SW480 (a), breast cancer line CAMA-1 (b), and monocytic line U937 (c) for surface staining with chicken anti-claudin peptide IgG CLDN1.1, 3.2, 3.3, and 4.3 are shown (see also Table 1). An anti-human Ep-CAM mouse monoclonal antibody 3B10 was used as positive control (upper panel). Faint line IgG of a nonimmunized chicken; bold line chicken anti-claudin IgG or anti-Ep-CAM mab.

The surface binding of claudin-specific chicken antibodies was verified by immunocytochemistry. IgG fractions purified from chicken anti-claudin 3 and 4 antibodies showed a surface staining of nonpermeabilized SW480 (Fig. 4, left panel) and CAMA-1 cells (Fig. 4, right panel). Cells incubated with the IgG fraction from a nonimmunized chicken did not show this cell surface–specific staining pattern (Fig. 4, lower row). The staining with IgG from claudin 3–specific peptides was more intense than with claudin 4–specific IgG. The surface stainings by anti-claudin IgGs showed a pronounced granularity that may indicate an occurrence of claudins 3 and 4 in some kind of surface clusters.

Fig. 4.

Binding of chicken anti-claudin IgGs to vital SW480 and CAMA-1 cells. Bright field microscopic images are depicted in the first and third vertical panels. Immunofluorescence images are shown in the second and fourth vertical panels. Staining was performed with chicken anti-claudin IgG against peptides CLDN3.2, 3.3, and 4.3, as shown on the left. Bound chicken IgG was visualized with Cy3-conjugated rabbit anti-chicken IgG. IgG from a nonimmunized chicken egg served as negative control. Cell staining was analyzed by immunofluorescence using a phase contrast microscope.

Staining of human cancer tissue sections by anti-claudin antibodies

We investigated whether claudin expression can be detected by the antibodies in human tissue sections derived from normal kidney and renal cancer. Sections from normal human kidney tissue (Fig. 5a), and human clear cell renal carcinoma (Fig. 5b) were stained with claudin-specific chicken IgG. Normal human kidney tissue was also stained with a monoclonal antibody directed against Ep-CAM. Both anti-Ep-CAM and anti-claudin 3 (CLDN3.3) antibodies stained epithelial cells of renal tubuli in a comparable fashion (Fig. 5a). The anti-claudin 3 IgG (CLDN3.2) showed an even staining of the cell membrane of human kidney carcinoma cells in tissue sections (Fig. 5b). No such staining was found with a nonimmune IgG fraction from chicken. The mouse control antibody MOPC21 also did not show staining under these conditions (data not shown).

Fig. 5.

Immunohistochemical staining of frozen tissues with chicken anti-human claudin IgG. Normal tissue from human kidney (right panel) was stained with mouse anti-Ep-CAM (3B10) antibody and anti-claudin 3 chicken IgG (CLDN3.3). Human renal carcinoma tissue (left panel) was stained with anti-claudin 3 chicken IgG (CLDN3.2). IgG from a nonimmunized chicken egg served as negative controls. Cell staining was analyzed by phase contrast microscopy.

For claudin 1, parallel stainings were performed on serial sections from human ovarian carcinoma tissue using chicken anti-CLDN1 and rabbit anti-CLDN1 antibody (Zymed, South San Francisco, CA, USA) which is directed against the intracellular C-terminal part of claudin 1. Slides used for these experiments were prepared from frozen tissue samples and air-dried before staining. Identical staining patterns were obtained with both antibodies directed against the intracellular and extracellular part of claudins (data not shown).

Claudin 3 expression on micrometastatic tumor cells from the bone marrow of breast cancer patients

Micrometastatic cells of epithelial origin can be frequently detected in bone marrow aspirates of cancer patients and have prognostic relevance for tumor progression [6, 50]. A standard technology to detect such disseminated tumor cells in bone marrow samples is immunocytochemical staining with the cytokeratin-specific monoclonal antibody A45/B/B3 [5, 63]. Here, we investigated by double immunofluorescence labeling whether the A45/B/B3-positive cells found among bone marrow cells of four breast cancer patients also express claudin 3. As shown in Fig. 6, essentially all cytokeratin-positive cells (green color) from four breast cancer patients were also stained by the polyclonal chicken anti-claudin 3 IgG raised against peptide 3.3 (red color). Staining for claudin 3 and cytokeratin was comparable to that of breast cancer line CAMA-1 with respect to intensity and distribution (left bottom panels). No such staining was observed with control antibodies (bottom right panels). The claudin 3–specific staining of disseminated tumor cells was concentrated at the cell membrane. Our data suggest that claudin 3 is still expressed on disseminated carcinoma cells, which are detached from the original tumor. An anti-claudin 3 antibody is useful for detection of micrometastatic cells in bone marrow samples and, in a therapeutic form, may reach a large proportion of disseminated tumor cells.

Fig. 6.

Double staining of cells isolated from bone marrow aspirates of four mamma carcinoma patients for claudin 3 and cytokeratin. The upper panels show in red claudin- and in green cytokeratin-specific staining of bone marrow cells of breast carcinoma patients BM1117, BM412, BM520, BM573 with corresponding bright field images. Air-fixed slides were prepared from mononucleated cells and stained with chicken anti-CLDN3.3 peptide IgG, and the mouse anti-cytokeratin antibody A45/B/B3 detected by Cy3-labeled anti-chicken and FITC-labeled anti-mouse antibodies, respectively. The lower left panels show claudin 3- and cytokeratin-specific staining of CAMA-1 breast cancer cells as positive control and the corresponding bright fields. Chicken control IgG, and mouse MOPC21 antibody were used as negative controls. Exposure times with Cy3 filter were 100 ms, with FITC filter 500 ms, and bright field 1 ms.

Discussion

We found that mRNAs encoding claudins 1, 3, 4, and 7 are all well expressed in a great variety of human carcinoma lines of different origins. Likewise, protein expression of claudins 3 and 4 was demonstrated by affinity-purified polyclonal anti-peptide antibodies on various cultured cancer lines, on renal cancer tissue, as well as on disseminated tumor cells found in the bone marrow of four breast cancer patients. The affinity-purified polyclonal antibodies reacted with predicted extracellular domains of claudins 3 and 4 and showed that claudins are principally accessible and distinguishable by antibodies on the cell surface. Because extracellular domains of claudins are thought to mediate homotypic and heterotypic association and formation of TJs [18], we consider it likely that our antibodies reacted with a subpopulation of claudins that is not acutely engaged in forming a TJ but rather is a population of “free” claudin. Moreover, it is likely that our antibodies could not recognize claudin proteins within TJs because of a shielding of their binding epitope in such structures. Immunofluorescence studies on single cells outside the compact structure of an epithelial cell layer allow the detection of claudin expression in general but do not differentiate between claudin freely expressed on the cell surface or in TJs. So far, only antibodies to intracellular portions of claudins were employed for TJ staining [14, 43, 64]. Such considerations may explain two findings of the present study. One is the relatively low level of claudin surface expression as seen in FACS histograms of cell lines showing relatively strong claudin mRNA signals. Secondly, our immunohistochemical staining of normal renal tissue did not reveal a selective staining of TJs but a rather homogeneous and weak overall surface staining. Because tumor cells express claudins but do not form classical TJs as found in normal epithelial tissue, tumor cells may have a considerable pool of free claudin that is amenable to extracellular antibody binding and immunotherapy. Therefore selective binding of claudin-specific antibodies to free claudin and claudin in TJs cannot be achieved in tumor tissue sections.

In these properties, claudins 3 and 4 are not unlike the Ep-CAM target which is being used for the development of various antibody-based therapeutics. Ep-CAM antibodies are being developed as cytotoxic murine, chimeric, and fully human IgGs [19, 44, 62], bispecific T-cell–recruiting antibodies [34], immunotoxins [10], and as conjugates with cytokines [54]. The subcellular distribution of claudins and Ep-CAM on epithelial cells may also not be very different. While claudin expression is largely confined to TJs [13], Ep-CAM expression was reported to be intercellular and basolateral [39, 45]. It is possible that claudins in healthy epithelium are even better shielded from the access by antibodies than Ep-CAM. For antibody-dependent cell-mediated cytotoxicity (ADCC) by a cytotoxic human IgG1 or murine IgG2a antibody, the formation of a cytolytic synapse between target cell and natural killer (NK) cell or other immune cells is a prerequisite [9, 38]. Synapse formation may require a certain target density and target availability. In normal tissue, Ep-CAM engaged in homotypic cell adhesion and claudins engaged in forming TJs may not be available in the critical density for cytolytic synapse formation. In contrast, Ep-CAM and claudins overexpressed on tumor cells, where they seem less involved in cell adhesion and TJ formation, may facilitate cytolytic synapse formation with immune cells as triggered by antibodies. Likewise, complement-dependent cytotoxicity (CDC) appears to require a critical target density and availability [23]. Natural cytotoxic mechanisms against tumor cells based on ADCC and CDC may therefore increase the therapeutic window of an antibody.

Claudin 4 has been described as a receptor for the cytotoxic Clostridium perfringens enterotoxin (CPE) [11, 55], suggesting that claudins in healthy tissue may be accessible to proteins even if those are components of TJs. Binding of CPE to pancreatic cancer cells led to an acute dose-dependent cytotoxic response, which was restricted to claudin 4–expressing cells and correlated with claudin 4 expression levels [40]. These findings suggest that claudin 4–expressing tumors can be targeted by an exogenous proteinaceous agent. While CPE can theoretically be used as a targeting device substituting for immunotherapeutics, there are general issues with bacterial proteins. They typically are of high immunogenicity, have toxic properties, and, in the case of CPE, are specific for only one claudin. It is therefore desirable to develop monoclonal antibodies to claudins that are of low immunogenicity, not toxic, and with specificity for individual claudin family members. We show here that such antibodies can be obtained. Affinity-purified chicken IgGs were developed that were monospecific for claudins 3 and 4, respectively, and reacted with predicted extracellular loop structures of the two claudins. Polyclonal chicken IgGs are not suitable for the development of human therapeutics but can establish that claudins principally are amenable to targeting by antibodies. With anti-claudin antibodies or fragments of these antibodies including VH or VL chains or complementarity-determining regions (CDR), the full spectrum of antibody-based therapies can be developed including human IgG1, immunotoxins, radioimmunoconjugates, superantigen fusions, and bispecific antibody formats. The high sequence identity between certain claudins (e.g., between claudins 3 and 4) may even allow generation of antibodies recognizing more than one claudin species. An anti-claudin antibody with, e.g., dual-target specificity might be more effective than a monospecific anti-claudin antibody because it minimizes the risk for development of escape mutants and could target a broader range of tumors in a larger segment of patients. Based on expression data, potential indications for anti-claudin antibody-based therapeutics include carcinomas of colorectal, breast, ovarian, and prostate origin. The study of other claudin proteins may identify more candidates for antibody-based therapies.

The overexpression of certain claudin proteins in tumors may increase the therapeutic window for antibody-based therapies. Hough et al. [22] showed that the panepithelial TJ molecule claudin 4 is overexpressed in carcinomas. Using serial analysis of gene expression to generate global gene expression profiles from various ovarian cell lines and tissues, a number of genes including claudin 3 and 4 were identified, which were up-regulated in ovarian cancer in contrast to nontransformed ovarian epithelia. The overexpression of claudin 3 and 4 at the protein level in tumor samples was further validated through immunohistochemical analysis [22]. Claudin 4 was also shown to be overexpressed in most pancreatic cancer tissues and cell lines [40], and in several other gastrointestinal tumors. Terris et al. [58] applied cDNA microarray analysis to characterize gene expression profiles of invasive pancreatic ductal adenocarcinomas. This confirmed that claudin 4 mRNA was up-regulated in these tumors.

Epithelial expression of claudin 3 and 4 was described by several authors [24, 42]. Two further studies showed that claudins 3 and 4 were up-regulated in epithelial tumor tissue. Long et al. [33] studied the mRNA expression of claudins 3 and 4 in prostate cancer tissue. Persistently high levels of claudin 3 expression were detected in prostate adenocarcinoma and in metastases. Also in this case, tumor cells could be killed by CPE [33]. Using oligonucleotide microarrays, Bhattacharjee et al. [4] analyzed mRNA expression levels in 186 lung tumor samples corresponding to 12,600 transcript sequences, including 139 adenocarcinomas resected from the lung. In this study, an overexpression of the mRNA encoding claudins 3 and 4 was significantly enhanced in lung metastases originating from colon carcinomas.

The overexpression of claudin proteins on tumors of various origins may indicate that certain members of this protein family provide a selection advantage for tumor cells possibly related to adhesion, migration, metastasis, angiogenesis, survival, or immune modulation. Future studies will attempt to correlate claudin expression levels on tumor cells with an impact on survival of patients. Antibody reagents of the type developed in this study will be useful reagents for such studies.

Abbreviations

ADCC

Antibody-dependent cell-mediated cytotoxicity

AP

Alkaline phosphatase

CDC

Complement-dependent cytotoxicity

CDR

Complementarity-determining region

CLDN

Claudin

CPE

Clostridium perfringens enterotoxin

dNTP

Deoxynucleoside-triphosphate

ELISA

Enzyme-linked immunosorbent assay

Ep-CAM

Epithelial cell adhesion molecule

FACS

Fluorescence-activated cell sorting

FCS

Fetal calf serum

HPLC

High-performance liquid chromatography

IgG

Immunoglobulin G

KLH

Keyhole limpet hemocyanine

NK

Natural killer cell

PBS

Phosphate-buffered saline

PCR

Polymerase chain reaction

PNPP

Paranitrophenylphosphate

RT

Reverse transcription

TJ

Tight junction