Cell transfection, biochemical, and electronmicroscopic studies suggest that Ten-m1 is a dimeric type II transmembrane protein. Expression of fusion proteins composed of the NH₂-terminal and hydrophobic domain of ten-m1 attached to the alkaline phosphatase reporter gene resulted in membrane-associated staining of the alkaline phosphatase. Electronmicroscopic and electrophoretic analysis of a secreted form of the extracellular domain of Ten-m1 showed that Ten-m1 is a disulfide-linked dimer and that the dimerization is mediated by EGF-like modules 2 and 5 which contain an odd number of cysteines.

Northern blot and immunohistochemical analyses revealed widespread expression of mouse ten-m genes, with most prominent expression in brain. All four ten-m genes can be expressed in variously spliced mRNA isoforms. The extracellular domain of Ten-m1 fused to an alkaline phosphatase reporter bound to specific regions in many tissues which were partially overlapping with the Ten-m1 immunostaining. Far Western assays and electronmicroscopy demonstrated that Ten-m1 can bind to itself.

All known pair rule genes code for transcription factors, except for a gene identified independently in two laboratories and designated ten-m (Baumgartner et al., 1994) and odz (Levine et al., 1994). ten-m and odz are identical genes and mutations lead to a pair rule phenotype (Baumgartner et al., 1994; Levine et al., 1994) similar to odd-paired in which every other segment is missing (Nüsslein-Volhard et al., 1995). Despite the fact that both reports showed identical sequences, Ten-m was described as a secreted Drosophila tenascin-like molecule (Baumgartner et al., 1994) and Odz as a type I transmembrane receptor (Levine et al., 1994). Tenascins are a family of extracellular matrix proteins with a modular structure composed of fibronectin type III (FNIII) repeats, EGF-like repeats, and a COOH-terminal fibrinogen-like repeat (Erickson, 1993). Biochemical studies using a Drosophila cell line indicated that Ten-m is a large secreted proteoglycan with chondroitinase ABC-sensitive chondroitin sulfate and/or dermatan sulfate side chains. The core protein was reported to contain EGF-like and FNIII repeats, but to lack the fibrinogen-like domain (Baumgartner et al., 1994). Odz was isolated as a novel phosphotyrosine-containing protein (Levine et al., 1994). A transmembrane region was predicted COOH-terminal of the EGF repeats, followed by the cytoplasmic domain containing several tyrosine kinase phosphorylation consensus sites (Levine et al., 1994). More recently, Wang et al. (1998) described a mammalian orthologue of Ten-m/Odz, termed DOC4 (downstream of chop), which is induced by the stress-induced transcription factor CHOP. The open reading frame of DOC4 shares 31% sequence identity and 50% sequence similarity with Ten-m/Odz. Furthermore, DOC4 contains a short stretch of hydrophobic amino acids ~400 amino acids COOH-terminal of the putative start codon. This together with the cell surface localization led to the suggestion that DOC4 may constitute a type II transmembrane molecule (Wang et al., 1998).

Ten-m/Odz, as well as DOC4, contains a stretch of eight consecutive EGF-like modules which are most similar to the EGF repeats of tenascins. EGF modules are structural units of proteins or parts of protein, located extracellularly. They can occur as isolated modules such as in reelin (D'Arcangelo et al., 1995) and in selectins (Whelan, 1996), or in arrays like in notch (Fleming et al., 1997) and tenascins (Spring et al., 1989). A conserved feature of the EGF domain in Ten-m/Odz, DOC4, and Ten-a, a Drosophila molecule related to Ten-m/Odz (Baumgartner and Chiquet-Ehrismann, 1993), is the substitution of a cysteine residue with an aromatic amino acid in two of the eight EGF-like modules. This leaves two cysteines with no intramodular partner. The importance of the integrity of the cysteine patterns in EGF-like modules is exemplified by the functional impairment of notch 3, which has been observed in patients with an autosomal dominant disorder causing stroke (Joutel et al., 1997). The molecular basis of this disease is predominantly the substitution of cysteines with other amino acids in the EGF modules of notch 3. The observation that the EGF-like modules of Ten-m/Odz with five cysteines are ontogenetically conserved indicates that they are able to fold into a structure which might be important for the function of the protein.

Many genes that control pattern formation are expressed at several different periods during development to function in a variety of processes both during embryogenesis and postnatal life. After the initial expression in seven stripes at the cellular blastoderm stage, ten-m/odz is downregulated and appears at later stages in the central nervous system (CNS),¹ dorsal vessel, trachea, and the eye and discs giving rise to the cephalic (antenna), ventral (wing), and dorsal (legs) thoracic appendages (Baumgartner et al., 1994; Levine et al., 1994). The highest level of Ten-m/Odz expression is observed in the CNS where the protein is deposited on the surface of axons (Levine et al., 1994; Levine et al., 1997). The Drosophila eye disk is another location where high levels of Ten-m/Odz are found in very distinct sites including the morphogenetic furrow, photoreceptor-like cells, and nonepithelial cells of the eye disc (Levine et al., 1997). The expression pattern of DOC4 in mammals is not well characterized but the presence of the mRNA has been demonstrated in the developing mouse brain (Wang et al., 1998).

Several mutations in the ten-m/odz gene have been identified, all resulting in embryonic lethality (Baumgartner et al., 1994; Levine et al., 1994). Due to the lack of viable hypomorphic mutations, it is not clear whether the protein executes an important function in all sites where it is expressed. One possible function for Ten-m/Odz comes from studies with DOC4 which has been isolated in search of GADD153/CHOP (growth arrest and DNA damage/ C/EBP homology protein)-induced mRNA. GADD153/ CHOP is responsive to many forms of stress, including alkylating agents, UV light, and conditions that trigger an ER stress response. For example, ER stress which occurs during ischemia alters proliferation of cells, induces cell death, and the expression of GADD153/CHOP (Zinszner et al., 1998). Recent studies have shown that GADD153/ CHOP exerts at least part of its function via the induction of DOC4 and other proteins (Wang et al., 1998).

As an initial step to obtain insight into the function of Ten-m/Odz–related proteins in mammals, we have cloned several cDNAs for mouse orthologues of Ten-m/Odz. We report here that at least four different cDNAs with similarity to the Drosophila ten-m/odz cDNA are expressed in mice. One of them, ten-m4, is identical to the DOC4 cDNA. The alignment of the four deduced mouse protein sequences indicated a strong conservation of the characteristic features for type II transmembrane molecules, which was also recognized for DOC4 (Wang et al., 1998). This predicted topological orientation was experimentally confirmed by expression studies of recombinant mouse Ten-m1 fusion proteins in mammalian cells. In addition, the recombinant production of the putative extracellular domain of Ten-m1 revealed the formation of dimeric structures. We provide experimental evidence which shows that the dimerization of Ten-m1 is mediated via the single cysteine residues in the EGF modules that lack their intramodular partners. Furthermore, we show that Ten-m1 is able to make homophilic interactions.

The ten-m2, ten-m3, and ten-m4 cDNAs were isolated by screening the brain cDNA library with the PstI/EcoRI fragment of DT19 (731–1,833 bp) at low stringency as previously described (Oohashi et al., 1994). Full-length cDNA fragments were isolated and characterized from the same mouse brain cDNA library. Sequence alignments were performed using the BLAST program (Altschul et al., 1994) and PileUp from the GCG (Genetics Computer Group) package.

The following oligolabeled probes were used: for ten-m1 a fragment ranging from nucleotide 75 to 1833, for ten-m2 a fragment ranging from nucleotide 1 to 2006, for ten-m3 an EcoRI fragment ranging from nucleotide 762 to 1408; for ten-m4 a fragment ranging from nucleotide 1 to 1108. The RNA loading was controlled by probing blots with a GAPDH cDNA probe.

Human embryonic kidney cells (HEK 293 cells; American Type Culture Collection) were transfected with the constructs in an eukaryotic expression vector containing a CMV promoter (pRC/CMV; Invitrogen) and a puromycin resistance gene. Puromycin-resistant clones were isolated by ring cloning. Positive clones were identified by SDS-PAGE and Coomassie blue staining according to standard protocols and maintained as described (Retzler et al., 1996). The AP expression was determined as described by Flanagan and Leder (1990). For determining whether the intracellular and transmembrane domain localized the AP module on the outer cell surface, cells were treated with trypsin for 30 min at 37°C and then allowed to adhere on gelatinized glass coverslips for 30 min, washed, fixed, and stained for AP activity.

Western blots for detecting recombinant proteins were performed by transferring proteins separated by SDS-PAGE to supported nitrocellulose (Bio-Rad) in Tris/glycine buffer containing 10% methanol for 1 h with 100 V with the Bio-Rad mini gel system. The blots were blocked with 1% BSA, incubated with an anti–calf intestine AP antiserum (Sigma Chemical Co.), and developed with AP-conjugated secondary antibodies.

Figure 1 Protein sequence of EGF-like modules and structure of mouse Ten-m 1–4. (A) EGF-like domains of mouse Ten-m 1–4, Drosophila Ten-m, and Drosophila Ten-a. Conserved cysteines are indicated by asterisks. Substitutions of cysteines by (more ...)

Figure 2: COOH-terminal domain of mouse Ten-m 1–4. Conserved amino acid residues are in shadowed boxes.

Figure 2 Expression of the NH2-terminal sequence of Ten-m1 linked to an AP module in HEK 293 cells. (A) Cartoon of Ten-m1 and of the AP fusion protein (ten-m1 ap). Symbols are the same as in Fig. 1 B. (B) Transfected HEK 293 cells show AP activity on the surface (more ...)

When the ten-m1 cDNA fragment DT19 (731–1833 bp) was hybridized to the brain cDNA library at low stringent condition, several clones showing a weak hybridization signal were isolated and sequenced. The deduced amino acid sequence of three clones contained the eight EGF-like repeats and part of the cysteine-rich domain of ten-m molecules. Despite the high homology of 63, 51, and 52% to Ten-m1, respectively, the three cDNA clones were clearly different from ten-m1 and from each other. To obtain full-length cDNAs, overlapping cDNA clones were isolated from the same library. The cDNA sequences were designated ten-m2, ten-m3, and ten-m4, which had open reading frames of 2,764, 2,715, and 2,771 amino acid residues, respectively, and the translated amino acid sequences showed an overall similarity between 56 and 70%.

None of the four ten-m sequences contained a signal peptide. Approximately 300–400 residues after the start codon, all four sequences showed a continuous stretch of 34 amino acids lacking charged residues, which could serve as transmembrane domain (Supplemental Figure 1). About 200 amino acids COOH-terminal of the hydrophobic amino acids were eight consecutive EGF modules (Fig. 1 A) followed by a large COOH-terminal sequence which was rich in cysteines and devoid of any known modular motifs (Supplemental Figure 2). The absence of a signal peptide and the presence of a stretch of hydrophobic amino acids suggested that the family of ten-m proteins may be expressed as type II transmembrane molecules. According to this model, the NH₂-terminal 300–400 amino acids serve as cytosolic domain. They showed on average the lowest level of identity between the four mouse ten-m sequences, ranging between 34 and 46%. The extracellular part of ten-m molecules would consist of a linker domain of ~200 amino acids, a region with eight EGF-like domains of ~250 amino acids, and a COOH-terminal domain of ~2,000 amino acids (Fig. 1 B). The linker domain of all mouse ten-m proteins contained several dibasic amino acid residues which could serve as potential sites for proteolytic processing of the molecules. One of these sites is conserved in all four mouse ten-m sequences and the Drosophila Ten-m/Odz and Drosophila Ten-a (Supplemental Figure 1). The Drosophila Ten-m/Odz shows higher identity to the mouse ten-m sequences in the linker domain (26–29%) than in the cytosolic domain (19–21%). Between the mouse ten-m sequences, the similarity of the cysteine-free linker domain ranged from 43 to 48%, the EGF domains ranged from 65 to 72%, and the large cysteine-rich COOH-terminal domain ranged from 58 to 68%. The similarity to Drosophila Ten-m/Odz over the entire COOH-terminal domain ranged from 30 to 33%.

When the ten-m protein sequences were compared with the cDNA and protein data bank, DOC4 (Wang et al., 1998) was identical to Ten-m4. In addition, the NH₂-terminal part of the human γ-heregulin, identified in the breast cancer cell line MDA-MB-175 (Schäfer et al., 1997), had 95% identity to the NH₂-terminal fragment of mouse Ten-m4/DOC4. The COOH terminus of the ten-m proteins contained several amino acid repeats which showed similarities with the rearrangement hot spot elements (rhs) of Escherichia coli (Feulner et al., 1990) and with a wall-associated protein (WAP) of Bacillus subtilis (Foster, 1993).

To obtain biochemical evidence for the membrane localization of the fusion protein, the expressing cells were analyzed by Triton X-114 phase partition experiments. Aqueous solutions of 0.7% Triton X-114 separate into two phases when warmed from 4 to 30°C. The dense Triton X-114 containing detergent-rich phase has been shown to be enriched in integral membrane and membrane-anchored proteins, whereas the Triton X-114–depleted phase is enriched in soluble proteins (Justice et al., 1995). In such experiments the fusion protein of 116 kD, which is the expected size for a protein consisting of the NH₂-terminal 370 amino acids of Ten-m1 and the 67-kD AP module, was distributed preferentially to the detergent-rich phase (Fig. 2 E, lane 8). The majority of the proteins released from the cell layer at 4°C with 0.7% Triton X-114 remained at 30°C in the detergent-depleted phase (Fig. 2 E, lanes 1 and 2).

These data show that the NH₂-terminal and the hydrophobic sequence of Ten-m1 can direct the AP to the plasma membrane. This, together with the finding that the AP activity can be found on the cell surface, suggests that Ten-m1 is a type II transmembrane protein.

Figure 3 Expression and electronmicroscopic analysis of the extracellular domain of Ten-m1. (A) Cartoon of Ten-m1 and the secreted COOH-terminal part of Ten-m1 (Ten-m1sec). (B) Coomassie blue staining of molecular mass standards (lanes 1 and 3) and purified (more ...)

Purified recombinant molecules were visualized by electron microscopy after spraying from glycerol spraying/ buffer mixtures (Fig. 3, C and D) and after adsorption to carbon films and negative staining (Fig. 3, E and F). By both techniques mainly particles with a structure compatible with dimeric molecules were visible. They consisted of two elongated globular domains connected by a thin extended rod. 24% of these dimeric particles were, in addition, visible in close proximity to each other. The size distribution of the spherical moiety exhibited Gaussian profiles with a long diameter d1 of 13.2 ± 1.3 nm and a short diameter d2 of 7.8 ± 1.3 nm. Values are corrected for overestimation of ~3 nm due to decoration with platinum. Negative staining revealed similar values (d1 12.7 ± 0.8 nm, d2 6.9 ± 0.7 nm), but in addition, each globular domain was resolved into three globular subdomains with similar sizes of 5.4 ± 0.9 nm (Fig. 3, E and F). The extended rod connecting the spherical domains had a total length of 5.9 ± 2.9 nm after negative staining. This rod appeared more variable in structure after rotary shadowing with a total length of 17.9 ± 4.9 nm where the middle part occasionally formed a short branch of 8.8 ± 2.8 nm extending sideways.

Figure 4 Expression of EGF-like domains derived from Ten-m1 and fused to an AP module. (A) Cartoon of Ten-m1 and the AP fusion proteins. In AP-3EGF the AP module is linked to the first three EGF-like domains and AP-8EGF to the eight EGF-like domains of Ten-m1. (more ...)

These data suggest that the EGF-like modules of Ten-m1 with an odd number of cysteines can form intermolecular disulfide bonds, leading to the homodimerization of Ten-m1.

Figure 5 Northern blot of tissues using ten-m cDNAs as probes. Poly(A)+ RNA was isolated from the tissues indicated and probed with ten-m1, 2, 3, and 4 cDNAs coding for the cytoplasmic and transmembrane domain and exposed for 4 d (A) or 24 h (B). Arrowheads (more ...)

To assess protein expression of Ten-m1, a specific antiserum was produced by immunizing rabbits with the recombinant extracellular domain of Ten-m1. The antiserum was further purified by affinity chromatography with covalently immobilized immunogen. The results obtained by the specific antibodies in Western blots depended strongly on the reducing agent in the sample buffer and the treatment of the samples (see Materials and Methods). After exhaustive reduction of detergent extracts from mouse brain with dithiothreitol, two major protein bands with apparent molecular masses of 270 and 225 kD were observed (Fig. 6, lane 1). Often also higher molecular mass proteins, probably representing incompletely reduced protein dimers, were recognized by the purified antibody (Fig. 6, lane 2), occasionally even after prolonged incubations in the reducing sample buffer. Whereas the 270-kD form might correspond to the complete Ten-m1 molecule, the 225-kD form might represent an alternatively spliced or proteolytically processed molecule. It is also possible that the 225-kD form is a soluble form of the molecule, since it migrated at a similar position as the recombinant extracellular domain of Ten-m1 (Fig. 6, lane 3). A clear pattern of protein bands recognized by the purified antiserum could only be observed in brain extracts, but not in extracts of other tissue such as thymus, lung, and kidney (not shown). A lower abundance of Ten-m1 protein in tissues other than brain would be in agreement with the results of the Northern blot assay.

Figure 6 Western blot of brain extracts with affinity-purified antibodies against Ten-m1. 3% Triton X-100 extracts of mouse brain (lanes 1 and 2) were precipitated with acetone. Dried precipitates were dissolved in sample buffer, either for 30 min at 70°C (more ...)

To localize Ten-m1 protein expression in mice, immunofluorescence assays were performed on a variety of tissue sections. Strong signals were observed in many regions of the mouse brain (Fig. 7, A–D). The cerebellum showed intensive staining in the molecular layer and weaker staining around cells in the granular layer. The staining around the cells appeared membrane associated and sometimes speckled. This localization strengthens the notion that Ten-m1 is a type II transmembrane protein with the COOH terminus on the extracellular side (antibodies were raised against the extracellular domain), as already suggested by sequence analysis and results obtained with recombinant fusion proteins. Purkinje cells and white matter were negative for Ten-m1 expression (Fig. 7, A and B). The hippocampus had strong staining in the molecular layer and the neuronal layer of the CA3 region. Weak staining was observed in the CA1 region and dentate gyrus (Fig. 7, C and D). Several other regions of the brain including the deep cerebellar nuclei, thalamic nuclei, cortex, and brain stem showed strong staining for Ten-m1 (not shown). Double immunostaining using anti–β3 tubulin antibodies which specifically label neurons revealed colocalization in many areas (not shown).

Figure 7 Immunolocalization of Ten-m1 in tissue sections. (A and B) Immunostaining of cerebellum. Strong staining is seen in the molecular layer (m) and weaker staining around cells in the granular cell layer (g). Purkinje cells (arrow in B) show no expression (more ...)

Expression of Ten-m1 was also seen in the outer and inner segments of the retina where rods and cones showed very strong staining (Fig. 7 E) and around the basal epithelial cells of the cornea (Fig. 7 F). The corneal surface showed a filamentous, strong staining (Fig. 7 F) which either could identify soluble Ten-m1 present in the tear film moistening the corneal surface or could be an artifact. Lung showed high expression in smooth muscle cell layers of bronchi, veins, and arteries and low expression on alveolar epithelial cells (Fig. 7 G). In kidney, the staining for Ten-m1 was restricted to the glomeruli (Fig. 7 H) and vessels in the medullar region (not shown). Immunostaining of adult testes revealed Ten-m1 expression around spermatides but no expression was detected in spermatogonia and mature sperms (Fig. 7 I). Staining of control sections with a nonimmune serum was negative (data not shown).

Figure 8 Binding pattern of AP-ten-m1 in tissue sections. (A) Cartoon of Ten-m1, the fusion protein composed of AP-ten-m1 and AP. Symbols are the same as in Fig. 1 B. (B) Coomassie blue staining of the purified AP-ten-m1 and AP separated by 7% SDS-PAGE under (more ...)

The binding of AP-ten-m1 in brain sections was high in the molecular layer of the cerebellum and around Purkinje cells. Little but significant staining was observed in the granular layer (Fig. 8 C). The hippocampus showed a clear signal in the molecular layer (Fig. 8 E) and a weak signal was visible in the CA3 region, CA1 region (not shown), and dentate gyrus (Fig. 8 E). Strong reactivity was also observed in many other regions of brain including cortex and several deep cerebellar nuclei (not shown). Kidney sections showed very intense staining in glomeruli (Fig. 8 G) and vessels of the medullar region (not shown). The lung sections had AP reactivity in the smooth muscle cell layers of bronchi, arteries, and veins (not shown). Incubation of tissues with the AP module did not reveal any staining reaction (Fig. 8, D, F, and H and data not shown).

Figure 9 Homophilic interaction of Ten-m1. Western blot of the recombinant extracellular domain of Ten-m1 (lanes 1, 4, 7, and 10), recombinant neurocan core protein (lanes 2, 5, and 8), and BSA (lanes 3, 6, and 9), separated on a 5% (Ten-m1 and neurocan) or (more ...)

Given the fact that Ten-m1 appears to interact in a homophilic fashion, it is noteworthy that by rotary shadowing and negative staining of purified Ten-m1 extracellular domain ~24% of the Ten-m1 particles showed a clear interaction with each other (see arrows in Fig. 3, C and E).

The mRNAs for all four ten-m genes were present in all tissues analyzed so far. The highest levels of ten-m mRNAs were found in brain similar to the expression of Ten-m/Odz in Drosophila (Baumgartner et al., 1994; Levine et al., 1994). In many tissues the mRNA bands appeared as different sizes, suggesting that the ten-m genes are alternatively spliced. This was supported by the isolation of further ten-m cDNA clones corresponding to the different mRNA bands seen in Northern analysis (Oohashi, T., and R. Fässler, unpublished observations). The expression of Ten-m1 protein was further demonstrated by Western blotting and immunohistochemistry. Immunostaining for Ten-m1 confirmed the widespread expression. Moreover, Ten-m1 was expressed in a very restricted manner in the tissues analyzed. For example, in kidney Ten-m1 is only found in the glomeruli and in vessels of the medulla, in lung the expression is particularly high in the smooth muscle layers of the bronchi, veins, and arteries and in the testes Ten-m1 is present in spermatides but not in spermatogonia, mature sperms, or Sertoli cells. The brain sections revealed a distinct signal around neuronal cells (hippocampus, cerebellum, cortex) and a very strong signal in fiber tract regions. A similar staining pattern was also observed in the CNS of developing and adult flies. Similarly to Drosophila ten-m/odz, ten-m1 mRNA is expressed in nerve cell bodies and the protein in axons, suggesting that Ten-m1 may also have an axonal localization signal (Zhou, X.-H., T. Oohashi, and R. Fässler, unpublished observations).

The lack of a signal peptide and the presence of a single stretch of hydrophobic amino acids COOH-terminal of the start codon suggested that the ten-m proteins in mouse are expressed as type II transmembrane proteins (Fig. 10).

Figure 10 Model of mouse Ten-m1. Two Ten-m1 protein chains are inserted into the plasma membrane as a type II transmembrane molecule. EGF-like modules 2 and 5 are engaged in intermolecular bonds. The COOH terminus is divided into three globular domains. (more ...)

Our finding that the NH₂-terminal part of Ten-m1 including the short hydrophobic domain has the ability to locate an AP reporter protein at the outside of the cell membrane suggests that this region of the protein contains all the information necessary to insert the protein into membranes as a type II transmembrane molecule. These results are different from the sequence interpretation of Odz (Levine et al., 1994) which described a transmembrane domain COOH-terminal of the EGF repeat domain. However, this sequence is interrupted by charged amino acids and, therefore, is not entirely hydrophobic, which is unusual for transmembrane domains. Furthermore, the transmembrane domain of Odz is poorly conserved in the mouse ten-m proteins. In addition to charged amino acids, all four mouse ten-m proteins contain a conserved potential N-glycosylation site in this region. Finally, the secreted form of the recombinant mouse Ten-m1 contained this stretch of amino acids but we never observed integration of the recombinant protein into the cell membrane of transfected HEK 293 cells. Therefore, we believe that even in Drosophila Odz this region cannot be used as a transmembrane domain for a type I insertion.

The ultrastructural analysis of the extracellular domain of Ten-m1 and the expression of recombinant fusion proteins composed of an AP reporter and either the first three or all eight EGF-like domains revealed that ten-m proteins are expressed as dimers covalently linked by a pair of free cysteine residues in EGF repeats 2 and 5. Negative staining of the extracellular domain in the EM revealed that the large COOH terminus of Ten-m1 is divided into three globular domains preceded by a fine rod which represent the EGF repeat domains. The EGF repeat domain of all ten-m proteins is composed of eight EGF-like domains which lack Ca²⁺ binding consensus motifs (Pan et al., 1993) and are most similar to the EGF-like repeats of the ECM molecule tenascin-C (Spring et al., 1989). The predicted length of the complete EGF repeat domain is 16 nm and of the first five EGF-like repeats 10 nm (Engel, 1989). This is in agreement with the EM observation that the extended rods which connect the two monomers are ~9 nm long. In EGF-like domains 2 and 5 the third cysteine is replaced by either a tyrosine or phenylalanine residue. This cysteine substitution is conserved in all four mouse ten-m proteins and in the Drosophila Ten-m/Odz. In a normal EGF repeat, the first cysteine would make a disulfide link to this third cysteine residue. Since it cannot be engaged in such a link in EGF-like domains 2 and 5 of the Ten-m proteins, it conceivably can form an intermolecular disulfide bond with another ten-m molecule. To our knowledge it has never been reported that an EGF repeat domain can covalently interact with another EGF domain. However, it cannot be excluded that an EGF-like domain with only five cysteine residues will loose the typical EGF-like folding pattern when it is engaged in intermolecular covalent interactions. Clearly, more structural studies are needed to resolve the folding pattern of the EGF-like repeats in its dimerized form.

The only other molecule which shares this incomplete cysteine pattern in modules within an EGF repeat array is Drosophila Ten-a (Baumgartner et al., 1993), which appears to have a topography very similar to ten-m molecules, but terminates shortly after the EGF repeat domain. The EGF repeat domain of Ten-m/Odz and Ten-a are more similar to each other than to any of the four mouse ten-m sequences (Oohashi, T., and R. Fässler, unpublished observations). Therefore, in addition to the ten-m family it is possible that a ten-a family of mammalian orthologues exists.

To be able to visualize potential Ten-m1 ligand(s) in tissue sections and show binding on Far Western blots, we expressed and purified a recombinant fusion protein in which the AP module was fused to the extracellular domain of Ten-m1. This fusion protein bound strongly to specific regions in several tissues which, especially in the nervous tissue, overlapped with the immunostaining. Far Western blots and electron micrographs indicated that like other membrane molecules in the nervous system, for example NCAM and NgCAM/L1 (Brümmendorf and Rathjen, 1996), Ten-m1 is able to interact homophilically. Such interactions could influence a wide variety of processes including cell migration, neurite extension, and fasciculation. However, since the staining patterns of immunohistochemistry and affinity histochemistry were not completely overlapping, heterophilic interactions of Ten-m1 with other, so far unidentified ligands are likely to occur also. Additional heterophilic interactions have been observed as well for NCAM and especially for NgCAM/L1 which is able to interact with several other molecules including laminin, proteoglycans, other neural cell adhesion molecules, and integrins (Brümmendorf and Rathjen, 1996).

The presence of an RGD containing motif in Drosophila Ten-m/Odz together with results from cell adhesion and cell spreading assays using Ten-m/Odz–derived peptides suggested an interaction with specific splice variants of the Drosophila PS2 integrins (Graner et al., 1998). We could not observe an altered adhesion or spreading behavior when we did adhesion assays with a β1 integrin– deficient cell line (Fässler et al., 1995) and recombinant mouse Ten-m1 (Fässler, R., unpublished observations). Furthermore, none of the mouse ten-m proteins contains an RGD motif which is frequently used for integrin engagement (Ruoslahti and Pierschbacher, 1987).

The mechanism by which ten-m molecules may transduce signals from the extracellular environment into the cell is not obvious. The activation of many signaling receptors is mediated by ligand-induced dimerization. Since ten-m molecules are already dimerized before ligand binding, an alternative activation could be associated with a conformational change induced by ligand binding which allows intracellular protein interactions and phosphorylation. A similar signaling cascade has been suggested for integrins (Schwartz et al., 1995). Furthermore, the cytoplasmic domains of the ten-m proteins contain several proline-rich regions which could serve as docking domains for intracellular SH3-containing proteins (Mayer and Eck, 1995).

Another way to transmit signals into the interior of the cell could occur via proteolytic cleavage of the ten-m molecules in the linker domain close to the cell surface. All mouse ten-m proteins contain several dibasic sites in the linker domain and one of them is conserved in all sequences as well as in the Drosophila Ten-m/Odz and Ten-a. Dibasic sites have been observed frequently as recognition sites for a family of mammalian proteases with similarity to bacterial subtilisin (Barr, 1991). A proteolytic cleavage in this domain would release the cytosolic part of the molecule from a dimeric into a monomeric state, which might alter its ability to interact with other molecules or its accessibility for kinases and phosphatases. Evidence for a proteolytic cleavage was reported for Drosophila Ten-m by Baumgartner et al. (1994). Another observation which indicates proteolytic processing events in the linker domain is the occurrence of a soluble form of γ-heregulin in the supernatant of the breast cancer cell line MDA-MB-175 (Schäfer et al., 1996). Since γ-heregulin and mouse Ten-m4 share marked homologies in their NH₂ termini, including their linker regions, the potential for shedding of the extracellular domain may be a characteristic of all the ten-m family members.

Do signaling cascades also modulate pair rule gene activity in Drosophila? During the syncytial blastoderm stage gap genes and pair rule genes are directly regulated by the spatial and temporal expression of upstream transcription factors. In the cellular blastoderm stage, when cell membranes form and cellular boundaries are established, the regulation of gene expression may become dependent on signal transduction. In elegant genetic and biochemical experiments it was shown that several pair rule genes including even-skipped, fushi tarazu, and runt are regulated by the soluble ligand unpaired which activates a Janus kinase (JAK; hopscotch)/signal transducer and activator of transcription (STAT; Stat92E) signaling pathway (Binari and Perrimon, 1994; Hou et al., 1996; Yan et al., 1996; Harrison et al., 1998). Although current models of the JAK/STAT signaling pathway propose that the activation of JAK, phosphorylation of STAT, and finally regulation of gene transcription (Darnell, 1997) are preceded by ligand-induced dimerization or multimerization of transmembrane receptors, experimental evidence for this speculation is lacking, and other signal transduction mechanism, like conformational changes in transmembrane molecules, cannot be ruled out.

In Drosophila, pair rule genes including ten-m/odz initiate a genetic program which is essential for the normal segmentation. The expression pattern of Ten-m/Odz in adult flies suggests further functions in many organs which have not yet been elucidated. Our data using immunostaining of various tissues derived from adult mice also indicate that the ten-m family has a widespread expression. At present, however, we are unable to exclude that the antiserum raised against the extracellular domain of Ten-m1 does not cross-react with other family members.

A possible function for Ten-m4 has been reported by Wang et al. (1998) who searched for stress response gene which are dependent on the activity of the transcription factor CHOP. Using representational difference analysis, they isolated the DOC4 cDNA which is identical to the ten-m4 cDNA. Although DOC4/Ten-m4 expression is induced by cellular stress, it is not clear which function is propagated by this protein. To obtain direct evidence for possible roles of ten-m proteins during development and disease we have isolated the mouse ten-m genes and have generated mutant mice. The analysis of these mice will provide further insights into the function(s) of these interesting molecules.

We thank Dr. Karl-Heinz Mann for amino acid sequencing, Hanna Wiedemann for electron microscopy, Dr. J. Flanagan for the AP module, Drs. Stefan Baumgartner, Anders Björklund, John Bateman, Ray Boot-Handford, and Cord Brakebusch for the stimulating discussion and for carefully reading the manuscript, and Yoshifumi Ninomiya for supporting Toshitaka Oohashi.