An Oligodendrocyte Cell Adhesion Molecule at the Site of Assembly of the Paranodal Axo-Glial Junction

The rapid rate of nerve impulse conduction, which is characteristic of myelinated fibers in the central nervous system (CNS), is dependent on the high density of sodium channels at the node of Ranvier (Ritchie and Rogart 1977). Therefore, clustering of these ion channels is a key event in the morphological and functional maturation of the vertebrate CNS. Oligodendrocytes, the myelin-forming glia of the CNS, are believed to have an important role in sodium channel clustering (Vabnick and Shrager 1998). One proposal is that oligodendrocytes exert their influence on neurons by means of a secreted factor (Kaplan et al. 1997). More recent work has highlighted the significance of axo-glial contact adjacent to the node as a prerequisite for the assembly of a functional node (Rasband et al. 1999). These paranodal axo-glial adhesive zones bear a strong morphological similarity to the septate junctions found in invertebrates (for review see Salzer 1997).

In addition to their putative involvement in defining the site of the node of Ranvier, paranodal axo-glial junctions are also believed to function as a barrier to ion flow in the mature nerve (Chiu and Ritchie 1981). However, in spite of their probable importance, the molecular basis of axo-glial interaction at the paranodal junctions is still poorly understood. Paranodin/Caspr1 (contactin-associated protein 1) is an axonal constituent of the axo-glial adhesion zone and has been proposed to play an important role in paranodal junction formation (Einheber et al. 1997; Menegoz et al. 1997; Peles et al. 1997a,Peles et al. 1997b). Support for this view has come from studies on the nerves of Shiverer (Shi) mouse mutants and of mice deficient in the enzyme UDP-galactose-ceramide galactosyltransferase (CGT) in which Caspr1 fails to concentrate at the paranodes (Dupree et al. 1999; Rasband et al. 1999). However, the glial counterparts of Caspr1 have yet to be identified.

Neurofascin is an ankyrin-binding, cell adhesion molecule of the L1 subgroup of the immunoglobulin G superfamily that has been implicated in neurite outgrowth, fasciculation, and interneuronal adhesion (Rathjen and Schachner 1984; Grumet et al. 1991; Volkmer and Rathjen 1992; Volkmer et al. 1992; Davis et al. 1993; Zhang et al. 1998). The neurofascin gene is subject to extensive alternative splicing (Hassel et al. 1997), and two of the best characterized products are a 186-kD isoform (NF186), typical of myelinated axons, and found at nodes of Ranvier and axonal initial segments, and a 155-kD isoform (NF155), which has been localized to unmyelinated axons in the peripheral nervous system (PNS; Davis et al. 1996). Compared with NF 186, NF155 contains an extra fibronectin type III (FNIII) repeat and lacks a mucinlike domain (Davis et al. 1996).

All functional studies of neurofascin have sought to elucidate its role in neurons (Rathjen and Schachner 1984; Rathjen et al. 1987a,Rathjen et al. 1987b; Brummendorf and Rathjen 1996; Brummendorf et al. 1998). However, there is some evidence that neurofascin might be expressed in cells that have been identified, on the basis of their morphology, as oligodendrocytes (Moscoso and Sanes 1995). We have confirmed this observation and have suggested that the specific isoform expressed in oligodendrocytes is NF155 (Collinson et al. 1998). Expression of neurofascin mRNA is first detectable at the initiation of myelination, but once the ensheathment of target axons begins in earnest, the amount of neurofascin mRNA in oligodendrocytes declines precipitously. This transient peak of expression at the onset of myelination in the CNS suggested that neurofascin might be a strong candidate to mediate and signal axon–glial interaction. Here, we show that NF155 is targeted to the paranodal loops of the sheath, where they interact with Caspr1 on the axon. Hence, NF155 is the first glial cell adhesion molecule of the paranodal junctional complex to be described.

Antibodies against neurofascin were generated in New Zealand white rabbits against three peptides coupled to keyhole Limpet hemocyanin by means of a terminal cysteine (Research Genetics). The first antibody (NFC1) was raised against a region at the extreme COOH terminus of the intracellular domain common to both NF186 and NF155 (CGNESSEATSPVNAAIYSLA), the second (NFF3) to the third FNIII domain specific to NF155 (CLWVSQKRQQASFPGDRPR), and the third (MNF2) to a sequence within the mucin domain specific to NF186 (CTESPTTTTGTKIHETAPDEQS).

Each antibody was affinity-purified by immunoabsorption to a column of peptide coupled to aminohexyl Sepharose 4B (Sigma Chemical Co.). Other primary antibodies used were against myelin-associated glycoprotein (MAG; mouse mAb; Boehringer Mannheim), ankyrin_G (mouse monoclonal; Zymed Laboratories Inc.), phosphorylated neurofilament M NFM (mouse monoclonal, RMO108; a gift of Dr. V.M.-Y. Lee, University of Pennsylvania, Philadelphia, PA) contactin (rabbit antiserum, a gift from Dr. J. Trotter, Department of Neurobiology, University of Heidelberg, Germany), and phosphotyrosine (mouse monoclonal MCPY; Upstate Biotechnology). The cDNA of the entire Caspr1 cytoplasmic region was used to generate rabbit and guinea pig antisera, as has been described elsewhere (Menegoz et al. 1997; Peles et al. 1997b). In brief, the primers that flank the cytoplasmic region were used in reverse transcriptase (RT)–PCR of RNA from adult rat brain stem and cerebellum. The resulting cDNA was cloned into pGEX4T2 (Amersham Pharmacia Biotech) and verified by sequencing. The fusion protein expressed in Escherichia coli (strain BL21; Stratagene) was purified on glutathione-Sepharose (Amersham Pharmacia Biotech).

Peroxidase-coupled antisera were purchased from the Scottish Antibody Production Unit. Antisera and affinity-purified antibodies (NFC1, 1:2,000; NFF3, 1:100; and MNF2, 1:100) were tested by Western blot analysis as described previously (Gillespie et al. 1994). For immunoprecipitation, optic nerves or brains were removed from 10- and 12-d-old mice and homogenized in 250 μl ice-cold lysis buffer (25 mM Hepes, 10 mM EDTA, 5 mM NaF, 15 mM Na₄P₂O₇·10 H₂O, 1 mM benzamidine, 0.5 mM Na₃VO₄, peptide inhibitors [P8340; Sigma Chemical Co.], and 1 mM PMSF, pH 7.4). The homogenate was solubilized for 5 min on ice by adding an equal volume of 2% Triton X-100 in lysis buffer. After centrifugation at 4°C for 15 min at 13,000 g, NFC1 or Caspr1 antibody at a dilution of 1:100 coupled to 5 mg of protein A agarose was added to the cleared supernatant and tumbled for 2 h at 4°C. The immune complexes were harvested by centrifugation and washed twice with lysis buffer and twice with 0.1% Triton X-100 in lysis buffer. Proteins were solubilized, resolved on 8% SDS polyacrylamide gels, and transferred electrophoretically to Protran nitrocellulose (Schleicher & Schuell) as previously described (Gillespie et al. 1994). The membrane was blocked overnight in 5% milk solids and probed with NFC1 (1:2,000), anti-Caspr1 (1:1,000), or anticontactin (1:1000) diluted in PBS containing 0.2% gelatin, 0.1% Tween 20, pH 7.4; followed by goat anti–rabbit IgG-HRP conjugate (1:1,000), or with MCPY diluted (1:1,000) in 3% milk solids in PBS, followed by goat anti–mouse IgG-HRP conjugate (1:2,000). Conjugates were detected using the ECL system according to the manufacturer's instructions (Amersham International).

Dorsal root ganglion (DRG) cultures were prepared following the procedure described by Bunge and co-workers (Kleitman et al. 1991). In brief, 12-mm glass coverslips, inserted into 4-well dishes (NUNC), were coated at room temperature for 1 h with Matrigel (Collaborative Biomedical products) diluted 1:30 in L15 medium. Spinal cords were removed from E16 Sprague-Dawley rat embryos, and the attached DRGs were removed and dissociated by trypsin treatment before plating on the matrigel-coated coverslips. Dissociated DRG were plated in C media (MEM, 10% FCS, 0.4% glucose, and 100 ng/ml NGF). After two cycles of antimitotic feeding (Kleitman et al. 1991), the cultures were reseeded with Schwann cells. Ascorbic acid (50 μg/ml) was added to the medium to trigger myelination. Primary Schwann cells were prepared from newborn rat sciatic nerves following the method described by Brockes (Brockes et al. 1979). Schwann cells were grown on poly-l-lysine–coated coverslips in DME containing 10% FCS.

CNS and PNS tissues were fixed by transcardial perfusion with 4% paraformaldehyde in 0.1 M sodium phosphate, pH 7.4. After extensive washing with 0.1 M sodium phosphate, pH 7.4, the tissue was cryoprotected by immersion for 15 min each in 5%, then 10% sucrose in 0.1 M phosphate buffer, pH 7.4, and finally overnight in 25% sucrose in 0.1 M phosphate buffer, pH 7.4, at 4°C. After embedding in OCT (Tissue TEK) blocks were frozen in isopentane cooled in liquid nitrogen. Sections (8–10 μm) were collected on 3-aminopropyltriethoxysilane subbed glass slides, and OCT was removed by washing in PBS (Sigma Chemical Co.). Sciatic nerve fibers were teased in 0.1% Triton X-100 in PBS. Before incubation with the NFF3 antibody, sections or teased sciatic nerve fibers were treated with Bouin's reagent for 1 min. Samples were blocked for 3 h in 0.2% Triton X-100, 0.2% gelatin in PBS, pH 7.4, containing either 10% nonimmune goat serum when using rabbit and mouse primary antibodies, or donkey serum when the sheep antibody was used. The antibodies were applied overnight in 4% nonimmune serum in the same buffer, and after washing them in buffer without serum, fluorescently labeled secondary antibodies were applied for 2 h in buffer containing serum, followed by further washes. Secondary antibodies were as follows: FITC-labeled goat anti–rabbit (Cappel Laboratories), TRITC-labeled goat anti–mouse IgG₁, TRITC-labeled donkey anti–rabbit (Jackson Laboratories), and FITC-labeled donkey anti–sheep (Jackson Laboratories). Further washes in blocking buffer minus serum were followed by several washes in PBS. Sections were mounted in Vectashield (Vector Laboratories). Images were captured with a Leica TCS4D confocal or Olympus BX60 microscope, and figures were produced with Adobe Photoshop.

Mice were perfused transcardially with heparinized (200 U heparin/100 ml) saline (0.89% NaCl) followed by freshly prepared 4% paraformaldehyde, 0.1% glutaraldehyde in 0.1 M sodium phosphate buffer, pH 7.4 for 2 h. The tissue was washed several times with 0.1 M sodium phosphate buffer, pH 7.4, containing 3.5% sucrose and stained with 0.25% tannic acid in the same buffer for 1 h at 4°C (subsequent steps were performed at 4°C). Aldehydes were quenched with 50 mM NH₄Cl in 0.1 M sodium phosphate buffer, pH 7.4. The tissue was washed four times in 0.1 M maleate buffer, pH 6.2, containing 4% sucrose, followed by 1% uranyl acetate in the maleate sucrose buffer for 1 h. The tissue was dehydrated to 100% ethanol (from 70% ethanol onward all steps were at −20°C), infiltrated with a 1:1 ratio of LR Gold (Agar Scientific Ltd.) and ethanol, followed by a 7:3 ratio of LR Gold and ethanol, and two changes of LR Gold for 2–3 h each. The tissue was infiltrated overnight in LR Gold containing 0.5% benzoin methyl ether and embedded in gelatin capsules. Polymerization was by UV irradiation at a wavelength of 365 nm for 24 h at −20°C.

Sections on formvar carbon–coated nickel grids were blocked for 40 min at room temperature with 1% BSA, 0.5% fish skin gelatin, 0.05% Triton X-100, 0.05% Tween 20 in 10 mM Tris buffer, pH 7.4, containing 500 mM NaCl, and incubated with NFC1 antibody in the same buffer at 4°C overnight. Grids were washed four times with the above buffer and incubated for 1 h with goat anti–rabbit IgG conjugated to 10 nm gold (1:20 Avrion). The grids were washed on drops of distilled water, fixed with 2.5% aqueous glutaraldehyde, and rinsed in a stream of distilled water. After postfixing for 15 min with 2% aqueous OsO₄, the grids were stained with uranyl acetate and lead citrate and examined at 80 kV in a Phillips Biotwin electron microscope.

RNA was prepared from adult mouse brains by TRIZOL extraction (Life Technologies) for RT-PCR. Reverse transcription was performed with 1 μg of RNA using SuperScript™II reverse transcriptase (400 U; Life Technologies) in the presence of random hexamers (Life Technologies). After incubation at 42°C for 45 min, reactions were terminated by heating at 80°C. DNA in 1 μl of the first strand reaction was amplified by PCR with Dynazyme (Flowgen). Primer sequences were selected from the published neurofascin (ankyrin-binding protein) sequence (available from GenBank/EMBL/DDBJ under accession No. L11002) (Davis et al. 1996) that would specifically amplify products from either the 155- or 186-kD splice variants of neurofascin (see Fig. 1). A probe specific for NF155 was amplified from within the third FNIII domain of neurofascin with a forward primer, 5′-CCTGAACAGCACAGCCATCAG-3′ (nucleotides 3,550–3,570) and a reverse primer 5′-GACCACAACCATCTCCTGCTTG-3′ (nucleotides 3,775–3,754). A probe specific for NF186 was amplified from within the mucin domain of neurofascin using a forward primer, 5′-TGTGAGCAGTACTGATGCTAC-3′ (nucleotides 4,189–4,209) and a reverse primer, 5′-GTGATGACGGTACTGCTGAATG-3′ (nucleotides 4,635–4,615). The products were cloned into the pGEM-T vector (Promega) and sequenced. The cDNA templates were linearized with either NotI or NcoI and transcribed with either T7 (antisense probes) or SP6 RNA (sense probes) polymerases (Boehringer Mannheim), respectively. The cRNA probes were labeled with [³⁵S]UTP and hybridized to sections (5–10 μm) of fresh mouse tissue that had been frozen in isopentane cooled to −70°C on liquid nitrogen as described (Collinson et al. 1998).

Neurofascin has been characterized as a neuronal cell adhesion molecule in the developing and mature CNS (Rathjen et al. 1987a,Rathjen et al. 1987b; Davis et al. 1996). However, we also demonstrated strong expression of the gene by in situ hybridization in oligodendrocytes (Collinson et al. 1998). Furthermore, it appeared that in the myelinating optic nerve, which does not contain neuronal mRNAs, the primary neurofascin transcript appeared to be NF155 rather than NF186 (Collinson et al. 1998). To establish if these two neurofascin isoforms were indeed expressed in a cell type–specific fashion in the CNS, we prepared probes for in situ hybridization and for immunocytochemistry that recognized the mucin (NF186) or the third fibronectin type III (NF155) domains (Fig. 1). In situ hybridization analysis of postnatal day 21 (P21) mouse brain with isoform-specific riboprobes demonstrated that NF186 mRNA was widely distributed with strong signals from certain neuron populations such as the hippocampus and the Purkinje cells of the cerebellum (Fig. 2 A). In contrast, NF155 mRNA was strongly expressed in white matter tracts (Fig. 2 B).

The differential expression of the two neurofascin isoforms was seen most dramatically in the retina and optic nerve. The neurons of the retina, including the ganglion cells, which project their axons to the optic nerve, were devoid of NF155 mRNA (Fig. 2 D), whereas each of the retinal neuron layers displayed strong signals with the NF186 probe (Fig. 2 C). In contrast, cells in the optic nerve itself, which were presumed to be oligodendrocytes, were strongly labeled with the NF155-specific riboprobe (Fig. 2 E). The use of a specific riboprobe confirmed the transient nature of NF155 expression during the most active period of myelination in the cerebellar white matter tracts that had been observed earlier with a nonspecific neurofascin probe (Fig. 2, F–H; Collinson et al. 1998). We found that, in a similar fashion, the expression of NF155 transcripts in the developing optic nerve also declined after a transient peak of expression (data not shown). All sense riboprobe controls yielded no discernible signal (data not shown).

To demonstrate that these cells were indeed oligodendrocytes, we generated neurofascin isoform–specific antibodies (Fig. 3 A). In optic nerve extracts, neither of these antibodies detected the minor 140-kD isoform of neurofascin found in other regions of the brain (Davis et al. 1996). NF186 protein was clearly neuronal, as demonstrated by strong expression in the plasma membrane and initial segments of the Purkinje cells of the cerebellum (Fig. 3 B, A and B) as previously shown by Bennett and co-workers (Davis et al. 1996). Double labeling for NF155 and an antibody against the oligodendrocyte marker MAG demonstrated that, at early stages of myelination, neurofascin was expressed in the cell bodies of oligodendrocytes (Fig. 3 B, C–E). Some punctate staining for NF155 was also observed which probably represented nascent paranodal labeling (see below). The NF155 antibody was shown to be specific for myelinating oligodendrocytes because, at the onset of myelination, those cell bodies detected by the NF155 antibody were always MAG-positive. This excludes expression of NF155 by astrocytes or oligodendrocyte precursors in the optic nerve.

In the myelinating optic nerve, NF155 is first observed at the cell body of the oligodendrocyte and in a diffuse pattern on the first processes of the myelin sheath as they engage the axon (Fig. 4 A, A and B). A dramatic restriction in the localization of the protein occurs over the next two days: the protein is no longer detectable in oligodendrocyte cell bodies, but it is concentrated in puncta on nerve fibers (Fig. 4 A, C and D). These foci of expression correspond to paranodes since they flank the nodes of Ranvier, at which ankyrin_G is concentrated in myelinated nerves (Fig. 4 B, A–C; Kordeli et al. 1995). In contrast, NF186 colocalizes with ankyrin_G, confirming its association with the axonal membrane at the node of Ranvier (Davis et al. 1996). Evidence that NF155 is a glial component of the paranodal axo-glial junction was provided by immunoelectron microscopy. In a longitudinal section (Fig. 5A, A′, B, and C) and in a tangential section (Fig. 5 D), the protein is seen to be localized at the base of the cytoplasm-filled paranodal loops, where they appose the axonal membrane. Although tyrosine phosphorylation inhibits the ankyrin-mediated association of neurofascin with the actin cytoskeleton (Tuvia et al. 1997), restriction of NF155 to paranodes in the optic nerve does not appear to require complete dephosphorylation since NF155 is phosphorylated at P12, a time when it has become restricted to the axo-glial junction (Fig. 6).

Caspr1/paranodin is known to be an axonal constituent of the paranodal axo-glial junction (Einheber et al. 1997; Menegoz et al. 1997; Peles et al. 1997a,Peles et al. 1997b). Further substantiation of the restriction of NF155 to the paranodal regions of mouse optic nerve axons was provided by its colocalization with Caspr1 (Fig. 7A and Fig. B), which was also observed at the paranodes formed by Schwann cells in mouse sciatic nerve (Fig. 7, C–E). Additionally, in the sciatic nerve, both antibodies decorated a helical ribbonlike structure (Fig. 7 E, arrow), which probably represents the interface between the inner mesaxon of the myelin sheath and the axon. In situ hybridization of the mouse sciatic nerve with an NF155-specific riboprobe demonstrated strong labeling of Schwann cells indicating that, as in the CNS, NF155 was a product of myelin-forming glial cells (Fig. 7 F). NF155 antibodies also stained the cytoplasm-filled channels in the sheath, the Schmidt-Lantermann incisures, where the protein colocalized with MAG (Fig. 7 G), but the apposed axon was not enriched in Caspr1 (data not shown). Immunostaining of cocultures of myelinating rat Schwann cells and dissociated DRG neurons (Fig. 7 H) revealed that the colocalized NF155 and Caspr1 at the paranodes could be resolved into a series of annuli, which probably reflects the lower density of packing of the paranodal loops at the axo-glial junction adhesion zone in this in vitro model of myelination by comparison with intact sciatic nerve fibers (Fig. 7C and Fig. H).

The Shiverer (Shi) mouse lacks the major structural protein myelin basic protein (Roach et al. 1985). Shi oligodendrocytes extend processes around axons in the CNS, but they are unable to produce compact myelin, neither do they assemble paranodal axo-glial junctions (Inoue et al. 1981; Rosenbluth 1981). The absence of paranodes is reflected by the abnormal distribution of Caspr1 (Rasband et al. 1999). Therefore, it was of interest to investigate how this disruption might affect the localization of glial neurofascin. In Shi mice, neurofascin displays an immature localization in that the protein is strongly expressed in oligodendrocyte cell bodies (Fig. 8G and Fig. J), and there is no indication that it is targeted to discrete domains that correspond to paranodes, as occurs in the nerves of wild-type littermates where neurofascin and Caspr1 colocalize (Fig. 8, A–F). In spite of their inability to establish a normal axo-glial junction, immunolabeling of Shi nerves for Caspr1 revealed that the protein was not distributed diffusely along the axon, but instead was concentrated in discrete patches (Fig. 8G, Fig. J, and Fig. M) where it frequently colocalized with NF155 (Fig. 8I, Fig. L and Fig. O). This colocalization suggested that Caspr1 and NF155 might still be apposed in the glial and axonal plasma membranes, albeit at an ectopic site. However, the fact that Caspr1 and contactin did not coimmunoprecipitate with NF155 from the brain using either neurofascin or Caspr1 antibodies suggests that Caspr1 and NF155 do not interact directly (data not shown).

The sites of interaction between the terminal loops of paranodal myelin and the axon are characterized by a narrow gap of 2.5–3 nm. These junctions represent highly specialized zones that are likely to have a distinctive complement of adhesion molecules. Furthermore, unlike the tightly compacted lamellae found in the rest of the myelin sheath, the paranodal myelin loops are filled with cytoplasm, which would facilitate the transduction of signals from these transmembrane proteins to the cell body of the myelin-forming glial cell. We have previously shown that the cell adhesion molecule E-cadherin is strongly expressed in these paranodal loops (Fannon et al. 1995). However, E-cadherin does not appear to be a component of the axo-glial junction, but is rather a constituent of adherens junctions at the proximal constrictions of the loops. Another member of the immunoglobulin superfamily, MAG, is also found in the paranodal loops in the PNS and in periaxonal membranes but is not concentrated at the junctional complex (Trapp et al. 1989a,Trapp et al. 1989b). Hence, the NF155 isoform of neurofascin is the first glial cell adhesion molecule of the axo-glial junction to be described.

What are the signals that initiate the development of this regional specialization? The axonal transmembrane proteins neurofascin and NrCAM, both members of the L1 cell adhesion molecule family, have been proposed as pioneer molecules in defining the site of node formation (Lambert et al. 1997). We have confirmed previous work that NF186 clusters at the node and is the axonal isoform of neurofascin (Davis et al. 1996). Localization of this isoform to the node of Ranvier is coincident with the first appearance of ankyrin_G in the myelinating optic nerve (data not shown). Hence, it is likely that these proteins play an important role in assembling the molecular machinery required for generating the action potential typical of saltatory nerve impulse conduction at the node as suggested previously by Bennett and co-workers (Davis et al. 1996; Lambert et al. 1997).

It seems likely that physical interaction between the oligodendrocyte and the axon also plays an essential part in defining the site of node assembly. The axo-glial adhesion zone and the invertebrate septate junction both contain members of the NCP (for neurexin IV, Caspr1/paranodin, and Caspr2) subgroup of the neurexin superfamily (Menegoz et al. 1997; Peles et al. 1997a,Peles et al. 1997b; Bellen et al. 1998; Poliak et al. 1999). Neurexin IV is a constituent of the pleated septate junctions of Drosophila that play an important role in mediating glial–glial interactions (Baumgartner et al. 1996), and Caspr1/paranodin has been shown to be a constituent of paranodal junctions in both the CNS and PNS (Einheber et al. 1997; Menegoz et al. 1997). In the developing optic nerve, it appears that the targeting of Caspr1 to prospective paranodes has a deterministic role (Rasband et al. 1999). Indeed, without correct axon–glial contact, ankyrin_G fails to localize to the node (Rasband et al. 1999). The clustering of sodium channels occurs after the paranodes have been defined and after ankyrin_G and NF186 have codistributed at the node (Rasband et al. 1999; and this work). Hence, one function of the paranodal junctions may be to prevent the diffusion of sodium channels from the nodes, where they are essential participants in the generation of the action potential. The interaction between Caspr1, neurofascin, and their respective ligands would play an important part in ensuring this restricted distribution.

The integrity of the paranodal adhesive junction between axons and myelin-forming glia is not only vital for the establishment of rapid rates of nerve impulse conduction during development, but it is also crucial for the maintenance of normal function in the mature nerve. Subtle derangements in the interactions between the paranodal loops of the myelin sheath and the axon represent the earliest structural correlates of decreased rates of nerve impulse conduction in myelinated fibers and may explain paralysis in some pathologically mild cases of demyelinating disease (Griffin et al. 1996). In support of the importance of the integrity of the axo-glial junction for normal function in myelinated fibers, a mouse mutant, which lacks the enzyme UDP-galactose-ceramide galactosyltransferase (CGT), and which is therefore unable to synthesize the major myelin constituent galactocerebroside, assembles compact myelin but has paranodal myelin loops that do not make structurally normal connections with the axon (Coetzee et al. 1996; Dupree et al. 1998, Dupree et al. 1999). Consequently, the conduction properties of the axons are severely compromised, probably as a result of the aberrant exposure of potassium channels in the juxtaparanodal domain of the axonal membrane to the perinodal ionic environment. This conclusion is supported by the positive effect of the potassium channel blocker 4-aminopyridine on the amplitude of nerve impulse conduction in the CNS fibers of CGT mutant mice (Coetzee et al. 1996).

In the myelinating optic nerve, the initiation of axo-glial adhesion assembly is a very early event, since paranodal junctions have been detected as soon as the first spiral of the oligodendrocyte process has ensheathed an axon (Wiggins et al. 1988). Hence, the early targeting of NF155 to the paranodal myelin loops in the optic nerve probably reflects the importance of this protein in the early establishment and stabilization of axon–glial contact. The mechanism by which this targeting is achieved has yet to be determined, but the COOH terminus of the neurofascins does contain a functional ankyrin-binding site that is regulated by tyrosine phosphorylation (Davis et al. 1993; Davis and Bennett 1994; Garver et al. 1997). Dephosphorylation is required for ankyrin binding, which would suggest that NF155 may only become fully competent to interact with the actin cytoskeleton after it has reached its paranodal destination, since phosphorylation of the protein in the optic nerve is still detectable after it has concentrated at the paranode. This would suggest that the actin cytoskeleton is more likely to be important for neurofascin's function at the paranode rather than in ensuring delivery of this cell adhesion molecule to the axo-glial junction.

The axonal ligand for NF155 has yet to be identified. At first sight, Caspr1 is a strong candidate, particularly in view of the fact that both proteins not only colocalize at normal paranodes, but they also codistribute at ectopic sites in the nerves of Shi mutant mice that are unable to assemble paranodes (Rasband et al. 1999). Furthermore, the two proteins codistribute in the PNS, where the inner mesaxon of the myelin sheath apposes the axon. However, these proteins do not colocalize at the Schmidt-Lantermann incisures, which are cytoplasm-filled channels in the compact peripheral myelin sheath, neither do they coimmunoprecipitate from the brain. Hence, it seems likely that, although members of a supramolecular adhesion complex, they are unlikely to interact directly. The other elements of the paranodal axo-glial junction remain to be identified. In view of the importance of this unusual type of mammalian adhesion complex for the cell biology of the neuron, it will be of great interest to further establish its molecular structure. The identification of the glial isoform of neurofascin as a constituent of the junction is a step towards this goal.

We thank Mary Davie (University of Edinburgh) for excellent technical assistance. Shi mice were generously supplied by Professor Ian Griffiths (Glasgow University, Glasgow, UK).

The Medical Research Council, the Multiple Sclerosis Society of Great Britain and Northern Ireland, and the Wellcome Trust are thanked for financial support. D.R. Colman is supported by the National Multiple Sclerosis Society and by grants from the National Institutes of Health (NS20147).