Neurothelin: molecular characteristics and developmental regulation in the chick CNS

The blood-brain barrier represents a complex system of active and passive exclusion and transport mechanisms subserving highly sensitive brain functions such as synaptic activity. Passage of blood components around or through endothelial cells forming the capillaries is eliminated or reduced by the presence of tight junctions and the low number of pinocytotic vesicles in endothelial cells (Bradbury, 1984). Exclusively endothelial cells rather than pericytes or astrocytes of CNS capillaries have been found to form the blood-brain barrier (Reese and Karnovsky, 1967). However, endothelial cells seem to depend on a direct cell contact with astrocytes or soluble astrocytic factors in order to express blood-brain barrier characteristics (DeBault and Cancilla, 1980; Maxwell et al. 1987; Arthur et al. 1987; Janzer and Raff, 1987). On the molecular level, the blood-brain barrier is still poorly defined, although hybridoma technology has provided immunological tools for the identification of distinct proteins that are possibly involved in formation of the blood-brain barrier and carrier functions (Jefferies et al. 1984; Risau et al. 1986).

We have generated the monoclonal antibody 1W5, which binds to the cell surface protein neurothelin (Schlosshauer and Herzog, 1990). Neurothelin is expressed in blood-brain-barrier-specific endothelial cells of the chick brain. In addition, it can be induced in neurothelin-negative endothelial cells, if nonvascularized mouse brain tissue is transplanted onto systemic blood vessels of the chorioallantoic membrane of the chicken in vivo. Under these conditions, blood capillaries of the chicken sprout into the mouse transplant and start to express neurothelin. Localization of neurothelin on the luminal surface of blood capillaries suggested that it might be directly involved in barrier/carrier functions. In addition, it has been found on neurons of the retina, which is part of the central nervous system but lacks blood vessels (Ramon Y Cajal, 1933). Therefore neurothelin appears to be an essential component of all CNS regions, being synthesized either by vascular endothelial cells or by neurons themselves if no blood vessels are present.

A cellular difference between the chick retina and brain is the absence of astrocytes in the retina (Ramon Y Cajal, 1933). This prompted us to speculate whether neurothelin expression in endothelial cells and neurons might be induced or suppressed by astrocytes. In the brain, astrocytes contact neuronal and endothelial cells (Rhodin, 1975).

The absence of astrocytes as well as contaminating cells of mesenchymal origin in the retina, make the chick retina an ideal model system to investigate regulatory cell interactions for neurothelin gene expression. In this study, we present data showing that purified astrocytes suppress to a certain degree neurothelin expression in retinal neurons in vitro.

For a number of cell surface proteins, post-translational modifications have been shown to influence the function of these proteins. Saccharide moieties have been found to be involved e.g. in adhesion processes (Stoolman, 1989; Sadoul et al. 1983) and affect the proteolytic resistance of the corresponding glycoproteins (Sharon, 1981). In addition, the function of a given cell surface protein can be regulated by the way the protein is associated with the cytoplasmic membrane. The lipid-anchored isoform of the neural cell adhesion molecule (N-CAM₁₂₀) has a high lateral mobility in the plane of the cytoplasmic membrane, whereas the transmembrane isoform (N-CAM₁₈₀) is preferentially located at the contact area between neighboring cells (Pollerberg et al. 1986, 1987). The latter one is associated with spectrin. Therefore N-CAM₁₈₀ has the potential to focus elements of the cytoskeleton at membrane regions where cell-cell interactions take place. To characterize neurothelin in more detail, a number of biochemical studies are presented concerning cell surface association and glycosylation.

Chemicals used were purchased from Merck (Darmstadt) if not stated otherwise. Homogenization and solubilization buffers contained the enzyme inhibitors 2,3-dehydro-2-deoxy- N-acetylneuraminic acid, aprotinin, leupeptin, pepstatin and phenylmethylsulfonyl fluoride (Schlosshauer, 1989). The 180 × 10³M_r isoform of the neural cell adhesion molecule was purified as described (Schlosshauer, 1989). mAb Dl (Schlosshauer, 1989), mAb 1W5 (Schlosshauer and Herzog, 1990); and mAb Q211 (Rohrer et al. 1985) (generously provided by Dr S. Henke-Fahle), serum against the glial fibrillary acidic protein (Dakopatts) and a polyclonal serum specific for the F11-antigen (Rathjen et al. 1987) (generous gift of F. Rathjen) were used at a concentration of approx. 1 –10 μg ml ⁻¹ for histological and biochemical investigations (16h, 4 °C). Fluorescein- and rhodamine-conjugated F(ab’)₂ fragment goat anti-mouse IgG (heavy and light chain) and peroxidase-conjugated F(ab)₂ fragment goat anti-mouse (Dianova, FRG) rhodamine B isothiocyanate (Sigma), laminin (Gibco), phosphatidylinosit-phospholipase C, endoglycosidase H and endoglycosidase F/N-glycosidase F were from Boehringer, test-neuraminidase (of Vibrio cholerae) from Behring. The glycan differentiation kit (Boehringer) contained the digoxigenin-conjugated lectins GNA (Galan- thus nivalis agglutinin specific for terminal mannose), SNA (Sambucas nigra agglutinin specific for sialic acid linked α (2 –6) to galactose), MAA (Maackia amurensis agglutinin specific for sialic acid linked α (2 –3) to galactose), PNA (peanut agglutinin specific for galactose β (1 –3) N-acetylgalactosamine), DS A (Datura stramonium agglutinin specific for galactose β (1 –4) N-acetylglucosamin) and the control glycoproteins carboxypeptidase Y, transferrin, fetuin and asialofetuin as well as alkaline-phosphatase-conjugated sheep Fab fragments anti-digoxigenin (Boehringer).

8-well slides (Flow) or 96-well microtiter plates were coated consecutively with poly-L-lysine and laminin (20 μg ml ⁻¹ PBS pH7.4, 2h, 37 °C, each). Astrocytes purified from neonatal (P2-P7) rat cerebral cortex (Raff et al. 1985) were plated as a confluent monolayer onto poly-L-lysine/laminin. Human skin fibroblasts and pig smooth muscle cells (all provided by Dr H. Hammerle, NMI/Reutlingen) were plated directly onto tissue culture plastic. Chick retina cells were added to the cultures one day after these cell types had been plated into microtiter dishes. Retinal and tectai cell cultures of chick embryos were obtained by trypsinization of the corresponding tissues (Schlosshauer et al. 1988; Schlosshauer, 1989) and plating single cells onto poly-L-lysine/laminin with or without astrocytes. The culture medium was based on F12 medium containing fetal calf serum (Walter et al. 1987).

For retinal-tectal cocultures, tectal cells were prelabeled as follows (F. Bonhoeffer, personal communication). Embryonic tectal tissue was cut into 300 μm² pieces using a tissue chopper (McIlwain). 20 μl rhodamine B isothiocyanate (RITC) in dimethylsulfoxide (50 mg ml ⁻¹) were added aseptically to 2 ml Hanks balanced salt solution (HBSS) by vortexing vigorously. 100 μl of this solution was diluted in 5 ml HBSS (final RITC concentration: 200 ng ml ⁻¹) and used for labeling tissue fragments or single cells (15min, 22 °C). Thereafter tissue fragments and cells were washed twice in HBSS before further processing (see above).

4% paraformaldehyde-fixed cryosections of chicken heads or retinae from different developmental stages were used for immunofluorescence staining as described (Schlosshauer et al. 1988). Incubation of paraformaldehyde-fixed cell cultures with monoclonal antibodies or polyclonal serum for immunofluorescence assays was performed for 16 h, 4 °C. The developmental expression of neurothelin in the retina was quantified using tissue homogenates (10 μg ml ⁻¹) from embryonic day E6, E10, E18, postnatal day P42 and adult. Retina tissues were homogenized at 4 °C (20 strokes, glass homogenizer), sonicated (three times, 20 s, 40 watts) and finally filtered through paper filters (Schleicher and Schüll) before protein quantification. For ELISA assays, horse-radish-peroxidase- conjugated secondary antibodies were employed using 1,2- phenylendiamine as a chromophore. Incubation times for antigen and antibody binding were 1 h each. Neurothelin data from in vitro assays were normalized for cell proliferation and cell surface expansion using two additional monoclonal antibodies specific for cell surface antigens (Dl, Q211; see above). In each of three independent assays quadruplets were determined.

Affinity purification of neurothelin from cytoplasmic membranes ofchick retina E9/E10, solubilized with 1 % Triton X100 in 10mM Tris-HCl, pH 7.4, 150 mM NaCl, was performed as described (Schlosshauer and Herzog, 1990). For characterization of cell surface proteins cytoplasmic membrane preparations (Hoffman et al. 1982) of chick retina E9 were used. Temperature-induced Triton X-114 phase separation (Bordier, 1981; Schlosshauer, 1989), analysis of lipid- anchored membrane proteins by phosphatidylinositphospho- lipase C (Wolff et al. 1989) and removal of peripheral cell surface proteins (Fujiki et al. 1982) at pH 11.5 using 100 mM Na₂CO₃ were performed as described earlier. For twodimensional gel electrophoresis (O’Farrell, 1975) Servalyte 3 –10, 4 –6 and 5 –8 (Serva) were employed in a ratio 1:2:2. Deglycosylation of neurothelin was performed according to the manufacturer’s instructions. For treatment with neuraminidase (Schlosshauer et al. 1984) and endoglycosidase F, neurothelin immobilized to mAb 1W5 conjugated CNBr- Sephadex beads (Pharmacia) was employed. Solubilized neurothelin, which had been purified by affinity chromatography and preparative gel electrophoresis, was used for endoglycosidase H digestion. To characterize polysaccharide side chains, purified neurothelin was dotted onto nitrocellulose strips and consecutively incubated with digoxigenin- conjugated lectins and HRP-conjugated anti-digoxigenin antibodies according to the manufacturer’s instructions (Boehringer).

Protein quantification was performed as described (Bradford, 1976; Lowry et al. 1951). Polyacrylamide gel electrophoresis (Laemmli, 1970), silver staining of SDS gels (Ansorge, 1985) and western blotting (Towbin et al. 1979; Hawkes et al. 1982) were employed for qualitative and quantitative analysis.

At early stages of embryogenesis (embryonic day 4: E4) neuroblasts of the central nervous system express neurothelin as revealed by indirect immunofluorescence of paraformaldehyde-fixed cryosections (Fig. 1A). During the next 5 days of cytodifferen- tiation, neurothelin expression declines gradually in differentiating neuroblasts. At E7 young blood vessels are still devoid of neurothelin. However, by E10 at a more advanced state of differentiation, blood vessels of the brain are clearly labeled by mAb 1W5 specific for neurothelin, whereas other tissue components such as neurons are neurothelin negative (Fig. IE). This cellular distribution of neurothelin remains constant and is maintained similarly after hatching (Fig. 1G). Therefore, during brain development, neurothelin expression is characterized by a switch from neuroblast to vascular endothelial cells.

The complementary developmental profiles of neurothelin expression in the two cell types raised the question whether an indirect interaction between angiogenesis and neural differentiation might exist. The microenvironment formed by glial cells as a connecting link between neural and endothelial cells could represent the suppressive element for neural neurothelin synthesis. Therefore we disintegrated the three-dimensional cytoarchitecture of the brain and investigated neurothelin expression of neurons in vitro. The nearest neighbor relationships between cells were eliminated by trypsinization and individualized brain cells were plated onto a polylysine/laminin substratum and immunolabeled after various times. To demonstrate that brain cells (E8/E9) express initially only minute amounts of neurothelin, cells of the tectum opticum were prelabeled with rhodamine-isothiocyanate and cocultured with unlabeled retinal cells as a positive control. After 14 h in vitro the coculture was immunolabeled with mAb 1W5. Rhodamine-labeled tectal cells (Fig. 2B) were essentially devoid of neurothelin in contrast to retinal cells (Fig. 2A) (see also Fig. 3). However, if tectal cells were kept in culture for several days, neurons of the tectum opticum became neurothelin positive. Fig. 2C shows a culture originating from the tectum only. Tectal neurons have partially aggregated, and long axon fascicles have grown between small cell colonies. Neural cell bodies as well as neurites are clearly labeled with mAb 1W5, although the labeling was not as pronounced as with retinal cells. The cellular identity of these cells was confirmed by the neuron specific antibodies mAb Dl and Q211 (Schlosshauer, 1989; Rohrer et al. 1985) as well as by morphological criteria. These data reveal that, once the regular microenvironment in vivo is destroyed, the developmental decrease of neurothelin expression in neurons can be reversed in vitro.

Immunofluorescence micrographs of mAb lW5-labeled chick retinae suggested that retinal neurons express more neurothelin at the end of histogenesis rather than in the beginning (Schlosshauer and Herzog, 1990). To confirm this finding quantitatively, ELISA assays were performed with retina samples from different developmental stages (Fig. 3). During embryogenesis the neurothelin concentration of the retina increases by a factor of 7. (Incubation time in ovo is 21 days). The highest value is reached after hatching at postnatal day 42 (P42), when the neurothelin concentration is one order of magnitude higher than at E6. In adulthood, a slight decrease in comparison to the posthatching period is evident resulting in a value 9 times higher than during the first week of incubation.

The quantification demonstrates that the developmental sequence in retinal neurons is in complete contrast to that of other CNS neurons, where neuro-thelin expression declines with time. Therefore, we questioned which cellular mechanism might be responsible for the difference and whether this difference might be based on distinct histological features of the retina.

One cellular difference between brain tissues such as the tectum opticum and the retina is the absence of astrocytes in the chicken retina (Ramon Y Cajal, 1933). This prompted us to speculate whether astrocytes might suppress neurothelin expression in neurons of the brain. In addition, the absence of astrocytes in the eye could also explain the synthesis of neurothelin in adult neurons of the retina.

To test the hypothesis of a suppressive action by the astrocytes, we cultured neurothelin-positive retinal celts either on a polylysine/laminin substratum or on a preformed monolayer of purified rat astrocytes. After in vitro incubation for two days, cultures were fixed and immunolabeled with mAb 1W5 for cytological analysis and ELISA. mAb Dl and mAb Q211 were employed to identify neurons. The same two antibodies which bind to cell surface glycoproteins and glycolipids, respectively, served also to normalize neurothelin data according to standard procedures (Lorenz, 1984) for neuron proliferation and cell surface expansion due to neurite outgrowth. These precautions were taken because neurothelin is also a cell surface protein (Schlosshauer and Herzog, 1990). Astrocytes were identified with the aid of antiserum specific for the glial fibrillary acidic protein (GFAP) (Fig. 4).

Quantitative evaluation of these cultures revealed that astrocytes reduce neurothelin expression in retinal cells from 100 % to 65 % (Fig. 5). This suggests that rat astrocytes are capable of modifying neurothelin gene expression in chick neurons to a certain extent.

In control cocultures retinal cells were plated onto fibroblasts and smooth muscle cells. In these cases, dishes had not been precoated with polylysine/laminin. Quantification of antigen expression after two days in vitro indicated that retinal cells expressed 7 –8% more neurothelin on fibroblasts and smooth muscle cells than on polylysine/laminin. In summary, this suggests that laminin does not specifically induces neurothelin synthesis. In contrast, astrocytes but not fibroblasts or smooth muscle cells appear to suppress neurothelin expression.

Immunolabeling of vital cell cultures indicated that neurothelin is a cell surface protein (Schlosshauer and Herzog, 1990). To determine whether neurothelin is inserted into or only superficially associated with the cytoplasmic membrane, three different biochemical protocols were employed (Fig. 6A). Temperature- induced detergent-phase separation enriches integral membrane proteins (Bordier, 1981). Using Triton X-114 to solubilize cytoplasmic membranes, a homogeneous solution is formed at 4 °C, whereas a phase separation occurs at 22 °C. Upon centrifugation the lower phase contains detergent and integral membrane proteins with hydrophobic domains. Other proteins remain in the upper phase. Investigating retinal membranes this way, in conjunction with western blot analysis, reveals that neurothelin is clearly enriched in the detergent phase (Fig. 6B).

To determine whether major neurothelin isoforms exist that are linked to the membrane by phosphatidyl- inositlipid, purified cytoplasmic membranes from neurothelin-rich chick retinae were exposed to phosphatidylinosit-phospholipase C. Phospholipase C has been described to free phospholipid-anchored proteins from membranes (Wolff et al. 1989). After enzyme digestion, solubilized and particulate components were separated by ultracentrifugation. Both fractions were subjected to western blot analysis using mAb 1W5 and an antiserum specific for the 135 × 10³ glycoprotein F11 (Rathjen et al. 1987) (Fig. 6C). F11, which served as a positive control, was preferentially found in the centrifugation supernatant, suggesting that a major fraction of F11 had been enzymatically detached from retinal membranes. Control samples without enzyme did not remove significant amounts of F11 from membranes (data not shown). In contrast, neurothelin remained completely in the membrane fraction (ultracentrifugation pellet). This suggests that neurothelin has no phospholipid anchor.

In the third approach, peripheral membrane proteins were removed from purified retinal cell membranes under strongly alkaline pH conditions (pH 11,5) (Fujiki et al. 1982). Solubilized and particulate fractions were analysed as before. Again, neurothelin was found only in the ultracentrifugation pellet but not in the supernatant (Fig. 6D).

In summary, these data indicate that neurothelin is an integral membrane protein. Presumably, no neurothelin isoforms exist that are typical for phosphatidyl-inosit anchored or peripheral membrane proteins.

The isoelectric point of neurothelin was determined by two-dimensional gel electrophoresis (O’Farrell, 1975). Detergent solubilized retinal membranes were subjected to isoelectric focusing in the first dimension and SDS-gel electrophoresis in the second dimension. Western blots of these gels were labeled with mAb 1W5, specific for neurothelin, or indian ink to reveal the complete protein pattern of the sample (Fig. 7). Neurothelin was found to represent a group of six major isoforms in the pH range of 6.2 to 6.6. The microheterogeneity is characterized by a more or less linear relation between relative molecular mass and isoelectric point: the higher the mass the lower the isoelectric point. Acidic side chains could be responsible for the increasing molecular masses and decreasing isolectric points (see below). On indian ink stained blots, no major proteins could be observed in the area of neurothelin spots, suggesting that neurothelin represents possibly only a minor protein fraction of the cytoplasmic membrane.

The observed microheterogeneity could be based on different glycomoieties attached to distinct isoforms of neurothelin. Three deglycosylation experiments were performed in order to reveal whether a reduction of the relative molecular mass could be induced. Affinity- purified neurothelin as well as the neural cell adhesion molecule (N-CAM) as a positive control were exposed to neuraminidase to remove the saccharide neuraminic acid. After SDS-gel electrophoresis no significant band shift was observed for neurothelin, whereas for embryonic N-CAM a reduction of the relative molecular mass of approx. 70 ×10³ occurred (Fig. 8A). Endoglycosidase H (Endo H) reduced the relative molecular mass of neurothelin from 41 ×10³ to 37 ×10³ (Fig, 8B), and Endoglycosidase F (Endo F) from 41 ×10³ to 25 ×10³ (Fig. 8C). These data indicate that neurothelin has some polysaccharide side chains of the high mannose type (split by Endo H). However, most side chains are of the complex type (split by Endo F) (Elder and Alexander, 1982). Intriguing is the drastic reduction of the relative molecular mass after Endo F treatment demonstrating that neurothelin is (1) highly glycosylated, and (2) has a relatively small proteinaceous backbone, which represents less than 60% of the glycoprotein.

After the general classification of polysaccharide side chains, a detailed analysis of neurothelin was performed using lectins. Affinity-purified neurothelin and four positive control glycoproteins (carboxypeptidase Y, transferrin, fetuin and asialofetuin) were dotted repeatedly onto nitrocellulose strips. Thereafter the strips were incubated with five lectins specific for different saccharide units (Fig. 9). Because the lectins were coupled to digoxigenin, a very sensitive detection reaction with a horse-radish-peroxidase-conjugated antibody specific for digoxigenin could be employed to determine the neurothelin saccharide composition. Neurothelin was recognized by Galanthus nivalis agglutinin, Sambucas nigra agglutinin and Datura stramonium agglutinin. Therefore, neurothelin side chains contain terminal mannose and galactose (1 –4)- N-acetyl-glucosamine but no galactose (1 –3)-N-acetyl- galactosamine. Sialic acid was found only coupled in 2 –6 positions with galactose but not in 2 –3 positions. Because neuraminidase treatment of neurothelin does not result in a significant reduction of the relative molecular mass (see Fig. 8A), the amount of sialic acid bound appears to be quite limited.

Glial cells appear to play a key role in neurothelin gene expression as has been initially suggested by histological and developmental data. Astroglia differentiate from radial glia which span the entire thickness of the early embryonic brain wall. The gradual transformation into astrocytes is accompanied by a switch of cytoskeleton proteins: vimentin in radial glia cells changes to glial fibrillary acidic protein in astrocytes (Voigt, 1989). Glutamine synthetase as astroglia marker is found as early as embryonic day 9 (E9) in the tectum opticum of the chicken. Other astrocyte-specific antigens (glial fibrillary acidic protein and carbonic anhydrase) become expressed during subsequent days of embryogenesis (Linser and Perkins, 1987). The temporal sequence of astrocyte differentiation coincides with decreasing neurothelin expression in neuronal cells of the brain. At E4, when no astrocytes have yet developed, all brain neuroblasts appear to express some neurothelin, whereas at E10 when astrocytes start to differentiate, tectal neurons are essentially neurothelin negative.

Although the neural retina is part of the central nervous system, in the retina developmental regulation of neurothelin is significantly different from that in the tectum opticum. At early stages, limited amounts of neurothelin are found in retinal neuroblasts. After completion of the histogenesis of the retina, substantially more antigen is expressed than at embryonic stages. This suggests that the retina represents the only exception so far known of the general rule that in CNS- neurons neurothelin expression declines with development.

Various reports have demonstrated glial-neuron interactions in vitro. Early response genes such as c-fos (Arenander et al. 1989) and glutamine synthetase (Wu et al. 1988) in astrocytes, as well as myelin basic protein P_o in Schwann cells (Lemke and Chao, 1988) can be induced by neurons. Conversely glial cells can affect neural cells. The glia-derived extracellular matrix protein tenascin is likely to inhibit specific differentiation processes (axonal branching) of motor neurons (Wehrle and Chiquet, 1990). Two myelin-specific proteins (M_r 35 × 10³ and 250 ×10³) inhibit neurite outgrowth on oligodendrocytes in vitro and in vivo (Caroni and Schwab, 1988; Schnell and Schwab, 1990). However, in these cases, molecular changes induced in the affected neurons have not yet been evaluated. We questioned whether glial cells also modify neurothelin expression in neurons.

The chick retina is unique in its cell type composition. In contrast to other CNS regions of the chicken, neither astrocytes nor oligodendrocytes are present in the retina (both are glial cells) (Ramon Y Cajal, 1933). In the chick brain, oligodendrocytes develop one week later than astrocytes. However, neurothelin suppression takes place one week prior to differentiation of oligodendrocytes (Linser and Perkins, 1987). Considering these temporal limitations, astrocytes appeared to be the most likely cells in the brain that might suppress neurothelin gene expression in neurons.

To investigate this issue, tectal tissue was disintegrated enzymatically to destroy regulatory cell contacts. During prolonged culturing of tectal cells, differentiating neurons that had lost their native cellular environment such as neighboring astrocytes, became neurothelin positive. Although this observation supported our working hypothesis, the experimental conditions made it difficult to evaluate specific effects of astrocytes. Therefore, we employed retinal neurons that become strongly neurothelin positive in vivo (Schlosshauer and Herzog, 1990). It was found that in vitro astrocytes partially suppress neurothelin expression in retinal cells.

Our cocultures did not lead to a complete suppression of neural neurothelin expression. One reason could be the employment of rat rather than chick astrocytes. However, the lack of reliable chick astroglia specific antibodies as well as technical limitations during the course of chick glia purification had to be taken into account. On the other hand, the successful induction of neurothelin in chick endothelial cells by virtue of rodent brain cells (Schlosshauer and Herzog, 1990) indicated that functional interactions between cells originating from different species do take place. In this case, changes in neurothelin expression have not been. revealed quantitatively. The limited quantitative effects in our astrocyte-retina cocultures are, however, in the same range as another glia-induced change in vitro. γ-glutamyl-transpeptidase activity rises by 34 % in endothelial cells under the influence of astrocytes (Maxwell et al. 1987).

The formation and maintenance of the blood-brain barrier by endothelial cells most likely depends on inductive signals from astrocytes. Therefore, the barrier cannot be considered functionally and morphologically stable. Disruption of the astrocyte-endothelial cell interactions, as in some tumors, also disintegrates the blood-brain barrier (Long, 1970; Stewart et al. 1985). Conversely, inductive effects of astrocytes on endothelial cells have been demonstrated in vitro and in vivo. In vitro, rat endothelial cells formed tight junctions as a crucial element of the blood-brain barrier only under the influence of astrocytes (Arthur et al. 1987). In vivo transplantation of purified astrocytes onto blood vessels of the extraembryonic chorioallantoic membrane of the chicken induced a functional analogue of the blood-brain barrier (Janzer and Raff, 1987).

Similarly, neurothelin expression can be induced in vivo. Transplantation of CNS tissue from embryonic mice onto the neurothelin-negative chorioallantoic membrane resulted in vascularization of the transplant by chicken blood vessels. In the presence of mouse brain cells, chicken endothelial cells started to synthesize neurothelin (Schlosshauer and Herzog, 1990). Since the instructive action of glial cells on endothelial cells has been reported in other studies (De Bault and Cancilla, 1980; Maxwell et al. 1987), it is possible that astroglial cells are responsible for neurothelin induction in endothelial cells in our transplantion experiments as well as during regular embryogenesis. This is supported by preliminary experimental data (Janzer and Schlosshauer, unpublished data). Consequently, astrocytes would fulfill dual functions with regard to neurothelin expression: suppression in neurons and induction in endothelial cells. Both interactions would take place coincidentally as judged from histological data, starting in the second week of incubation (Schlosshauer and Herzog, 1990). From histological data, three principles can be deduced: (1) neurothelin is present in all CNS regions, (2) in individual CNS regions neurothelin is not expressed in endothelial and neural cells simultaneously, (3) neurothelin is developmentally regulated and this implies generally a cell-type switch (exception: retina). Therefore, neurothelin expression appears to be strictly controlled spatiotemporally.

Neurothelin has been identified as an integral membrane protein. Neither cell-surface-associated nor lipid- anchored isoforms have been found. The lateral diffusion of the latter are generally more pronounced, whereas transmembrane proteins with cytoplasmic domains are likely to interact with the cytoskeleton and are therefore less mobile (Pollerberg et al. 1986). For the transmembrane isoform of the neural cell adhesion molecule N-CAM ₁₈₀, an interaction with spectrin was evident (Pollerberg et al. 1987). N-CAM₁₈₀ was found to be concentrated at contact sites between neighboring cells in vitro (Pollerberg et al. 1987). Similarly, neurothelin is preferentially located at cytoplasmic membrane contacts of blood-eye barrier forming pigment epithelial cells (Schlosshauer and Herzog, 1990). Being an integral membrane protein, neuro-thelin has structural features that allow a nonhomogeneous distribution on the cell surface. Future experiments will hopefully reveal whether the specific cell surface distribution is governed by interactions with the cytoskeleton.

Neurothelin is a glycoprotein containing sulfated glucuronic acid (Schlosshauer and Herzog, 1990), galactose, mannose, (V-acetyl glucosamine and sialic acid. Different glycomoieties are possibly responsible for the existence of various isoforms of the antigen, which have been identified by two-dimensional gel electrophoresis. Endoglycosidase digestion indicated that 40 % of the antigen is composed of polysaccharides.

A pronounced glycosylation has been indicative of functional modifications of distinct glycoproteins. The homophilic binding activitiy of N-CAM is reduced with increasing glycosylation of the adhesion molecule (Sadoul et al. 1983). The lymphocyte homing receptor gp 90^MEL–14 is a lectin that binds to a distinct saccharide of a glycoprotein (Stoolman, 1989). Here carbohydrates are directly involved in the specific adhesion of lymphocytes to vascular endothelial cells of lymph nodes. It will be of special interest to evaluate whether the high rate of glycosylation has some impact also on the function of neurothelin. It is consistent with histological data that neurothelin could fulfill receptor functions in transport processes. Whereas neurothelin expression and proposed transport processes could be realized by endothelial cells in the adult brain, neuroblasts in the early embryonic brain and retinal neurons in the avascular retina would perform such transport functions. Obviously, the nervous system can not afford any topological gap in neurothelin distribution in the investigated CNS regions. Additional biochemical characterization as well as in vitro assays promise to evaluate structural-functional relationships of neurothelin.

The generous support of Dr F. Bonhoeffer is gratefully acknowledged. I thank I. Sommer, J. Vanselow and M. Wild for excellent technical assistance and Drs F. Bonhoeffer, D. Dütting, U. Egert, C. Garner, H. Hammerle and B. Müller for inspiring discussions.