Polarized distribution of Na,K-ATPase in honeybee photoreceptors is maintained by interaction with glial cells

The plasma membrane of most cells is organized into distinct domains characterized by differences in morphology, molecular composition and functional properties (Almers and Stirling, 1984; Simons and Fuller, 1985; Handler, 1989). The mechanisms involved in the establishment and maintenance of cell polarity have been the subject of intense research in recent years. These studies have provided evidence that both the submembrane cytoskeleton and cell-cell interactions contribute to the segregation and retention of membrane proteins in distinct domains (Rodriguez-Boulan and Nelson, 1989; Wollner and Nelson, 1992). Moreover, tight junctions may function as barriers to limit the diffusion of membrane constituents (Gumbiner, 1987). Most of these finding have been obtained from cultures with homogeneous cell lines. Much less is known about the establishment and maintenance of cell polarity in situ in complex tissues composed of different and interacting cell types, e.g. in nervous tissue consisting of neurons and glial cells.

The arthropod retina provides an ideal system for studying how cell polarity is maintained in nervous tissue because its organization is exceptionally clear. The arthropod retina is composed of just one type of neuron, the photoreceptors, and glial cells and both are arranged in a regular pattern. Moreover, the photoreceptors are highly asymmetric in both structure and function (Carlson and Chi, 1979). The rhabdomere, the light-receptive compartment, is composed of densely packed microvilli containing the photopigment rhodopsin (Langer and Thorell, 1966; Boschek and Hamdorf, 1976), the enzymatic machinery of photo-transduction (Schneuwly et al., 1991) and the light-dependent channels (Bacigalupo and Lisman, 1983; Payne and Fein, 1984; Bacigalupo et al., 1991; Hardie and Minke, 1992). The remaining part of the cell is poor in rhodopsin (Chi and Carlson, 1979; Suzuki and Hirosawa, 1991) and has metabolic functions, since it contains endoplasmic reticulum, Golgi complexes and a large number of mitochondria as well as other cell organelles (Carlson and Chi, 1979). The Na,K-ATPase is a heterodimeric integral membrane protein that consists of a ∼100 kDa catalytic α-subunit highly conserved in evolution (Lebovitz et al., 1989; Takeyasu et al., 1991) and a smaller noncatalytic β-subunit (Sweadner, 1989; Horisberger et al., 1991 for reviews). In arthropod photoreceptors, the Na,K-ATPase is highly active restoring ion gradients after light-induced influx of Na⁺ and effux of K⁺ (Tsacopoulos et al., 1983; Coles and Orkand, 1985; Coles et al., 1987; Hamdorf et al., 1988; Jansonius, 1990; Baumann et al., 1991). The Na,K-ATPase is thought to be concentrated in the nonreceptive compartment (Dimitracos and Tsacopoulos, 1985; Jansonius, 1990) as it is in vertebrate photoreceptors in which sodium pumps are restricted to the inner segment (Stirling and Lee, 1980; Ueno et al., 1980; Stahl and Baskin, 1984; McGrail and Sweadner, 1989). However, firm evidence for a polarized distribution of Na,K-ATPase in arthropod photoreceptors is lacking.

In this study, we have investigated the distribution of Na,K-ATPase in honeybee photoreceptors and the role of glial cells in positioning the Na,K-ATPase on the photoreceptor surface. Using immunocytochemical techniques, we confirm that the Na,K-ATPase is concentrated on the non-receptive surface of honeybee photoreceptors. We report further that, on the nonreceptive surface, the Na,K-ATPase is restricted to the areas in close contact with the glial cells. We have analyzed the distribution of the Na,K-ATPase in the absence of photoreceptor-glia contact and after experimentally remodeling photoreceptor-glia contact in order to assess the function of glial cells in positioning the Na,K-ATPase on the photoreceptor surface. The results of these experiments suggest that the glial cells play a direct role in the segregation of the Na,K-ATPase on the photoreceptor surface. Ca²⁺-independent adhesion of glial cells to the photoreceptor surface may trap the Na,K-ATPase molecules at the sites of photoreceptor-glia contact.

Honeybees (Apis mellifera) were obtained from a commercial dealer. They were kept in the dark at 35°C to 37°C and fed a 50% sugar solution.

Experiments were carried out on slices of retinal tissue or on dissociated retinal tissue. The preparation of 400 to 800 μm thick slices from the head of the honeybee drone was described in detail by Bertrand et al. (1979). The technique of Tsacopoulos et al. (unpublished) was used for dissociating drone retina; retinal tissue was treated with 4 mg/ml Pronase (Calbiochem, Bad Soden, FRG) in Ringer solution (see below) for about 20 min and then gently triturated through a fire-polished glass pipette.

Retinal tissue slices and dissociated retina were prepared for electron microscopy as described in detail previously (Baumann and Walz, 1989). They were fixed in 4% paraformaldehyde, 3.5% glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.4) for 2-3 h, and kept in the fixative with 1% tannic acid for a further hour (dissociated retina) or overnight (slices) at 4°C. The specimens were washed in cacodylate buffer, postfixed with 2% OsO₄, stained en bloc with 2% aqueous uranyl acetate, dehydrated in a graded ethanol series, and embedded in Epon or Spurr’s resin. Ultrathin sections were stained with uranyl acetate and lead citrate and examined with a Zeiss EM 10C electron microscope.

A worker bee was frozen in liquid nitrogen and its head removed. The isolated head was homogenized in 200 μl of buffer solution containing 150 mM NaCl, 50 mM Tris-HCl (pH 7.5), 1 mM EDTA and 1% Triton X-100 and then mixed with 50 μl of sample buffer (62.5 mM Tris-HCl pH 6.8, 10% glycerol, 2% SDS, 650 mM β-mercaptoethanol, 0.025% bromophenol blue). Samples (1/4 and 1/100) of this homogenate were fractioned on a SDS/5% to 15% polyacryamide gel, electrotransferred onto nitrocellulose paper and subjected to immunodetection. In an additional experiment, 10 retinae were isolated without fixation, homogenized in 50 μl of 5 times-diluted sample buffer and subjected to SDS-PAGE and immunoblot analysis.

The honeybee sodium pump was identified on the blots by the binding of a monoclonal antibody, IgG 5, specific to the sodium pump α-subunit (Takeyasu et al., 1988). This was accomplished in a series of washes/hybridizations (1 h each) in phosphate-buffered saline (PBS). Briefly, the nitrocellulose paper carrying fractionated honeybee proteins was soaked in PBS containing 5% skim milk to block nonspecific binding, washed in PBS, hybridized with IgG 5 (10 μg/ml), washed in PBS, hybridized with goat anti-mouse IgG conjugated to alkaline phosphatase (Sigma, Deisenhofen, FRG; 1:200 dilution), washed in PBS and then soaked with nitro blue tetrazolium/bromochloroindoyl phosphate for 10-20 min in a reaction buffer containing 100 mM Tris-HCl (pH 9.5), 100 mM NaCl and 5 mM MgCl2.

Retinal slices were fixed with 2% paraformaldehyde, 0.15% picrinic acid in 0.1 M phosphate buffer, pH 7.2, for 2 h at 4°C. The specimens were infiltrated with phosphate-buffered 25% sucrose overnight at 4°C and rapidly frozen in melting isopentane (Wolfrum, 1990). Sections (6-8 μm) were cut in a cryostat and collected on poly-L-lysine-coated coverslips.

Procedures for immunolabeling were similar to those used by Klein and Zimmermann (1991) for immunolocalization of vacuolar-type H+-ATPase in insect sensilla. The sections were successively incubated in: (1) 0.01% Tween-20 (Sigma) in PBS for 10 min; (2) 50 mM NH4Cl in PBS for 5 min to block free aldehyde groups of the fixative; (3) PBS for 10 min; and (4) in blocking solution (BS) containing 1% normal goat serum (Amersham, Braunschweig, FRG), 0.8% bovine serum albumin (Fluka, Neu-Ulm, FRG) and 0.1% gelatin (Sigma) in PBS for 10 min. The sections were incubated in a humidified atmosphere overnight at 4°C with the monoclonal antibody IgG 5 (20 or 40 μg protein/ml in BS). Control sections were incubated in BS or in BS with normal mouse serum (∼400 μg protein/ml). After washing three times in PBS to remove unbound antibodies, the sections were placed for 1 h at room temperature on fluorescein/sheep anti-mouse IgG (Sigma) or on tetramethylrhodamine/goat anti-mouse IgG (Sigma). Sections were washed again and mounted in Mowiol 4.88 (Farbwerke Hoechst, Frankfurt, FRG) containing 2% n-propyl-gallate to retard photobleaching, and examined with a Zeiss Axiophot. Kodak TMax400 film was used for photographic documentation.

Retinal tissue was fixed in 3% paraformaldehyde, 0.2% glutaraldehyde in 0.1 M phosphate buffer, pH 7.2, for 2 h at 4°C. The specimens were washed in phosphate buffer, dehydrated in a graded ethanol series and infiltrated and embedded in LR White. Ultrathin sections were cut on an ultramicrotome and collected on Formvar-coated nickel grids.

Selected grids were preincubated according to steps (2) to (4) of the immunofluorescence protocol. The grids were then incubated for 55-70 h at 4°C in a moist chamber with primary antibody (40 μg protein/ml) in BS. Control sections were incubated in normal mouse serum (∼400 μg protein/ml) in BS. After extensive washing in PBS, secondary antibody (goat anti-mouse IgG; Amersham) coupled to 1 nm or 10 nm gold was added for 3-5 h at 4°C. Sections were washed and 1 nm gold label was silver-enhanced by the method of Danscher (1981).

Slices of drone retina were incubated for 2 h in normal Ringer solution (270 mM NaCl, 10 mM KCl, 10 mM Tris-HCl, 1.6 mM CaCl₂, 10 mM MgCl₂, pH 7.4) or in Ca2+-free Ringer solution (2 mM K ₂EGTA, without CaCl₂, otherwise identical), or for 1 h in hyperosmotic Ringer solution (400 mM sucrose, otherwise identical) at room temperature and with oxygenation. Incubation proceeded under red light (> 610 nm), which elicits no electrical response of the photoreceptor cells (Bertrand et al., 1979). The specimens were then prepared for conventional electron microscopy and immunogold electron microscopy as described above. Tissue slices treated with hyperosmotic Ringer were fixed in solutions containing additionally 400 mM sucrose.

Cryostat sections (see above) were pretreated with Tween-20 and NH₄Cl and washed in PBS (steps (1)-(3) of the immunofluorescence protocol). Sections were labeled for 1 h with biotinylated WGA (Sigma) diluted 1:10 in PBS, washed extensively in PBS, incubated with fluorescein-conjugated ExtrAvidin (Sigma) diluted 1:40 in PBS, washed again in PBS and mounted in Mowiol 4.88.

An ommatidium in the honeybee compound eye contains nine photoreceptor cells that are arranged in a rosette called the retinula (Fig. 1). Only eight photoreceptors are observed in any transverse section of a retinula because the 8th and 9th photoreceptor are tiered. The light-sensitive rhabdom in the center of the retinula consists of microvilli contributed by each of the photoreceptor cells. Pigmented glial cells completely ensheathe the retinula. The glial cells send fin-like processes into the clefts between the photoreceptors towards, but not reaching, the rhabdom (Fig. 2A,B).

The plasma membrane of the cell bodies of the honeybee photoreceptors can be divided into three domains with respect to the distribution of membrane proteins (Fig. 2A,B). The most conspicuous is the rhabdomeric domain whose microvillar membranes are packed with photopigment molecules (Perrelet et al., 1972). Junctions of the zonula adherens type (Baumann, 1992a) separate the rhabdomeric membrane domain from the rest of the plasma membrane; this has no microvilli and is presumably insensitive to light. The nonreceptive surface may be further sub-divided into a major part bounded by pigmented glial cells and smaller glia-free domains. At the glia-bounded domain, the photoreceptor faces the pigmented glial cells across an extracellular space which is 10-15 nm wide and filled with amorphous material (Fig. 2B,C). Typically, numerous mitochondria are arranged along the glia-bounded domain (Fig. 2A). Some areas of the nonreceptive photoreceptor surface, however, do not face glial cells (arrows in Fig. 2A,B). This glia-free surface is found adjacent to the zonulae adherentes, where the glial processes do not reach, and at a few small patches further away from the rhabdom. At the gliafree membrane domains, the photoreceptor faces a slightly extended extracellular space. By electron microscopy, we have not observed any junctional structures at the borders between glia-bounded and glia-free membrane domains (Fig. 2C).

We have used a monoclonal antibody, IgG 5, to identify the α-subunit of the honeybee Na,K-ATPase. Specificity of this monoclonal antibody to the α-subunit has been very well established: (i) IgG 5 was made against affinity-purified chicken kidney Na,K-ATPase α-subunits and cross-reacts very weakly with mammalian α-subunits (Takeyasu et al., 1988); (ii) IgG 5 recognizes all known chicken Na,K-ATPase α-subunit isoforms (α1, α2 and α3) expressed individually from cloned isoform-specific cDNAs in immuno-precipitation and immunoblot analysis (Takeyasu et al., 1988; Kone et al., 1991); and (iii) IgG also specifically labels the gene product encoded by the Drosophila melanogaster Na,K-ATPase α-subunit (Lebovitz et al., 1989).

The specificity of the antibody for the sodium pump α-subunit of honeybee head and retina was confirmed by immunoblot analysis of tissue homogenate (Fig. 3). As seen in lanes 1 and 3, the antibody labeled one band with a molecular mass of ∼100 kDa. This was also the case when the sample was overloaded (lane 2) and when Drosophila melanogaster head (gift from Dr Vassin’s laboratory) was used (data not shown). Specificity of IgG 5 binding was demonstrated in control experiments with IgG 24, which recognizes the chicken sodium pump β-subunit and does not cross-react with any other species (Fambrough and Bayne, 1983; Tamkun and Fambrough, 1986); IgG 24 did not label any antigen on blots of honeybee retina and brain (data not shown).

The monoclonal antibody IgG 5 was used for indirect immunofluorescence staining of cryostat sections of the honeybee drone retina. The antibody labeled the outline of the retinula rosette (Fig. 4A) and the outline of the photoreceptor axons (Fig. 4C). The rhabdom, whose contour is identified in the corresponding Nomarski-contrast micro-graph (Fig. 4B), did not display immunoreactivity. Faint immunofluorescence staining was observed at sites where glial cells contact each other. However, in Fig. 4A staining of the sites of glia-glia contact is outshone by the intense staining of the retinula periphery. Labeling of the glial cell surface was most conspicuous in sections close to the corneal lens that do not include photoreceptors (data not shown). No labeling was detected on control sections that were incubated with normal mouse serum instead of antibody to sodium pump α-subunit.

Since the cryostat sections were relatively thick (6-8 μm), the absence of immunostaining at the rhabdom could have resulted from poor penetration of the antibody into this structure. To test this, ultrathin sections of plastic-embedded drone retina were prepared for immunogold electron microscopy. Positive staining of the rhabdom with anti-bodies to actin (Arikawa et al., 1990; O. Baumann, unpublished data), the myosin-homologue ninaC (Porter et al., 1992) and the carboxyl terminus of rhodopsin (De Couet and Tanimura, 1987) demonstrates that antibodies have access to the microvillar interior in ultrathin sections of plastic-embedded insect retina. Thus, if Na,K-ATPase is present in the rhabdom it should be detected by indirect immunogold labeling with IgG 5, whose binding epitope is within the cytoplasmic domain of the sodium pump α-sub-unit (Lemas et al., 1992).

Fig. 5 shows two ultrathin sections of drone retina reacted with IgG 5; primary antibody binding was detected by 1 nm secondary gold and subsequent silver enhancement. Consistent with the results of the immunofluorescence labeling, no binding other than rare background staining could be detected over the rhabdom. Intense, specific immunogold staining was confined to the sites of photo-receptor-glia contact (Fig. 5A,B). As shown below, this intense staining at the sites of photoreceptor-glia contact can be accounted for by sodium pumps on the surface of photoreceptors but not of glial cells. The glia-free domains of the nonreceptive photoreceptor surface displayed immunoreactivity barely above background (arrows in Fig. 5A,B). Weak labeling was found at the sites of glia-glia contact.

These results demonstrate that the photoreceptors have a high concentration of Na,K-ATPase localized at the sites of photoreceptor-glia contact.

We examined the distribution of the Na,K-ATPase in isolated retinulae to investigate the role of glial cells in the positioning of Na,K-ATPase on the photoreceptor surface.

Dissociation of retinal tissue into single retinulae, complexes of glial cells and crystalline cones was accomplished by enzyme treatment and mechanic trituration. As shown in Fig. 6, the photoreceptors retained their retinula arrangement and their characteristic cytoarchitecture and they were still linked by zonulae adherentes. Adjacent photoreceptors often adhered closely to each other by their nonreceptive surface near the zonula adherens. Most of the nonreceptive surface, however, was exposed to the surrounding medium and clean of glia fragments.

IgG 5 intensely labeled the photoreceptor surface when dissociated retinal tissue was used for immunogold electron microscopy; the plasma membrane of glial cells was barely stained (Fig. 7). This finding demonstrates that the photoreceptors contribute most to the intense anti-Na,K-ATPase staining at the sites of photoreceptor-glia contact in intact retinal tissue. However, the staining pattern of isolated retinulae was different from that of retinulae in situ. In isolated retinulae, IgG 5 labeled the entire nonreceptive photoreceptor surface (Fig. 8). Gold label was also found at the microvillar bases, whereas, consistent with the data on intact retina, the rhabdom was not stained. The simplest interpretation of these findings is that removal of glial cells from the photoreceptor surface caused a redistribution of the Na,K-ATPase within the plasma membrane. The pump molecules may have spread over the entire photoreceptor surface except the microvillar membranes.

To investigate further the role of glial cells in the positioning of Na,K-ATPase on the photoreceptor surface, we experimentally remodeled photoreceptor-glia contact by incubation of retinal tissue slices in hyperosmotic saline and looked for a possible redistribution of Na,K-ATPase.

Electron microscopy of tissue slices after a 1-h incubation in hyperosmotic, oxygenated saline showed severe cell shrinkage and rearrangement of sites of photoreceptor-glia contact (Fig. 9). The fin-like glial processes were retracted from the rhabdom and, thus, the photoreceptors displayed more extensive areas of free surface close to the zonula adherens. However, some glial fins did not retract but spread into numerous thin, finger-like processes still adhering to the photoreceptor surface. Often, several glial fingers were positioned next to each other on the photoreceptor surface leaving little space between adjacent processes. These glial processes are seen in cross-section as oval profiles aligned along the photoreceptor surface (Fig. 9B).

Immunogold labeling of retinal tissue after hyperosmotic treatment revealed weak anti-Na,K-ATPase staining of the extensive glia-free domains but intense staining of the glia-bounded domains of the photoreceptor surface (Fig. 10). Significantly, the gold label was strictly localized to the photoreceptor-glia contact when several profiles of glia fingers were aligned on the photoreceptor surface (Fig. 10B,C). These results demonstrate that remodeling of photoreceptor-glia contact causes a corresponding redistribution of sodium pumps on the photoreceptor surface.

Studies on vertebrate epithelia have demonstrated that the establishment and the maintenance of cell-cell contacts and of the polarized distribution of surface proteins requires a millimolar Ca²⁺ concentration in the extracellular medium (Nelson and Veshnock, 1987; Vega-Salas et al., 1987). To examine the Ca²⁺-dependence of photoreceptor-glia contact and of the Na,K-ATPase distribution in drone retina, we have incubated tissue slices in Ca²⁺-free Ringer solution. Incubation proceeded for 2 h to obtain sub-micromolar Ca²⁺ concentrations in the extracellular space (Ziegler and Walz, 1990).

Electron-microscopic examination showed that incubation in Ca²⁺-free medium had little effect on cell-cell contacts in the drone retina; photoreceptors and glial cells adhered closely to each other, neighboring photoreceptors were linked by adherens junctions and the rhabdomeric microvilli were tightly packed (Fig. 11A). However, photoreceptor morphology was strikingly altered; the cytoplasm was denser and long, lamellar protrusions extended between glial cells (Fig. 11B).

Immunogold localization of Na,K-ATPase showed no obvious effect of Ca²⁺-free Ringer on the distribution of Na,K-ATPase in the photoreceptors; gold label was almost exclusively localized to the glia-bounded surface of the photoreceptors, including the lamellar protrusions (Fig. 12A,B).

To investigate whether the polarized distribution of Na,K-ATPase is the result of nonspecific aggregation of membrane proteins at the sites of photoreceptor-glia contact, we have localized bulk glycoproteins by binding of a lectin, wheat germ agglutinin (WGA). Binding of biotin-conjugated WGA to cryostat sections was visualized by fluorescein-labeled ExtrAvidin.

Cross-sectioned ommatidia revealed intense, highly localized WGA fluorescence at the periphery of the rhabdom (Fig. 13A). In a focus series, WGA fluorescence was visualized as bands extending alongside the rhabdom. Comparing the position of WGA fluorescence with the organization of ommatidia in Fig. 1, we conclude that WGA was bound to the glia-free photoreceptor surface next to the ZA. This conclusion is supported by analysis of WGA binding to retinae incubated in hyperosmotic saline (Fig. 13C,D). These results indicate the presence of bulk glycoproteins on the glia-free surface of the photoreceptor. Thus, non-specific aggregation of membrane proteins at the glia-bounded surface cannot account for the polarized distribution of Na,K-ATPase.

However, WGA fluorescence was weak or negative at other surface domains, the sites of photoreceptor-glia contact and the rhabdom. This does not necessarily indicate the absence of WGA-binding sites because WGA might not have access to the extracellular side of the plasma membrane if membranes adhere tightly to each other.

Our data demonstrate that the Na,K-ATPase is asymmetrically distributed in the photoreceptors of an arthropod, the honeybee Apis mellifera. The Na,K-ATPase is concentrated on the nonreceptive surface but virtually absent on the light-sensitive, microvillar surface, a distribution analogous to that of vertebrate photoreceptors (Stirling and Lee, 1980; Ueno et al., 1980; Stahl and Baskin, 1984). These findings provide a morphological basis for the previous hypothesis on the location of the sodium pump in arthropod photoreceptors, which was based on physiological and biochemical data. Dimitracos and Tsacopoulos (1985) studied the recovery of energy-dependent processes in bee photoreceptors after anoxic treatment and reported that sodium pumping recovers before phototransduction. The authors concluded that the sodium pumps are supplied with ATP prior to the phototransduction machinery as the pumps are located in the plasma membrane close to mitochondria; namely, the nonreceptive domain. Jansonius (1990) suggested that the microvillar membrane of blowfly photoreceptors is poor in sodium pumps as rhodopsin already accounts for 65% of the microvillar membrane particles. Moreover, Walz (1985) has provided physiological evidence that the sodium pump is also localized to the nonreceptive surface in the microvillar-type photoreceptor of the leech. Thus, the Na,K-ATPase appears to lie on the nonreceptive surface of all photoreceptors.

We have found, further, that the distribution of sodium pumps in honeybee photoreceptors is more complex than previously expected. On the nonreceptive photoreceptor surface, the Na,K-ATPase is restricted to those areas that are in intimate contact with glial cells. Additionally, we have provided evidence suggesting that the glial cells play a crucial role in the positioning of the sodium pumps on the photoreceptor surface and retain the pumps at the contact sites.

The mechanisms involved in the maintenance of membrane protein domains have been studied previously in polarized absorptive and secretory epithelia of vertebrate origin. Some mechanisms that have been proposed include: (a) tight junctions restricting the lateral diffusion of membrane constituents and, thus, acting as barriers between membrane domains; (b) vesicular transcytosis of misplaced proteins; (c) linkage of membrane proteins to the submembrane cytoskeleton; and (d) retention of membrane proteins at sites of cell-cell contact by their attachment to cell adhesion molecules (Almers and Stirling, 1984; Simons and Fuller, 1985; Rodriguez-Boulan and Nelson, 1989; Wollner and Nelson, 1992, for reviews). There is evidence for each of these mechanisms. However, their contributions may vary depending on cell type and origin.

Three observations on the drone retina indicate that the specific contact between photoreceptors and glial cells is essential for the maintenance of the polarized Na,K-ATPase distribution on the photoreceptor surface. First, the distribution of sodium pumps on the photoreceptor surface correlates strictly with the sites of photoreceptor-glia contact. Secondly, remodeling of photoreceptor-glia contact by incubation in hyperosmotic saline causes a redistribution of sodium pumps on the photoreceptor surface corresponding to the redistribution of glial cells. This finding suggests that the pump molecules are trapped at the sites of photoreceptor-glia contact and dragged along the photoreceptor surface as the contact sites move. Finally, in isolated retinulae the sodium pumps are found over the entire nonreceptive surface and at the microvillar bases. One interpretation of this redistribution is that removal of the glial cells from the photoreceptor surface releases the restriction on the lateral diffusion of the Na,K-ATPase and the pump molecules are able to spread in the plasma membrane. Even under these conditions, sodium pumps are excluded from the microvillar membranes proper; this supports the view that the microvillar membrane represents a microdomain with limited lateral diffusion of proteins (Goldsmith and Wehner, 1977).

Ultrastructural analysis of the drone retina did not provide any evidence for junctional complexes at the boundary between glia-free, Na,K-ATPase-negative and glia-bounded, Na,K-ATPase-positive membrane domains. Thus, the Na,K-ATPase must be maintained at the sites of photoreceptor-glia contact by some other form of cell-cell contact. We suggest that the pump molecules are directly or indirectly linked to cell adhesion molecules mediating photoreceptor-glia contact. Significantly, the extracellular space between photoreceptors and glial cells is of constant width and filled by an amorphous material, indicating a tight molecular interaction between the two cell membranes. WGA labeling indicates further that not all the glycoproteins of the nonreceptive photoreceptor surface are aggregated at the sites of photoreceptor-glia contact and so there must be molecular specificity in the attachment of the Na,K-ATPase to cell adhesion molecules.

Recent studies on the distribution of Na,K-ATPase in polarized vertebrate epithelial cells have also provided evidence for a direct role of cell-cell contact in the development and maintenance of cell polarity. In most polarized epithelia, the Na,K-ATPase is localized to the basolateral domain; namely, the sites of cell-cell and cell-substratum contact (Almers and Stirling, 1984; Simons and Fuller, 1985). Analysis of conversion of nonpolarized single cells to polarized epithelial layers has demonstated that the accumulation of sodium pumps on the basolateral surface coincides with the formation of cell-cell contact (Nelson and Veshnock, 1986; Wang et al., 1990). Moreover, McNeill et al. (1990) have transfected nonpolarized fibroblasts with cDNA encoding the Ca²⁺-dependent cell adhesion molecule uvomorulin. They reported that the expression of uvomorulin induces the formation of cell-cell contacts and the concentration of Na,K-ATPase at these contact sites. In agreement with our study, the retention of sodium pumps in transfected fibroblasts at the sites of cell-cell contact occurs in the absence of tight junctions. Finally, Nelson et al. (1990) have shown in kidney epithelial cells that the Na,K-ATPase is part of a protein complex containing uvomorulin and the cytoskeletal proteins ankyrin and spectrin. There are, however, distinct differences between the molecular mechanisms responsible for the polarized distribution of Na,K-ATPase in vertebrate epithelial cells and the drone retina. In vertebrate epithelia, homophilic Ca²⁺-dependent binding between uvomorulin on adjacent cells induces and maintains cell surface polarity of Na,K-ATPase (McNeill et al., 1990). In the honeybee retina, cell-cell contact and the polarized Na,K-ATPase distribution on the photoreceptor surface are independent of extracellular Ca²⁺. Moreover, the positioning of the Na,K-ATPase on the photoreceptor surface requires specific recognition between the photoreceptor surface and the glial cell surface, since the Na,K-ATPase is localized at sites of photoreceptor-glia contact but not photoreceptor-photoreceptor contact. We conclude that the Na,K-ATPase is retained at the site of photoreceptor-glia contact by a heterophilic, Ca²⁺-independent mechanism.

In addition, the submembrane cytoskeleton of vertebrate epithelia and drone retina may differently contribute to the retention of the Na,K-ATPase in surface domains. In vertebrate epithelia, the submembrane meshwork of spectrin and ankyrin participates in the development and the maintenance of a polarized Na,K-ATPase distribution as it links the Na,K-ATPase molecules to uvomorulin (Nelson et al., 1990; McNeill et al., 1990). In drone retina, using a polyclonal antiserum to Drosophila α-spectrin (Byers et al., 1987) and indirect immunofluorescence, spectrin was found at the sites of photoreceptor-glia contact and all sites of glia-glia contact (Baumann, 1992b). Unfortunately, we could not obtain immunogold labeling at the EM level and, therefore, the immunolabeling experiment did not resolve whether spectrin resides in the photoreceptor or the glial cell, or in both, at the sites of photoreceptor-glia contact. However, EM visualization of a membrane skeleton, like that of erythrocytes, under the glia membrane but not under the photoreceptor surface (Baumann, 1992b), raises the question of whether spectrin is present at all in the photoreceptor cortex. Thus, a linkage of Na,K-ATPase to spectrin appears unlikely in this system.

Finally, the finding that glial cells maintain protein topography on the neuronal surface is not peculiar to the drone retina but compatible with a recent study on dorsal root ganglion neurons. This has demonstated that the clustering of voltage-dependent Na⁺ channels at the nodes of Ranvier requires the interaction with Schwann cells (Joe and Angelides, 1992). In contrast, in vertebrate photoreceptors, the segregation of the Na,K-ATPase on the inner segment is independent of cell-cell contact but critically dependent on the attachment to the spectrin-based submembrane skeleton (Madreperla et al., 1989).

The disposition of the Na,K-ATPase on the nonreceptive surface of the photoreceptor represents an important aspect of photoreceptor function. Steady illumination of arthropod photoreceptors produces a maintained depolarization by opening of cation-specific channels (Hardie, 1991; Ranganathan et al., 1991). As the main light-induced current is carried by Na⁺ (Fulpius and Baumann, 1969; Hardie, 1991), the cytoplasmic Na⁺ concentration increases in response to light. This light-induced gain of cytoplasmic Na⁺ has been measured directly in honeybee photoreceptors (Coles and Orkand, 1985; Coles et al., 1987) and amounts to 23 mM for a 90 s light stimulus. The photoreceptors also lose an almost equal amount of K⁺ (Coles et al., 1987). The spatial segregation of the light-dependent, Na⁺-permeable channels on the microvillar membrane (Payne and Fein, 1984) from the Na,K-ATPase on the nonreceptive membrane implies that Na⁺ passing the light-dependent conductances have to diffuse across the cell to be pumped out of the cell. On the other hand, there must be a flux of Na⁺ within the narrow extracellular space from the site of the nonreceptive photoreceptor surface to the rhabdom. Significantly, neighboring photoreceptor cells are not connected by junctional complexes that might act as barriers to the intercellular diffusion of ions. Tracer studies on arthropod retina have shown that ions can pass the adherens junctions and readily exchange between all extracellular compartments (Perrelet and Baumann, 1969; Shaw, 1978; Chi and Carlson, 1981).

The localization of the Na,K-ATPase at the sites of photoreceptor-glia contact raises an additional point. Neurons require a major part of their energy consumption for Na⁺ transport (Hansen, 1985) but their energy metabolism may largely depend on the interaction with glial cells, since the glial cells may supply metabolic substrate (Kuffler and Nichols, 1976). The nutritive function of glial cells has been documented best in the honeybee retina because the segregation of metabolic functions between neurons and glial cells is exceptionally complete in this system (Tsacopoulos et al., 1987, 1988; Coles, 1989). Glial cells of the honeybee retina contain a large amount of glycogen but few mito-chondria (Perrelet, 1970; Tsacopoulos et al., 1987). The photoreceptors, in contrast, have almost no glycogen but numerous mitochondria, most of them located close to the sites of photoreceptor-glia contact (Perrelet, 1970; Baumann and Walz, 1989) and, hence, close to the Na,K-ATPase (present study). Light stimulation of the photoreceptor cells activates glycogen metabolism in the glial cells and a carbohydrate substrate is transfered to the photoreceptors (Tsacopoulos et al., 1987, 1988). Since more than 50% of the extra energy consumption during a light stimulus is accounted for by the Na,K-ATPase (Tsacopoulos et al., 1983), the proximity of the glycogen store, the mitochondria and the Na,K-ATPase seems convenient. This proximity minimizes the diffusional distances for the carbohydrate substrate and for the large number of ATP molecules required by the sodium pump. The results presented here complement current views on the tight metabolic interactions between photoreceptors and glial cells, as we have demonstrated that the suppliers of metabolic substrate, the glial cells, play a direct role in the positioning of the sodium pumps, the main comsumers of ATP, on the photoreceptor surface.

We thank Mrs. Birgit Lautenschläger for excellent technical assistance, Dr Bernhard Zimmermann and Dr Siegfried Seidl for their generous gift of secondary antibodies and for advice on immunocytochemistry, Dr Bernd Walz and Dr Jonathan A. Coles for critically reading the manuscript, and Dr Jenny Kien for linguistic corrections. We are especially grateful to Dr Marcos Tsacopoulos for showing his technique for dissociating drone retina. This work was supported by a grant from the Deutsche Forschungsgemeinschaft (Ba 1284/1-1 to O. Baumann) and by a NIH grant (RO1-GM-44373 to K. Takeyasu).