Membrane traffic in polarized neurons in culture

Neurons are highly asymmetric cells equipped with long extensions: axons and dendrites. The dendrites receive and process information and the axons transmit it to the target cell at specialized sites: the synapses. This functional polarity is illustrated by an asymmetric distribution of many cell components. Although there are some exceptions, in most neuronal cells ribosomes (Barlett and Banker, 1984) and RNA are segregated (Davis et al., 1987; Steward and Banker, 1992) into the dendrites and the cell body, whereas synaptic vesicles are in the axon. The cytoskeleton is also organized in a polarized fashion: phosphorylated MAP IB, tau and neurofilament H are exclusively axonal, while MAP 2 and dystrophin are somatodendritic (Sato-Yoshitake et al., 1989; Kosik and Finch, 1987; Shaw et al., 1985; Caceres et al., 1984; Lidov et al., 1990). Plasma membrane components are segregated as well: the transferrin receptor is localized on the dendrites (Cameron et al., 1991; Parton et al., 1992), neurotransmitter receptors are in the postsynaptic area of the dendrite (Killisch et al., 1991), and acetylcholinesterase (G4 form) and TAG-1 (axonin-1) are present on the axonal plasma membrane (Rotundo and Carbonetto, 1987; Ruegg et al., 1989; Furley et al., 1990). Endocytosis takes place in the dendrites, the cell body and the presynaptic terminals but not from the axon shaft (Parton et al., 1992).

Little is known about how this morphological and functional polarization is accomplished. Studies on neuronal polarity have been hampered by technical limitations. Firstly, in situ studies are difficult because of the complexity of the brain. Secondly, no polarized neuronal cell lines exist. We have analyzed neuronal membrane sorting in hippocampal neurons in culture (an example of such a cell is shown in Fig. 1). Experimentally, we used the same strategies that have been used successfully to understand polarized membrane traffic in epithelial cell.

In epithelial cells, the sorting between apical and basolateral plasma membrane proteins can be executed at two positions: at the trans-Go\gi network (direct sorting) and at the endosomes/plasma membrane (indirect sorting), (Simons and Wandinger-Ness, 1990; Wessels et al., 1990; Jalal et al., 1991; Schell et al., 1992; Nelson, 1992). Studies with viral glycoproteins show that Madin-Darby canine kidney (MDCK) cells predominantly separate their membrane proteins at the trans-GtAgi network into vesicles that are destined either for the apical or the basolateral domain (direct sorting). Hepatocytes target most, if not all, proteins to the basolateral domain first. From here, apically destined proteins are rerouted via transcytosis to the apical site of the cell (indirect sorting). The intestinal epithelial cell line Caco-2 uses both the direct and the indirect pathways. The sorting machinery of both the trans-GcAgi network and the plasma membrane/endosome seems to recognize the same signals, since most polarized membrane proteins have the same distribution in MDCK and hepatocytes. Fig. 2 shows schematically the direct and indirect pathways of epithelial cells.

The analysis of mutated viral-envelope proteins and endogenous plasma membrane proteins led to the concept that specific signals determine vectorial polarized traffic. The first signal was identified in glycoproteins that are anchored to the plasma membrane via a glycosyl-phos- phatidylinositol (GPI). In epithelial cells, GPI-linked proteins were found exclusively in the apical membrane (Lisanti et al., 1990). The GPI-anchor is added in the endoplasmic reticulum to those proteins that contain a cleavable hydrophobic signal for GPI attachment at their C terminus (Lisanti and Rodriguez-Boulan, 1990). Adding this signal to basolaterally sorted proteins redirects them to the apical membrane. However, apical targeting signals can also reside in the protein itself. For example, the nerve growth factor (NGF) receptor is a transmembrane protein, so does not contain a GPI anchor, but is apically directed in MDCK cells (Le Bivic et al., 1991). Some GPI-anchored proteins contain an additional apical targeting signal in their ectodomains, which, in the absence of the GPI anchor still directs them apically (Lisanti and Rodriguez-Boulan, 1990).

Basolateral targeting signals have also been discovered. One signal has been identified within a 14 amino acid segment of the poly-immunoglobulin receptor (Casanova et al., 1991). Other signals have been found in the cytoplasmic tail of several basolateral proteins. Upon removal of the signals, the proteins are targeted to the apical site (Townbridge, 1991; and see reviews by Hopkins, 1991, 1992). Very interestingly, in several cases a correlation exists between basolateral sorting signals and signals for clathrin- mediated endocytosis (Mostov et al., 1992; Brewer and Roth, 1991).

Since neurons can be cultured in vitro as polarized cells, it is now possible to test whether the sorting mechanisms discovered in epithelial cells are also utilized by polarized neurons. Such studies, recently performed in our laboratory, are described below.

In culture fetal rat hippocampal neurons establish axonal and dendritic arbors and a molecular organization similar to their counterparts in situ (Dotti et al., 1988; Banker and Waxman, 1988).

The same viruses that have been successfully used to study protein sorting in epithelial cells were employed to determine the sorting of newly synthesized membrane glycoproteins (Dotti and Simons, 1990; Dotti et al., 1993). Polarized neurons were infected with the temperature-sensitive mutant VSV ts045. The location of the viral glycoprotein during its biosynthetic route is dependent on the temperature. At the non-permissive temperature of 39°C, viruses can infect the cells. However, the synthesis of the vesicular stomatitis virus (VSV) glycoprotein is arrested in the rough endoplasmic reticulum. At 20°C the glycoproteins can leave the rough endoplasmic reticulum, but are arrested in the trans-Golgi network. When the temperature is raised to the permissive temperature of 32°C, a synchronous and massive transport from the Golgi apparatus to the plasma membrane takes place. After various times post-infection, neurons were fixed and processed for immunofluorescence. At the end of the 20°C block, the VSV glycoprotein was only observed in the cell body, consistent with a location in the Golgi complex. After one hour of incubation at the permissive temperature (32°C) intense labelling of both the cell body and the dendrites was observed. However, axons were devoid of labelling. Similarly, upon infection with Semliki Forest virus, the viral glycoproteins were only seen in the somatodendritic domain.

When hippocampal cells were infected with the avian fowl plague virus (FPV), the hemagglutinin protein was preferentially inserted into the axonal plasma membrane. Four hours after infection with the wild-type FPV virus, hemagglutinin fluorescence was predominantly observed in the cell body, in accordance with a distribution in the Golgi complex. One hour later, staining was present in all the axons as well. Dendritic labelling was variable; some dendrites were completely devoid of labelling, others showed staining in their proximal parts, which might represent either viral protein processing in the dendritically located Golgi cistemae or mis-sorting of the hemagglutinin. These immunofluorescence experiments were confirmed by quantitative immunoelectron microscopy. The illustration in Fig. 3 summarizes the findings of the neuronal infection approach.

These results show that the viral glycoproteins are transported from the trans-Golgi network to the axonal or dendritic plasma membrane via a direct pathway. It was thus suggested that the mechanism used by neurons to deliver newly synthesized viral glycoproteins to the axons and dendrites was similar to that operating in MDCK cells. In these cells the hemagglutinin protein is sorted directly to the apical membrane and the VSV glycoprotein to the basolat- eral domain.

To analyze whether the correlation in the sorting of viral proteins between epithelial cells and neurons would also hold for the sorting of glypiated proteins, the distribution of Thy-1, a GPI-linked protein, was investigated in hippocampal neurons in culture (Dotti et al., 1991). In polarized epithelial cells GPI-anchored proteins are preferentially targeted to the apical surface. First of all we demonstrated that in the hippocampal neurons Thy-1 was indeed membrane bound via a GPI-anchor. Secondly, immunofluorescence and immunoelectron microscopy revealed the exclusive distribution of Thy-1 on the axonal surface. This result strengthened our early hypothesis that the axonal domain is the analogue of the apical domain of the epithelial MDCK cells and that the dendritic plasma membrane is equivalent to the basolateral membrane. As discussed below, subsequent work has supported, as well as contradicted, this rule.

In support of the axonal/apical and dendritic/basolateral rule are studies concerning the localization of several GPI- linked and integral membrane proteins. GPI-anchored (like 5’-nucleotidase) TAG-1, F3/F11 and acetylcholinesterase are present in the axonal domain (Grondal and Zimmermann, 1987; Dodd et al., 1988; Furley et al., 1990; Faivre- Sarrailh et al., 1992; Rotundo and Carbonetto, 1987). Moreover, a neuronal GPI-linked protein, an isoform of N-CAM, when expressed in epithelial cells follows the apical route (Powell et aL, 1991). The similarities also extend to certain integral membrane proteins. For example, the transferrin receptor, which is present in the basolateral plasma membrane of epithelial cells (Fuller and Simons, 1986) is restricted to the dendritic plasma membrane of neurons (Cameron et al., 1991; Parton et al., 1992). The voltagedependent sodium channel, preferentially restricted to the axon hillock and the nodes of Ranvier (Srinivasan et al., 1988; Joe and Angelides, 1992), is routed to the apical domain when expressed in MDCK cells (Angelides and Wible, unpublished data).

Contradictions to the rule axonal/apical-dendritic/baso- lateral have been described as well. The GPI-linked protein F3/F11, which is axonal in neuronal granule cells, is uniformly distributed in neuronal Golgi cells (Faivre-Sarrailh et al., 1992). Thus, one conclusion from these results would be that there is no uniformity on how different neuronal cells sort the same protein. Therefore, and because of this lack of uniformity, epithelial and neuronal cells cannot be compared. However, the absence of unity in the sorting mechanism of a given protein is also true for epithelial cells. In FRT cells, a polarized epithelial cell line derived from thyroid tissue, GPI-anchored proteins are either unsorted or routed to the basolateral domain (Zurzolo et aL, 1993). The glycoproteins of certain viruses also follow divergent routes in different epithelial cells. The glycoproteins of Sindbis and Semliki Forest viruses, which are normally sorted to the basolateral domain of MDCK cells, are apically delivered in FRT cells (Zurzolo et al., 1992). The distribution of the human low-density lipoprotein (LDL) receptor constitutes another example of sorting divergence among epithelia. In transgenic mice, the receptor is found in the basolateral surface of hepatocytes and intestinal epithelial cells, but apical in the renal tube epithelium (Pathak et aL, 1990).

The distribution of the Na⁺,K+-ATPase in neuronal and epithelial cells also contradicts the hypothesis of an axonal/apical-dendritic/basolateral rule (Pietrini et aL, 1992). When the neuron-specific Na⁺,K⁺-ATPase was expressed in polarized epithelial cells, it was found on the basolateral surface. The endogenous protein is, however, uniformly distributed on the neuronal surface. The differences in distribution of this protein between epithelia and neurons may be due to specific interactions with the cytoskeleton. In epithelial cells newly synthesized Na⁺,K⁺- ATPase is delivered to both the apical and the basolateral domains. Upon insertion into the apical domain it is selectively removed, while in the basolateral membrane it is stabilized via interactions with the polarized cytoskeletal ankyrin/fodrin complex (Hammerton et aL, 1991; Nelson and Veshnock, 1987; Nelson and Hammerton, 1989). The importance of the ankyrin/fodrin complex for the localization of the Na⁺,K⁺ pump is illustrated by the fact that a reversal of the ankyrin/fodrin polarity changes the polarity of Na⁺,K⁺-ATPase as well (Gundersen et aL, 1991). The unpolarized distribution of Na⁺,K⁺-ATPase in hippocampal neurons might therefore by related to a lack of polarization of this cytoskeletal complex.

After leaving the irans-Golgi network, different proteins are packed into vesicles destined for either the axonal or the dendritic plasma membrane domain. How are these different vesicles transported to their correct destinations? Newly synthesized membrane proteins are transported by fast- axonal transport, a microtubule-dependent event (Hammerschlag and Brady, 1988; Sheetz et al., 1989; Schroer, 1992). Nothing is known on how newly synthesized membrane proteins reach the dendritic plasma membrane. To determine whether or not microtubules are involved in the delivery of both axonal and dendritic exocytic vesicles, the effect of nocodazole on the delivery of viral envelope proteins was studied (C. G. Doth and K. Simons, unpublished data). The addition of nocodazole inhibited transport of influenza hemagglutinin out of the cell body to the axon, resulting in an intense staining of the cell body. The dendritic delivery of the glycoprotein from VSV was affected similarly; in 71% of the infected cells dendritic transport was abolished and again intense labelling of the cell body was observed in these cells. Although the transport from the Golgi apparatus to the dendrites was inhibited, budding of the VSV virus still occurred in the cell body area. This indicates that the inhibition of viral protein delivery was due to the lack of microtubules and not caused by a functional impairment of the Golgi apparatus. The effects of nocodazole treatment were reversible. Already 1.5 hours after removal of nocodazole the VSV glycoprotein was observed again in the dendrites and the hemagglutinin of FPV in the axon. These results suggest that the delivery of both axonal and dendritic exocytic vesicles is mediated by microtubuli.

Since the peripherally located microtubules of axons and dendrites are of uniform polarity, with their plus ends directed distally (Baas et al., 1988), our results pose a challenge to the newly formed axonal and dendritic plasma membrane proteins: how to choose the right microtubules. The answer to that question appears complicated and several specific molecular interactions have to come into place simultaneously. A simple scenario would be as follows. Axonal and dendritic membrane proteins are packed in different vesicles at the trans-Golgi network, each one having specific proteinaceous and lipidic partners, as demonstrated for apical and basolateral vesicles of epithelial cells (Wandinger-Ness et al., 1990; Kurzchalia et al., 1992). After budding from the tran,s-Golgi network, different microtubule-vesicle-binding motors with a plus-end, kinesin-like, directed force will associate with axonal and dendritic vesicles (Hirokawa et aL, 1991). This is quite likely considering the number, at least 35, of members in the kinesin family (Goldstein, 1991). Some of them are known to associate specifically with a certain class of vesicles like synaptic vesicles (for the kinesin Unc104) or vesicles containing neurotransmitter channels (Hall and Hedgecock, 1991; Gho et al., 1992). The last condition would then be the recognition of the axonal and dendritic plus- ended microtubules. The different MAPs that associate with axonal or dendritic microtubules (Kosik and Finch, 1987; Sato-Yoshitade et al., 1989; Caceres et al., 1984) can act as such recognition molecules.

Although the axonal and dendritic plasma membranes have different compositions, the plasma membrane is in principle continuous. The two membrane domains still maintain their unique features, indicating that diffusion obviously does not occur. In epithelial cells, tight junctions, located at the border between the apical and the basolateral domain, prevent mixing of the different membrane components (van Meer and Simons, 1986; Gumbiner, 1987). To determine whether a diffusion barrier is present in neurons, experiments similar to those described by van Meer and Simons (1986) were performed (Kobayashi et al., 1992). Vesicles containing water-insoluble fluorescent lipids were added to influenza virus-infected polarized neurons. Taking advantage of the fusogenic activity of the influenza hemagglutinin protein at low pH, fluorescent vesicles fused exclusively to the axonal domain. Uninfected neurons never showed any labelling, indicating that the fusion process was indeed hemagglutinin dependent. Although the fluorescent lipids were mobile in the axonal domain, as shown by energy transfer and fluorescence recovery after photobleaching, diffusion of labelling to the cell body or dendrites was never observed; the fluorescence terminated abruptly at the axon origin (Fig. 4). Since the hemagglutinin was inserted along the entire axonal surface, from the axon origins to their ends, it is unlikely that the lack of labelling in the cell soma is due to the slow diffusion of the labelled lipids. Even four hours after fusion, labelling of the somatodendritic domain was not observed (T. Kobayashi and C. G. Dotti, unpublished observation). The diffusion barrier was not located in the axonal cytosol; fusion of liposomes containing the fluorescent water-soluble calcein resulted in labelling of the entire cell cytosol, including the somatodenditic part. By fusing both symmetrically and asymmetrically labeled vesicles, which contain the fluorescent lipids in both leaflets or only the inner leaflet of the vesicle, respectively, it became clear that the diffusion barrier was in both leaflets of the plasma membrane. This is in contrast to the barrier of epithelial cells, which is present in the outer leaflet of the plasma membrane only (van Meer and Simons, 1986).

Although the precise molecular composition of this diffusion barrier is unknown, it must block the diffusion of plasma membrane proteins as well because many, including GPI-linked proteins, are highly mobile in the plasma membrane (Zhang et al., 1991; Ishihara et al., 1987), but do not mix with the dendritic domain (Dotti et al., 1991). It is plausible that a submicroscopic organization is responsible for such a fence. However, its true nature is still a mystery.

Morphological polarity of hippocampal neurons in culture is visible early after plating. Between two and three days of culture, the future axon can be distinguished from the other outgrowing neurites (Dotti et al., 1988). We referred to those young polarized neurons as ‘stage 3’ neurons. This morphological polarization is not yet fixed in this phase of development. Amputating the major neurite, which will develop into the axon, resulted in the outgrowth of one of the remaining smaller neurites (Dotti and Banker, 1987). Lack of sorting of certain molecules accompanies this absence of morphological commitment. Thus, Thy-1, which is axonal in the mature hippocampal neurons, was found in axons and dendrites of stage 3 cells (C. Dotti, R. Parton and K. Simons, unpublished results) and the same was found when the distribution of viral glycoproteins was analyzed (Dotti and Simons, 1990). Other examples of proteins that show a polarized distribution in the mature but not in the young cells are the transferrin receptor, the cytoskele- tal proteins MAP 2 and tau and the small GTP-binding protein rab8 (Dotti and Simons 1990; Dotti et al., 1990, 1991; Cameron et al., 1991; Kosik and Finch, 1987; Huber et al., 1993). Even GAP-43, a protein specifically involved in the early stages of axonal growth, has recently been shown to be present in axon and dendrites (van Lookeren Campagne et al., 1992; however, see Goslin et al., 1988). It is therefore obvious to ask when and how molecular sorting first occurs in neuronal cells. Will epithelial cells help to disclose the mechanisms underlying the initial stages of neuronal sorting? In epithelia, the development of polarity seems to be induced by cell-cell and cell-substratum contact (Gumbiner and Simons, 1986). This results in the gradual redistribution of ankyrin and spectrin, which leads to a polarized organization of Na⁺,K⁺-ATPase and probably triggers other intracellular mechanisms involved in the polarized delivery of newly synthesized proteins (Rodriguez-Boulan and Nelson, 1989; McNeill et al., 1990). During neuronal development, numerous cell contacts take place. However, these are transitory. The fixed neuronal circuitry, in the form of synaptic appositions, occurs at later times. Thus, one may speculate that molecular sorting would take place after the final synaptic contacts have been established. Whether electrical activity or cellular interactions are responsible for the triggering of molecular sorting must be tested experimentally.

Neurons polarize in a very complicated fashion. Some neurons are unipolar, some are bipolar and some are multipolar. Even among the same type, there are numerous differences: in the type of neurotransmitter, in the length of the axon, in the target cells, in the growing environment and in the number of dendrites. As they are such complex and heterogeneous cells, the mechanisms of sorting will be varied and complex. However, it is still possible that some of the basic mechanisms are conserved. To analyze this, we asked if the mechanisms of sorting of epithelial cells also existed in a neuronal cell type. We found some striking similarities but also divergences. Nevertheless, and independently of resemblances and differences, subordination to epithelial cell biology has been rewarding and we have answered some specific neurobiological questions. More importantly, new questions have been engendered and it may be that neurons, because of their special morphological features, become useful for studying some problems of intracellular sorting that are difficult to approach in epithelial cells; for instance, the direct observation of membrane traffic, as recently shown during endocytosis (Parton et al., 1992).

Special thanks are given to Liane Meyn for her excellent technical assistence and to Kai Simons for his constant encouragement and advice. We also thank Anne Walter (Cell Biology Program, EMBL) for help in the preparation of the manuscript, and Toshihide Kobayashi and Lukas Huber (EMBL), who worked on some of the experiments described here. Meltsje de Hoop is a recipient of a Human Frontier Science Program fellowship (LT 118/92).