Making the connection – shared molecular machinery and evolutionary links underlie the formation and plasticity of occluding junctions and synapses

More than 20 years ago, Woods and Bryant noticed that, seemingly diverse, cell–cell junctions all had a member of the membrane-associated guanylate kinase (MAGUK) protein family as a main component (Woods and Bryant, 1993). This finding, coupled with other observations, suggested a common evolutionary origin for vertebrate epithelial tight junctions, which were known at the time to possess the MAGUK zona occludens protein 1 (ZO-1), invertebrate epithelial pleated septate junctions (pSJs) – which were known to contain the MAGUK Disc large (Dlg), and synaptic junctions – which were known to contain the MAGUK postsynaptic density protein 95 (PSD-95) in mammalian brain (Woods and Bryant, 1993). In this Hypothesis, we revisit this topic as the last two decades of research have provided considerable molecular and functional evidence that these junctions and, in addition the paranodal junction, are all derived from an ancestral epithelial pSJ containing MAGUKs in complexes with members of two families of cell adhesion proteins, the neurexin superfamily and the immunoglobulin (Ig) cell adhesion molecule (IgCAM) superfamily. We consider evidence that the mechanisms controlling the plasticity of epithelial junctions through these proteins arose early in evolution and that these mechanisms formed the basis of synaptic plasticity in the later-emerging synapse. We propose that findings regarding how the formation of epithelial occluding junction formation and plasticity is regulated, is relevant to the formation and plasticity of synapses, and vice versa.

There is ample evidence that epithelia were the first tissues to develop in metazoans and we should look to the emergence of epithelia during metazoan evolution for the probable origins of the septate and related junctions (Tyler, 2003). Two types of septate junction have been described in invertebrates, the pSJ and the smooth septate junction (sSJ) with their names reflecting the difference in appearance of the junctional septa in oblique sections (Izumi and Furuse, 2014). In Drosophila, the organism in which septate junctions have been best described, pSJs occur in ectodermally derived epithelia and between glial cells at the blood–brain barrier, whereas sSJs are found in endodermally derived epithelia. Compared with the sSJ, the pSJ is the junction that has been characterized more extensively on molecular and genetic levels. Septate junctions are not found in vertebrate epithelia but a septate junction, the paranodal junction, connects myelinating glial cells to axons at the nodes of Ranvier in vertebrates (Schnapp et al., 1976). In transmission electron microscopy (EM) images, the pSJ has a ladder-like appearance between cell membranes that are about 15–18 nm apart (Green and Bergquist, 1982). Choosing the well-characterized Drosophila pSJ as the ‘gold standard’, we can consider structural and molecular equivalents along the branch points of metazoan evolution. The list of proteins associated with fly pSJ continues to grow and we will focus on the best-characterized members of that list (Limmer et al., 2014). At the core of the junction is a set of Drosophila proteins shown to be immobile in photobleaching studies; it includes the neurexin superfamily member neurexin IV (Nrx-IV), the L1 IgCAM family member neuroglian (Nrg), coracle (Cora), the Drosophila band 4.1 protein, and the ATPα and nervana 2 (Nrv2), subunits of the ion pump Na⁺/K⁺ ATPase (Baumgartner et al., 1996; Genova and Fehon, 2003; Lamb et al., 1998; Laval et al., 2008; Oshima and Fehon, 2011). Other IgCAMs associated with the pSJ include, contactin (Cont), fasciclin3 (Fas3), lachesin (Lac) and, possibly, fasciclin2 (Fas2) (Fig. 1, Table 1) (Faivre-Sarrailh et al., 2004; Llimargas et al., 2004; Narasimha et al., 2008; Strigini et al., 2006; Tonning et al., 2005). Not surprisingly, the majority of pSJ-associated proteins contain domains that are involved in cell–cell adhesion, with the Na⁺/K⁺ ATPase being a notable exception. Na⁺/K⁺ ATPase is required for pSJ formation but this function might be independent to that of an ion pump (Genova and Fehon, 2003; Paul et al., 2007). Interestingly, studies in vertebrate epithelial cells indicate that the Na⁺/K⁺ ATPase can contribute to cell–cell adhesion through β-subunit trans-dimerization with neighboring cells (Vagin et al., 2012).

Although pSJ-like structures have been reported in metazoans that lack true epithelia, such as sponges, these early-branching metazoans lack some of the known molecular components of the pSJ (Ganot et al., 2015; Green and Bergquist, 1982; Ledger, 1975). Trichoplax adhaerens is the only extant member of the early-branching metazoan phylum Placozoa and a very simple organism that consists largely of two epithelial sheets, the dorsal and the ventral epithelia (Smith et al., 2014; Srivastava et al., 2008). Detailed EM studies on Trichoplax are lacking, but septa-like structures between cell membranes that are ∼20 nm apart have been reported in the ventral epithelium (Ruthmann et al., 1986). Trichoplax appears to have the majority of the molecular components found at the Drosophila pSJ, and, at the time of its branching-off, it is likely that metazoans had evolved a structure that is molecularly similar to the current pSJ (Ganot et al., 2015). Neural tissues of the kind seen in later-emerging metazoans are not present in early-branching metazoans, such as Trichoplax and the sponges, supporting the idea that the septate junction arose before the synapse. The early-branching ctenophores have a neural system, but this is molecularly very different from the later-emerging nervous system and appears to be a case of convergent evolution (Moroz, 2015).

An important distinction in cell junction organization in epithelia between vertebrates and invertebrates is the nature of the occluding junction that blocks paracellular flow between apical and basolateral epithelial surfaces. In most invertebrate epithelia, this function is performed by a pSJ that is located basally to the adherens junction, whereas in vertebrates, it is carried out by a tight junction apical to the adherens junction. Why the change in junctions? The answer may lie in the emergence of the axo–glial paranodal junction and/or the synapse during evolution, leading to a shift in the expression of septate junction proteins to non-epithelial cells in a case of gene co-option. One mechanism of gene co-option is the acquisition of novel cis-regulatory elements in a gene, leading to its expression in new tissues (True and Carroll, 2002). In what has been termed gene sharing, the ancestral function can either be maintained along with the function in the new tissue, or being lost (True and Carroll, 2002). Gene co-option may have commenced before the vertebrate-invertebrate divergence, but has followed a different course in the two lineages, with maintenance of the ancestral epithelial function in invertebrates (gene sharing) but loss of this role in vertebrates.

Drosophila provides an example of how pSJ formation in an epithelium is blocked during normal embryonic development through the shutdown of a few genes (Narasimha et al., 2008). The amnioserosa is a transient epithelium in the fly embryo that never forms pSJs, and this appears to be due to a lack of the transcription factor Grainyhead. Misexpression of Grainyhead in the amnioserosa causes expression of the pSJ proteins Dlg, Nrx-IV, Fas3, Cora and the claudin-like protein sinuous (Sinu) (Narasimha et al., 2008). The transcription factor Grainyhead-like 2 is required for tight junction formation in mice, suggesting an ancient regulation of occluding junctions by the Grainyhead family of proteins (Senga et al., 2012; Tanimizu and Mitaka, 2013). An interesting possibility is that prior to the split between invertebrates and vertebrates, an ancestral Grainyhead factor regulated expression of pSJ genes, but for some of these genes, this control was co-opted in the vertebrate lineage by the requirement for these proteins in neuronal, muscle and/or glial cells. With the loss of these proteins from the vertebrate epithelium, a new occluding junction emerged, the tight junction. It would be of interest to examine a panel of epithelial tissues and cell lines to see if a particular pSJ protein or proteins is consistently missing or expressed at low levels. A combination of bioinformatics, evolutionary comparisons and promoter studies could then be used to determine if Grainyhead factor binding sites in such a gene had been converted to regulatory sequences that promote gene expression in other tissues.

An alternative, but not mutually exclusive explanation for the appearance of the tight junction is differing functional requirements following the vertebrate-invertebrate divergence. For example, the challenge of living with an open circulation in invertebrates may have favored the maintenance of the robust pSJ, whereas a more plastic occluding junction might be desirable during vertebrate development.

The molecular structure of the tight junction is quite distinct from that of the pSJ but they both contain members of three protein families: claudin/claudin-like, the MAGUKs (represented by ZO-1, ZO-2 and ZO-3 in tight junctions) and IgCAMs – represented by the junctional adhesion molecules (JAMs) – in tight junctions (Fig. 1, Table 1) (Behr et al., 2003; Nelson et al., 2010; Niessen, 2007; Wu et al., 2004). The Na⁺/K⁺ ATPase is required for tight junction formation, but in most epithelia is localized to the basolateral membrane, separate from the apical tight junction (Rajasekaran et al., 2001; Violette et al., 2006). An exception are the retinal pigment epithelial cells, where the Na⁺/K⁺ ATPase is apically localized and required for tight junction structure and function (Rajasekaran et al., 2003).

Some of the tight junction components may have been present prior to the vertebrate-invertebrate divergence. For example, claudins are an essential component of tight junction paracellular barrier function and Drosophila has multiple claudin-like molecules, three of which have been shown to be part of the pSJ and are required for its proper development (Behr et al., 2003; Nelson et al., 2010; Wu et al., 2004). However, it should be noted that the evolutionary relationship between the vertebrate claudins and claudin-like proteins has not been well defined (Ganot et al., 2015). Drosophila also has a ZO-1 orthologue, which is localized apically to the septate junction at the adherens junction, in the vicinity of where the tight junction would form in vertebrates (Choi et al., 2011). It remains uncertain whether any true tight junctions exist in invertebrates because, although junctions that structurally resemble tight junctions have been observed, they have not been characterized molecularly (Lane and Chandler, 1980).

With the emergence of neural tissue, the pSJ adapted to function as a glial–axonal and glial–glial junction (as described above) as well as a synapse. The cleft between membranes at the synapse is ∼15–25 nm, which is similar to that of the pSJ cleft and considered optimal for synaptic function (Savtchenko and Rusakov, 2007). Transmission EM of the well-characterized synapses at the Drosophila neuromuscular junction (NMJ) reveals the presence of regularly spaced electron-dense material in cross section and a honeycomb-like appearance in oblique sections, which both are similar in appearance to the pSJ (Prokop, 1999). Similarly, rat hippocampal synapses exhibit intermittently occurring complexes in the synaptic cleft (Zuber et al., 2005). However, in Drosophila, when comparing the NMJ synapse to that of the pSJ, a number of changes in the adhesive apparatus are apparent (Banovic et al., 2010; Chen et al., 2012; Koper et al., 2012; Sun et al., 2011; Xing et al., 2014). One important distinction is the presence of neurexin-1 (Nrx-1) and the absence of Nrx-IV. Nrx-1 is a fly orthologue of a family of three mammalian neurexins (namely NRXN1, NRXN2 and NRXN3) that are quite distinct from Drosophila Nrx-IV and its vertebrate ortologue Caspr (officially known as CNTNAP1). During evolution, the synaptic neurexins appeared before Nrx-IV/Caspr, but no function for these molecules prior to their use in the synapse has been identified so far (Ganot et al., 2015; Moroz et al., 2014; Nichols et al., 2006). Another change in the fly NMJ synapse compared to the pSJ is the appearance of a family of non-catalytic cholinesterase-like molecules, the neuroligins, that are distinct from gliotactin, a non-catalytic cholinesterase-like molecule found at the tricellular pSJ (Genova and Fehon, 2003; Schulte et al., 2003). The neuroligins are important components of synapses and form transynaptic connections with neurexin (Banovic et al., 2010; Chen et al., 2012; Sun et al., 2011; Xing et al., 2014). In contrast, at the pSJ, there is no indication that Nrx-IV binds to gliotactin. Neurexin–neuroligin interactions are complex in mammals with the mammalian neurexin genes each encoding a long and a short version, α-neurexin (NRXN1) and β-neurexin (NRXN2), respectively. The neurexin and neuroligin genes also generate protein variants by alternative splicing (Wei and Zhang, 2010). Cora and the IgCAMs Fas2 and Fas3 persist at the Drosophila NMJ synapse together with the MAGUK Dlg (Chen et al., 2005; Kose et al., 1997; Lahey et al., 1994; Schuster et al., 1996a). It is worth noting that the mammalian excitatory synapse has a similar collection of pSJ-like proteins as the Drosophila NMJ synapse (Table 1 and Fig. 1) (Giagtzoglou et al., 2009).

Na⁺/K⁺ ATPase subunits are widely expressed along neural membranes in both Drosophila and vertebrates, and could therefore be components of a junctional complex at the synapse and at the pSJ (Lebovitz et al., 1989; Mobasheri et al., 2000; Sun et al., 1998). In line with this, Drosophila with mutations in the ATP synthase α-subunit (ATPα) show overgrowth of the NMJ, a phenotype that could reflect increased plasticity (Trotta et al., 2004). The α3-subunit of vertebrate Na⁺/K⁺ ATPase is enriched at synapses of the central nervous system and serves as a receptor for agrin, a heparan sulfate proteoglycan involved in synapse formation both in the brain and at NMJs (Hilgenberg et al., 2006).

Overall, synapses show many molecular parallels with the pSJ and, in terms of adhesion, can be thought of as a ‘modified’ pSJ. Layered onto this modified pSJ, is the large set of proteins, including receptors and ion channels, that constitutes the functional apparatus of the synapse. Many of these proteins are present at the protosynapse – an ancestral molecular machinery – in early-branching metazoans where, presumably, they had different roles prior to their recruitment into the emerging synapse (Alie and Manuel, 2010; Ryan and Grant, 2009; Sakarya et al., 2007). An indication that protosynapse proteins may have been associated with the pSJ during the transition to the tight junction in vertebrates is the finding of a cluster of synaptic molecules associated with the tight junction in cultured human epithelial cells (Tang, 2006).

The basket cells – inhibitory GABAergic interneurons – of the vertebrate cerebellar cortex provide additional evidence for the emergence of synapses from the pSJ. The axon branches of these cells envelop the Purkinje cell body in a basket-like arborization, with axo-axonal septate junctions between basket cell branches and between branches and the beginning of the Purkinje cell axon (Gobel, 1971; Sotelo and Llinas, 1972). In contrast to the pSJs that link epithelial cells, septate junctions of basket cells are short and found in arrays of three or more between two neighboring axons. These septate junctions have some properties of synapses but do not appear to function as synapses. For instance, although they contain post-synaptic voltage-gated K⁺ channels and the MAGUK PSD-95, they do not have voltage-gated Na⁺ channels (Laube et al., 1996). Furthermore, although there are synaptic vesicles in the vicinity of the septate junctions, they do neither cluster at the junctions, nor do the junctions exhibit a post-synaptic density by EM (Gobel, 1971). It is possible that these junctions, which may electrically inhibit Purkinje cells (Korn and Axelrad, 1980), represent an intermediate in the derivation of synapses from pSJs.

Further evidence that synapses have derived from pSJs comes from the discovery of a synapse that is particularly similar to a septate junction. The calyceal synapses of vestibular hair cells in the mouse show a high similarity to septate junctions in EM microscopy and contain both Caspr and contactin, with loss of Caspr resulting in the separation of the membranes at the synapse (Lysakowski et al., 2011; Sousa et al., 2009). Recently, it has been demonstrated that a contactin5 (CNTN5)–Caspr4 complex together with the IgCAM NgCAM-related cell adhesion molecule (NrCAM) guides the formation of synapse-to-synapse (axoaxonic) synapses between GABAergic interneurons and sensory motor synapses in the spinal cord, and there are likely to be other examples of ‘transitional' synapses that would benefit from further molecular characterization (Ashrafi et al., 2014).

Previous review articles have already commented on a probable common ancestor for the pSJ and the paranodal junction, and this will not be discussed in detail here (Banerjee et al., 2006; Hortsch and Margolis, 2003). The pSJ and paranodal junctions are similar in that they share a core adhesive complex that – in Drosophila – is composed of three proteins: Nrx-IV (Caspr in vertebrates), Nrg (neurofascin in vertebrates, herafter referred to as NF155), and contactin (Cont; and CNTN1–6 in vertebrates) (Fig. 1, Table 1). However, two distinctions are the narrower cleft between cell membranes in the paranodal junction and the presence of Gliotactin at the tricellular septate junction in the fly (Banerjee et al., 2006; Hortsch and Margolis, 2003). Gliotactin is required for correct formation of the Drosophila septate junction (Genova and Fehon, 2003; Schulte et al., 2003) but might not be required at the paranodal junction because tricellular adhesions are not present. As discussed above, the Na⁺/K⁺ ATPase is an essential component of the pSJ; but, although, this enzyme is enriched in glial cells at the paranodal junction, we are not aware of it having been characterized with regard to a role at this junction (Ariyasu et al., 1985; Ogawa et al., 2006).

We have argued above that occluding junctions and synapses have their origins in an ancestor that was largely composed of epithelial cells. The ability to achieve diversity in epithelial development requires plasticity in the form of making, breaking and modifying cell–cell adhesions (Lecuit, 2005). Similarly, synaptic plasticity during development, learning and memory requires the making and breaking of adhesions across the synaptic cleft (Giagtzoglou et al., 2009). Below, we consider how the shared gene families might have first developed mechanisms to regulate epithelial formation and plasticity that were later used again for similar functions at the synapse. Rather than an exhaustive consideration of all factors, we will focus on a few mechanisms that occur in epithelia as well as synapses. We will not consider here the importance of cadherins in both epithelial and synaptic plasticity, as this Hypothesis focuses on parallels between occluding junctions and the synapse. For a thorough review of synaptic cell adhesion the reader is referred to Missler et al., 2012.

Fas2 is a Drosophila IgCAM and an orthologue of the mammalian neural cell adhesion molecule 1 (N-CAM-1) that is expressed both pre- and post-synaptically at the NMJ (Schuster et al., 1996a). Imaging of live embryos (Kohsaka et al., 2007) revealed that Fas2 accumulates postsynaptically when filopodia extending from motor neurons make contact with myopodia from muscles at the nascent synapse (Kohsaka et al., 2007). This postsynaptic clustering of Fas2 is dependent on presynaptic Fas2, suggesting that homophillic adhesion between Fas2 molecules on opposite sides of the synapse is one of the early events in synaptic development. Fas2 subsequently recruits the MAGUK Dlg, which is a Fas2-binding protein, to the early synapse, which may anchor Fas2 to the cytoskeleton (Thomas et al., 1997; Zito et al., 1997). Dlg itself is required for development of normal synaptic structure, and at the mature larval NMJ, synaptic localization of Fas2 is dependent on Dlg, as it is disrupted in dlg mutants in Drosophila (Guan et al., 1996; Lahey et al., 1994; Thomas et al., 1997; Zito et al., 1997).

Synaptic plasticity at the Drosophila NMJ involves synaptic expansion because the NMJ must keep up with the enormous growth of the muscle during larval development (Schuster et al., 1996a). Downregulation of the synaptic levels of Fas2 leads to synaptic sprouting, suggesting that weakening of adhesion across the synapse is required for synaptic growth (Schuster et al., 1996a,,b). One mechanism by which the synaptic levels of Fas2 are controlled is through phosphorylation-dependent regulation of Dlg localization by the kinases Ca²⁺/calmodulin-dependent protein kinase II (CaMKII) and Partitioning-defective-1 (PAR-1) (Koh et al., 1999; Zhang et al., 2007). Phosphorylation of Dlg by either of these kinases impedes its localization at the NMJ, and this subsequently reduces synaptic Fas2 levels, which may have a bearing on plasticity.

Despite the absence of synapses in Trichoplax, a gene encoding a Fas2-like protein has been found in this organism, and Fas2 is likely to have started out as an epithelial junction protein before assuming a synaptic role (Srivastava et al., 2008). Emerging findings indicate that epithelial plasticity in Drosophila is also regulated by modulation of Fas2 and Dlg, demonstrating that a particular IgCAM or MAGUK can have similar functions at both the synapse and the epithelial membrane. CaMKII-mediated phosphorylation appears to be capable of modulating the localization of Dlg at the epithelial membrane in the Drosophila embryo, whereas studies on the Drosophila follicular epithelium during oogenesis are revealing roles for Fas2 in epithelial plasticity (Szafranski and Goode, 2004, 2007; Wang et al., 2011). In the follicular epithelium, Fas2 is a component of a basolateral complex that matures into a pSJ in the late egg chamber (Muller, 2000; Szafranski and Goode, 2004, 2007). During oogenesis, a cluster of cells, the border cells, delaminates from the follicular epithelium and migrates through the egg chamber during epithelial–mesenchymal transition (EMT), a form of epithelial plasticity (Montell et al., 2012). Similarly to its function at the synapse, Fas acts as an inhibitor of this epithelial plasticity, and is required for correct Dlg localization in the follicular epithelium (Szafranski and Goode, 2004, 2007).

Another form of follicular epithelial plasticity inhibited by Fas2 is the conversion of cuboidal epithelium to a flattened, squamous epithelium (Gomez et al., 2012). This morphogenesis involves shrinkage of the epithelial lateral membrane, which is inhibited by Fas2-mediated cell–cell adhesion. The inhibition is released by endocytosis of Fas2 and, endocytosis of Fas2 at the synaptic membrane of the NMJ might contribute to synaptic plasticity (Mathew et al., 2003). Furthermore, Dlg, the interaction partner of Fas2, functions as a tumor suppressor and is required for epithelial integrity in several different types of epithelium during Drosophila development, including the follicular epithelium, embryonic epidermis and imaginal discs (Bilder et al., 2000). Thus, downregulation of Fas2 and Dlg function is associated with both synaptic and epithelial plasticity.

As discussed above, Fas2 and Dlg are involved not only in synaptic plasticity but also the initial formation of the synapse; thus, the question arises whether these proteins also promote formation of the pSJ. To our knowledge, Fas2 has not been evaluated in Drosophila regarding a role in pSJ formation but the IgCAMs contactin, Nrg, Lac have shown to be involved (Faivre-Sarrailh et al., 2004; Genova and Fehon, 2003; Llimargas et al., 2004; Strigini et al., 2006). That Fas2 is interacting with pSJ components and involved in pSJ formation is suggested by the finding that Fas2 mutants exhibit the same defect in tracheal epithelia as mutants in several pSJ proteins, including Pickel (also known as Megatrachea), ATPα, Lac, Nrx-IV, Cora and Sinu (Tonning et al., 2005).

Evaluating the contribution of Dlg to pSJ formation is complicated by its role in specification of apicobasal polarity (Gibson and Perrimon, 2003). Compared to some other pSJ proteins, which form a highly immobile heteromeric complex, Dlg is highly mobile (Laval et al., 2008; Oshima and Fehon, 2011). Dlg is not required for formation of the immobile pSJ protein complex, but regulates its positioning (Oshima and Fehon, 2011). In contrast, the MAGUK varicose (Vari) is required for formation of the immobile pSJ complex (Laval et al., 2008). Similar to the role of Dlg in Drosophila epithelial cells, photobleaching experiments indicate a dynamic behavior of Dlg at the Drosophila NMJ (Zhang et al., 2007). Oshima and Fehon have demonstrated a mobility of pSJ components before junction formation in the late embryo, suggesting the existence of a plastic, immature pSJ (Oshima and Fehon, 2011). We propose that Dlg and its partner Fas2 are components of this immature pSJ, and regulate its plasticity prior to the emergence of the final, stable barrier. As suggested by studies on the follicular epithelium, the immature pSJ might show plasticity to enable epithelial morphogenesis, with the barrier function of the immobile pSJ setting in later (Gomez et al., 2012; Muller, 2000; Szafranski and Goode, 2004, 2007).

SynCAM1 (also known as CADM1) is a vertebrate IgCAM that shows some functional similarities to Fas2. SynCAM1 interacts with the MAGUK CASK, is involved in synapse formation and can promote synapse formation even in non-neuronal cells (Biederer et al., 2002; Fogel et al., 2011). It also inhibits synaptic plasticity and SynCAM1-knockout mice show improved spatial learning (Robbins et al., 2010). Thus, a variation in the expression levels of SynCAM1 might affect synaptic plasticity in a way that is similar to that of Fas2 at the Drosophila NMJ.

JAMs are members of the IgCAM superfamily with two Ig-like domains and, thus, can form both homophilic and heterophilic interactions (Garrido-Urbani et al., 2014). JAM-A is found together with its binding partner the MAGUK ZO-1 in the lamellipodia of adjacent cells that make nascent points of contact (Ebnet et al., 2001). JAM-A participates in the recruitment of the complex between Bazooka (Baz)/Par-3 (also known as PARD3 in mammals), Par-6 (also known as PARD6 in mammals) and aPKC (Baz/Par-3–Par-6–aPKC) and, as polarity emerges, forms a complex with the claudins, which also bind to ZO-1 (Ebnet et al., 2001; Furuse et al., 1994; Itoh et al., 1999, 2001). In addition to recruitment of tight junction proteins, ZO-1 anchors the JAM-A-containing adhesion complex to the cytoskeleton by directly binding to actin and possibly also through protein 4.1 (Fanning et al., 2002; Mattagajasingh et al., 2000). The importance of JAM-A in tight junction assembly is underscored by the finding that expression of mutant versions of JAM-A that cannot localize at cell–cell contacts causes a defect in tight junction formation (Rehder et al., 2006).

As a link to the cytoskeleton, ZO-1 is likely to be an important stabilizer of the tight junction, and perturbations of ZO-1 are associated with a disruption of tight junction function. Knockdown of ZO-1 affects the tight junction barrier function, leading to a permeability increase for large solutes (Van Itallie et al., 2009). Furthermore, enhanced ZO-1 phosphorylation has been observed in MDCK cells with increased barrier permeability, and alterations in ZO-1 function are associated with EMT and tumorigenesis (Harhaj and Antonetti, 2004; Stevenson et al., 1989). Disassembly of tight junctions by PKC isoforms might involve mislocalization of ZO-1 because activation of PKC signaling through various means has been shown to disrupt ZO-1 positioning at cell borders (Harhaj and Antonetti, 2004). As seen with the immature pSJ and the synapse, these various results indicate that regulation of a MAGUK affects the stability of a junctional structure. For an additional consideration of how junctions of varying stability might be established and regulated see Boxes 1 and 2.

Thus far, we have focused on the parallels between the junctional complex of proteins at vertebrate and invertebrate synapses and the invertebrate pSJ. However, are there further parallels between the organization of the vertebrate and invertebrate synaptic membranes and those of invertebrate epithelial cells, from which they were derived? To address this issue, we consider the Drosophila NMJ bouton – a swelling at the tip of the axon where multiple synapses with the muscle are formed – and compare it to an epithelium. Epithelial cells show a polarized distribution of proteins in the apicobasal axis and, in Drosophila, five membrane domains are discerned; the free apical and the marginal zone (which, together, constitute the apical membrane), and the adherens junction, lateral and basal domains (Fig. 2A) (Tepass, 2012). Both the presynaptic membrane and postsynaptic membrane of the bouton also show a polarized distribution of proteins, although the postsynaptic polarity is more apparent. Depending on the protein distribution, the postsynaptic membrane can be divided into four domains – active zone, periactive zone, peribouton zone and surrounding muscle membrane (hereafter referred to as the perisynaptic zone) (Fig. 2B) (Ruiz-Canada et al., 2004; Sone et al., 2000). The postsynaptic membrane that surrounds the active zone is long and folded into a structure in the periactive zone that has been termed subsynaptic reticulum (SSR). Many of the proteins that show a polarized distribution in the epithelial membrane also show a polarized distribution in the postsynaptic membrane. Both the apical membrane of epithelial cells and the active zone membranes of the synapse are sites of dynamic interaction with the cellular environment. The apical membrane and presynaptic active zone membrane are sites of endocytosis and exocytosis, whereas the postsynaptic active zone membrane and the apical membrane contain signaling receptors. The adherens junction in epithelial cells is actin rich, similar to the peribouton region of the bouton (Ramachandran et al., 2009; Tepass, 2012). The lateral membrane is the site of the septate junction, whereas the active and periactive zones of the bouton are the location of the synapse (Banerjee et al., 2006; Tepass, 2012). Finally, comparing the basal membrane of epithelia cells with the bouton, receptors for the extracellular matrix, such as the integrins and dystroglycan, that are localized basally in Drosophila epithelia are largely found in the peribouton and perisynaptic zones at the NMJ (Beumer et al., 2002, 1999; Bogdanik et al., 2008; Deng et al., 2003; Sone et al., 2000; Tanentzapf et al., 2000) (Fig. 2).

The extent of the epithelial apical membrane is regulated by a complex interplay between several groups of proteins, that has been best characterized by genetic studies in Drosophila (Rodriguez-Boulan and Macara, 2014; Tepass, 2012). The Scribble complex, of which Dlg is a member, is required for correct pSJ formation and regulates the size of the apical domain. Dlg is found at the lateral membrane in Drosophila epithelia and in the active zone and peribouton region of the NMJ (Lahey et al., 1994; Sone et al., 2000; Woods and Bryant, 1991). Loss of epithelial Dlg leads to disruption of polarity and expansion of apical proteins along the lateral membrane and, in an interesting parallel, loss of Dlg at the NMJ leads to an expansion of the active zone into the periactive zone and the reduction of the SSR (Bachmann et al., 2010; Bilder and Perrimon, 2000; Bilder et al., 2003; Lahey et al., 1994; Mendoza-Topaz et al., 2008; Tejedor et al., 1997; Thomas et al., 1997) (Fig. 2B,D).

A second group of proteins participating in apicobasal polarity is the Baz/Par-3–Par-6–aPKC complex, which also functions in patterning of the postsynaptic membrane. This complex is involved in establishing the F-actin-rich adherens junction in epithelial cells and, similarly, is required for F-actin accumulation in the peribouton zone at the NMJ (Ramachandran et al., 2009; Ruiz-Canada et al., 2004; Tepass, 2012). Phosphorylation through aPKC regulates localization of Baz in the apicobasal axis of epithelial cells and, similarly, its distribution at the NMJ (Ramachandran et al., 2009; Rodriguez-Boulan and Macara, 2014; Ruiz-Canada et al., 2004; Tepass, 2012).

The Crumbs protein complex and the Coracle group of epithelial polarity proteins are two additional entities that regulate apicobasal polarity (Rodriguez-Boulan and Macara, 2014; Tepass, 2012). The Crumbs protein complex member Stardust/PALS1 (also known as MPP5 in mammals) is undetectable at the NMJ and, therefore, the complex might not be functioning here (Bachmann et al., 2004). The Coracle protein group members Cora and Na⁺/K⁺ ATPase were shown to regulate synaptic development at the NMJ but whether they function as part of a complex that polarizes the postsynaptic membrane remains unaddressed (Chen et al., 2005; Trotta et al., 2004).

We have explored the parallels between occluding junctions and the synapse with regard to their formation, plasticity and molecular composition. Our intention in considering the shared evolutionary origins of these various junctions is to encourage researchers to a broader use of available literature when developing their ideas. For example, further consideration of the complex protein–protein interactions that mediate the emergence of apicobasal polarity might provide additional clues on how the organization of the synapse is established. In a similar vein, McLachlan and Heiman have proposed that comprehension of epithelial morphogenesis can help us understand the shaping of dendrites (McLachlan and Heiman, 2013). By comparing the different junctions, one can make specific predictions that may fill in gaps in our knowledge. For example, there is evidence that PTEN regulates the distribution of Baz/Par-3 at the postsynaptic membrane by dephosphorylating it, raising the question whether PTEN-mediated dephosphorylation also regulates its distribution in the apicobasal axis of epithelial cells (Ramachandran et al., 2009). There is convincing evidence that a trimeric protein complex composed of Nrx-IV (Caspr), Nrg (NF155) and contactin (CNTN) contributes to septa found in septate and paranodal junctions, so we predict that the axo-axonal junction of basket cells also contains members of these three families. Indeed, basket cell axons in contact with Purkinje cells show strong Caspr3 immunoreactivity (Banerjee et al., 2006; Spiegel et al., 2002). Aside from the utility to experimental biologists, considering the evolutionary aspects of synapse origins is fascinating. Starting from a simple cellular junction that helped to hold our ancient ancestors together, an exquisitely modulated cell–cell communication device has appeared that underlies the remarkable complexity of the nervous system.

We thank members of N.H.’s and C.K.’s labs for discussions and Esther Verheyen for comments on the manuscript.