Organization and Function of the Blood–Brain Barrier in Drosophila

Formation of the blood–brain barrier

All Drosophila glial cells, except the midline glia, express the repo gene (Xiong et al., 1994; Lee and Jones, 2005). Several repo-positive glial cell layers cover the Drosophila nervous system. Two distinct cell types, the perineurial and the subperineurial glia form the outer cell layers. They separate axons and neuronal cell bodies, which are surrounded by cortex and wrapping glia, from the high potassium concentration in the hemolymph to allow normal electrical conductance. To characterize individual glial cells we followed either a flip-out strategy and expressed the Flp recombinase in all glial cells of flies carrying different flip-out constructs or we generated MARCM clones [using either a repoflp strain (Silies et al., 2007) or a repoGal4 UASflp strain]. These procedures result in the random, but specific labeling of few glial cells in every animal. The MARCM technology requires cell division to generate labeled cell clones whereas the flip-out technology can also be applied in postmitotic cells (Struhl and Basler, 1993; Lee and Luo, 1999). In previous reports these outer glial cell layers were often considered as single morphological unit termed surface glia, peripheral glia or perineurium and no comprehensive assignment of the morphological and functional diversity has been made (Bellen et al., 1998; Leiserson et al., 2000; Sepp et al., 2000; Bainton et al., 2005; Schwabe et al., 2005; Freeman and Doherty, 2006).

The neural lamella

A dense network of extra cellular matrix, called neural lamella, surrounds both the central and the peripheral nervous system and can be detected by a protein trap insertion into the collagen IV locus (Fig. 1A). The neural lamella can be detected from embryonic stage 16 onwards (supplemental Fig. 1A,B, available at www.jneurosci.org as supplemental material). No distinction can be made between the matrix covering the CNS and peripheral nervous system, indicating that the neural lamella is a continuous, relatively unstructured layer. In mutants with reduced hemocyte motility no lamella is formed (supplemental Fig. 1C, available at www.jneurosci.org as supplemental material) (Olofsson and Page, 2005).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Anatomy of the peripheral nerve. Orthogonal sections through stacks of confocal images of nerves of third instar larvae stained for HRP expression (blue), Repo expression (red), and GFP expression (green). A, The GFP-gene trap insertion in Collagen IV (Viking) labels the neural lamella. B, The Gal4 driver strain c527 activates expression of CD8:GFP predominantly in the perineurial cells, which are just below the neural lamella. C, The SPG-Gal4 driver activates CD8:GFP expression in the subperineurial cells that tightly encircle the axonal fascicles (blue). Note the occurrence of a glial cell nucleus in the fascicle (B, C). D, The GFP-gene trap insertion in neurexinIV (#454) labels a thin stripe along the entire axonal fascicle. E, Repo-Gal4 activates CD8:GFP expression in all glial cells. Note, that some GFP expression is also visible within the nerve. F, A GFP-gene trap insertion in the nervana2 gene labels glial cell membranes within the fascicle. Scale bars: 2 μm.

The perineurial layer

The perineurial glial cells are located just below the neural lamella. We did not identify a gene trap exclusively expressed in these cells. However, the Gal4 driver c527 (Hummel et al., 2002) shows preferential activity in this cell type as can be seen in orthogonal sections of third instar peripheral nerves (Fig. 1B). Cells within the perineurial glial cell layer of the CNS exhibit a star-like shape with many thin cell protrusions within the perineurial layer itself, which never invade into the neural tissue (Fig. 2A,B). Based on several MARCM clones comprising only perineurial glial cells (Fig. 2B), we conclude that these cells can divide considerably during larval stages. In the PNS, only few perineurial glial cells are found in the first instar stage. In third larval instar nerves, perineurial cells divide and cover the nerve. Here they appear elongated with fine cell processes (Fig. 3C,f, arrowhead).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Morphology of perineurial and subperineurial glial cells. To label individual cells or cell clones Flp expression was induced in glial cells (repoFlp; repoGal4 UASactin::GFP; Gal80 FRT19A/FRT19A). Larval nervous systems were stained for Repo expression (red) and GFP expression (green), and neurons were labeled with anti HRP (blue). A, MARCM cell clone of perineurial glial cells in the third instar larval brain. Note the extensive cell protrusions generated by these cells (arrowhead). B, A large MARCM clone consisting exclusively of perineurial cells in the abdominal part of the ventral nerve cord of a third instar larval brain. Perineurial glial cells show many fine filopodia-like cell protrusions (arrowhead). C, Small flip-out clone of subperineurial glial cells (repoGal4; UASflp; UAS>CD2 y⁺>mCD8GFP). In this confocal section, two of the nuclei can be identified. In contrast to the perineurial glia, the subperineurial cells never form lateral filopodia-like extensions (arrowhead). D, D', The large and flat subperineurial glial cells are covered by an outer layer of glial cells, the perineurium. D', In an orthogonal section, as indicated by the white line in D, Repo-positive glial nuclei are seen apically to the GFP expressing glial cells (arrowhead). E, Projection of confocal stacks of a third instar brain lobe stained for Repo (red), HRP (blue) and expression of NeurexinIV::GFP fusion protein that is confined to the septate junctions of subperineurial cells. Scale bar, 85 μm. F, F', CNS of a stage 16 embryo stained for Neurexin::GFP and Repo expression. Note that some Repo-positive nuclei are located apically to the GFP expressing subperineurial cells.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Glial cells in the peripheral nervous system. To label individual cells or cell clones Flp expression was induced in glial cells of animals carrying a UAS>CD2 y⁺>mCD8GFP construct. Larval nervous systems were stained for HRP expression (blue), Repo expression (red), and GFP expression (green). The top panel shows an overview of a ventral nerve cord with attached peripheral nerves. The boxed areas are shown in higher magnification in A–C. A, B, Peripheral nerve with a GFP labeled single wrapping glial cell. This glial cell spreads over at least 200 μm. The position of four orthogonal sections is indicated by small letters (a–d). The wrapping glial cells can form very thin processes that extend over long distances. C, C', The upper nerve shows a wrapping glia that has covered almost the entire fascicle. The lower nerve shows a perineurial glia clone. The arrowhead denotes a fine cell process. C', Different confocal z-section of the same region as shown in C. Note the processes of the perineurial glial cell that cover the nerve. The small letters indicate the position of the orthogonal sections (e, f). Fine perineurial cell processes are indicated by an arrowhead (C', f). g, h, Two additional examples of perineurial glial cells.

The subperineurial layer

The second glial cell layer is formed by the subperineurial cells. These cells specifically express the moody gene (Bainton et al., 2005; Schwabe et al., 2005) and we used moody promoter sequences to generate the subperineurial glia (SPG)-Gal4 driver (Fig. 1C). Generally, the subperineurial cells form a thin layer below the perineurial cells (Figs. 1C, 2D,F). In the CNS, single subperineurial cells labeled by the flip-out technique appear always as very large, hexagonally arranged cells often extending 40–80 μm in diameter (Fig. 2C–E). Subperineurial cells in the PNS are very large as well and can form autocellular junctions (see below).

The wrapping glia

The final glial cell layer in the PNS comprises the wrapping glia. During late embryonic and early larval stages, these glial cells initially contact fascicles of sensory and motor axons and only later during the third larval stage ensheath almost every single axon resembling the appearance of Schwann cells in Remak bundles of the mammalian PNS (see below) (Nave and Salzer, 2006). Along the peripheral nerves, only three nervana2-Gal4 (Sun et al., 1999) positive wrapping glial cells are found (Fig. 1F, supplemental Fig. 2, available at www.jneurosci.org as supplemental material). These cells extend long and thin processes that wrap individual axons (Fig. 3A,B,a–d). In a given section of a peripheral nerve two to three glial cell processes can be detected (see Figs. 3, 6) (see below). In the CNS, the inner glial layer is more complex and comprises cortex glia that insulate neuronal cell bodies as well as the initial segments of the axons and neuropil glia that ensheathes the axon fascicles and presumably contacts individual synapses in dendritic compartments (Younossi-Hartenstein et al., 2003; Pereanu et al., 2005). Further details on these glial cell classes will be presented elsewhere.

Sensory and motor axons are individually wrapped in the periphery

The PNS harbors two distinct axonal classes: the sensory axons that project into the CNS and the motor axons that project outwards from the CNS to the muscle fibers in the periphery. To address whether wrapping glial cells keep sensory and motor components separate, we used either a Gal4 driver that is active in most sensory neurons (43Gal4) and counterstained with anti-Fas II antibodies to label motor axons (Fig. 4A,C) or we used a gene trap insertion into the fas II gene mimicking fas II gene expression in all motoneurons (#397) (U. Lammel, unpublished observation) and then counterstained with anti-Futsch antibodies that label all sensory axons and only few motor axons (Hummel et al., 2000) (Fig. 4B,D). Confocal analyses of such embryos indicate that indeed the motor and sensory axons are kept in distinct fascicles (Fig. 4A–D). To test the organization of sensory and motor axons in third instar larvae we used a fasGal4 driver to label motor axons and a chaGal4 driver to visualize sensory axons (Fig. 4E,F). Again, confocal analyses demonstrated a clear separation of the two modalities within the segmental nerves.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Sensory and motor axons project in distinct fascicles. Preparations of stage 16 nervous systems stained for sensory and motor axons. Anterior is up. A, C, Embryo carrying the 43Gal4 insertion that directs GFP expression to many sensory neurons (green) and some neuropile glia in the CNS (ng) counterstained for the expression of the Fasciclin II protein (red) which is found on all motor axons as well as on some interneurons (in) in the CNS. Note that within the PNS sensory axons and motor axons are running in distinct trajectories (arrowheads). B, D, Embryo carrying a Fasciclin II GFP gene trap insertion (green) counterstained with Mab 22C10 labeling the Futsch protein (red) that is expressed in the peripheral sensory neurons and their axons. A similar separation of sensory axons and motor axons as in A and C can be seen. E, Third instar larvae expressing GFP in motor axons directed by a fasII:Gal4 driver (green). Motor axons are still found in one part of the nerve (sensory axons are labeled with 22C10; Futsch protein, red; all axons are counterstained with anti HRP in blue). F, Third instar larval nerves expressing GFP in sensory axons directed by the cha:Gal4 driver (green) do not intermingle with motor axons (red, anti-Fas II staining). The dotted line indicates the position of the orthogonal section shown on the right.

Ultrastructural analysis

To corroborate the above findings we determined the ultrastructural appearance of the different glial cells at embryonic and larval stages. At the electron microscopic level, the structure of the neural lamella appeared identical in the CNS and the PNS (Fig. 5). In embryonic stages, only few perineurial cells were identified. They generally develop numerous processes that strictly stay within the glial layer and mediate extensive cell–cell contacts among the perineurial cells (Fig. 5C,D). In contrast to later developmental stages, the perineurial cells do not cover the entire circumference of the nervous system (Figs. 5A,F,G, 6).

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

Ultrastructural morphology of perineurial and subperineurial cells. A–G, The figure shows electron micrographs of stage 16 embryonic nerves (A, B), first larval instar nerves (E–G), and third instar ventral nerve cord (C, D). A, In a peripheral nerve of a stage 16 nerve, perineurial (pg) and subperineurial glial cells (spg) are tightly associated with the axonal fascicles. The wrapping glia (wg) does not ensheath individual axons. The boxed area is shown in magnification in B. B, No glial processes can be detected within the fascicle. The axons are in direct contact with the subperineurial glia (spg) that forms a thin layer around the fascicle (arrowheads). C, In the CNS of a third instar larvae the subperineurial glia (light blue shading) is characterized by its flat appearance. Septate junctions can only be recognized in this glial layer (arrowhead). A thick neural lamella (nl) covers the nervous system. D, In the perineurial glial cell layer (light green shading), numerous cell protrusions can be seen (asterisk). These processes never invade into the subperineurial layer. E, In a first instar larval nerve, septate junctions are formed by the subperineurial cells. The image corresponds to the dotted area in F. F, Only few perineurial glial cells are found along the nerve. They do not fully cover the subperineurial cells as they do in later developmental stages. The wrapping glia has not yet started to individually ensheath every axon. G, The different glial cells are highlighted by shading. nl, Neural lamella; pg, perineurial glia; spg, subperineurial glia; sj, septate junction; wg, wrapping glia; ax, axons. Scale bars: 1 μm.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6.

Three glial cell layers are present in the larval nerve. A, Electron micrograph of a third larval instar peripheral nerve. The neural lamella (nl) has the same size throughout the nerve circumference. Below are the perineurial glial cells (pg, one cell is highlighted by light blue shading) that show various indentations and profiles of small cell protrusions. The fascicle is encircled by one subperineurial glial cell that forms only short processes toward the axon fascicle (spg, arrows, the cell is labeled in blue). The subperineurial glia forms autocellular septate junctions (boxed area and magnification; white arrowhead points to septate junctions; black arrowhead points to septate junction-free cell–cell contact). Within the fascicle, usually two to three glial cell profiles can be detected (one profile is highlighted in red). B, Third instar larval nerves were stained for HRP (blue), Repo expression (red), and Dlg:GFP expression (green). Scale bar, 1 μm. B, C, Lateral and orthogonal (C) view of a larval nerve expressing a Dlg:GFP fusion protein under the control of the SPG:Gal4 driver. GFP expression accumulates at the septate junctions formed by the subperineurial glial cell.

Below the perineurial cells lies the subperineurium. Electron microscopic analyses confirm the flat cell shape of the subperineurial glia and demonstrate that these cells establish septate junctions with each other (Figs. 5E, 6A, inset) (see below). In the CNS, the subperineurial cells are not in contact with axons, which instead are wrapped by neuropile glia. In the PNS of the first instar larvae, however, the subperineurial glial cells can be in direct contact with axons as they form a sheath around the entire nerve (Fig. 5F,G).

In late embryos, the wrapping glia separates sensory and motor axon fascicles. The wrapping glia has not yet started to ensheath individual axons (Fig. 5A,B). In first instar larvae, the wrapping glia starts to ensheath individual peripheral axons (Fig. 5F,G). During the third instar larvae stage, the wrapping glia has individually ensheathed most axons and the contact between axons and subperineurial cells is nearly completely lost (Fig. 6).

In the periphery, the subperineurial glial cells form autocellular junctions

Septate junctions formed by the subperineurial cells have long been considered to be the structural basis for the blood–brain barrier (Treherne, 1962; Treherne and Pichon, 1972; Carlson et al., 2000). Injection of electron opaque tracer molecules demonstrated that these molecules could easily penetrate through the neural lamella, but did not progress beyond the junctional complexes (Lane and Treherne, 1972). More recently, it was shown that the G-protein coupled receptor Moody is expressed specifically by subperineurial glial cells and that loss of moody function results in both a reduced formation of septate junctions and a leaky blood–brain barrier (Bainton et al., 2005; Schwabe et al., 2005).

In support of this we found septate junctions prominently in the subperineurial glial layer in serial EM sections in both, the CNS and the PNS (Figs. 5, 6; supplemental Fig. 3, available at www.jneurosci.org as supplemental material). Tracking of glial cell membranes in serial electron microscopic cross sections of peripheral nerves shows that generally one subperineurial cell is detected which exhibits prominent autocellular septate junctions (Fig. 6). To further support the notion that the subperineurial glial cells generate septate junctions we expressed a GFP-tagged Discs large protein that in epithelial cells faithfully localizes to septate junctions. After expression of the GFP::DlgS97 fusion (Bachmann et al., 2004) in the subperineurial glial cell layer (using the SPG-Gal4 driver), the GFP::DlgS97 fusion protein was recruited into septate junction-like structures that resemble those labeled by the neurexinIV gene trap insertion (Figs. 6B,C, 7). In the periphery, septate junctions are visualized as a thin single line closely following the axonal fascicles, and only rarely we noted ring-like structures around the nerve representing contact zones of two subperineurial glial cells (Fig. 7A,B), resembling the junctional organization in trachea (Fig. 7C) (Ribeiro et al., 2004). This further demonstrates that only a few subperineurial cells populate the nerve and that they predominantly form autocellular septate junctions as suggested in the electron microscopic analyses.

• Download figure
• Open in new tab
• Download powerpoint

Figure 7.

Septate junction formation in the PNS. Septate junction formation is followed using the neurexinIV GFP gene trap insertion (green) and confocal analysis. Glial cell nuclei are stained in red (Repo staining) whereas axonal membranes are shown in blue (HRP staining). A, Confocal analysis of the entire nerve, with only the NeurexinIV::GFP expression shown in B. The dotted lines indicate the orthogonal sections shown under (a–e). In level c, septate junctions are formed between two glial cells resulting in a ring-like structure. C, Note the extended and broadened expression of NeurexinIV::GFP, which is in contrast to the discrete NeurexinIV::GFP expression in a corresponding contact point in the tracheal system.

Physiological role of the different barrier layers

The different glial cells analyzed above comprise the functional blood–brain barrier. The integrity of this barrier can be measured after injection of labeled dextran molecules into the hemolymph (Bainton et al., 2005; Schwabe et al., 2005). Furthermore, the use of differently sized dextran allows addressing a possible size selectivity of the barrier. To determine the kinetics of dextran uptake we used a Zeiss 5 Live LSM (see Materials and Methods). This allowed us to directly follow and quantify the dextran uptake in living embryos.

When we injected a fluorescently labeled 10 kDa neutral dextran into stage 17 wild-type embryos, only little amount of fluorescence could be detected in the ventral nerve cord with the laser settings used in the experiment (Fig. 8A). Almost no increase in the level of fluorescence can be detected within a 30 min observation period. To genetically remove all components of the blood–brain barrier, we used homozygous mutant glial cells missing (gcm) embryos. In these animals, all glial cells, except the midline glial cells, are routed toward a neuronal fate (Hosoya et al., 1995). When we injected labeled 10 kDa dextran into stage 17 homozygous mutant gcm embryos, high levels of fluorescence were detected within the nerve cord immediately after injection and no increase in the level of fluorescence were seen within 30 min (Fig. 8D,P). This reflects the complete loss of the blood–brain barrier in gcm mutants.

• Download figure
• Open in new tab
• Download powerpoint

Figure 8.

Different glial layers contribute differently to the blood–brain barrier. Differently sized fluorescent-labeled dextran molecules were injected into stage 17 embryos of the following genotypes: w¹¹¹⁸; gcm^N7–4; nrxIV⁴³⁰⁴; sinu^NWU7, and moody^Δ17. Images shown were taken 2 min after injection. The differently sized dextrans are indicated; the images after injection of 10 and 70 kDa dextran were taken with a Zeiss LSM 5 Live microscope. The scan mode was set at a speed of four frames per second. The images after injection of 500 kDa dextran were taken with a Zeiss LSM 510 Meta. Homozygous embryos were recognized using GFP-labeled balancer chromosomes. A–C, A w¹¹¹⁸ embryo is used as wild-type control. D–F, In gcm mutants, all dyes readily penetrated into the nerve cord. G–I, In nrxIV mutant embryos, larger dextran molecules did not penetrate as easily. J–O, Dye penetration in sinuous mutants is similar to penetration in moody mutants. P, Quantification of 10 kDa dextran uptake over 30 min in the mutants indicated.

To address the function of the subperineurium we analyzed mutants known to affect the formation of septate junctions and compared their phenotypic consequences with those of gcm mutants. Mutants that remove important structural components of the septate junctions in epithelia such as neurexinIV mutants also show a severe disruption of the integrity of the blood–brain barrier (Baumgartner et al., 1996; Schwabe et al., 2005; Strigini et al., 2006). Indeed, fluorescently labeled 10 kDa dextran penetrated with very similar kinetics into the ventral nerve cords of nrxIV or gcm mutant embryos (Fig. 8G,P). Because on the EM level the morphology of the outer glial layers seems to be intact in nrxIV mutants in addition to the lack of septate junctions, this suggests that most of the blood–brain barrier function is conveyed by the subperineurial glial cells and here in particular by the septate junctions formed between these cells (supplemental Fig. 4, available at www.jneurosci.org as supplemental material). Compared with neurexinIV mutants, we also determined weaker uptake of 10 kDa dextran into the CNS of moody mutant embryos which correlates with the reduced number of septae found in the septate junctions of this mutant (Schwabe et al., 2005) (Fig. 8M,P).

Septate junction independent blood–brain barrier components

To test for other putative components of the blood–brain barrier in addition to septate junctions, we used differently sized labeled dextran molecules. After injection of a labeled 70 kDa dextran into gcm mutants high levels of fluorescence were immediately detected in the CNS (Fig. 8E). However, in contrast to the injection of a 10 kDa dextran, it took ∼20 min before comparable amounts of fluorescence were detectable in mutant neurexinIV embryos demonstrating that additional barrier mechanisms exist (Fig. 8H, supplemental Fig. 5, available at www.jneurosci.org as supplemental material). Commercially available dextrans of 10 or 70 kDa are either neutral or negatively charged. To compare the role of charge in penetration of the blood–brain barrier we injected both forms of 70 kDa in neurexinIV mutant embryos. As no difference was observed in these experiments we conclude that charge is only a minor factor regarding the tightness of the septate junction independent component of the blood–brain barrier (data not shown).

To test how the blood–brain barrier seals the nervous system from even larger particles, we injected a 500 kDa anionic dextran with a size of ∼26 nm in diameter (Luby-Phelps et al., 1986). In comparison, GFP has a diameter of ∼10 nm (Ormo et al., 1996). When gcm mutant embryos were injected, the neuropile immediately took up the injected dextran indicating that even large particles can easily penetrate into the nervous system in the absence of glial cells (Fig. 8F). When we injected 500 kDa dextran into neurexinIV mutant animals, dye penetration was significantly reduced, corroborating the notion that not only septate junctions of the subperineurial cells contribute to the physiology of the blood–brain barrier (Fig. 8I). When we injected the 500 kDa dye in moody mutant animals, dye penetration was further reduced (Fig. 8O). In summary, the formation of the blood–brain barrier appears to depend mostly on the integrity of the septate junctions formed by the subperineurial glial cell layer, but large molecules appear to be retained also by other barriers.

Screening for genes required for blood–brain barrier formation

The above findings allow us to test for further components involved in the formation or maintenance of the blood–brain barrier. In a first step we asked whether additional proteins associated with septate junctions in epithelial cells are required to set up a functional barrier. We injected labeled dextran molecules into embryos lacking neurexinIV, neuroglian, nervana2, contactin or coracle. For coracle and nervana2 mutants we obtained penetration phenotypes comparable with what is seen for mutant neurexinIV animals. Embryos lacking neuroglian function have a slightly less pronounced penetration phenotype suggesting that septate junctions in the nervous system are organized in a similar manner as in the epithelial tissues (supplemental Fig. 6, available at www.jneurosci.org as supplemental material). Contactin was shown previously to be required for normal barrier function and for the normal cell surface expression of NeurexinIV (Faivre-Sarrailh et al., 2004). In our hands, we found that the blood–brain barrier of embryos homozygous for a contactin deficiency was less affected as in mutant neurexinIV embryos suggesting that Contactin does fulfill an auxiliary function in the blood–brain barrier as it was described for the epithelium (Faivre-Sarrailh et al., 2004) (supplemental Fig. 6, available at www.jneurosci.org as supplemental material).

In addition, we tested whether Claudin-type proteins that, based on work in mammalian systems confer size selectivity to endothelial or epithelial barriers (Tepass, 2003; Furuse and Tsukita, 2006; Van Itallie and Anderson, 2006), are required for Drosophila blood–brain barrier formation. In the Drosophila genome six different Claudin-like proteins have been identified (Wu et al., 2004). Two of these proteins, Sinuous and Megatrachea, have been shown to have a function in the epithelial septate junction dependent paracellular diffusion barrier (Behr et al., 2003; Wu et al., 2004). Both proteins are also required for the integrity of the blood–brain barrier and homozygous mutant embryos allowed penetration of the 10 and 70 kDa dyes comparable with moody mutant embryos, whereas penetration of the 500 kDa dextran was not observed (for sinuous see Fig. 8K,L,P). To test a possible contribution of the other four Claudin-like proteins we injected 10 kDa dextran into embryos carrying chromosomal deficiencies removing these genes (for details, see Materials and Methods). In all cases no disruption of the blood–brain barrier was observed, suggesting that these Claudin-like proteins are not needed for blood–brain barrier function (data not shown).