Brain endothelial cells and the glio-vascular complex
- First Online:
- Cite this article as:
- Wolburg, H., Noell, S., Mack, A. et al. Cell Tissue Res (2009) 335: 75. doi:10.1007/s00441-008-0658-9
- 1.2k Views
We present and discuss the role of endothelial and astroglial cells in managing the blood-brain barrier (BBB) and aspects of pathological alterations in the BBB. The impact of astrocytes, pericytes, and perivascular cells on the induction and maintenance of the gliovascular unit is largely unidentified so far. An understanding of the signaling pathways that lie between these cell types and the endothelium and that possibly are mediated by components of the basal lamina is just beginning to emerge. The metabolism for the maintenance of the endothelial barrier is intimately linked to and dependent on the microenvironment of the brain parenchyma. We report the structure and function of the endothelial cells of brain capillaries by describing structures involved in the regulation of permeability, including transporter systems, caveolae, and tight junctions. There is increasing evidence that caveolae are not only vehicles for endo- and transcytosis, but also important regulators of tight-junction-based permeability. Tight junctions separate the luminal from the abluminal membrane domains of the endothelial cell (“fence function”) and control the paracellular pathway (“gate function”) thus representing the most significant structure of the BBB. In addition, the extracellular matrix between astrocytes/pericytes and endothelial cells contains numerous molecules with inherent signaling properties that have to be considered if we are to improve our knowledge of the complex and closely regulated BBB.
KeywordsAgrinAstrocytesBlood-brain barrierExtracellular matrixTight junctions
The endothelial cell of brain capillaries
Establishment of the barrier during development is accompanied by changes in the phenotype of the brain endothelial cells. For example, the HT7-antigen/neurothelin/EMMPRIN (extracellular matrix metalloproteinase inducer) is up-regulated during BBB maturation (Schlosshauer and Herzog 1990; Seulberger et al. 1992), and the endothelial antigen MECA-32 is expressed throughout the vasculature, but is down-regulated in the brain from embryonic day 17 on, thus representing an indicator of vessel leakiness (Hallmann et al. 1995). Indeed, fenestrated endothelial cells in the circumventricular organs and in the choroid plexus maintain their reactivity against MECA-32. Concerning the development of the barrier function of brain capillaries (Wakai and Hirokawa 1978; Kniesel et al. 1996; Dziegielewska et al. 2001; Engelhardt 2003; Liebner and Engelhardt 2005), BBB tightness is probably not just “switched on” at a specific time point during brain angiogenesis, but rather the tightening of the barrier occurs as a gradual process, which is independent from vascular proliferation and begins late during embryogenesis when angiogenesis is not complete (Risau et al. 1986). On the other hand, an unequivocal correlation between tight junction structure and measured permeability has not been shown so far (Møllgard et al. 1980; Saunders et al. 2000). The molecular mechanisms involved in barrier maturation are still poorly understood. Transplantation studies showing that vessels derived from the coelomic cavity gain BBB characteristics when growing into an ectopic brain transplant have indicated that the development of BBB characteristics in endothelial cells is not pre-determined, but rather induced by the neuroectodermal microenvironment (Stewart and Wiley 1981).
Much work has been carried out on adhesion molecules in BBB endothelial cells and cannot be reviewed appropriately here. In particular, the neuroimmunological literature mainly focuses on the mechanism of leukocyte diapedesis through the endothelium during inflammation and the participation of immunoglobulin superfamiliy members, intercellular and vascular cell adhesion molecules (ICAM-1, VCAM-1), the platelet endothelial adhesion molecule (PECAM), and the integrins and selectins, including the involved signal cascades (see Ebnet and Vestweber 1999; Engelhardt and Wolburg 2004; Döring et al. 2007; Woodfin et al. 2007; Couty et al. 2007; Weber et al. 2007).
We will therefore restrict this section to information concerning cadherins and integrins in the context of the BBB.
The role of β-catenin in endothelial cells, and in particular in the BBB, is not well defined (Liebner and Engelhardt 2005). However, evidence has recently accumulated showing the substantial function of β-catenin in stabilizing vascular structures. β-Catenin deficiency in mice leads to embryonic lethality and vascular fragility (Cattelino et al. 2003). β-Catenin is the key molecule controlling the Wnt/wingless pathway during which Wnt ligands bind to a receptor (“frizzled”) leading to the stabilization of β-catenin by inhibiting the β-catenin-degrading complex. Accumulated β-catenin translocates into the nucleus and activates transcription factors. By this mechanism, the genes for tight junction molecules might be among those transcribed as an effect of the Wnt-signaling pathway (Liebner and Engelhardt 2005). To date, however, no direct information is available on the role of Wnt-signaling in the BBB; nevertheless, some information has been obtained regarding the role of Wnt signaling during angiogenesis (Maretto et al. 2003; Goodwin et al. 2006). In addition, Wu et al. (2007) have recently described the involvement of the Wnt-signaling pathway in the induction of matrix metalloproteinases (MMPs), which steadily modify the extracellular matrix during T cell transmigration in inflammation.
Integrins are cell-surface transmembrane proteins comprising non-covalently linked α- and β-heterodimers that are involved in the cell-extracellular matrix recognition and intercellular adhesion (Fig. 3). Although integrins are of eminent significance for the formation of the vasculature (Hynes 2002), their role for the BBB has not been established. In mammals, 18 different α-subunits and eight different β-subunits are known and can be assembled in at least 24 heterodimeric integrins, from which eight have been identified in endothelial cells (for a review, see Hodivala-Dilke et al. 2003). Integrins are double-faced molecules that recognize extracellular epitopes, frequently by means of an arginine-glycine-asparagine (RGD) binding motif. They are conformationally changed, not only in the extracellular binding domain, but also in the cytoplasmic tails of the two heterodimers. This change in the three-dimensional conformation leads to signaling that activates the actin cytoskeleton via linker molecules such as paxillin, vinculin, talin, or α-actinin. The most important role for endothelial integrins seems to be during angiogenesis. A large number of studies with integrin antagonists or integrin gene ablation has found an involvement of integrins in angiogenesis (Del Zoppo and Milner 2006). Mice lacking αv or β8 integrin subunits suffer from vessel dilation at early developmental stages and cerebral hemorrhage (Bader et al. 1998). This defect has been attributed to failure of an adequate αvβ8-based interaction between endothelial cells and astrocytes during cerebrovascular development (McCarty et al. 2002). Accordingly, during focal cerebral ischemia, vascular integrins have been found to be down-regulated in the rat brain, again demonstrating the significance of integrins for vascular integrity (Burggraf et al. 2008). In glioma endothelial cells (Gladson 1996) and in pulmonary artery endothelial cells activated by tumor necrosis factor-α (Gao et al. 2002), αvβ3 integrin has been demonstrated to be up-regulated and to be involved in the spreading and migration of these endothelial cells.
Morphology of tight junctions
Epi- and endothelial cells are interconnected by tight junctions as specialized contact zones (Fig. 2e). They are responsible for the polarization of the cell and the separation of an apical from a basolateral membrane domain (“fence function”), and for the restriction of the paracellular pathway (”gate function”). Reese and Karnovsky (1967) were the first to define the endothelial tight junctions as the site of the BBB proper. Classically, and up to now, the freeze-fracturing technique has been the method of choice to describe the morphology of tight junctions (Staehelin 1974; Nagy et al. 1984; Wolburg et al. 1994; Piontek et al. 2008).
The structure of the tight junction in endothelial cells of BBB capillaries has been found to be the most complex among the entire vasculature of the body (Nagy et al. 1984). This is in nice correlation with the hypothesis of Claude (1978) who has proposed a logarithmic relationship between the number of tight junction strands and transcellular electrical resistance. In addition, the association of tight junction particles with the P-face or E-face of the membrane has been described as decisive for the quality of the endothelial permeability barrier in the brain (Wolburg et al. 1994; Fig. 4a). The BBB tight junctions are unique among all endothelial tight junctions in that their P-face particle association is as high as or even slightly higher than their E-face particle association. Interestingly, the P-face/E-face particle ratio of BBB tight junctions continuously increases during development (Kniesel et al. 1996). In cell culture, the freeze-fracture morphology of BBB endothelial cells is similar to that of non-BBB endothelial cells indicating that the association of the strand particles with the membrane leaflets reflects the quality of the barrier and is under the control of the brain microenvironment (Fig. 4b). Interestingly, the positive correlation between the P-face particle association and the function of the tight junctions also seems to be valid for epithelial tight junctions (Inai et al. 2008).
Molecular composition of tight junctions
In recent years, the molecular biology of tight junctions has been found to be extremely complex (Fig. 3). Most data have been established in epithelial cells, probably because the regulation of tight junctions in endothelial cells of the BBB is considerably more complex than that in epithelial cells. If epithelial cells are separated from the microenvironment within the tissue and cultured in vitro, they develop tight junctions that fulfil all morphological and functional features typical for an epithelial barrier in vivo. In contrast, if BBB endothelial cells are isolated from their brain microenvironment, tight junctions and other barrier parameters differ from those found in vivo (Wolburg et al. 2006). Generally, the molecular components identified at tight junctions can be separated into different classes based on their structures and functions. First, there are the integral membrane proteins occludin and the members of the claudin family, and, as detected more recently, Ig-superfamily members such as the junction-adhesion molecules of the JAM group, the endothelial cell-selective adhesion molecule ESAM, and the coxsackie- and adenovirus receptor CAR (for details, see Wolburg et al. 2006). Second, there are adaptor proteins that are distinguished according to their function as being first or second order adaptors. First order adaptors are based on their direct association with the integral tight junction proteins via PDZ domains and include, for example, ZO-1, ZO-2, and ZO-3 (see below). Second order adaptors are based on their indirect association with the integral tight junction proteins and include, for example, cingulin or the newly described cingulin-related junction-associated coiled-coil protein (JACOP; see below).
In the last ten years, knowledge about the molecular composition and regulation of the tight junctions has rapidly extended (Tsukita et al. 2001; Wolburg and Lippoldt 2002; Gonzales-Mariscal et al. 2003; Matter and Balda 2003; Dejana 2004; Bazzoni and Dejana 2004; Wolburg et al. 2006; Förster 2008). Occludin and the claudin family are the most important membranous components, both of which are proteins with four transmembrane domains and two extracellular loops. A molecular concept of the way that the single occludin and claudin molecules form lipid raft domains and the manner in which their extracellular loops may interact with each other to establish a paracellular barrier is just beginning to emerge (McCaffrey et al. 2007; Piontek et al. 2008).
Occludin was the first tight-junctional transmembrane molecule discovered (Furuse et al. 1993). It was initially isolated from junction-enriched membrane fractions of the chicken liver as a transmembranous tight junction protein of approximately 65 kDa and exists in several isoforms. Human, murine, and canine occludins are closely related, showing approximately 90% identity. Surprisingly, the tight junctions in occludin-deficient mice (Saitou et al. 2000) are not affected with regard to morphology and transepithelial resistance as measured in small and large intestine epithelial cells compared with wild-type mice; however, the mice develop chronic inflammation and hyperplasia of the gastric epithelium, calcifications in the brain and around brain vessels, thinning of bones, postnatal growth retardation, testicular atrophy, and abnormalities in sexual behavior. Saitou et al. (2000) have concluded that occludin probably has a function in tight junction modulation via the induction of intracellular signaling. Moreover, occludin is not required for the formation of tight junction strands. In a number of reports, posttranslational modifications of occludin, such as phosphorylation of its cytoplasmic domains or binding to an ubiquitin-ligase (Traweger et al. 2002), have been described as features of the regulation of tight junctions. For example, serine phosphorylation has been shown to stabilize occludin in its membrane-bound location (Sakakibara et al. 1997). Accordingly, dephosphorylation of occludin has been reported in an experimental model of multiple sclerosis known to evoke disturbance of the BBB (Morgan et al. 2007). In contrast, an increase of occludin phosphorylation has been described after treatment with vascular endothelial growth factor (VEGF; Antonetti et al. 1999) suggesting that hormonal and mechanical changes are able to increase paracellular permeability by an early increase of occludin phosphorylation and a subsequent decrease of the occludin content. As will be described later in the caveolin section, this caveolae-associated protein can also bind to occludin, thus participating in the stabilization of tight junctions. Taken together, it seems that mature cells need occludin to regulate tight junctions rather than to establish their barrier properties.
The claudin family
The claudins are the tight junction molecules that seem to fulfil the task of establishing barrier properties (Morita et al. 1999a; Tsukita and Furuse 1999; Tsukita et al. 1999). Claudins share with occludin their overall organization of four transmembrane domains but have no sequence homology to occludin. The first claudins identified were isolated from junctional fractions of chicken liver and were called claudin-1 and claudin-2 (Furuse et al. 1998). Since then, a number of related proteins has been identified, and at present, the claudin family contains more than 20 members (Wu et al. 2006). Claudins are now believed to be responsible for the regulation of paracellular permeability through the formation of homotypic and heterotypic paired strands (Piontek et al. 2008). Ion selectivity is achieved through the selective expression and combination of distinct claudins in certain tissues (Tsukita and Furuse 2000). Therefore, unsurprisingly, the claudins are not randomly distributed throughout the organs, but at least in part show a tissue-specific expression pattern.
Tight-junction-negative L-fibroblasts transfected with claudin-1 or claudin-3 form tight junctions that appear in freeze-fracture replicas to be associated with the P-face (Furuse et al. 1999). When transfected with claudin-2 or claudin-5, the cells form tight junctions associated with the E-face (Furuse et al. 1998; Morita et al. 1999b). Claudin-3 (Wolburg et al. 2003), claudin-12 (Nitta et al. 2003), and claudin-5 have been reported to occur in the BBB. The last-mentioned seems to be constitutively expressed in all endothelial cells including the BBB endothelial cells (Liebner et al. 2000a). The P-face/E-face ratio of about 55%/45% in BBB endothelial cells in vivo (Kniesel et al. 1996; Fig. 4a) suggests that the degree of association with one or the other leaflet roughly reflects the stoichiometry of claudin expression in the tight junctions. Indeed, in non-BBB endothelial cells, tight junctions are almost completely associated with the E-face, and claudin-3 is rarely or not expressed.
Peripheral membrane components at tight junctions
Transporters in the BBB endothelium
The exchange of dopamine between the blood and brain belongs to the best-known examples of the metabolic BBB. Dopamine cannot directly be taken up by the endothelial cells, but L-DOPA is transported by an amino acid transporter into the endothelial cell, where it is converted enzymatically to dopamine by the DOPA-decarboxylase within the endothelial cell (Goldstein and Betz 1986). Unsurprisingly, most interest of the pharmaceutical industry is focused on these transporters, including the multidrug resistance complex in order to overcome the BBB for successful delivery of drugs against neurological disorders (Begley and Brightman 2003; Mann et al. 2003; Reichel 2006; Gaillard et al. 2006; Smith and Gumbleton 2006; Deeken and Löscher 2007). Several independent carrier systems for the transport of hexoses (glucose, galactose), basic, acidic, and neutral amino acids (including L-DOPA), monocarboxylic acids (lactate, pyruvate, ketone bodies), purines (adenine, guanine), nucleosides (adenosine, guanosine, uridine), amines (choline), and ions have been described (for overviews, see Mann et al. 2003; De Boer et al. 2003).
Among these transporters, the glucose transporter is of special importance because glucose is the main energy source of the brain (Fig. 6; for a comprehensive overview, see Qutub and Hunt 2005). The 55-kDa form of the glucose transporter isoform GLUT1 as one of five members of a supergene family of sodium-independent glucose transporters is highly restricted to the capillary endothelial cells in the brain (Maher et al. 1994; Bauer et al. 1995; Mann et al. 2003). The density of glucose transporter molecules as detected by quantitative immnogold-labeling is three to four times higher in the abluminal than in the luminal membrane (Farrell and Pardridge 1991; Bolz et al. 1996; Dobrogowska and Vorbrodt 1999). However, as Simpson et al. (2001) have pointed out, the method of the way that the distribution of GLUT1 has been measured plays a decisive role in the determination of the luminal to abluminal ratio: cytochalasin binding to luminal and abluminal membrane fractions gives a luminal to abluminal ratio of 2:1, whereas the performance of Western blot analysis with an antibody against the 20 C-terminal amino acids indicates a ratio of 1:5, but use of an antibody against an intracellular loop of GLUT1 suggests a ratio of 1:1. These different results have been interpreted as suggestive of either the presence of an additional glucose transporter isoform in the luminal membrane or the masking of a C-terminal epitope in the luminal, but not in the abluminal membrane (Simpson et al. 2001).
Another important endothelial gene product is P-glycoprotein, which is required for the differentiation of the BBB (Schinkel et al. 1994) and is developmentally regulated (Virgintino et al. 2008); it seems to ensure the rapid removal of toxic metabolites from the neuroectoderm before the BBB has fully differentiated (Begley 2004). P-glycoprotein is an ATP-binding cassette (ABC-) transporter for the acquisition of the multidrug resistance phenotype and, together with the multidrug resistance-associated protein Mrp1 (Qin and Sato 1995), has been identified in brain endothelial cell membranes. Importantly, these proteins are responsible for the active extrusion of nonpolar molecules out of endothelial cells and are therefore the focus of research on drug delivery to the brain (for a review, see Turcotte et al. 2006; Fig. 6). Interestingly, P-glycoprotein has been described to be co-expressed together with caveolin in the endothelial cells of brain microcapillaries (Virgintino et al. 2002a; Jodoin et al. 2003), and Barakat et al. (2007) have demonstrated a direct interaction of caveolin-1 with P-glycoprotein: the down-regulation of caveolin-1 enhances, and the transfection of caveolin-1 inhibits the transport activity of P-glycoprotein. However, only a fraction of P-glycoprotein may be associated with caveolin-1 suggesting that the expression of both proteins may be under the control of defined metabolic requirements (Demeule et al. 2000).
Barrier permeability is determined by both tight-junction-controlled paracellular and caveolae-mediated transcellular permeability. In addition, receptor-mediated endocytosis can occur, as morphologically represented by clathrin-coated pits and vesicles. The literature regarding these vesicular structures is vast (see, for example, Stan 2002; Simionescu et al. 2002; Tuma and Hubbard 2003; Parton and Richards 2003; Parton and Simons 2007; Head and Insel 2007). Caveolae have been suggested as sites of endothelial transcytosis, endocytosis, and signal transduction, and as docking sites for glycolipids and glycosylphosphatidylinositol-linked proteins (Parton et al. 1994; Simionescu et al. 2002). Among the components of the caveolar membranes are the receptors for low-density lipoprotein (not only present in coated pits), high-density lipoprotein, transferrin, insulin, albumin, ceruloplasmin and advanced glycation end products , interleukin-1, vesicle-associated membrane protein-2 (VAMP-2), and caveolin-1/2. Regarding the BBB, caveolin-1 has been detected not only in endothelial cells, but also in astrocytes and pericytes (Virgintino et al. 2002b). Caveolin-1 binds to cholesterol and fatty acids and forms high-molecular-weight oligomers. Among the molecules forming signaling complexes at caveolin-1 as a multivalent docking site are heterotrimeric G-proteins, members of the mitogen-activated protein kinase pathway, src tyrosine kinase, protein kinase C, and the endothelial nitric oxide synthase (NOS). All these molecular complexes are organized in lipid-based microdomains or rafts (as summarized in Parton and Richards 2003). The involvement of caveolin-1 at least in nitric oxide and calcium signaling processes has been impressively documented in the caveolin-1-deficient mouse (Drab et al. 2001). The function of caveolae as organelles of endocytosis still remains controversial. Internalization of caveolae has been demonstrated to be regulated by phosphorylation (Parton et al. 1994). On the other hand, the caveolin-1-deficient mouse seems to have unaffected transendothelial transport (Drab et al. 2001). Caveolin-1 has been proposed not to be required for endocytosis but would stabilize endocytic raft domains decreasing endocytosis (Nabi and Le 2003).
In the endothelium, the relationship of paracellular and transcellular permeability is of crucial importance for the regulation of overall transendothelial permeability. VEGF is well known as being identical to the permeability growth factor, VPF (Senger et al. 1983). VEGF plays a central role in triggering angiogenesis and vascular permeability. Endothelial cells under the influence of VEGF have previously been shown to form fenestrations (Fig. 2a), caveolae, and caveolin-1- and VAMP-positive vesiculo-vacuolar organelles (Roberts and Palade 1995; Esser et al. 1998). In addition, the src-suppressed C-kinase substrate (SSeCKS) in astrocytes has been reported to be responsible for the decreased expression of VEGF and the increased release of the anti-permeability factor angiopoietin-1 (Lee et al. 2003). In parallel, Lee et al. (2003) have demonstrated that SSeCKS overexpression increases the expression of tight junction molecules and decreases paracellular permeability in endothelial cells. Indeed, VEGF has been shown to induce the phosphorylation of occludin and ZO-1; this could result in both the dissociation of caveolin from the junction (Antonetti et al. 1999) and its targeting to the luminal membrane. In the presence of VEGF, vascular endothelial cells in culture down-regulate ZO-1, relocate ZO-1 from the membrane to the cytoplasm, and lose their transendothelial resistance (Ghassemifar et al., 2006). The VEGF receptor 2, also known as Flk-1, is closely associated with caveolin-1, the main molecule of caveolae (Labreque et al. 2003), and has also been shown to co-precipitate with occludin (Nusrat et al. 2000). The monocyte chemoattractant protein -1 (MCP-1), now known as the chemokine CCL2, has been reported to be co-localized with caveolin-1 during its transit across BBB endothelial cells (Ge et al. 2008) and to alter the expression of both tight-junction-related proteins and caveolin-1 (Song and Pachter 2004). Thus, tight junctions could play a role as a “sink” for caveolin-1, or, put another way, occludin-bound caveolin-1 may be a stabilizer of tight junctions. Accordingly, the down-regulation of caveolin-1 expression by the treatment of cultured microvascular endothelial cells from mouse brain with short interfering RNA has been accompanied by the reduced expression of the tight junction molecules occludin and ZO-1 and by the dissociation of occludin from the cytoskeleton framework (Song et al. 2007); Song et al. (2007) assume a positive functional correlation of caveolin-1 and endothelial barrier integrity. In another study, the induced breakdown of the BBB (decrease of occludin and claudin-5) in the cold-injury model in rats was accompanied by increased caveolin-1 expression (Nag et al. 2007). At first glance, the two investigations contradict each other; however, in the latter, no experiment was performed to determine the compartment to which the up-regulated caveolin was targeted. Therefore, we have to be aware of different functions of caveolin-1 in different subcellular compartments.
Once dissociated from occludin, caveolin-1 might increasingly bind to the dystrophin-dystroglycan complex (DDC), in particular to NOS (Michel 1999), However, any information about a connection between NOS and endothelial dystroglycan is lacking to date. On the other hand, endothelial NOS is present in BBB endothelial cells, and its activity increases in permeable blood vessels (De la Torre and Stefano 2000; Nag et al. 2001). In addition, nitric oxide donors have been shown to disrupt the BBB (Mayhan 2000). Accordingly, the well-known cytokine erythropoietin seems to protect the BBB against VEGF-induced permeability by reducing the level of NOS and by stabilizing tight junctions (Martinez-Estrada et al. 2003).
The glio-vascular complex
The glio-vascular complex (Fig. 3) is involved in the regulation of blood flow and nutrient supply within the CNS. This regulation includes (1) the control of perfusion parameters differentially realized in specific brain regions according to local requirements, (2) the maintenance of energy supply from the blood to neuronal and synaptic metabolism via glial cells, and (3) the protection of the nervous parenchyma from alterations in blood composition, in particular from neurotoxic compounds including reactive oxygen species. The network of neuro-glio-vascular interactions (the “neurovascular unit”) is involved in manifold metabolic dependencies between neurons and glial cells on the one hand and glial cells and vascular cells on the other hand (for comprehensive overviews on the neurovascular unit, see Iadecola 2004; Simard and Nedergaard 2004). The direct interface between the neuroglial compartment and the vascular compartment is established by the perivascular glial endfeet, forming the glial limiting border. Astrocytes are now generally accepted to play a decisive role not only in the maintenance of the barrier properties of the endothelial cells of brain microcapillaries (Wolburg et al. 1994; Liebner and Engelhardt 2005; Abbott et al. 2006), but also in the control of cerebral blood flow (Mulligan and MacVicar 2004; Takano et al. 2006). An interesting correlation exists between astroglial differentiation and BBB maturation. Astroglial differentiation can be morphologically recognized as the polarization of astrocytes, which arises concomitantly with the maturation of the BBB (Nico et al. 2001; Brillault et al. 2002). It is not maintained by reactive astrocytes (Saadoun et al. 2002) and is difficult if not impossible to establish in culture (Nicchia et al. 2000, 2004).
Furthermore, in addition to the astrocytes, pericytes play an ill-defined role in the organization of the perivascular complex. During early development, the adhesion between endothelial cells and pericytes might be the result of the release of chemotactic factors by endothelial cells. This might induce the migration of pericytes toward the endothelial cell wall and subsequent maturation of the vessels by an increased production of extracellular matrix components elicited by the action of activated transforming growth factor-β and other proteins (Folkman and D’Amore 1996). Amongst these, platelet-derived growth factor (PDGF)-B, a high affinity ligand for the receptor tyrosine kinase PDGF-Rβ present on pericytes, is produced by endothelial cells during development. PDGF-B has been shown to be involved in vascularization of the brain, since disruption of the PDGF-B gene leads to pericyte loss, endothelial hyperplasia, and lethal microaneurysm formation during late embryogenesis (Lindahl et al. 1997). Fine structural investigation of the endothelial cells in these PDGF-B- and PDGF-Rβ-deficient mice shows a malformation of the brain endothelial cells as characterized by the folding of the luminal surface (Hellström et al. 2001). Interestingly, this increase in the luminal surface is also a typical feature of the blood vessels of the pecten oculi, which is a convolution of vessels within the vitreous body of the avian eye (Wolburg et al. 1999). In these vessels, the pericytes die by apoptosis during development (Gerhardt et al. 2000). Thus, both the physiological loss of pericytes in the pecten oculi and the pathological loss of pericytes in the PDGF-Rβ-deficient mouse lead to a characteristic alteration of the shape of endothelial cells suggesting a role for pericytes in the morphogenesis of microvessels. An overview on the complex interrelationships between pericytes and the endothelium describing metabolic, signaling, and mechanical roles of the pericytes is given by Ramsauer (2006).
Implications of the molecular composition of the glio-vascular complex for the integrity of the BBB
The polarity of astrocytes implies a molecular and structural heterogeneity of specific membrane domains on the astroglial surface. At points at which the processes of glial cells projecting to the brain surface or to blood vessels (so-called endfeet) contact the superficial or perivascular basal lamina (glia limitans superficialis et perivascularis), the glial membrane is studded with numerous square arrays or orthogonal arrays of intramembranous particles (OAPs) that can only be visualized by means of the freeze-fracture technique (Fig. 2f; for an overview of the literature, see Wolburg 1995). At points where the contact of the glial cell membrane with the basal lamina is lost by bending away into deeper parenchymal regions of the neuropil, the density of OAPs is dramatically reduced. The establishment and maintenance of this kind of glial polarity is suggested to be dependent on the extracellular matrix of the basal lamina.
Aquaporins in astrocytes and endothelial cells
Within the CNS, astrocytes and related macroglial cells such as retinal Müller cells, cerebellar Bergmann glial cells, and ependymal cells are the only elements expressing OAPs (Wolburg 1995). OAPs are now well-known to contain at least the water channel protein aquaporin-4 (AQP4; Figs. 2f, 5a). Aquaporins mediate water movements between the intracellular, interstitial, vascular, and ventricular compartments, which are under the strict control of osmotic and hydrostatic pressure gradients (Nicchia et al. 2004; Papadopoulos et al. 2002; Badaut et al. 2002; Amiry-Moghaddam and Ottersen 2003; Nase et al. 2008). However, AQP4 does not seem to be inserted into the membrane as a single water channel molecule but forms functional complexes with other membrane proteins such as the DDC (Nicchia et al. 2008). Furthermore, the distribution of the potassium channel Kir4.1 and K+ conductivity is similar to that of the DDC and AQP4 (Blake and Kröger 2000; Amiry-Moghaddam and Ottersen 2003; Connors et al. 2004; Nagelhus et al. 2004; Warth et al. 2005; Fig. 5a,c; see below). This co-distribution as described in astrocytes and retinal Müller cells may enable these cells to respond to the potassium uptake with water influx (Nielsen et al. 1997; Kofuji et al. 2000; Connors and Kofuji 2002). In addition, laminin has been described to induce the aggregation of the potassium channel Kir4.1 and AQP4 via the DDC in cultured astrocytes (Guadagno and Moukhles 2004). On the other hand, Ruiz-Ederra et al. (2007) have shown, in a recent study, that the elctrophysiological properties of Kir4.1 are unaltered in astrocytes of the AQP4-deficient mouse suggesting that the colocalization of both channels need not imply functional interdependence. However, knockdown of AQP4 by small interfering RNA is accompanied by a down-regulation of volume-regulated anion channels suggesting the existence of raft-like membrane domains in which close functional interactions between the water and anion channels play fundamental roles in regulating water homeostasis in the brain (Benfenati et al. 2007).
The involvement of AQP4 in OAP formation has been demonstrated by the absence of OAPs in astrocytes of the AQP4-deficient mouse (Verbavatz et al. 1997), by the formation of OAPs in Chinese hamster ovary cells stably transfected with AQP4 cDNA (Yang et al. 1996), and by the immunogold fracture-labeling technique showing that AQP4 is a component of the arrays (Rash et al. 1998). Moreover, Nielsen et al. (1997) have been able to demonstrate, by immunogold immunocytochemistry, that the distribution of AQP4-related immunoreactivity is identical to that of the OAPs and is restricted to glial membranes.
For several years, a positive relationship between the OAP-based polarity of astrocytes and the quality of the BBB has been postulated (Wolburg 1995; Nico et al. 2001). Indirect evidence of this relationship has been provided by the observation that, under brain tumor conditions with a leaky BBB, the OAP-related polarity of glial cells decreases (Neuhaus 1990). In addition, the AQP4 content as detected by immunocytochemistry increases (Saadoun et al. 2002; Warth et al. 2004). The apparent contradiction of the up-regulation of AQP4 and the down-regulation of AQP4-positive OAPs in glioma cells can be resolved on the assumption that, under glioma conditions, AQP4 exists separated from the OAPs in the glial membrane and is no longer restricted to the glial endfeet membranes.
Aquaporins other than AQP4 in the CNS include AQP1 and AQP9. AQP9 is expressed by other cells in the organism and the brain including astrocytes and glioma cells. AQP9 mediates transport not only of water, but also of monocarboxylates, glycerol, purines, pyrimidines, and other solutes (Badaut and Regli 2004; Warth et al. 2007). AQP1 is known as the main water channel of the epithelial cells of the choroid plexus (Speake et al. 2003; Johansson et al. 2005) but is also present in endothelial cells. In cell culture of microvessel endothelial cells from rat brain, AQP1 expression is up-regulated during passages, but down-regulated if cocultured with astrocytes (Dolman et al. 2005). The low level of AQP1 expression in BBB endothelial cells seems to be maintained by the presence of astrocytes because the expression level of AQP1 in non-brain endothelial cells is higher than that in brain endothelial cells. Endothelial cells in the olfactory region of the nasal cavity of the rat have been shown to be leaky and to express massive amounts of AQP1 (Wolburg et al. 2008), indirectly suggesting a negative correlation between barrier property and AQP1 expression. Kaneko et al. (2007) have reported that, under hypoxia, the levels of AQP1 mRNA and protein expression significantly increase in cultured retinal microvessel endothelial cells. This once again supports the view that AQP1 expression in endothelial cells is associated with compromised barrier properties. In addition, a reduction of AQP1 expression inhibits tube formation suggesting an involvement of AQP1 in hypoxia-inducible angiogenesis (Kaneko et al. 2007).
The question as to whether AQP4 is expressed by endothelial cells in brain microvessels is still under debate. Most researchers believe that endothelial cells do not express AQP4 (Nielsen et al. 1997; Nicchia et al. 2004; Dolman et al. 2005), but some reports claim a low level of AQP4 (Amiry-Moghaddam et al. 2004). In any case, endothelial cells never form OAPs; if they express AQP4, this water channel would be formed as a non-OAP molecule.
Extracellular matrix, DDC, and MMPs
In the microvascular bed of the brain, the restriction of AQP4 immunoreactivity to the endfoot membrane is currently believed to be dependent on the extracellular heparan sulfate proteoglycan, agrin (Warth et al. 2004; Rascher-Eggstein et al. 2004), which is deposited by both the astrocytes and the endothelial cells. Agrin was originally characterized as an essential molecule for clustering acetylcholine receptors at the motor endplate (Nitkin et al. 1987; Bezakova and Ruegg 2003) but has also been described as being important within the CNS, particularly for the integrity of the BBB (Barber and Lieth 1997; Berzin et al. 2000; Smith and Hilgenberg 2002; Fig. 5b). The agrin splicing variant Y0Z0 has been reported to be specifically present in the endothelial cell basal lamina of CNS capillaries (Stone and Nikolics 1995). If agrin is absent from the basal lamina, AQP4 immunoreactivity is randomly found across the entire surface of the cell (Rascher et al. 2002; Rascher-Eggstein et al. 2004). This suggests that agrin is responsible for the restriction of AQP4 molecules to the glial endfoot membrane. Recently, we have been able to demonstrate that agrin has a direct influence on the insertion of AQP4 into the membrane of the astrocyte, on the AQP4 expression level, and on the water transport capacity of cultured astrocytes challenged by hypotonic stress (Noell et al. 2007).
However, agrin has no binding site for AQP4 but can bind to α-dystroglycan (Gee et al. 1994). Alpha-dystroglycan is a member of the DDC, which has been best investigated in the muscular system. The complex is localized in the cell membrane and links components of the extracellular matrix to the sarcolemma, providing stability and structural integrity during contraction and perhaps a way for transducing signals (Durbeej et al. 1998; Winder 2001; Ehmsen et al. 2002). However, dystrophin and its truncated isoforms such as Dp260, Dp140, Dp116 and Dp71 have also been found in the CNS (Lidov et al. 1993; Blake et al. 1999; Blake and Kröger 2000; Haenggi et al. 2004). Moreover, dystroglycan has been described in the CNS, including synapses, astrocytes, and endothelial cells (Tian et al. 1996; Zaccaria et al. 2001; Calogero et al. 2006). In dystrophin-deficient mice, Nico et al. (2004) have observed an increase in vascular permeability, a loss of some tight-junctional components, and a reduction in the expression of AQP4. Although components participating in the muscular DDC are different from those in the DDC in the astrocyte endfeet, the main molecules and functions are identical. As depicted in Fig. 3, the DDC is expressed in the astrocyte endfoot membrane. The presence of dystroglycan in endothelial cells is contested and under current debate (for a review, see Haenggi and Fritschy 2006). Concerning the BBB, the expression of dystroglycan has been reported in a light-microscopical immunohistochemical study (Zaccaria et al. 2001) but has not been confirmed by immunogold labeling (K. Wolburg-Buchholz and H. Wolburg, unpublished). Nevertheless, dystroglycan expression has been observed during vessel growth in several experimental paradigms (Hosokawa et al. 2002) suggesting that endothelial dystroglycan might play a role in angiogenesis. However, it seems not to be involved in the maintenance of the BBB. In astrocytes, actin filaments of the cytoskeleton are either linked to dystrophin or utrophin, which is also called dystrophin-related protein. Dystrophin is connected via its amino-terminal domain to dystroglycan, which consists of an α- and a β-subunit. Both subunits are encoded by a single gene and are formed by cleavage of a precursor protein into two mature proteins that form a tight noncovalent complex (Holt et al. 2000; Winder 2001). The transmembrane protein β-dystroglycan anchors α-dystroglycan to the cell membrane and the cytoskeleton via its linkage to the C-terminal domain of dystrophin. The brain-selective deletion of the dystroglycan gene has been described as causing brain malformations such as the disorganization of cortical layering and the aberrant migration of granule cells (Moore et al. 2002). In human glioma, Calogero et al. (2006) have reported a differential alteration of dystroglycans: whereas the level of β-dystroglycan expression is unchanged, α-dystroglycan is strongly reduced, but not around blood vessels. In addition to dystroglycan, further proteins such as the dystrobrevins and syntrophins are also connected to the C-terminus of dystrophin and enable the DDC to interact with channel molecules (Fig. 3). Dystrobrevin has been immunolocalized at glial and endothelial cells in the rat retina (Ueda et al. 2000a) and rat cerebellum (Ueda et al. 2000b). A subunit of syntrophin, α1-syntrophin, contains a PDZ-binding domain in its C-terminal domain, which is connected both to AQP4 (Neely et al. 2001) and to the inward rectifier potassium channel Kir4.1 (Connors et al. 2004). Laminin has been described to induce the aggregation of Kir4.1 and AQP4 via the DDC in cultured astrocytes (Guadagno and Moukhles, 2004).
Laminins as the major proteins of all basal laminae form a large family of glycoproteins. Laminins are heterotrimers composed of an α, β, and γ chain. Only two isoforms, namely laminin 8 (composed of laminin α4β1 and γ1) and laminin 10 (composed of laminin α5β1 and γ1) are found in the endothelial basal laminae of most tissues including the CNS (Sorokin et al. 1994; Hallmann et al. 2005). Blood vessels in the CNS have also been reported to express laminins 1 and 2 (Jucker et al. 1996), which are not detected in the basal laminae of blood vessels elsewhere. The distribution of laminin in brain microvessels has been clarified by investigating the laminin isoforms of the basal lamina of the BBB during experimental allergic encephalomyelitis (EAE), the widely used animal model of multiple sclerosis. The localization of inflammatory cuffs surrounding postcapillary venules during EAE has allowed a clear distinction of endothelial cell and astroglial basal laminae, demonstrating that the basal lamina of endothelial cells contains laminins 8 and 10, whereas the astroglial basal lamina contains laminins 1 and 2 (Sixt et al. 2001). Previously, molecular mechanisms specifically involved in the penetration of inflammatory cells across the astroglial basal lamina have been shown in EAE. The extracellular matrix receptor β-dystroglycan localized in the astrocytic endfoot membrane (Tian et al. 1996) is selectively cleaved by macrophage MMP2 and MMP9 entering the CNS during EAE (Agrawal et al. 2006). MMP2 and MMP9 activity has been shown to be required for leukocyte migration across the astroglial basal lamina but dispensable for leukocyte entry across the endothelial BBB and the endothelial basal lamina. MMP expression as part of the regulation of T cell transmigration (Wu et al. 2007) and of angiogenesis (Masckauchán et al. 2006) has recently been shown to be induced by the Wnt signaling pathway (see above).
In the last few years, the regulation of MMPs has turned out to be one of the most powerful mechanisms during both normal development and pathological procesess in the brain (Rascher-Eggstein et al. 2004; Rosenberg 2005; Fig. 3). Virgintino et al. (2007) have reported that, in growing microvessels in the developing human brain, the MMP-2 production by pericytes and endothelial cells is associated with vessel sprouting. After the formation of inter-endothelial junctions, oxidative stress induced by cerebral ischemia has been shown to disrupt the BBB, and antioxidants such as superoxide dismutase and catalase attenuate BBB disruption (Lagrange et al. 1999; Kim et al. 2001; Krizbai et al. 2005; Schreibelt et al. 2007a, 2007b). Furthermore, oxidative stress-induced depletion of intracellular glutathione has been demonstrated to affect endothelial cell permeability by means of the modification of adhesion molecules such as cadherins, integrins, and tight junction proteins (Krizbai et al. 2005; Usatyuk et al. 2006). All these alterations observed after the administration of reactive oxygen species might be explained as the effect of MMP activation associated with enhanced tyrosine phosphorylation of tight junction proteins (Haorah et al. 2007a). Moreover, the relationship between the activation of MMPs and BBB injury has been confirmed in stroke patients (Haorah et al. 2007b); alcohol has been implicated by Haorah et al. (2007b) as a risk factor for developing stroke, as it diminishes barrier tightness via the activation of protein tyrosine kinase and MMPs. Furthermore, after induction of ischemic stroke in the rat by means of middle cerebral artery occlusion (MCAO), Amantea et al. (2008) have reported activation of MMP-2 and MMP-9. Guo et al. (2008) have described, following induction of MCAO, a reduction of MMP-9 expression, associated with an amelioration of BBB dysfunction, after exercise of the rats in the treadmill. MMPs have been shown to be released from all cells participating in the vascular complex, including endothelial cells, pericytes, and astrocytes. According to the scenario established by Rosenberg (2005), astrocytes release proMMP-9, which is transformed into active MMP-9 by MMP-3, which in turn is released from the pericytes. In addition, Zozulya et al. (2008) have shown that endothelial cells also produce and release MMP-9, and that this release is dependent on pericytes. The presence of astrocytes diminishes the release of endothelial MMP-9. Recently, Förster et al. (2006, 2007) have noted that glucocorticoids such as hydrocortisone and dexamethasone can reduce the level of MMP9 by transcriptional activation of the tissue inhibitor of metalloproteinase-1. This might stabilize the endothelial barier function, including the expression of occludin.
Although MMPs have long been known to cleave all compounds of the extracellular matrix including agrin, the observations of VanSaun and Werle (2000) in the muscular system and of Solé et al. (2004) in the CNS are the only reports that have focused on the cleavage of agrin by MMP-3 (Fig. 3). In the context of this section, in which the role of agrin in the induction of the OAP/AQP4-based polarity of astrocytes and the role of this in the maintenance of the barrier properties in endothelial cells has been described, the presence or regulation of MMP-3 within the perivascular complex seems to be of eminent importance. Indeed, in cerebral ischemia displayed by the model of spontaneously hypertensive rats with middle cerebral artery occlusion (Yang et al. 2007) and in multiple sclerosis patients (Rosenberg et al. 1996; Kanesaka et al. 2006), MMP-3 has been found to be up-regulated, and its level increases in serum, respectively. In addition, an inhibitor of MMPs prevents MMP-induced tight junction degradation (Rosenberg and Yang 2007; Yang et al. 2007). In the MMP-3 knockout mouse, the expression of the lipopolysaccharide-induced opening of the BBB and the degradation of the tight junction proteins claudin-5 and occludin occur to a lower extent than in the wild-type mouse (Gurney et al. 2006). However, the relationship between MMPs and tight junctions is complicated by the observation that claudin-1 and claudin-5 are able to promote the activation of pro-MMP2 (Miyamori et al. 2001) suggesting that certain BBB-related molecules are involved in the activation of MMPs. HT7, also called neurothelin or EMMPRIN, which is associated with the normal BBB (Schlosshauer and Herzog 1990; Seulberger et al. 1992), is also present on the surface of tumor cells stimulating adjacent cells to produce MMPs (Sameshima et al. 2000).
In the 123 years since the discovery of the BBB by Paul Ehrlich (1885), BBB research has focused on the morphological description of the barrier by using mainly conventional histological and electron-microscopical methods plus methods to demonstrate the tightness of the barrier against a variety of low and high molecular weight substances (for many reviews, see Nag 2003). Tight junctions have been described as a network of protein particles, as seen by using freeze-fracture electron microscopy in the 1960s. It required 30 additional years and a great methodological advance in molecular biology to discover protein components of the tight junction complexes. Further time was then required for these tight junction complexes to be recognized as highly dynamic structures. However, to date, we have no clarification as to whether the four-fold membrane-spanning tight junction molecules are folded in a way similar to that of the gap junction proteins as is commonly suggested, whether tight junctions consist of a lipidic backbone stabilized by transmembrane proteins as proposed by Wolburg et al. (2006), or, finally, whether each of these folding types can transform into the other as a result of regulatory processes. In any case, the BBB tight junctions should not be regarded as isolated barrier molecules, but as highly dynamic structures that are under the close regulation of the brain microenvironment. The players in this regulatory scenario include adhesion molecules such as the cadherins and the catenins, probably the Wnt signalling pathway, the tight junction molecules occludin and the claudins with all their associated scaffolding proteins, the integrins and the micro-cosmos of the extracellular matrix, and extracellular MMPs. As we have shown in some detail, the heparansulfate proteoglycan agrin may play a decisive role in modulating the architecture of the astroglial endfoot membrane; this is of fundamental importance for the homeostasis of ions and water in the brain. The complexity of the regulatory processes at the BBB seems to be as advanced as those that are in the focus of the majority of neuroscientists, namely synaptic and even cognitive processes. Probably an increase in BBB permeability is not always an indication of pathological processes (stroke, inflammation, tumor, neurodegenerative diseases) but may belong to the basic set of facilities of the brain itself to obtain access to hematogenous factors in the circulation in order to modulate neuronal circuits. Investigations of such processes might form the basis of “vascular neurobiology”. However, at present, such subtle regulatory events are not detectable by any clinically used barrier imaging. Nevertheless, an understanding of all the glial, endothelial, and extracellular components of the BBB that may operate together or separately at certain periods during development or during pathological disturbances or in various combinations will be one of the greatest challenges in the future for this growing and exciting interdisciplinary field of research.