Endothelial barriers: from hypothetical pores to membrane proteins^*

Introduction

Starling (1896) recognized that the diffusive and osmotic components of water and solute movements across capillary walls were sensitive to the molecular size of the solutes. Fifty years ago Pappenheimer et al. (1951) proposed that the wide range of properties of different capillaries in filtration and diffusion of water and hydrophilic solutes could be accommodated within a porous membrane model. They concluded that only a minute fraction of the wall of capillaries need be involved in such exchanges, and that pore size could be modelled in terms of cylindrical pores of about 6 nm diameter or rectangular slits of about 4 nm width. Subsequent studies added to the model a much smaller number of ‘large pores’ or ‘leaks’ with diameters perhaps as large as 20 nm (Grotte, 1956; Mayerson et al. 1960). Despite shifts in opinion on the sizes and numbers of pores (Paaske & Sejrsen, 1989), two-pore models have proved robust and continue to provide the basis for discussion of hydrophilic permeability in non-brain capillaries (Rippe & Haraldsson, 1994).

The purpose of this introduction to the symposium is to examine the extent to which the application of microscopy and molecular biology has illuminated the nature of small capillary pores. Discussion is largely limited to continuous non-brain capillaries and paracellular pathways in non-inflamed vessels; substantial consideration of blood–brain barrier capillaries, transendothelial pathways and inflammatory events are left to other contributors. In all cases the strength of a structural model needs to be evaluated in terms of how well it accounts for the size, frequency and other known parameters of the pore, and the extent to which it offers mechanisms for physiological regulation.

The location of small pores

Small pore dimensions are well within the resolving power of electron microscopy, so the introduction of thin-sectioning techniques for biological electron microscopy made it possible to ask testable questions about the localization of pores within endothelia. Obvious candidate structures are the paracellular junctions, well known as permeability barriers in epithelia, and the 70-nm vesicles and pits that constitute the most distinctive structural feature of most endothelial cells (Fig. 1). Controversy over the relative importance of these two sets of structures in both small-pore and large-pore properties (reviewed by Palade et al. 1979; Crone, 1984; Rippe & Haraldsson, 1994; Michel, 1996) has given way to a consensus that brings endothelia back into the main stream of epithelial cell biology. The 70-nm vesicles and pits are caveolae (caveolin-coated vesicles), common to many cell types, that serve roles in ligand binding, transport, cell signalling and membrane transformation (Anderson, 1993; Dvorak et al. 1996; Predescu et al. 1997; Galbiati et al. 2000; Vogel et al. 2001). There is good evidence that caveolae are centrally involved in the large-pore properties of endothelia and in responses to permeability-modifying signals (Feng et al. 1996; Vogel et al. 2001).

The paracellular clefts have long appeared the most likely candidate as a site for small pores for several distinct reasons:

1. Calculations of pore area for various types of capillaries range from about 90% of the paracellular cleft cross-sectional area in mesenteric capillaries to about 10% in skeletal muscle capillaries and virtually zero in blood–brain barrier capillaries (Crone, 1984; Curry, 1986; Rippe & Haraldsson, 1994). Thus the area of the structure fits the range of pore areas.

2. They contain junctional complexes that are acknowledged in epithelia and blood–brain barrier endothelia as the structures restricting paracellular transport both in terms of molecular size and transepithelial flux.

3. They are structurally continuous between the capillary lumen and the tissue extracellular space, an essential feature for hydraulic and osmotic flow of water and solutes.

However, the paracellular clefts of epithelia are far less permeable than those of non-brain endothelia. The restrictive element in the epithelial paracellular cleft is the tight junction, so demonstration of structural differences and larger pore size in endothelial tight junctions is an important test of a paracellular pore model.

Microscopic structure and permeability of endothelial tight junctions

Karnovsky (1967) demonstrated that, unlike epithelial tight junctions, the junctions of cardiac muscle capillary endothelium were permeable to horseradish peroxidase and showed regions in which the membranes at tight junctions were separated by a gap of 4 nm. Wissig (1979) confirmed the existence of the gap and demonstrated graded permeability to haem peptides of various molecular masses. However, some competent electron microscopists were unable to demonstrate these gaps (Simionescu et al. 1975). Initially this was attributed to regional variation in the muscle microvasculature (Crone, 1984). However, examination of ferrocyanide-stained junctions of placental capillaries using a goniometer stage showed that the 4-nm gap was present in more than 90% of endothelial junction profiles if tilted into alignment with the electron beam (Firth et al. 1983). Subsequently, this result was extended to cardiac muscle capillaries (Fig. 2) and mesenteric capillaries though not to blood–brain barrier capillaries, in which tight junctions are composed of multiple continuous strands and show no evidence of a 4-nm gap (Ward et al. 1988; Schulze & Firth, 1992; Adamson & Michel, 1993). Furthermore, freeze-fracture replication of capillaries from muscle and placenta showed the tight junction strands as rows of p-face particles rather than the more or less continuous ridges seen in epithelia or brain capillaries (Simionescu et al. 1975; Firth et al. 1983).

Proposals that the 4-nm gap of the tight junctions of non-brain endothelial capillaries corresponded to small pores were rejected because pores would be expected to occupy about 10% of the paracellular cleft area in muscle capillaries rather than the 90% implied by the 4-nm gap, though the gap could possibly act as a pore for water but not solutes (Karnovsky, 1967; Wissig, 1979; Firth et al. 1983; Adamson & Michel, 1993). Serial sectioning of tight junctions in several capillary endothelia showed that there were intermittent discontinuities at which the membrane separation widened, sometimes to the full 18–20 nm of the non-junctional parts of the paracellular cleft (Bundgaard, 1984; Adamson & Michel, 1993); Wissig & Williams (1978) had previously suggested this possibility. In effect, the narrow clefts in the tight junctions could be by-passed through the wide discontinuities (Fig. 3). Endothelial junctions in capillaries typically consist of 1–4 cross-linked strands, and tortuous paths could be traced through the discontinuities across the full thickness of the tight junction (Adamson & Michel, 1993). The most likely interpretation was that the discontinuities represented the open parts of the junction's ‘shutter’ that limits the fraction of the paracellular cleft available as small pores. Such a view had already been expressed on physiological grounds before the evidence on junctional discontinuities appeared (Crone, 1984). This model implied that the pores themselves lay elsewhere in the cleft.

Evolution of the fibre matrix model

Although the discussion of pore size by Pappenheimer et al. (1951) was biophysical rather than structural, they commented: ‘In view of the small area in the capillary walls available for the interchange of the substances so far investigated it is of interest to consider the possibility of a fibrous matrix between endothelial cells. Structures of this type might offer long narrow slits rather than cylindrical pores as channels for filtration and diffusion.’ The possibility that the pore properties of the endothelial paracellular cleft might reside in the body of the cleft rather than the tight junction strands was developed into a proposal that a glycoprotein meshwork in the paracellular clefts or in a surface glycocalyx might function as a three-dimensional molecular sieve (Curry & Michel, 1980). This ‘fibre matrix’ model offered a solution to some discrepancies between pore size and hydraulic conductivity associated with two-dimensional pore models and also suggested a mechanism for the ‘protein effect’– the diminished selectivity of capillary walls if albumin is removed from their environment (Michel & Phillips, 1985; Curry, 1986; Michel, 1988). Theoretical analysis of the effects of such a matrix do not offer decisive evidence for placement on the lumenal or ablumenal side of the tight junction in the absence of detailed information on the size and distribution of the matrix fibres (Michel, 1996; Weinbaum et al. 1992). Detailed examination of electron micrographs depicting tracers held up at the cleft shows various depths of tracer penetration and suggests a progressive restriction rather than a sharply defined planar barrier. The main shortcoming of the fibre matrix concept, at least from the standpoint of an anatomist, was its lack of precision over the site of the fibre matrix and the membrane glycoprotein species involved.

The presence of regularly structured elements in the endothelial paracellular cleft was reported independently from two laboratories using non-standard tissue preparation methods – ferrocyanide or tannic acid-osmium staining and freeze-substitution (Firth et al. 1983; Frøkjær-Jensen, 1990). These regular arrays of linkers or posts, spaced at about 10 nm centre–centre, spanned the wide parts of the paracellular cleft between and around the tight junction strands (Fig. 4), and resembled the intercellular posts independently proposed on theoretical grounds as an adhesion mechanism (Silberberg, 1988).

An elegant three-dimensional analysis has been developed on the basis of the accumulated data on tight junction, adherens junction and other cleft structures, incorporating cleft-spanning matrix elements and three alternative types of discontinuities in the tight junction strands. A combination of large gaps in the tight junction strands (40–90 × 22 nm), in series with a matrix of fibres in the wide region of the cleft, provides a better model of small pore properties than any single sieve-like or matrix element (Weinbaum et al. 1992).

Immunohistochemical probing with antibodies directed against adhesion molecules shows that the regions occupied by linkers are the locations of the newly characterized endothelium-specific adhesion molecule VE-cadherin (cadherin-5) (Leach et al. 1993). Endothelial tight junctions are thus embedded in belts of VE-cadherin and associated cytoplasmic proteins, homologous with the epithelial adherens junction, within which they interact with catenins and upstream regulatory elements (Fuchs et al. 1996; Rabiet et al. 1996; Braga et al. 1999; Leach et al. 2000). Cadherin arrays in epithelial adherens junctions are associated with cell adhesion and are prerequisites for tight junction formation (Meyer et al. 1992). VE-cadherin appears to have similar roles in non-brain endothelia (Dejana et al. 1995) but in addition is a component of the mesh of a paracellular fibre matrix appropriately placed to be a significant component of the small pore system. Several lines of direct evidence show that VE-cadherin plays a central role in the determination and regulation of microvascular permeability (Leach et al. 1993; Dejana, 1996; Gotsch et al. 1997; Corada et al. 1999; Dejana et al. 1999; Wong et al. 1999; Dye et al. 2001). Blood–brain barrier vessels have been reported to express E-cadherin (Rubin, 1992) and N-cadherin (Schulze & Firth, 1993) rather than VE-cadherin.

Tight junction molecules and restriction of small pore area

This model of a small pore system based on an interaction between tight junction and adherens junction elements incorporates observed features of the tight junction strands, notably the 4-nm membrane separation and various sizes of discontinuities, that are not characteristic of classical epithelial tight junctions. In the last decade the tight junction of epithelia has been progressively mapped in terms of molecular structure and relationships. Although the arrangement of known molecular components has not been elucidated for any tight junction, consideration of known differences between epithelial and endothelial isoforms of tight junction molecules may provide hints of what is generic to tight junctions and what may account for the special features of their endothelial manifestations. There may be particular interest in any differences in tight junction molecular expression between non-brain and blood–brain barrier endothelia, as these could discriminate between those differences associated with permeability and those related to the endothelial cell lineage.

The most fundamental integral proteins of the epithelial tight junction are probably the claudins, a growing family of tetra-spanning proteins (Furuse et al. 1998; Tsukita & Furuse, 2000). Endothelial cells are unusual in expressing a cell type-specific claudin (known as claudin-5). It is interesting that claudin-5 is expressed in all segments of the vascular systems of the brain and the lungs, whereas in kidney the arterial and arteriolar endothelium is claudin-5-positive but capillaries and veins are negative (Morita et al. 1999). Tight junctions are known to be present in some segments of organ vascular beds that lack claudin-5, so it is likely that at least one other claudin species, possibly the rather widely distributed claudin-1, may be found in endothelial cells. The presence of claudin-5 in both brain and non-brain endothelia suggests that its expression is associated with endothelial lineage rather than with leaky tight junctions. Permeability characteristics of specific endothelial tight junctions may depend critically on how the molecular building blocks are arranged and may reflect the interactions of several claudin or non-claudin species.

Another (non-homologous) tetra-spanning epithelial tight junction protein, occludin, is also expressed in some endothelia. Although reported as absent in most endothelia (Furuse et al. 1993; Hirase et al. 1997), occludin is expressed by human placental microvascular endothelial cells in the larger villi but not in terminal villi (Leach et al. 2000). In primary cultures of endothelial cells from the same source occludin expression is up-regulated when cyclic AMP-stimulated cells differentiate to form junctional complexes (Dye et al. 2001). Occludin is also strongly expressed by brain endothelial cells (Hirase et al. 1997; Saitou et al. 1997), so it is possible that the occludin/claudin-5 combination may be significant in the establishment of blood–brain barrier properties.

Junction adhesion molecule (JAM) represents a third class of tight junction-associated integral protein. Unlike the tetra-spanning claudins and occludin, JAM is a mono-spanning member of the immunoglobulin superfamily with two extracellular immunoglobulin loops (Martin-Padura et al. 1998). JAM dimers in epithelia and endothelia function as adhesion molecules and appear to be particularly involved in tight junction assembly and resealing (Liu et al. 2000). As JAM is also found on leucocytes, it is not surprising that it forms part of the sequence of adhesion molecules involved in diapedesis of leucocytes across blood vessel walls (Martin-Padura et al. 1998; Johnson-Léger et al. 2000).

Intracellular regulatory proteins more or less specifically associated with tight junctions include ZO-1 (Stevenson et al. 1986), ZO-2 (Gumbiner et al. 1991), ZO-3 (Haskins et al. 1998), cingulin (Citi et al. 1988), 7H6 (Zhong et al. 1993) and symplekin (Keon et al. 1996). ZO-1 is particularly interesting because it occurs as alternatively spliced isoforms differing by an 80-amino acid α domain (Willott et al. 1992) located within the C-terminal 943-amino acid, proline-rich portion. The α-containing ZO-1α+ isoform is expressed at the tight junctions of most epithelia, whereas ZO-1α– is restricted to endothelia, including blood–brain barrier vessels, and to two atypical epithelia: seminiferous tubular (Sertoli) cells and renal podocytes (Balda & Anderson, 1993). There is no apparent correlation between junction structure or tightness and which ZO-1 isoform is expressed; it is possible that the α– isoform may correlate with tight junction lability or some other aspect of interaction with the cytoskeleton (Balda & Anderson, 1993). The remaining N-terminal 793-amino acid portion is homologous with a protein associated with the barrier-forming septate junctions of invertebrates, suggesting a shared evolutionary origin either of these two types of junctions or of their signal systems (Willott et al. 1993). The 7H6 antigen is of some interest as induction of its expression is associated with increased tightness (Satoh et al. 1996) and it is associated with the junctions of brain endothelial cells. Symplekin is confined to epithelia and absent from all vascular endothelia examined, including blood–brain barrier vessels (Keon et al. 1996). The other molecules in this diverse collection are also expressed in sites other than tight junctions and are not likely to be important in endothelial pore properties.

Junction molecules, cell lineage and paracellular permeability

This review has summarized evidence that broadly supports a model of the capillary small pore system in which the tight junction defines pore area per unit endothelial area while a fibre matrix, probably occupying part of the wide regions of the paracellular cleft, is the element restricting pore size. On the basis of such evidence Weinbaum et al. (1992) developed a quantitative model that suggested the numbers of tight junction proteins that would need to be omitted to provide gaps of appropriate size. Occludin, the first tight junction membrane protein, was reported the following year (Furuse et al. 1993), and has subsequently been joined by the claudins and JAM. However, none of these components appears to be unique to the endothelial tight junction except claudin-5, and that is found in tight blood–brain barrier capillaries as well as in leaky capillaries known or presumed to have discontinuous tight junctions. We do not know whether such discontinuities are maintained by particular configurations of the three known families of tight junction membrane proteins, or whether they depend on other membrane molecular species so far unreported. Most of the non-membrane proteins associated with the tight junction are involved in interactions with actin or with cell signalling cascades, but no candidate molecule has been implicated in the manipulation of the tight junction into a discontinuous form. Neither has any molecular explanation yet been offered for the 4-nm gap consistently found at the tight junctions of leaky capillaries but absent from blood–brain barrier capillaries and epithelia. Much more knowledge is needed of the structural biology of the tight junction membrane proteins and the rules governing their assembly into strands.

By contrast, the current wave of discovery of new junction molecules has had little impact on our understanding of the wide regions of the paracellular cleft. Although widely accepted as a model, the fibre matrix still has only one component that is properly characterized. VE-cadherin is apparently unique to endothelia with small pore properties and can redistribute in response to permeability-modifying ligands. Quantitative information on the organization of the extracellular domains of VE-cadherin at realistic packing densities would provide information that could be analysed for biophysical compatibility with the known properties of the fibre matrix.