I.  INTRODUCTION

Top Previous Next References

In this review we consider both the "basal" permeability of microvessels to fluid and hydrophilic solutes in normal (undisturbed) tissues and increased microvascular permeability such as occurs during the early stages of acute inflammation.

Thus, in the first part of this review, we address the classical question of relating functional measures of microvascular permeability in "normal" tissues to the ultrastructure of the microvascular wall. We extend the discussion to consider the contribution of endothelial cell membrane permeability to microvascular permeability to water and small solutes. We also reexamine the question of how macromolecules are transported through microvascular walls with particular emphasis on recent work on the endothelial plasmalemmal vesicular system. In the second part of the review we examine the phenomena of increased microvascular permeability and consider the intracellular signaling mechanisms that enable the endothelial cells to bring this about.

With the focus on the cellular and molecular basis of microvascular permeability, the content of this review differs from that of two Physiological Reviews articles that discussed different aspects of microvascular exchange some 5-6 years ago. In the first of these, Aukland and Reed (17) considered the exchange of fluid between the microvasculature and the interstitium and were concerned more with the properties of the interstitium than with pathways through microvascular endothelium. In the second review, Rippe and Haraldsson (246a) discussed how microvascular permeability to macromolecules could be described in terms of convection and diffusion through two populations of pores in microvascular walls. Although they considered how the "large pores" might be interpreted in ultrastructural terms, since their review was published there have been considerable advances in our general knowledge of vesicular transport and of the molecular constituents of the caveolae (or uncoated vesicles) of endothelial cells. We have, therefore, reexamined Rippe and Haraldsson's conclusions in the light of this recent work.

Many of the general characteristics of basal microvascular permeability appear to be well established. Surprisingly, this information is not so widely known, although it forms a basis for understanding the role of microvascular exchange in many physiological systems. We, therefore, begin our discussion of the permeability of normal vessels with a summary of this basic information.

	II.  MICROVASCULAR PERMEABILITY IN NORMAL (UNDISTURBED) TISSUES

Top Previous Next References

Although changes in microvascular permeability are often reported in terms of changes in the fluxes of fluid or solute between the blood and the tissues, the functional measures of microvascular exchange that represent the properties of microvascular walls are the permeability coefficients. Because microvascular exchange is largely (if not entirely) passive, the permeability coefficients relate the net fluxes of fluid (J_v) and solute (J_s) to the differences in pressure (P) and concentration (C) that drive them through microvascular walls. Four permeability coefficients are of interest: the hydraulic permeability (or hydraulic conductance or conductivity) (L_p), the diffusional permeability (to a particular solute) (P_d), the solvent drag or ultrafiltration coefficient (_f), and the osmotic reflection coefficient (_d). For "ideal" solutes (i.e., those with activity coefficients of unity), the osmotic reflection coefficient _d is equal to the ultrafiltration coefficient _f. The permeability coefficients are defined in Table 1. For a full discussion of their significance, the reader is referred to Curry (48).

In addition to the membrane coefficients, the term clearance is often used to describe microvascular exchange. The clearance of a substance from one compartment to another is defined as the net flux of material divided by the solute concentration in the compartment from which it is being cleared. Thus, if a solute is being cleared from the blood as it flows through a microvascular bed into the tissues, the clearance of the substance is equal to the flux of solute from blood to tissues (J_s) divided by the arterial concentration of solute (C_a), i.e., J_s/C_a. When the flux is unidirectional (as it may be for a tracer diffusing into a large tissue store), the clearance may approximate to the product of the diffusional permeability and the surface area of the microvascular walls through which exchange occurs. This approximation is only valid if the permeability to the substance is so low that the mean solute concentration in the microvessels (C_c) is equal to C_a. This is a reasonable assumption to make for large molecules but not for small hydrophilic molecules, where there may be large differences between the arterial and mean microvascular concentrations of the diffusing solute. The difference between C_a and C_c is determined by the ratio of the permeability-surface area product to the flow through the microvessels. This difference becomes negligible at high perfusion rates when clearance does approximate closely to the product of permeability and the exchange surface area as the arteriovenous concentration difference approaches zero (e.g., Renkin, Ref. 237). The interpretation of clearances of large or intermediate-sized molecules as permeability coefficients is also complicated. Here, transport may be dominated by convection rather than diffusion so that the clearance is a function of both _f and P_d and varies with the net fluid filtration rate from plasma to the tissues. Thus clearances, though convenient measures of microvascular exchange, need to be interpreted with caution.

These different experimental preparations have their advantages and disadvantages. Thus, although measurements made on the intact regional circulation of a human subject suffer from uncertainties surrounding the exchange surface area of microvascular wall and the values of the transcapillary differences in pressure and concentration, they usually involve minimal interference with the microvessels themselves. Thus these studies can provide valuable information concerning microvascular exchange under basal conditions. At the other extreme are measurements on single vessels. Here the surface area of the vessel can be measured directly, as also can the difference in pressure and concentration across the vessel walls. The disadvantages of studies on single vessels, however, are 1) that they involve direct interference with the vessels involved, and 2) they are usually restricted to a small number of convenient vessel types (e.g., mesenteric vessels). Direct interference with a vessel whether it be exposure to light or micromanipulation might be expected to increase permeability. For this reason, the early measurements on single microvessels in frog mesentery were regarded with suspicion, particularly as values of L_p and P_d to small hydrophilic solutes appeared to be an order of magnitude higher than estimates of L_p and P_d based on measurements on intact microvascular beds of skeletal muscle. This concern was allayed, however, when it was shown that measurements of L_p (52) and P_d to potassium ions (79) in single muscle capillaries were an order of magnitude lower than their values in mesenteric vessels and similar to values based on indirect measurements on the intact muscle microcirculation. The differences lay in the different permeabilities of microvessels in different vascular beds and were not the consequence of exposure or manipulation of the tissues. Interestingly, the reflection coefficient of mesenteric capillaries to macromolecules such as serum albumin is similar to that of muscle microvessels.

Although the rapid growth of endothelial cell biology is largely a result of experiments on cultured endothelial cells in vitro, there are serious limitations to the use of monolayers of cultured endothelial cells for gaining direct information about vascular permeability. The most widely reported permeability measurement on monolayers of cultured endothelium is P_d to serum albumin, and mean values are usually in the range of 10⁶ cm/s (11). This is two orders of magnitude greater than estimates based on the flux of albumin through the walls of intact microvessels. Estimates of the reflection coefficients of cultured monolayers of endothelial cells to macromolecules are too low for plasma proteins to exert a significant osmotic pressure across them (302). Reports of monolayers of cultured endothelium with high values of  to macromolecules appear to be based on an erroneous calculation of  (290, 291). These results clearly indicate that monolayers of cultured endothelial cells do not reflect the permeability characteristics of microvascular endothelium in vivo. For this reason, we have restricted discussion of permeability properties and the values of permeability coefficients to measurements made on either single vessels or on intact microvascular beds. We have, however, referred extensively to work on cultured endothelium in section VII, where signaling mechanisms within the endothelial cells are discussed.

Figure 1 shows how P_d to hydrophilic solutes in the microcirculation of skeletal muscle declines as solute molecular size increases. The values for P_d have been plotted on a logarithmic scale, and it is seen that the decline of P_d is maintained until the molecular radius reaches 3.6 nm (the Stokes-Einstein radius of serum albumin).

View larger version (9K): [in this window] [in a new window]   Fig. 1. Relation between permeability (Pd) of microvessels in skeletal muscle to hydrophilic solutes and solute molecular radius. Permeability has been plotted on a logarithmic scale to show range of values, although it is possible that values of Pd for the largest molecules are overestimates (see text). [Data from Renkin (238a).]

Values of P_d for molecules larger than serum albumin appear to decrease much less rapidly with increasing molecular size, suggesting that either the pathways or mechanisms concerned in the transport of macromolecules differ from those for smaller solutes. The decline in P_d of the smaller molecules is partly accounted for by the decrease in their free diffusion coefficients in aqueous solutions (D) as their molecular size increases. This is not, however, the entire story. In Figure 2, the ratio of P_d to D for each of the molecules shown in Figure 1 has been plotted against molecular radius. It is seen that P_d/D falls by more than an order of magnitude as molecular radius increases from 0.23 to 3.6 nm. If P_d declined only as a consequence of the fall in D, the ratio P_d/D would be constant. The decline of P_d/D with molecular size was first described by Pappenheimer et al. (222) as "restricted diffusion." They correctly interpreted it to be a consequence of the diffusion of the solutes through water-filled pores or channels with diameters or widths that were up to an order of magnitude larger than the diameter of the diffusing molecule. They pointed out that as the molecular diameter increased, so the degree of exclusion of the molecules from the pore and the viscous drag on the diffusing molecule would also increase.

View larger version (10K): [in this window] [in a new window]   Fig. 2. Restricted diffusion of hydrophilic solutes at walls of microvessels in skeletal muscle. Values of Pd shown in Figure 1 have been divided by free diffusion coefficient (D) of same solute. If fall in Pd with increasing molecular radius were due to reduction in D alone, then Pd/D would be a constant. It is seen that Pd/D falls by nearly 2 orders of magnitude as molecular radius rises from 2.4 to 36.

Values for P_d of the largest molecules shown in Figure 1 (and corresponding values of P_d/D in Fig. 2) are probably overestimates because they are based on measurements of transport between the plasma and the lymph. This method requires that steady-state transport is established, and such a steady state is difficult to achieve in microvascular beds such as those of muscle where L_p is low (see sect. V and Renkin and Tucker, Ref. 245).

Figures 1 and 2 describe relations between solute permeability and molecular size in microvessels in mammalian skeletal muscle. Similar relations are found in other types of microvessels, although the absolute values of permeability to the smallest molecules may vary by several orders of magnitude. The microvessels in very different tissues, however, have rather similar values for _f or _d to macromolecules. The wide range of values of L_p and the relatively constant value of  to serum albumin in different types of microvessel are shown in Figure 3. Each point in this diagram represents the mean value of L_p and  to albumin for a different type of microvessel or microvascular bed.

View larger version (8K): [in this window] [in a new window]   Fig. 3. Reflection coefficient to serum albumin (albumin) and hydraulic permeability (Lp) in different microvascular beds. Each point represents mean value of albumin and Lp for microvessels in a particular tissue: , cat hindlimb; , rat hindlimb; , dog lung; , dog heart; , frog mesentery; , rabbit salivary gland; , dog small intestine; , dog glomerulus; , rat glomerulus. [From Michel (186).]

The L_p value has been plotted on a logarithmic scale so that values covering three orders of magnitude can be displayed. It is seen that there is no correlation between  to albumin and L_p. The value of  is as high in those vessels with L_p values in the range of 4 × 10⁶ cm·s¹ · cmH₂O¹ as it is in those vessels with L_p values of 10⁸ cm·s¹ · cmH₂O¹. The conclusion from Figure 3 is that variations in L_p in different microvessels are not accompanied by variations in their leakiness to macromolecules. If we think of the pathways through microvascular walls as water-filled pores, then it would seem that the variations in L_p are accounted for by variations in the number of pores per unit area of wall in different vessels, but the diameter of the pores (which determines the reflection coefficients to macromolecules) is fairly constant.

If the differences in permeability of different microvascular beds are the consequence of variations in the numbers of channels (of constant selectivity) per unit area of wall in different microvessels, we might anticipate that L_p is proportional to P_d for small hydrophilic solutes. This is shown in Figure 4. Here again, each point represents the mean value of L_p for a particular type of microvessel plotted against mean P_d to inulin (open symbols) or mean P_d to sodium or potassium (solid symbols) for the same vessel. Both scales are logarithmic so that values covering two orders of magnitude can be shown. The lines relating L_p to P_d have been drawn to have slopes equal to unity, indicating direct proportionality. One conclusion from Figure 4 might seem to be that water and hydrophilic solutes cross microvascular walls by the same pathways. As we shall see, this is not entirely true, and a small fraction (10%) of the L_p may be accounted for by a path from which hydrophilic solutes are excluded. It does suggest that the dominant pathway for fluid and solute exchange is one which they share, and this does appear to be true. Direct proportionality between L_p and P_d is also consistent with the hypothesis that variations in these coefficients are the result of variations in pore or channel numbers per unit area of vessel wall and not variations in pore size. If increases in channel dimensions were responsible for increases in permeability, L_p would be expected to rise more rapidly that P_d and the slope of the relations in Figure 4 would be greater than one.

View larger version (10K): [in this window] [in a new window]   Fig. 4. Relations between hydraulic permeabilities (Lp) of microvessels in different tissues and their diffusional permeabilities to small and intermediate-sized molecules (Pd). Solid circles are values for Lp plotted against values for Pd to either sodium or potassium from same vessel. Open circles are corresponding data for Pd to inulin. Values of Lp and Pd have been plotted on logarithmic scales to show a range of 2 orders of magnitude. Lines are not regression lines but have been constructed through points with a slope of unity to indicate direct proportionality.

We have indicated that general characteristics of the permeability coefficients that are illustrated in Figures 1-4 can be interpreted in terms of pores or channels of constant selectivity. When we come to consider the real ultrastructure of the microvascular walls, we are presented with a greater challenge. The walls of the vessels with high values of L_p (>10⁶ cm · s¹ · cmH₂O¹) are fenestrated endothelium, whereas those of vessels with lower L_p values are continuous endothelium. Very different pathways are primarily responsible for the transport of fluid and hydrophilic solutes through these two types of endothelia. In fenestrated endothelia, this pathway is through the fenestrae, whereas in continuous endothelium, it is through the intercellular clefts. It is, therefore, very surprising that the clear differences in morphology are not accompanied by a qualitative change in the properties of the permeability coefficients. The permeabilities of fenestrated vessels to intermediate-sized molecules are characterized by restricted diffusion, and their reflection coefficients to macromolecules are similar to those of vessels with continuous endothelium. There is little difference in the permeability characteristics of vessels from the salivary gland (which fall at the bottom end of the range of L_p for fenestrated vessels) and those of mesenteric capillaries (which have high permeabilities for vessels with continuous endothelium). It would seem that the presence of fenestrations has an effect that is quantitative rather than qualitative on the permeability characteristics of a vessel. Fenestrations increase the L_p and the P_d to small hydrophilic solutes without changing the L_p/P_d or  to macromolecules such as albumin. Furthermore, as the number of fenestrations per unit area of vessel wall increases, both L_p and the P_d to small hydrophilic solutes increase in direct proportion (152).

Thus the molecular sieving characteristics of microvascular walls appear to be determined by the properties of some structure that is common to (or very similar in) both fenestrated and continuous endothelium. Curry and Michel (56) suggested that it was the luminal glycocalyx that acted as the molecular filter in both types of vessels, and it seems likely that this might be so. They also suggested that differences in L_p in different types of vessel with continuous endothelium might be determined by differing extents to which the intercellular clefts were open. It was suggested that in all these vessels, the ultrafiltration properties of the vessel wall ( to macromolecules; degree of restricted diffusion to small solutes) were largely determined by the luminal glycocalyx. In section III, we discuss how reasonable this view has proven to be and how recent experimental work coupled with mathematical modeling studies have been used to explore these ideas.

	III.  PRINCIPAL PATHWAYS FOR WATER AND SMALL HYDROPHILIC SOLUTES

Top Previous Next References

For water and water-soluble solutes, many of the principal permeability properties of the capillary wall can be described in terms of flow through water-filled cylindrical pores or rectangular slits through the vessel wall. Pore theory describes the resistance to water flow in terms of the viscous energy dissipation assuming laminar Poiseuille flow within a pore, the resistance to diffusion in terms of the additional drag on a spherical molecule moving within the pore relative to movement in free solution, and the selectivity of the membrane in terms of steric exclusion at the pore entrance (48, 187, 294). For each population of pores within the membrane forming the capillary wall, a hydraulic conductivity L_p can be calculated from Poiseuille's Law
where n is the density of pores (number/cm²) of radius R,  is water viscosity, and x is the membrane thickness. The equivalent relation for long rectangular slits used to describe the space between adjacent endothelial cells, and with x measured as the depth of the cleft from lumen to tissue, is
where L is the total slit length per unit area of vessel wall and f is the fraction of the length of the slit open to a width W. The L_p of the whole membrane is the sum of the individual pathway L_p values each weighted by area of the pathway relative to total membrane area.

The corresponding relations for the solute permeability coefficients (P_d) are

where D_pore or D_slit are the solute diffusion coefficients within the pore or slit,  is the solute partition coefficient and f is the fraction of slit which is open. The whole membrane permeability coefficient also is the sum of the individual pathway coefficients. Specific relations for diffusion coefficients and solute partition coefficients in cylindrical pore and rectangular slits are given in Table 2. Implicit in the relations for P_d are the definitions of the pore area per unit membrane area (fractional area for exchange) for each pore population (Ap) equal to nR² or WfL and a pore area per unit cleft depth (Ap/x) equal to nR²/x or WfL/x. A key assumption is that the models are one dimensional. They do not account for lateral spreading at the pore entrance or exit, or within the membrane, and they do not account for interactions between pores. More recent work has highlighted the importance of these interactions in two- and three-dimensional models (see sect. III, F-I).

In a pore, the solute partition coefficient is a measure of the area available to solute within the pore entrance relative to water. The osmotic reflection coefficient () of a porous membrane is a measure of the selectivity of the membrane to a particular solute that depends only on pore size, and not pore number or membrane thickness, and is given by the relation (48)
When there are several pathways in parallel, the membrane reflection coefficient is the sum of the individual coefficients weighted by the fractional contribution of each pathway to the membrane hydraulic conductivity. A fundamental assumption when using classical pore theory is that the same structures that determine the selectivity of the membrane also determine the primary resistance to water flow and solute diffusion. Thus a constraint on the interpretation of measured values of the reflection coefficient is that for a pore or slit within the endothelial barrier to be the primary selective barrier, it must also be the largest diffusion barrier in the pathway (with a corresponding large concentration drop across the barrier) (50).

There is now abundant data demonstrating that the magnitude of transcapillary flows of water, and the selectivity properties of many microvessels, to solutes ranging in size from electrolytes to plasma proteins and having both continuous endothelium and fenestrated endothelium, can be described in terms of three porous pathways in parallel: an exclusive water pathway across the endothelial cells, a population of small pores with a radius of 4-5 nm, and a population of larger pores with 20- to 30-nm radius (246a, 294). Estimates of the relative number of small pores to large pores fall in the range of 4,000 to 1 (skeletal muscle) to <1,000 to 1.

There are also well-documented limitations to these idealized pore models. For example, the pore sizes and pore densities that describe the water flows and the selectivity of the wall to macromolecules do not always account for the measured fluxes of solutes across the capillary wall (50, 186). Also, macromolecular transport is not always coupled to water flows as expected in large pores (242, 244). Furthermore, as shown in section IIIC, the average number of large pores in any one capillary is small. It is likely that there are many vessels with no large pores at all, and the pore model fails to account for the transport of macromolecules in these vessels. An even more fundamental problem is that real pores having the size and properties described by the pore theory (structure offering a uniform resistance to flows across the entire thickness of the membrane) have not been found and probably do not exist. However, just as the idea of the Stokes radius of a molecule undergoing free diffusion provides an effective description of the overall resistance to diffusion of complex ions, sugars, and proteins in solution, the pore theory is a useful starting point to evaluate the possible cellular and molecular structures that actually determine the permeability properties of the wall.

As a first step to relate these pore structures to the ultrastructure of the microvessel wall, it is useful to update calculations by Renkin and Curry (240) describing the exchange properties of a typical microvessel within mammalian skeletal muscle. Given a typical hydraulic conductivity of the order of 1 × 10⁸ cm · s¹ · cmH₂O¹ for skeletal muscle (see sect. I), pore theory enables the calculation of the pore density, which accounts for the permeability properties of the porous pathways in a typical vessel 5 µm in diameter and 1,000 µm in length, with a total surface area for exchange of ~16,000 µm², formed by 32 endothelial cells each with a mean area of 500 µm²/cell. The estimate of pore density depends on the value of the length of the exchange pathway across the wall, x, which characterizes the porous pathway. For a value characteristic of a straight channel across the average thickness of an endothelial cells (0.2 µm), this vessel would contain the equivalent of 9,000-11,000 small pores of 5-nm radius and 2.2 large pores of 30-nm radius in parallel with an exclusive water pathway. For a longer pathway (up to 0.8 µm or longer through the cleft between cells), the vessel would have to contain four times this pore density. These pore densities correspond to pore areas ranging from <0.01% to close to 0.04% of the total capillary surface area.

As a further step to understand the cellular and molecular basis for the permeability properties of continuous endothelium, we review attempts to relate these calculated pore sizes and densities to the structure of the microvessel wall. Pappenheimer et al. (222) estimated that the fractional area of the small pores (Ap/As) was <1% of the total microvessel surface area and suggested that the small pore system lay within the interstices of the intercellular cement that was assumed to be present in the intercellular junctions. The model of a fiber matrix within part of the cleft provides a quantitative description of almost the same idea and is developed throughout this section.

One estimate of the maximum area for exchange occupied by the space between endothelial cells can be calculated from morphological estimates of the length of the line of contact per unit area per cell (L) multiplied by the width of the wide part of the intercellular junction (W) (see slit pore theory above). The value of W, measured as the average spacing between the walls of adjacent endothelial cells in the interendothlial cleft (close to 20 nm), is a remarkably consistent figure in regions more than 30 nm away from the junctional strand in normal endothelium (9, 27, 279). This constant spacing has been explained on theoretical grounds to reflect the presence of spacer molecules that span the wide part of the cleft (120, 278), and there is some experimental evidence to support the presence of such structures (260, 275). The value of L, estimated to lie in the range 1,200-2,000 cm/cm² (5, 28) in skeletal muscle, heart muscle, and mesenteric capillaries depends on endothelial cell size and shape and on the extent of interdigitation of cells.

Together, these values of L and W provide maximum estimates of the area for exchange between cells of the order of 0.4% of the total capillary surface area. Thus the actual area of the cylindrical small pores described above for water and solutes in skeletal muscle (<0.01-0.04% of the total surface area) leads to the estimate that from 2.5 to 10% of the length of cell-cell contact between adjacent endothelial cells needs to be effectively open to accommodate the surface area for exchange estimated above (52, 79, 149, 226). The calculation is based on the assumption that flows through the cleft are essentially one dimensional. This means that there is an exact correspondence between the magnitude of flows through the cleft and the fraction of open junction. Because of the nature of the structure of the junctional strands, this simple correlation is no longer thought to exist (see Fig. 5 and sect. III, F-I).

View larger version (17K): [in this window] [in a new window]   Fig. 5. Diagrammatic representation of 3 possible regimes for flow through intercellular cleft. A: cleft entirely open and flow at constant velocity through its entire length. B: 1-dimensional flow limited to area of discontinuities in tight junction and wide region of cleft immediately above and below break. C: 2-dimensional convergent and divergent flow through wide regions of cleft above and below discontinuities in tight junction. [From Adamson and Michel (9).]

Although the idea that only a small fraction of the area between adjacent cells was needed for transport was developed over 25 years ago at the Benzon Symposium on Capillary Permeability (149), the interpretation of this result in terms of the ultrastructure of the junction remains a fundamental problem. The simplest interpretation is based on the assumption that the fraction of the junction that is effectively open is determined by breaks or discontinuities in the junctional strands. In freeze fracture, the junctional strands are seen as lines of particles in the membranes of the adjacent cells. These form an interconnected network that is most complex in arteriolar and true capillaries and less complex in venular capillaries (279, 282). In endothelial cells, the number and complexity of junction arrangement is generally less than in epithelial membranes. The location of the junctional strands in an electron micrograph of the cleft between adjacent cells corresponds to the position of the so-called "tight junction" region between the cells.

The hypothesis that up to 10% of the junction length might be open for exchange of solutes leads to the reasonable expectation that breaks in the junctional strand should be observable in conventional electron microscopy. During the late 1960s and 1970s, investigators demonstrated the penetration of electron-dense molecular probes (mol wt 10-45,000) across the junction strands in sections from mammalian skeletal muscle (279, 318). In addition, investigations of junctional ultrastructure suggested that, at the level of the tight junction, membranes of adjacent endothelial cells did not fuse but were separated by a space 4-8 nm wide. All studies were carried out on randomly selected sections, and none provided data on the frequency of the breaks. Thus a fundamental question for over two decades has been whether there are breaks in the junctional strand with a size and frequency consistent with the expectations of the pore models described above. For example, Simionescu et al. (282) found no leakage sites for tracer probes larger than 3 nm in radius in true capillaries and argued that the leakage sites described using electron-dense tracers in previous experiments were present only in venular capillaries.

It was also recognized by Wissig (318) and Wissig and Williams (319) that the apparent penetration of a tracer across a junctional strand when observed in a single random section did not prove the existence of a break in the strand at that position. Tracer may have crossed the strand at a break on either side of the section and spread by following a tortuous path or by lateral diffusion. The same arguments also compromise attempts to estimate break frequency from tracer labeling. On the other hand, breaks smaller than the thickness of one section are not likely to be found in the absence of tracer. Thus the evidence suggesting the presence of breaks, and the ultrastructure of the breaks in the junctional strand, based on random sections where adjacent serial sections were not examined, and where vessel type was not known, was severely compromised. Experiments based on serial sections, in the presence of tracer, and with section thickness adjusted to match possible break size were required. Both the sample sizes and technical limitations raise formidable sampling issues in capillaries having the permeability properties of the continuous endothelium in mammalian muscle, lung, and skin.

Before 1993, the most detailed evaluation of these questions in mammalian microvessels has been carried out by Bundgaard (27), who used serial section methods in rat heart capillaries and venules, but without the use of a tracer. In a total of 69 segments within true capillaries and venules representing a total length of junctional strand close to 40 µm [calculated from an average of 15 serial sections/segment, each 40-50 nm (0.04 µm) in thickness], Bundgaard (27) reported only 6 sections with detectable breaks in the junctional strand: 3 in capillaries and 3 in a single venule. The reconstructions demonstrated that at a break, the intercellular cleft is open to a width close to 20 nm at the level of the discontinuity. This is the same as the width in the cleft far from the junctional strand and shows no restrictive structure at the level of the break.

The break frequency observed by Bundgaard (27) using conventional thin sections (1/13.3 µm in true capillaries or 1/6.9 µm overall) was far too high to represent large pores, even though they were close to the size of large pores (40-50 nm long and up to 20 nm wide). On the other hand, Bundgaard (27) suggested that the observed break frequency was too low to account for normal permeability of heart microvessels to small solutes. This argument can be summarized as follows. For a break to be detected using sections 40-50 nm thick, it must be longer that the section thickness. If the break length detected by Bundgaard is set at 80-100 nm (twice the length of a single section), the total length of discontinuities measured in his serial section experiments is no more than 1.5% of the total strand length. Given that Bundgaard measured a mean interendothelial cleft depth close to 0.8 µm, this figure is less than one-third that estimated previously (96) as necessary to account for the flux of small solutes across the microvessel wall. Bundgaard (27) also recognized that conventional thin sections may fail to detect breaks in the junction strand that are smaller than 40-50 nm long. To test the latter possibility, he reexamined a small sample of 16 ultrathin serial sections of average thickness 12.5 nm, collected previously to investigate vesicles in muscle capillaries. In these ultrathin sections, Bundgaard found two regions, more than one section thick, where the adjacent endothelial cell membranes were separated by a distance greater than 4 nm but less than the 20 nm characteristic of the larger breaks. Because the larger discontinuities apparently did not account for sufficient open junction, Bundgaard (27) suggested that the open portion of the junction might lie within very small breaks of the order of 12.5-15 nm long and 4-20 nm wide distributed along the junctions.

When published in the early 1980s, Bundgaard's data (27) provided new support for the idea that both the extent of the open junction and the magnitude of the size-limiting structure of the interendothelial cleft might reside within the fine structure of the junctional strand. The model, known as the "constricted slit model," described tracer penetration through two barriers in series: the wide portion of the cleft modeled as a parallel sided slit up to 20 nm wide and accounting for more than three-fourths of the cleft depth and a narrow slit up to 8 nm wide that was assumed to form the principal size- limiting structure (46). Subsequent analysis has exposed limitations in the arguments used to support this model. One important limitation is that the resistance to diffusion of molecules larger than 1-nm radius can be significant within the wide part of the cleft; the resulting concentration gradients (a type of unstirred layer effect on either side of the junctional strand) mean that the 6- to 8-nm slit is much less effective as a molecular sieve than expected from its dimensions alone (50, 57).

A second problem is that tracer studies in mammalian microvessels continue to show that most of the length of the junction is effectively impermeable to small electron-dense tracers (197, 232, 309, 310). On the other hand, it has been recognized recently that the calculation of the effective area of open junction associated with the breaks described in conventional sections may be significantly underestimated. This is because two-dimensional spreading of water and solute flows on the luminal and abluminal sides of breaks increases the effective area for exchange in the breaks by two- to fourfold, as explained below. Finally, the problem that there is no structure that might form a molecular sieve at the level of larger breaks in the junction strand is resolved if structures at the entrance to the endothelial cleft, formed by the endothelial cell glycocalyx, form the primary molecular sieve in transcapillary pathways (56, 185, 186). The key to these new developments has been the use of serial section methods, with and without tracer, in microvessels of frog mesentery. These vessels have continuous endothelium, but the fraction of open junction is larger than in mammalian muscle capillaries and therefore more easy to analyze. We review this work in the next sections and then use these new insights to reevaluate studies in mammalian microvessels where the sampling problems remain unresolved.

Adamson and Michel (9) investigated the ultrastructure of true capillaries in frog mesentery using serial sections. The strategy to use frog mesenteric microvessels exploited the observation that these vessels have mean hydraulic conductivities and permeabilities to small solutes that are 3-10 times higher than those in muscle capillaries of frog and mammals. Thus it was predicted that the size of the breaks and/or frequency of breaks in the junctional strand would be larger than in mammalian heart and skeletal muscle (46, 50). In the microvessels used for ultrastructural analysis, the hydraulic conductivity was measured before each vessel was fixed under well-controlled experimental conditions. Thus these studies provide a unique picture of the ultrastructure of vessels of known permeability. In addition, in some vessels, the low-molecular-weight electron-dense tracer lanthanum, perfused for short periods up to 15 s, was used to identify specific pathways for small solutes across the microvessel wall. The approach extended previous investigations of the ultrastructure of junctions, the distribution of electron-dense tracers in the junction, studies of transport by vesicles, and investigations of the structure of cell surface glycocalyx in frog mesenteric microvessels (7, 37-39, 156, 169). These earlier studies, and independent investigations by Bundgaard and Frokjaer-Jensen (28), demonstrated that, within the limitations of the methods available, the continuous capillaries of frog mesentery shared the same ultrastructure characteristics as continuous capillaries in mammalian vessels.

Figure 6 shows details of serial section reconstructions in vessels with L_p values ranging from 2.2 to 3.6 cm · s¹ · cmH₂O¹, perfused with lanthanum tracer. In six capillaries perfused with lanthanum, Adamson and Michel (9) found that the tracer filled the entire depth of the junction from lumen to tissue to mark clear passage of this low-molecular-weight tracer at 11 sites within a total strand length of 23.5 µm. Five of the regions were completely delimited within the serial sections; in five others, only the first or last section lay within the leaky region, and at one site, the tracer penetrated the junction for 22 consecutive sections. The breaks account for 2.52 µm or 10.6% of the length of junction examined. In the absence of the tracer, Adamson and Michel (9) also reconstructed from serial sections on three capillaries, three breaks of 0.14, 0.14, and 0.17 µm long within a total capillary length of 13.36 µm. Although the fraction of open junction here is only 3.4%, it is not significantly different from the higher figure that was found in the presence of the lanthanum tracer. Thus Adamson and Michel (9) showed that in frog mesenteric capillaries, the breaks in the junctional strands between the endothelial cells are longer and occur more frequently than in capillaries of cardiac or skeletal muscle. This difference in ultrastructure of the tight junction is consistent with the L_p and P_d to small hydrophilic molecules being higher in mesenteric capillaries than in the capillaries of either of the other tissues.

View larger version (22K): [in this window] [in a new window]   Fig. 6. Reconstructions of discontinuities in tight junction from serial sections of intercellular cleft in frog mesenteric microvessels perfused without lanthanum in perfusate (top) and after 10-s perfusion with lanthanum before fixation (bottom). [Modified from Adamson and Michel (9).]

Following the lead of Ward et al. (311), Adamson and Michel (9) also examined the tight junctions between the endothelial cells on a tilting stage and found that the outer leaflets of the cell membranes were not fused but separated by an electron-lucent region with a mean width of 2.3 nm. By demonstrating that the lanthanum ions made their full contribution to the osmolarity of their perfusates, they confirmed that the lanthanum ions were in solution and that the failure of lanthanum to penetrate this potential pathway through the junction was not due to the formation of large ion complexes. They concluded that if a pathway is present through the tight junction, it appears to have a lower permeability than the dimensions of the electron-lucent region would suggest. The principal pathway through the intercellular cleft lies through the breaks or discontinuities in the junctional strands. Recent work on the molecular structure of the tight junction is consistent with the observations that the outer leaflets of the cell membranes are not fused. The transmembrane protein associated with tight junctions has been identified as occludin (85a). This is a 64-kDa protein, and structural models, based on its novel 504-amino acid sequence, indicate that four hydrophobic transmembrane helices allow each protein to form two extracellular loops of 44 and 45 amino acids. It is proposed that the tight junction is formed by the interlocking loops of occludin molecules from adjacent cells. If this model is correct, then the distance between the outer leaflets will be determined by the lipophilic properties of the outer section of each loop. Such properties may allow the outer section of the loop of an occludin molecule to enter the bilayer of the adjacent cell. If the loops of occludin molecules from adjacent cells remain entirely extracellular but interlock snugly, then the separation of the outer leaflets would be ~2 nm. It is of considerable interest that this is the value of cell separation reported by Adamson and Michel (9).

In the region of the discontinuity, the adjacent endothelial cell membranes were separated by 20 nm, which was the same value as that for the rest of the wide part of the cleft. As in rat heart muscle microvessels (27), there was no obvious structure to restrict solute movement at the level of the discontinuity in tight junction when the junction break was at least 40-50 nm long. Figure 5C shows streamlines for the two-dimensional water flow through the discontinuities in the junctional strand compared with the one-dimensional models considered previously. The key feature is that the water flow is a two-dimensional regime that converges on the discontinuity in the junctional strand on the luminal side of the junction and diverges from the discontinuity on the abluminal side. The flow on either side of the discontinuity is described by the hydrodynamics of flow through a thin layer of fluid, rather than Poiseuille flow in a pore and is obtained by solving the Laplace equation for two-dimensional pressure distribution within the cleft in the region of a discontinuity in the junctional strand. It is also assumed that the flows through individual breaks do not interact. Phillips et al. (228) expressed the L_p of capillary wall through which fluid permeated by this pathway as C × L_p (Poiseuille), where C is a factor >1 that depends on the size of the discontinuity relative to the depth of the cleft () and the depth of the strand within the junction expressed relative to junction depth (µ). The L_p (Poiseuille) described the flow through the parallel sided slit in which there is no spreading of flow (1-dimensional model) and was calculated from the one-dimensional slit theory. The expression for C (when the strand is at the center of the cleft, µ = 0.5) is
where K is a complete elliptic integral of the first kind and  = (1  cosh )²/4 cosh . For a slit 20 nm wide which occupies 10% of the junction length, L_p (Poiseuille) is 3.4 × 10⁷ cm · s¹ · cmH₂O¹ and C = 2.14, so the predicted L_p of the vessel wall is 7 × 10⁷ cm · s¹ · cmH₂O¹. The measured L_p values of the vessels in these studies fell in the range of 2.2-3.6 × 10⁷ cm · s¹ · cmH₂O¹. Thus the measured L_p values can be accommodated by flow through the observed breaks, even before the additional resistance to water flow associated with structures that may form the molecular sieve are added. An understanding of the two-dimensional spreading of flow to amplify the effective area available for exchange has provided a new way to investigate structure-function relationships in the interendothelial cleft over the past 5 years. All previous theory was restricted to "one-dimensional" models such as the pore theory and the one-dimensional slit model (Fig. 5, A-C).

The diffusion of small solutes through breaks in the junctional strand is described by a two-dimensional diffusion equation whose form is identical to the equation for the pressure distribution determining water flows in Figure 5. Thus the solute permeability coefficients of the vessel wall in which there is a two-dimensional diffusion profile are described by the relation
where C accounts for the two-dimensional spread of solute to increase flux flow through the discontinuity and has the same value as for water flow and P_slit is fLD_slit/DX, the permeability of a rectangular slit assuming one-dimensional diffusion across the capillary membrane. For potassium ion, sodium chloride, and glucose, the calculated solute permeability coefficients are 50.4 × 10⁵, 39.2 × 10⁵, and 14 × 10⁵ cm/s, respectively. These values are close to the largest measured values for these solutes in frog mesenteric capillaries (in cm/s): 67 × 10⁵ (potassium), 44 × 10⁵ (sodium chloride), and 10 × 10⁵ (glucose) (45, 47). These calculations demonstrate that some of the largest measured permeability coefficients of the mesenteric microvessels to small solutes can be accounted for if close to 10% of the junction is actually open in these vessels and the effective open area approaches close to 30% taking into account spreading. On the other hand, the permeability coefficient to albumin predicted from this model is 7 × 10⁶ cm/s, which is 20-30 times larger than the measured values (54). Thus, to account for the measured permeability coefficients to larger solutes, the molecular sieve must significantly restrict albumin but have only a minor resistance to small solutes.

The principal hypothesis to describe the molecular filter within the pathway associated through the breaks in the junctional strands is the fiber matrix model of capillary permeability (56). Specifically, the molecular filter is assumed to be a fiber matrix associated with the endothelial cell glycocalyx and possibly extending into the intercellular cleft. On the luminal side of the cleft the presence of a glycocalyx layer on the endothelial cell surface was first described based on staining experiments using ruthenium red and Alcian blue for cell surface glycoprotein (158). These experiments suggested the layer extended into the luminal caveolae and outer regions of the intercellular clefts. Electron micrographs of microvessels perfused with solutions containing native ferritin suggested that, where the luminal contents had been adequately fixed, the ferritin concentration was greatly reduced close to the luminal surface of the endothelial cells. Quantitative evidence that ferritin was excluded from the luminal caveolae was reported by Loudon et al. (156) as well as Clough and Michel (40), strengthening the idea that the glycocalyx could act as a barrier to the diffusion of macromolecules. More accurate estimates of the possible thickness of the endothelial cell glycocalyx were provided by Adamson and Clough (7) in frog mesenteric capillaries. Using cationic ferritin, they visualized the outer surface of the glycocalyx that was up to 100 nm from the endothelial cell surface (average thickness 60 nm) when the vessel was perfused with plasma. The glycocalyx was thinner in the presence of albumin-Ringer perfusate (31 nm). These observations were consistent with the hypothesis that plasma proteins were adsorbed to the endothelial cell glycocalyx and form part of the structure forming the molecular filter at the cell surface (49, 185, 186, 263, 264).

Further evidence for the role of absorbed macromolecules on the cell surface in the formation of a molecular filter was the observation that perfusion of microvessels with cationized ferritin in Ringer solution, without albumin, formed a matrix layer on the cell surface and restored the hydraulic conductivity and selectivity of the vessel wall to that when albumin was present (192, 303). Adamson (4) also demonstrated that enzymatic removal of the glycocalyx, using pronase, increased the hydraulic conductivity of frog mesenteric capillaries by 2.5-fold. Thus, in the first quantitative description of the fiber matrix model of capillary permeability, Curry and Michel (56) accounted for the exclusion of solutes from the matrix in the principal hydrophilic pathway in terms of solute exclusion by a network with fibers characteristic of glycoproteins on the endothelial cell surface. The nature of fibers associated with the endothelial cell surface and the cleft entrance is not well understood, but the side chains of glycosaminoglycans that are likely to form part the cell glycocalyx have a characteristic molecular radius close to 0.6 nm. Another possible site for this ultrafilter could be just within the luminal aspect of the intercellular clefts. Regularly arranged electron densities have been demonstrated in this region by Schultze and Firth (275), and these could represent fibers of a molecular filter.

Using the stochastic theory of Ogston et al. (209a), Curry and Michel (56) described the solute partition coefficient  and the reduction in solute diffusion coefficients of the matrix relative to the free diffusion coefficient (D_fiber/D_free) in terms of the fraction of the matrix volume occupied by fiber (V_f) and the fiber radius (r_f), as shown in the expression in Table 3 (random fiber arrangement). The membrane coefficient and the solute permeability (P_d) and reflection coefficients () were then described by the relations


where A_fiber is the area of fiber filled pathway, x is the length of the pathway, and the term K_p is a measure of the resistance of the fibers to water flow and was calculated from the semi-empirical relation for K_p in Table 3 known as the Carman Kozeny equation. The main assumption in this form of fiber matrix theory was that the fiber matrix formed the primary resistance to solute and water movement. Thus, in contrast to the pore theory, where a given pore radius accounted for both the resistance of the pore to flow, and also the area available for exchange (taking into account pore density), the geometry of the water pathway could be described in terms of the area available for exchange (A matrix) and membrane thickness x independently of the fiber parameters (V_f and r_f). The principal water pathway was assumed to be via the interendothelial cleft, but other structures (including fenestrations or pathways formed by the coalescence of cytoplasmic vesicles) could be included in the area for exchange.

Using these relations, Curry and Michel (56) found that a network of random fibers occupying 5% of the pathway volume accounted for the measured reflection coefficient to albumin and measured L_p of frog mesenteric capillaries if the fiber matrix was present throughout the interendothelial cleft (i.e., the junction strands was open to 50-90% of the line of its length and the matrix filled the whole cleft depth). Further calculations showed that the measured permeability coefficients for solutes (ranging in size from 0.2 to 3.5 nm) were also consistent with this model with a cleft length close to 0.8 µm (50). Thus, although the initial idea for the model suggested the matrix lay only at the endothelial surface, the first quantitative analyses argued for an extension of a similar type of matrix throughout the cleft.

These calculations were made before the observations of Adamson and Michel (9) that breaks in the junctional strand would account for only 10% of the actual length of the junction, and a maximum of only 20-30% of the junction being effectively open, taking into account two-dimensional spreading in the absence of matrix. Thus a junction break model with only 20-30% of the strand open for exchange and with matrix throughout the cleft would underestimate measured water flows at least threefold. In addition, a more critical evaluation of flows through fiber matrices, and the predictions of the Carman-Kozeny Equation by Levick (150), indicated that the Carman Kozeny equation underestimated the resistance to flow in matrices with fiber volumes in the range 1-10%. One problem was that the geometric factor in the equation (the Kozeny coefficient) was not a constant value [equal to 5 as used in Curry and Michel (56) but increased significantly at void volumes between 90 and 99%]. These considerations led to a revised form of the fiber matrix model for water and solute flows through the interendothelial cleft of frog mesenteric capillaries that incorporated the junction geometry of Adamson and Michel (9) and the fiber matrix only at the cleft entrance.

Figure 7 shows an extension of the junction-break model of Figure 5 to include a fiber matrix layer 100 nm thick at the surface of the capillary. The matrix is shown as a highly idealized ordered array of fibers spanning the cleft entrance and with a fiber spacing () of 7 nm, close to the size of albumin (Stokes radius, 3.5 nm). The most accurate description of the resistance to water flows through an array of cylinders that are not constrained within a cleft is given by Sangani and Acrivos (258). For fiber volume <0.7, Tsay and Weinbaum (299a) summarized the results from these authors using the relation
Here  is the spacing between the fibers, r_f is the fiber radius, and D is related to the fractional fiber volume by the relation  = [(/V_f)^1/2  2] × r_f. When the fibers lie within a rectangular slit of width W, the effect of the cleft wall must also be taken into account. Tsay and Weinbaum (299a) showed that the effective conductivity of the matrix within the walls (K_eff) was related to the value of K_p for the unconstrained fibers described by the relation
This relation predicts the specific conductance of a regular array of fibers, 0.6 nm in radius, all lying perpendicular to the direction of flow, and spaced to just exclude albumin (fractional fiber volume of 1.7% is 7 × 10