INTRODUCTION AND OVERVIEW

Atherosclerosis is a disease of the coronary, carotid, and other proximal arteries that involves a distinctive accumulation of low-density lipoprotein (LDL) and other lipid-bearing materials in the arterial wall (1). The disease tends to be localized in regions of curvature and branching in arteries where fluid shear stress (shear rate) and other fluid mechanical characteristics deviate from their normal spatial and temporal distribution patterns in straight vessels (2). Because of the association of disease with regions of altered fluid mechanics, the role of blood flow in the localization of atherosclerosis has been debated for many years (3, 4). Among the first mechanisms proposed to relate blood flow to the localization of atherosclerosis was one in which the fluid (blood)-phase resistance to transport of LDL or other atherogens was controlled by the local wall shear rate (3). Studies by Caro & Nerem (5), however, suggested that the uptake of lipid in arteries could not be correlated with fluid-phase mass transport rates, leading to the conclusion that the wall (endothelium) and not the blood was the limiting resistance to transport. This implied that fluid-flow effects on macromolecular transport were mediated by direct mechanical influences on the transport systems of the endothelium. Somewhat later, attention was drawn to the fact that accumulation of macromolecules in the arterial wall depends not only on the ease by which materials enter the wall, but also on the hindrance to passage of materials out of the wall offered by underlying layers (6, 7). This brought into focus the possibility that the subendothelial intima and medial layers could be important structures contributing to local macromolecular uptake patterns.

In keeping with this brief historical outline of mass transport processes in relation to atherosclerosis, the present review focuses first on the possible role of fluid (blood)-phase mass transport phenomena in the localization of arterial disease; it then directs attention to the arterial wall, discussing at length the processes that mediate transport across the endothelium. Throughout these initial sections describing basic transport processes and systems, the possible influences of fluid mechanics, particularly shear stress, are emphasized. Next, a review of macromolecular uptake patterns in relation to the localization of atherosclerosis is presented. The uptake patterns are then interpreted in light of basic mass transport processes and associated fluid mechanical influences in a final synthesis section in which four mass transport mechanisms that may play a role in the localization of atherosclerosis are described.

FLUID-PHASE RESISTANCE TO TRANSPORT

Solutes that are transported from the bulk fluid (blood) phase to the endothelial surface will encounter a variety of surface boundary conditions depending on the nature of the molecule being transported. For example, a species may be consumed over the entire endothelial surface by enzyme-catalyzed surface reactions (e.g., the hydrolysis of ATP to ADP), it may be transported across the entire endothelial surface because it is soluble in the plasma membrane (e.g., oxygen and other blood gases), or it is solubilized in vesicles that can be transported across the entire surface (e.g., albumin and LDL may be transported in part by this mechanism). On the other hand, many hydrophilic solutes that cannot cross the plasma membrane will be confined to transport through interendothelial junctions that are typically 20 nm in width (at the wide part of the junction) in regions of undamaged endothelium and occupy only a small fraction of the total endothelial surface area (8). Hydrophilic solutes below the size of albumin (7.0-nm molecular diameter) are transported primarily through this intercellular junction pathway. There is also a dominant transport pathway for large molecules (including albumin and LDL) through “leaky junctions” associated with cells in a state of mitosis or apoptosis (9). These leaky junctions have dimensions on the order of 30–1000 nm and occupy a very small fraction of the total surface area. Because of these diverse surface boundary conditions, we consider two cases in describing fluid-phase transport: (a) transport to the entire endothelial cell surface and (b) transport to discrete cellular junctions.

Transport to the Entire Endothelial Cell Surface

Here, we discuss three common situations for transport that utilize the entire endothelial surface: the reactive surface, the permeable surface, and the reactive wall. We then develop simple analytical criteria that allow us to estimate the importance of fluid-phase transport relative to other transport processes.

Referring to Figure 1, we assume that the species of interest is transported from the blood vessel lumen, where its bulk concentration is C_b, to the blood vessel surface, where its concentration is C_s, by a convective-diffusive mechanism that depends on the local fluid mechanics and can be characterized by a fluid-phase mass transfer coefficient k_L. The species flux (mass flow rate divided by surface area) in the blood phase is given by

At the endothelial surface, the species may undergo an enzyme-catalyzed surface reaction, which can be modeled using classical Michaelis-Menten kinetics, with a rate given by where V_max is the maximum rate (high C_s) and k_m is the Michaelis constant. When C_s k_m, as is often the case, then the reaction rate is pseudo–first order and with the rate constant for the surface reaction given by k_r = V_max/k_m.

At steady state, the transport to the surface is balanced by the consumption at the surface so that

It will be convenient to cast this equation into a dimensionless form by multiplying it by d/D, where d is the vessel diameter and D is the diffusion coefficient of the transported species in blood or the media of interest. Equation 4 then becomes where is the Sherwood number (dimensionless mass transfer coefficient) and is the Damkholer number (dimensionless reaction rate coefficient). Solving Equation 5 for the surface concentration, one finds When Da_r Sh, and the process is termed “wall-limited” or “reaction-limited.” On the other hand, when Da_r Sh, and the process is termed “transport-limited” or “fluid-phase-limited.” It is in this transport-limited case that the surface concentration, and in turn the surface reaction rate, depend on the fluid mechanics, which determine the Sherwood number (mass transfer coefficient). It will therefore be useful to compare the magnitudes of Da_r and Sh to determine whether fluid mechanics plays a role in the overall transport process of a surface reactive species.

Many species will permeate the endothelium without reacting at the luminal surface (e.g., albumin, LDL) and their rate of transport (flux) across the surface layer can be described by

where P_e is the permeability coefficient and C_w is the wall concentration beneath the endothelium (Figure 1). If the resistance to transport offered by the endothelium is significant, then it is reasonable to assume So that at steady state when the fluid and surface fluxes balance, Multiplying Equation 13 by d/D to introduce dimensionless parameters and then solving for the surface concentration leads to where Sh was defined in Equation 6 and is a Damkholer number based on endothelial permeability. Equation 14 shows that when Da_e Sh, fluid mechanics again becomes important in limiting the transport.

Oxygen is transported readily across the endothelium, but unlike most proteins, it is rapidly consumed by the underlying tissue. In this case, it is fair to neglect the endothelial transport resistance (assume C_w = C_s), and then by equating the rate of transport to the wall with the (zeroeth order) consumption rate within the wall, we obtain

where is the tissue consumption rate and T is the tissue thickness (distance from the surface to the minimum tissue concentration; see Figure 1). For the specific case of O₂ transport, it is conventional to replace concentration (C) with partial pressure (P) through the Henry's law relationship, C = KP, where K is the Henry's law constant. Invoking this relationship and rearranging Equation 16 into a convenient dimensionless form, we obtain where Sh was defined in Equation 6 and Da_w is another Damkholer number based on the wall consumption rate Clearly, when Da_w Sh, the process is wall limited. But, as Da_w → Sh, the process becomes limited by transport in the fluid phase (P_s → 0) and fluid mechanics plays a role. Because we are treating the tissue consumption rate as a zeroeth order reaction, the case Da_w > Sh is not meaningful (P_s < 0). In reality, as Sh is reduced, the tissue consumption rate must fall due to the lack of oxygen supply from the blood.

In each of the cases outlined above, the transport becomes fluid-phase-limited, and the fluid mechanics may be influential when the Sherwood number is reduced below the Damkholer number. Clearly, species with a high Damkholer number are most likely to be limited by the fluid phase. As described in detail elsewhere (10), Da_e = 0.02–1.0 for LDL and 0.027–0.10 for albumin, assuming that the entire surface is available for the transport of these species. In reality, only a fraction of the transport of albumin or LDL is likely to cross the entire surface (vesicular transport fraction), so the values of Da_e mentioned above should be considered upper bounds for the vesicular transport pathway. Species with much higher Damkholer numbers that are truly associated with the entire endothelial cell surface are ATP (Da_r = 17.7) and oxygen (Da_w = 10.8–49.0). So it is clear that oxygen and ATP are more likely than albumin and LDL to be fluid-phase limited. Because albumin and LDL are also transported through intercellular junctions, the possibility that transport to the junctions is fluid-phase-limited is considered in a later section.

For smooth, cylindrical tubes (a model of straight blood vessels) with well-mixed entry flow, the classical Graetz solution (11) shows that for fully developed transport, where the Sherwood number is minimum, Sh = 3.66 (constant wall concentration) or Sh = 4.36 (constant wall flux). It is shown elsewhere (10) that for parameters characteristic of the human aorta treated as a straight tube, the following values of the Sherwood number are predicted at the end of the aorta: Sh = 31.1 (O₂), 41.8 (ATP), 79.2 (albumin), 114 (LDL). These straight tube estimates of Sh suggest that O₂ and ATP might be fluid-phase-limited (Sh < Da), whereas albumin and LDL would not.

Nonuniform geometries associated with the localization of atherosclerotic plaques in arteries induce fluid mechanical features that can affect fluid-phase transport (Sherwood number), as described in detail by Tarbell & Qiu (10). Flow through a sudden expansion (e.g., at the anastomosis of a vascular graft), a stenosis or flow constriction (e.g., vessel with an atherosclerotic plaque), or a bifurcation, may induce axial flow separation. In such geometries, Sh is reduced near the separation point, where radial velocity away from the wall impedes transport, and increased near the reattachment point, where radial velocity toward the wall enhances transport. These separation zones are also regions of low fluid wall shear stress, which is identically zero at points of separation and reattachment.

In curved vessels and bifurcations, secondary flows (flows in planes perpendicular to the principal flow direction) can also alter fluid-phase transport rates. For conditions characterizing the curvature of coronary arteries over the surface of the heart and the transport of oxygen, Qiu & Tarbell (12) computed Sh = 2 at the inner wall (closest to the center of curvature where atherosclerotic plaques form) and Sh = 55 at the outer wall, a ratio of more than 25. Under the same flow conditions, the mean (time-averaged) wall shear stress was less than two times greater on the outside wall than on the inside wall. In the curved geometry, the secondary flow is directed away from the wall at the inner curvature, inducing a reduction in Sh, and toward the wall at the outer curvature inducing an elevation in Sh. Further studies of oxygen transport in an anatomically realistic right coronary artery model accounting for out-of-plane curvature showed a similar drastic reduction in the Sherwood number at the inner wall (13, 14). Because Sh Da_w for oxygen at the inner wall of these coronary artery models, hypoxia is predicted to occur at the inner wall where atherosclerotic plaques occur. For similar reasons, hypoxia might be expected to occur on the outer wall of a bifurcation (wall opposite the flow divider where atherosclerotic plaques are localized), and this has in fact been observed in experiments in the dog carotid bifurcation (15). The role of hypoxia in atherogenesis will be discussed further in a later section.

Transport to Endothelial Junctions

When transendothelial transport is confined to intercellular junctions, the fluid-phase transport problem involves a boundary surface that is impermeable almost everywhere except at the discrete junctions. To visualize this, in Figure 2 the intercellular junctions are exaggerated and idealized as slits between cells that are either parallel (sides of the cells) or perpendicular (ends of the cells) to the flow direction. This emphasizes the fact that endothelial cells are elongated in the direction of flow (16) so that most of the intercellular junction is in the parallel configuration. The problem of transport to a rectangular slit in the perpendicular configuration at high Peclet number ( is the wall shear rate, a is the slit half-width, and D is the solute diffusivity) has a well-known solution due to Léveque (17) that predicts the surface flux (to a sink at zero concentration) to be proportional to Pe^1/3. The Leveque solution is accurate for Pe > 100, but for tight junctions (a = 10 nm) and reasonable shear rates ( = 1000 s⁻¹), Pe < 1 for a broad range of solutes, including LDL (D = 2.06 × 10⁻⁷ cm²/s). Ackerberg et al. (18) presented experimental data for the heat transfer analog of the perpendicular slit problem and extended the range down to physiologically relevant Peclet numbers (10⁻⁴). Hodgson & Tarbell (8) solved the parallel slit configuration problem numerically for both specified concentration and specified flux conditions at the slit surface. There was little influence of the slit boundary condition on the Sherwood number (Sh = k_ca/D; k_c = mass transfer coefficient based on the slit area, A_c), which for the fixed concentration boundary condition was well predicted by the following correlation of numerical results for Pe < 10:

where Sh₀(= 0.299) is the Sherwood number at zero flow. The value of Sh₀ was within 2.26% of the theoretical prediction of Saito (19) for the problem of diffusion to an infinitely long slit. As shown in detail by Hodgson & Tarbell (8), the Sherwood number for slits parallel to the shear direction is lower than the Sherwood number for perpendicular slits at all Pe > 0. The two solutions are, of course, identical at Pe = 0, where Sh = Sh₀. For a sink having the shape of a circular disk of radius a, at Pe = 0, Saito (19) showed that Sh₀ = 1.27, but the dependence on shear rate (Pe) has not been determined for the circular disk sink.

To determine whether fluid-phase transport to intercellular junctions (modeled as slits) can limit overall transport, it suffices to compare the lowest possible value of the Sherwood number (when Pe = 0) to the appropriate Damkholer number for the solute of interest. Because the Damkholer numbers described previously were all based on the total surface area, it is necessary to multiply the slit mass transfer coefficient (k_c) by A_c/A_total to generate the k_L used in the definition of the Sherwood number (Equation 6). For cylindrical tubes with longitudinal slits, A_c/A_total = 2a/w, where w is the width of an endothelial cell (order 10 μm) and a is the slit half-width. Now the Sherwood number (Equation 6) is given by

It is interesting to note that this Sherwood number is proportional to the vessel diameter (d) and does not depend on the junction dimension (a). In the largest artery (d = 2.5 cm), Sh = 1495, and in the smallest capillary (d = 5 μm) Sh = 0.299. Because the Damkholer numbers for albumin and LDL are at most 0.1 and 1.0, respectively, it is clear that fluid-phase transport of these solutes to intercellular junctions is not a limiting factor in arteries, but may be important in capillaries.

The above considerations of the discrete nature of endothelial junctions as transport sinks did not include the effects of convective flux of solute into the junction associated with pressure-driven flow across the wall. The volume flux (J_v) or superficial velocity in arteries is typically of the order 10⁻⁶ cm/s (20) based on the filtration flow rate divided by the total surface area of the vessel (A_total). But, the transmural flow is confined primarily to the intercellular junctions such that the fluid velocity into the junction (v) is given by

Because the area ratio (A_total/A_c) may be as high as 10³, the local velocity entering the intercellular junction could reach 10⁻³ cm/s. This locally elevated fluid velocity might be expected to convect solute into the junction at a substantial rate. Although this convective flux seems to not have been investigated thoroughly in the literature, Hodgson & Tarbell (8) have estimated that it may be significant for large solutes such as LDL. This additional fluid-phase-transport mechanism reinforces the conclusion of the preceding paragraph that fluid-phase transport to intercellular junctions is not a limiting transport process in arteries.

Concentration Polarization

The convective flux of plasma proteins to and within the intercellular junction associated with the enhanced velocity described by Equation 21 could induce “concentration polarization”—the accumulation of solute on the surface or within the intercellular junctions because of the high resistance to transport offered by the endothelium (21, 22, 23). Elevated protein concentration at the surface could lead to higher protein transport across the endothelium by diffusive mechanisms as well as to reduced volume flux (J_v) because of the resistance of the concentrated protein layer and the locally elevated osmotic pressure at the endothelial surface. If significant concentration polarization layers exist on endothelial surfaces, then increases in the fluid shear stress of blood flow would be expected to reduce the polarization layer (reduce the maximum protein concentration at the surface), leading to elevation of J_v and reduction of the diffusive component of J_s. Furthermore, these shear stress effects would be rapidly reversed after returning shear stress to its original level.

Lever et al. (24) measured the change in J_v across rabbit common carotid arteries perfused with solutions containing 1% or 4% albumin when the lumenal flow shear stress was increased from zero to a level in the range of 0.57–1.65 dyne/cm². There was an approximately 30% increase in J_v associated with shear stress that was reversible within 30 min after removal of shear stress. However, there was no correlation between the J_v elevation and the shear stress level, and the protein concentration had no effect on the degree of J_v elevation either. An analysis based on concentration polarization in endothelial junctions suggested that this mechanism could not completely account for the observed changes in J_v with shear stress.

Sill et al. (25), using bovine aortic endothelial cells cultured on porous supports, observed that a step change in shear stress from 0 to 10 dyne/cm² induced a 2.16-fold increase in hydraulic conductivity (L_p = J_v/ΔP), which occurred gradually over a period of 60 min. When shear stress was removed, L_p remained elevated for an additional 120 min. Other tests were performed to show that the endothelial monolayers were not damaged by shear stress and that L_p could be reversed by addition of chemical agonists. Chang et al. (26), using the same cell culture model, showed a gradual 4.7-fold increase in L_p over a 3-h period of exposure to a step change in shear stress from 0 to 20 dyne/cm². The increase in L_p in response to shear stress could be blocked completely by preincubating the cells with a nitric oxide synthase (NOS) inhibitor. These cell culture studies suggest that increases in L_p in response to shear stress are associated with mechano-chemical transduction of shear stress by endothelial cells and not concentration polarization.

Williams (27) measured hydraulic conductivity in capillaries of the frog mesentery perfused with 1% albumin and observed that a change in shear stress from about 9 to 18 dyne/cm² led to about a 5-fold increase in L_p of venular capillaries, a 2.5-fold increase in L_p of true capillaries, and no increase at all in L_p of arteriolar capillaries. The differential behavior of these capillaries in response to increases in shear stress suggests endothelial-dependent mechano-chemical transduction as the controlling mechanism.

It should also be noted that most studies of the permeability of the endothelium to macromolecules have shown that increases in blood flow (shear stress) lead to an increase in permeability, not a decrease as would be predicted by concentration polarization. For example, Caro & Nerem (5) showed a weak positive dependence of endothelial permeability to lipoprotein-linked cholesterol in excised dog carotid arteries, and similar results were shown for albumin in the same model (28). Yuan et al. (29) showed that increases in flow through coronary venules of pigs increased albumin permeability by a nitric oxide–dependent mechanism.

There have been several computational modeling studies of concentration polarization of LDL at the luminal surface of an artery [e.g., (30)] that have been reviewed by Ethier (14). In these theoretical studies, transendothelial volume flux and solute flux were assumed to be distributed over the entire surface of the vessel wall. By setting J_v sufficiently high and the LDL diffusion coefficient sufficiently low it was possible to demonstrate conditions under which the wall concentration of LDL would be elevated above the bulk concentration. But, as suggested above, this type of model is not consistent with most observations of the effect of increasing shear stress on endothelial permeability and hydraulic conductivity. A more realistic model of the concentration polarization process would confine the volume flux to intercellular junctions and leaky junctions and distribute the LDL flux into a fraction associated with vesicular transport that would access the entire surface and a fraction associated with leaky junctions that would access only a very small fraction of the surface. Studies of this type remain to be investigated.

Solutes Limited by the Fluid Phase

In a brief summary of the considerations of this section, it appears that oxygen, and possibly ATP, low-molecular-weight species that are consumed readily by the arterial wall and that access the entire endothelial surface, are the most likely solutes to be limited by fluid-phase transport. This gives rise to the prediction of hypoxic regions in arteries where there is flow separation or secondary flow that correspond to locations where atherosclerotic plaques are often localized. The mechanisms by which hypoxia can induce atherosclerosis are discussed in a subsequent section. High-molecular-weight materials such as LDL and albumin and other hydrophilic solutes that are transported through intercellular junctions and leaky junctions and only access a small fraction of the endothelial surface, are not limited by the fluid-phase, but by the endothelium.

ENDOTHELIAL/INTIMAL RESISTANCE TO TRANSPORT

A number of recent review papers have treated endothelial transport (31, 32, 33) and may be consulted for additional information. The endothelium lines all blood vessels from the largest arteries and veins down to the smallest capillaries. Much of what is known about the nature of transendothelial transport derives from studies of the microcirculation. But atherosclerosis is a disease of the larger arteries, and thus endothelial transport in arteries is the focus of this section.

To characterize transport across the endothelium, the following membrane transport equations are usually employed (34):

where J_v is the volume flux across the endothelium, J_s is the solute flux, ΔP is the pressure differential across the endothelium, Δπ is the corresponding osmotic pressure differential, ΔC is the solute concentration differential, σ is the reflection coefficient, L_p is the hydraulic conductivity, P₀ is the diffusive permeability, and is the mean intramembrane solute concentration, which depends in a complex way on the relative importance of convection (second term in Equation 23) and diffusion (first term in Equation 23). ΔP and Δπ are driving forces for volume transport (water flow), which convects solute; ΔC is the driving force for diffusive solute transport; and P₀, L_p, and σ are transport properties. It is conventional to think of membrane transport properties as constant, but a remarkable feature of the endothelium is that its transport properties are sensitive to both the chemical and mechanical environment within which it resides.

The more rigorous form of Equation 23 was developed by Patlak et al. (35) and can be represented as follows:

In Equation 24, P_e is the apparent permeability that is usually measured experimentally, and the assumption has been made that the concentration on the bottom side of the endothelium (C_w in Figure 1) is much less than on the top (C_w C_s). The first term is the diffusive contribution and the second term is convective. The relative importance of diffusion to convection is determined by the factor Z, defined as where N_Pe is a Peclet number defined as

Starling's Law Revisited

Equation 22, with σ = 1, is often referred to as “Starling's law,” as it expresses Starling's pioneering hypothesis that fluid movement across microvascular walls is determined by the transmural difference in hydrostatic and oncotic pressures (36). It has become a general principle of cardiovascular and renal physiology that as capillary pressure drops below plasma oncotic pressure (about 25 mm Hg), the net driving force for volume flux changes from positive (filtration) to negative (reabsorption). Studies by Intaglietta & Zweifach (37), however, challenged the notion that significant reabsorption occurs at the venous end of the microcirculation where the osmotic pressure exceeds the capillary pressure. Their investigations indicated that fluid reabsorption was demonstrable in short-term experiments, where protein does not attain a steady state distribution in the extravascular space, but not on a long-term or steady state basis. This concept was demonstrated more clearly by Michel & Phillips (38) who performed experiments on isolated, perfused, frog mesenteric microvessels and observed transient reabsorption over a period of 15–30 s after the capillary pressure was dropped below the oncotic pressure, as expected from Starling's law. When the measurements were extended for 2–5 min after the drop in pressure, however, there was no evidence of reabsorption, only a small filtration flow. Additional experiments using frog mesenteric microvessels (39) and bovine aortic endothelial cells in vitro (40) have provided further support for the transient reabsorption/steady state filtration concept.

The first plausible model that could predict steady state filtration when the Starling forces (ΔP − Δπ) predict reabsorption was presented by Michel & Phillips (38). They developed a steady state, one-dimensional, transport model based on classical equations (Equations 22 and 24, above), but with the novel assumption that the pericapillary concentration, C_w, that determines π_w in Equation 22, was given by a mixing condition expressed as

where J_s is the flux of the oncotic solute. In general, C_w differs from the global tissue concentration that is fixed by bathing solutions or external reservoirs in typical experiments. The resulting volume flux equation, only admits positive solutions for J_v (filtration only).

A much more detailed microstructural model of the endothelial transport barrier accounting for the surface glycocalyx layer and the fine structure of the interendothelial junction has provided further insights into the Starling mechanism (9, 41). This model predicts that the Starling forces are determined by the local differences in hydrostatic and oncotic pressures across the surface glycocalyx layer. We return to a discussion of this model in a later section.

Pathways for Transport Across the Endothelium

The basic membrane transport equations presented in the preceding section can be used to define the transport properties (L_p, P_e, σ) based on experimental measurements of transendothelial fluxes in the presence of known driving forces (ΔP, Δπ, ΔC). In principle, they are based on the assumption that water and solutes share a single pathway across the endothelium. When this assumption is not valid, inconsistencies in the analysis of experimental data may arise. For example, Dull et al. (42) used Equation 24 to fit measurements of J_v and J_s for albumin across BAEC monolayers in vitro and found that the equation could only be satisfied for σ < 0. This contradiction suggests at least two pathways for transport across the endothelium.

In the early 1950s, Pappenheimer et al. (43) formulated the pore theory of capillary permeability that predicted the transport of small hydrophilic solutes through water-filled channels or pores of radius ∼4 nm. The pores were assumed to be morphologically represented (in continuous endothelia) by the clefts or junctions between endothelial cells. Later, Grotte (44) presented evidence for a two-pore barrier partitioning blood from lymph. Grotte observed that large-molecular-size dextrans appeared in canine leg lymph in a concentration that decreased rapidly as a function of increasing molecular size for molecules <4.5 nm in radius. Larger molecules still appeared in lymph, but at concentrations that were only slightly affected by molecular size. Grotte (44) therefore suggested that a few “large pores” or “capillary leaks” (1/30,000 of the small pores) with a radius of 25–60 nm would account for the transport of plasma proteins. The nature of these pores or pathways across the endothelium still has not been completely resolved, although considerable progress has been made as described below.

Intercellular Junction Transport

As suggested by Pappenheimer et al. (43), transport of water and hydrophilic solutes below the size of albumin are believed to be dominated by the junction between endothelial cells. In this section, the molecular structure and architecture of these junctions are described. Michel & Curry (33) provide a more extensive review of the developments leading up to our current understanding of the structure and function of the interendothelial junction and may be consulted for additional details.

Figure 3 is a schematic diagram of the intercellular junction showing details of the molecular structure of the tight junction and the adherens junction. Several protein components of the tight junction have been identified in recent years, and although the list is almost certainly incomplete, the structure and organization of the tight junction is becoming clearer (45, 46). In transmission electron micrographs, the tight junction appears as a series of very close points of cell-to-cell contact, sometimes referred to as “kisses,” that circumscribe the cell. The first transmembrane protein identified in the tight junction was the 65-kDa protein occludin (47), which is believed to play a key role in providing the tight junction seal. The protein has two extracellular loops that interact with the extracellular domain of molecules anchored to the apposing cell membrane to contribute to the tight junction seal. The cytoplasmic domain of occludin binds ZO-1, a cytoplasmic plaque protein that plays an important role in organizing the paracellular seal. ZO-2 is another well-characterized protein in the cytoplasmic plaque that links the tight junction to cytoskeletal filaments including actin. In 1998, the first members of the claudin family of tight junction proteins were identified (48). These 23-kDa integral membrane proteins, with four transmembrane domains, bear no sequence similarity to occludin, but are incorporated into the tight junction strand.

The adherens junction is a cell-cell adhesive junction that binds cells together and connects the cytoskeleton (actin) to the plasma membrane (49). Adherens junctions may be considered epicenters for signal reception, transduction, and response to local patterning cues. The adhesive components of adherens junctions are formed by type I cadherins (calcium-dependent adhesive proteins) that belong to the larger cadherin superfamily. Type I cadherins, of which VE-cadherin is the prototype, consist of five tandem extracellular 110–amino acid repeats, a transmembrane domain, and a highly conserved cytoplasmic region. Cadherins mediate highly specific homophilic adhesion with cadherins from apposing cells.

Although adhesive associations of the ectodomain occur spontaneously, the cytoplasmic domain is required to support sustained adhesion. The cytoplasmic domain governs the clustering of cadherins into a cooperative zipperlike arrangement and the tethering of cadherins to the cytoskeleton. The cytoplasmic domain of VE-cadherin forms a stable complex with major proteins: the catenins (alpha, beta, gamma) and plakoglobin. The cadherin-catenin complexes are bound to alpha-actinin, vinculin, radixin, and the actin cytoskeleton.

The plasma membrane of endothelial cells is covered by a thin layer of macromolecules commonly referred to as the glycocalyx. This surface layer is also believed to fill the entrance to the intercellular junction as suggested in Figure 3. The major molecular components of the glycocalyx are carbohydrates and glycoproteins, such as glycosaminoglycans (GAG), proteoglycans (PG), and glycolipids (50). A significant component of the proteoglycans of the endothelial glycocalyx are glucidic residues, as evidenced by the binding of lectins to the endothelium (51). The two most prevalent GAG carriers on the endothelial cell surface are syndecans and glypicans. Syndecans, the most common GAG carrier, contain a transmembrane core protein that is directly attached to the membrane phospolipid (52). On the other hand, glypicans are attached to the membrane via a glycosyl phosphatidyl inositol linkage. Syndecan-1, the most prevalent of four syndecan families on the vascular endothelium, has multiple forms based on the types of heparan sulfate GAG attached to the extracellular domain, but can also carry chondroitin sulfate (53, 54). The length of heparan sulfate is variable and typically depends on the abundance of N-acetylglucosamine residues integrated in the chain (55). There have been attempts to define the location of heparan sulfate and chondroitan sulfate GAGs along the core protein, but the location of either chain is not highly predictable (56).

The thickness of the glycocalyx layer has been controversial. As reviewed by Pries et al. (57), several electron microscopy studies that are subject to artifacts indicate a thickness of less than 100 nm. Indirect measurements in capillaries based on flow resistance measurements after heparinase treatment, plasma labeling, and hematocrit measurements indicate surface layer thicknesses of 0.5 μm–1.1 μm.

Figure 4 (adapted from Reference 58) provides a conceptual view of the interendothelial junction that captures the major features and translates the molecular descriptions of the preceding sections into a more quantitative form suitable for mathematical modeling of transport rates. In this three-dimensional view, the upper and lower planes separated by a distance 2B represent the surfaces of two adjacent endothelial cells and the intervening structures constitute the intercellular junction or “cleft.”

The lumen side of the channel is filled with fiber matrix (bars perpendicular to the channel walls) representing the glycocalyx that extends into the channel but not all the way to the “tight junction.” The fibers have a radius of 0.6 nm and are spaced 7-nm apart (close to the size of albumin) in a regular array. The tight junction is represented by the parallel walls in the middle of the channel that are separated by an open slit of width 2b_s. The slit in the tight junction that allows passage of smaller molecules can be replaced by a set of small pores that might more realistically depict small channels in the tight junction strand composed of pairs of proteins or oligomers of proteins (59). The model also incorporates breaks in the tight junction of width 2d, which repeat over a distance 2D as the tight junction forms a belt around the endothelial cell. The model treats the tight junction as a single belt with breaks, whereas freeze fracture electron microscopy studies reveal several discontinuous belts in close proximity (33). Typical values for the parameters in this model have been developed based on data from frog mesenteric capillaries (39): 2B = 20 nm, 2b_s = 1.5 nm, 2d = 150 nm, 2D = 4300 nm. The fiber matrix in the entrance region extends 150 nm, and the distance from the end of the fiber matrix to the tight junction is at least 25 nm. The actual thickness of the tight junction is of the order 10 nm, but is neglected in the mathematical treatment. The channel beyond the tight junction toward the tissue side may be thought of as representing the “adherens junction,” with protein structures that are important in maintaining the spacing between cells (2B), but offering little resistance to transport. Clearly, there are many parameters in this geometric model of the interendothelial junction that would be expected to vary with vessel type and species.

The model predicts that most of the water flow (volume flux) and larger solute flux (up to the size of albumin) pass through the breaks in the tight junction (2d × 2B) and not the high resistance slits in the tight junction (2b_s). The glycocalyx serves as the primary molecular sieve. It offers little resistance to small solutes (<1-nm radius), but as the solute size increases to approach the fiber spacing of 7 nm, the matrix resistance increases to become the major resistance to larger solutes the size of albumin. Whereas the tight junction provides a very high resistance to water flow, in the region of a break in the tight junction, the glycocalyx accounts for a substantial resistance to water flow (up to 80%). The model also demonstrates that the oncotic gradient across the entire capillary wall (π_s–π_w) is not the determinant of water flow as suggested by Equation 22 and Starling's law. Rather, it is the oncotic gradient (albumin concentration gradient) across the surface glycocalyx. The albumin concentration just beyond the glycocalyx layer is largely uncoupled from the concentration on the tissue side by convective flow through the breaks in the tight junction that prevents albumin back diffusion toward the glycocalyx layer.

Leaky Junctions

Intercellular junctions, as described in the preceding section, would not normally allow significant passage of LDL (diameter ∼23 nm) even through breaks in the tight junction because the wide part of the cleft is expected to be on the order of the LDL dimensions. Thus, normal intercellular junctions would not constitute a “large pore.” Weinbaum et al. (60), building on earlier experimental observations by Gerrity et al. (61), proposed that the large pore was an infrequent, transiently leaky interendothelial cleft associated with a tiny fraction of cells (one cell in 10³–10⁴) that were in the process of cell turnover; i.e., either cell division or cell death. These cells had leaky junctions because they were either dying and being sloughed off by healthy neighboring cells or had poorly formed junctions because they were in the process of dividing. Chien and coworkers, in a series of studies in the rat aorta (62) reported that approximately one cell in 3000 had a leaky junction and this leakage would last on average about one hour before a well-formed new junction was established. In the rat aorta, cells in mitosis accounted for about 26% of all leakage sites, and dead or dying cells for 37% of leakage sites. The dimensions of the leaky junctions were estimated from HRP staining under electron microscopy to have a minimum width of 80 nm and a maximum width of 1330 nm for mitotic cells, and a minimum width of 15 nm and a maximum width of 1000 nm around dying or dead cells (62).

In the rabbit aorta, however, mitotic cells accounted for only about 10% of enhanced permeability sites, and the distribution of replicating endothelium did not correlate with the distribution of sites of elevated LDL permeability or early lesions (63). In the rabbit aorta, 31% of the sites of elevated LDL permeability were associated with subendothelial white cells, suggesting a possible association of these cells with leaky junctions. It appears, however, that these studies in the rabbit aorta did not assess the contribution of dying cells to enhanced permeability.

Another possible mechanism for the generation of a transient large pore is the transient disruption of endothelial cell plasma membranes (64). Cells that have had their plasma membrane disrupted (wounded cells) but that are not dead are capable of resealing their membranes and continuing vital function. Such wounded endothelial cells have been observed in the aortas of rats and varied significantly between 1.4% and 17.9% of the total aortic endothelial cell population. Wounded cells were heterogeneously distributed, being found in distinct clusters often in the shape of streaks aligned with the axis of the vessel or in the shape of partial or complete rims surrounding bifurcation openings, such as the ostia of the intercostal arteries. In fact, 80% of mitotic cells were identified as wounded. Although the relationship of these wounded cells to sites of enhanced permeability has not been studied directly, the pattern of cell wounding in the aorta bears a superficial resemblance to the pattern of enhanced permeability (e.g., concentrated around ostia). The association with mitotic cells also suggests an association with enhanced permeability. The fraction of wounded cells (6.5% on average in the aorta), however, is remarkably high—orders of magnitude higher than the fraction of cells in mitosis or apoptosis. So it is not at all clear what fraction of wounded cells might be leaky and whether or not they would leak through their intercellular junctions or across their abluminal membranes.

Although tight junctions are generally regarded as forming nearly continuous belts around endothelial cells, electron microscopic freeze-fracture studies show that tight junctions are discontinuous at tricellular corners where the borders of three cells converge. It has been hypothesized that tricellular corners are potential sites for the transient opening and closing of the paracellular pathway and possible avenues through which white cells migrate during inflammatory reactions (65). A strong association between migrating neutrophils and tricellular corners has been demonstrated in some studies, suggesting that these sites may also constitute a large pore pathway for macromolecules. Gaps ranging in size from 0.25 to 2 μm have been observed in human umbilical vein endothelial cell (HUVEC) monolayers in vitro. Thus, tricellular corners could also contribute to a large pore system.

Vesicles

Clathrin-coated vesicles are well-characterized endocytic organelles that provide an efficient pathway for taking up specific macromolecules from the extracellular fluid, a process known as receptor-mediated endocytosis (66). The macromolecules bind to complementary cell-surface receptors, accumulate in coated pits, and enter the cell as receptor-macromolecule complexes in endocytotic vesicles. Because extracellular fluid is trapped in coated pits as they invaginate to form coated vesicles, substances dissolved in the extracellular fluid are also internalized by the process of fluid-phase endocytosis. The uptake of cholesterol bound to LDL into endothelial cells occurs via the binding of LDL molecules to LDL receptors in clathrin-coated pits. After shedding their clathrin coats, the endocytic vesicles deliver their contents to endosomes and the receptors are returned to the plasma membrane. Because these vesicles can transport LDL into the cell, they could constitute a “large pore” system if they were likely to deposit their contents on the abluminal side of the cells. This seems unlikely to be the case, at least for LDL, because a number of studies have shown that the endothelial permeability to unmodified LDL is the same as to modified forms of LDL that do not bind to the LDL receptor (31).

Capillary endothelium is rich in homogeneously sized (60–80 nm) membrane-bound vesicles, which have been termed plasma-lemmal vesicles or caveolae (67). Most capillary caveolae are individual structures that invaginate from either the luminal or abluminal plasma membrane. Sometimes caveolae form aggregates of two or three interconnected vesicles or appear as free structures in the endothelial cell cytoplasm without connection to the plasma membrane. Caveolae lack a visible cytoplasmic coat by electron microscopy, but a 21-kD integral membrane protein, caveolin-1, is associated with their cytoplasmic face. Caveolae may be responsible for transcytosis, the process by which plasma proteins are transported across the endothelium. Transcytosis is an active process in which caveolae that open to the vascular lumen take on plasma proteins as cargo then separate from the plasma membrane, pass across the endothelial cytoplasm, and fuse with the abluminal membrane where they discharge their contents. Persuasive evidence for the involvement of these vesicles in macromolecular transport comes from studies using the cholesterol scavenger, filipin, to inhibit caveolae. Schnitzer et al. (68) showed that not only did filipin remove caveolae from cultured pulmonary microvascular endothelial cells, but it also inhibited transport of albumin across monolayers of these cells. Other studies with N-ethylmaleimide (NEM), which inhibits the fusion of vesicles, have demonstrated inhibition of albumin transport (69). Whereas LDL binds to a surface receptor in clathrin-coated pits, albumin binds to the surface of endothelium through specific albumin-binding proteins (gp60, gp30, and gp18), gp60 (albondin) being associated with transcytosis via noncoated plasma-lemmal vesicles (68). Albondin is well expressed in microvasclar beds, but its abundance in arterial endothelium has not been described.

Fluid Mechanical Effects on Endothelial Transport

Intercellular junctions are the principal pathway for the transport of water and hydrophilic solutes below the size of albumin. Albumin itself likely traverses the endothelium through a variety of pathways, including intercellular junctions, leaky junctions, and vesicles. LDL and high-molecular-weight materials have limited access to the normal intercellular junction and must utilize leaky junctions or vesicles to cross the endothelium. Each of these transport pathways can be influenced by fluid mechanical forces, particularly shear stress, acting on endothelial cells. In this section, the overall effects of fluid mechanical forces on endothelial transport rates are reviewed first, and then the specific influences of fluid mechanics on individual transport pathways are considered.

The first unambiguous demonstration of a direct effect of fluid shear stress on endothelial transport was reported by Jo et al. (70) using bovine aortic endothelial cell (BAEC) monolayers cultured on a porous substrate mounted on the wall of a parallel plate flow chamber. This study demonstrated a tenfold increase in albumin diffusive permeability (P₀) within 1 h after the onset of 10 dyne/cm² steady shear stress. The permeability returned to preshear baseline levels within 2 h after the removal of shear stress. Sill et al. (25) demonstrated a direct effect of shear stress on the hydraulic conductivity (L_p) of BAEC monolayers, demonstrating a significant increase in L_p after one hour of exposure to 20 dyne/cm² steady shear stress.

More recent studies using the same BAEC model have shown that the L_p response is mediated by nitric oxide (NO) production in response to steady shear stress (26), whereas the albumin P₀ response to steady shear stress is independent of NO (71). These findings suggest that albumin has access to different physical pathways across the endothelium than water and that these pathways are controlled by distinct biochemical signaling mechanisms in BAECs. It must be remembered, however, that P₀, the diffusive permeability, is measured in the absence of water flux. Whether albumin transport coupled to water flux is enhanced by shear-induced increases in L_p has not been determined. It is known from static experiments (no shear stress) in BAECs, in which water and albumin flux were measured simultaneously, that albumin P_e increases as J_v increases, but with a negative reflection coefficient (s), suggesting that water and albumin do completely share a common pathway (42).

McIntire et al. (72) exposed bovine brain microvascular endothelial cell monolayers grown on porous polycarbonate filters to steady shear stress of 1 or 10 dyne/cm² for up to 72 h and monitored the P₀ of dextrans up to a molecular weight of 2 million. Maximal increases in P₀, as high as 76-fold for the largest dextran, were observed between 10 and 30 h with a return nearly to baseline after 48 h.

The most recent study of BAEC monolayer transport properties in response to shear stress showed that when BAECs were exposed to steady shear stress of 20 dyne/cm² or oscillatory shear stress of 10±10 dyne/cm², they displayed a similar L_p response: 3- to 3.5-fold increase in L_p after 3 h of exposure. But when the shear stress amplitude was increased so that a reversing oscillatory shear stress of 10±15 dyne/cm² was imposed, there was no increase in L_p above baseline during 3 h of exposure (73). This surprising behavior seemed to be mediated by a dramatic upregulation of NO production under the reversing oscillatory shear conditions that suppressed the L_p increase. It therefore appears that for BAEC monolayers there is a biphasic response of L_p to NO: At low levels, increases in NO induce L_p increases, whereas at high levels of NO, increases in NO suppress L_p. Such biphasic behavior may help to explain why seemingly opposite responses of vascular transport properties to changes in shear stress have been observed in some animal studies. For example, Kurose et al. (74) and Baldwin et al. (75) showed that when NO production was suppressed by blocking NOS, venules of the rat mesentery became more leaky to albumin.

It is also interesting to note that in vitro studies of transport using cell types other than BAECs have shown diverse responses. Lakshminarayanan et al. (76), using a bovine retinal microvascular endothelial cell (BREC) model, showed a dramatic increase in L_p in response to 20 dyne/cm² steady shear stress that could be completely blocked by a NOS inhibitor—similar to the BAEC response. On the other hand, the same experiments with a HUVEC model showed a small reduction in L_p in response to the same shear stress stimulus (77), even though it is well known that HUVECs produce abundant NO in response to steady shear stress (78). This diversity of responses to shear stress was paralleled by the response of these same in vitro models to vascular endothelial growth factor (VEGF). Chang et al. (79) observed that L_p and albumin P₀ were increased significantly after 3 h of exposure to 100 ng/ml VEGF in BAEC and BREC monolayers, but were unaffected by the same stimulus in HUVEC monolayers. It is therefore clear, and not surprising, that the transport properties of endothelial monolayers depend on the cell source.

Since 1991, there have been a number of additional reports of flow- or shear-dependent endothelial transport properties based on studies using intact vessels and in vivo preparations. Lever et al. (24) observed that the L_p of rabbit carotid arteries in an ex vivo flow loop increased by 30% after 20 min of exposure to a step change in shear stress of about 1 dyne/cm². Although this relative level of increase in L_p in response to shear stress is much less than described above for BAEC monolayers in vitro, it must be remembered that the medial layer of the arterial wall, which is not present in the in vitro models, is believed to contribute about 50% of the overall resistance to volume flux in an artery (20).

Although it had been hypothesized earlier (80), only recently have several studies reported an effect of flow on capillary permeability. Shibata & Kamiya (81) measured local capillary permeability of Cr-EDTA (MW 341) in the rabbit tenuissimus muscle at various capillary blood flow levels using an indirect tissue clearance method. They observed a significant positive correlation between capillary P_e and the mean capillary red cell velocity (a tripling of P_e as red velocity varied from 0.5 to 2.5 mm/s). Yuan et al. (29) measured P_e of albumin in isolated cannulated coronary venules at various intraluminal perfusion velocities and observed a 47% increase in P_e when velocity increased from 7 to 13 mm/s. Caldwell et al. (82) determined the uptake of the fatty acid iodophenylpentadecanoic acid (IPPA) in the coronary microcirculation of dogs exercising on a treadmill. The data could best be fit by a model in which the capillary P_e was taken to be a positive linear function of the local flow rate. Kajimura & Michel (83) observed a positive correlation between flow velocity and potassium ion permeability in single perfused venules of rats that was mediated by nitric oxide. Williams (27) measured L_p after step changes in shear stress in arterioles, capillaries, and venules of the frog mesentery using the modified Landis technique. The response of vessels was graded across the capillary bed, with arteriolar capillaries demonstrating no response, true capillaries a moderate response, and venular capillaries a strong response (fivefold increase) to a step change in shear stress. This diversity of responses is reminiscent of the differences between BAECs/BRECs and HUVECs observed in vitro as described above. It will be important in the future to determine the intracellular signaling mechanisms and molecular components of the intercellular junction that control these diverse responses.

The studies described above showing shear stress alterations of L_p and P_e were not designed to determine the specific transport pathway being affected by shear stress. However, there have been a number of studies specifically addressing the effects of shear stress on tight junctions, adherens junctions, vesicles, and the leaky junction–related processes of cell turnover and cell death. Most of these studies, described below, did not determine transcellular transport rates simultaneously, and so a direct association between altered transport pathways and altered transport rates can only be inferred.

DeMaio et al. (84) determined the effect of steady shear stress on the occludin and ZO-1 content of BAEC monolayers over a 3-h period. Immunofluorescence staining showed that these two proteins were predominantly localized to the cell boundaries as expected for tight junction proteins. Three hours of exposure to 10 or 20 dyne/cm² shear stress had no effect on ZO-1 content, but induced a 44% or 50% reduction in occludin content. Parallel transport experiments showed that 20 dyne/cm² shear stress induced a 4.7-fold increase in L_p after 3 h. Occludin expression and L_p were dependent on tyrosine kinase activity because erbstatin A attenuated both the shear-induced decrease in occludin content and the increase in L_p. Shear stress increased occludin phosphorylation after only 5 min of exposure, and continued to induce phosphorylation out to 3 h. The shear-induced increase in occludin phosphorylation was attenuated with dibutyrl (DB) cAMP, a reagent that has been shown to reverse the shear-induced increase in L_p (26). It appears that alterations in occludin content and phosphorylation are mechanisms by which shear stress alters BAEC L_p. The phosphorylation event occurs earliest and is associated with alterations in L_p before there is a change in occludin content, suggesting that the phosphorylation event is central to the transport response.

In a similar study, it was demonstrated that VEGF increases occludin phosphorylation in BREC monolayers within 15 min (85) and increases L_p after 30 min (86). In a related transport study, DeMaio et al. (87) measured BREC L_p and P_e of 70-kDa dextran simultaneously before and after administration of VEGF. VEGF induced a 3.5-fold increase in both water and solute flux after 120 min, but the reflection coefficient (σ, Equation 22) remained constant (about 0.7) while the transport fluxes were increasing. Taken together, these studies of the effect of VEGF on BREC transport suggest that the tight junction disassembles in response to VEGF by increasing either the number of breaks or the length of breaks in the tight junction (recall Figure 4) as occludin is phophorylated. When this happens, there is increased water flux and associated solute flux through the breaks, but the glycocalyx, which serves as the solute filter and determines the reflection coefficient, remains unaffected. Considering the strong parallels between shear stress and VEGF in altering tight junction proteins and transport, it is plausible that shear stress alters the tight junction in a similar manner.

Schnittler et al. (88) exposed HUVEC monolayers to 20 dyne/cm² steady shear stress and examined the junctional intensity of VE-cadherin, plakoglobin and platelet endothelial cell adhesion molecule 1 (PECAM 1), which is a Ca²⁺-independent transmembrane adhesion molecule. Depletion of extracellular Ca²⁺ caused the disappearance of both VE-cadherin and plakoglobin from junctions, but did not affect the distribution of PECAM-1. Cells stayed fully attached in low-Ca²⁺ media, but began to dissociate from neighboring cells after 15 min of 20 dyne/cm² steady shear stress. Stability to shear stress was restored with addition of Ca²⁺ in cells with normal plakoglobin content, but not in plakoglobin-deficient cells. This study shows that maintenance of intercellular adhesion under shear stress requires the junctional location of cadherins associated with plakoglobin. Beta-catenin cannot functionally compensate for the junctional loss of plakoglobin.

It is well known that endothelial monolayers maintained in static culture assume a cuboidal, cobblestone morphology, whereas they elongate and align in the direction of flow over a 24–48-h period when exposed to moderate or high steady shear stress (89, 90). During this morphological remodeling period, it is expected that the intercellular junctions remodel as well. To investigate this, Noria et al. (91) exposed confluent monolayers of porcine aortic endothelial cells (PAECs) to 15 dyne/cm² steady shear stress for 0, 8.5, 24, and 48 h and performed immunostaining and Western blotting for VE-cadherin, alpha and beta catenin, and plakoglobin. Under static conditions, staining for all proteins was intense and peripheral, forming a nearly continuous band around the cells at cell-cell junctions. After 8.5 h of shear stress, staining was punctate for all proteins, suggestive of a partial disassembly of the adherens junction. By 48 h, when cell shape change was completed, intense staining was recovered for all proteins, suggesting that the adherens junctions had been reassembled. Western blot analysis indicated that protein levels of VE-cadherin, alpha catenin, and plakoglobin decreased, whereas beta catenin increased after 8.5 h of shear. At 48 h, all protein levels were upregulated except plakoglobin, which remained below control levels.

Because intercellular adhesion at adherens junctions is thought to be a prerequisite for the assembly of other junctional complexes, including the tight junction (92), it is likely that tight junctions break down or are pulled apart as the adherens junction disassembles and then reform as the adherens junction reassembles. The in vitro transport study of McIntire et al. (72) that showed maximal increases in endothelial transport after 10–30 h of steady shear stress exposure with a return to baseline values by 48 h is consistent with a transient breakdown and reassembly of intercellular junctions. This is also consistent with a hypothesis of Friedman & Fry (93) suggesting that an adaptive response of the endothelium to changes in hemodynamics (shear stress) leads to a transient increase in uptake of atherogenic molecules during the period of adaptation. But it should be remembered that the tight junction can be modified very quickly after a change in shear stress. The study by DeMaio et al. (84), described previously, showed occludin phosphorylation as early as 5 min after exposure to shear stress, suggesting that tight junction remodeling may be initiated before the adherens junction is altered. Further studies are required to assess the dynamics of adherens junction remodeling at short times.

The preceding discussion has focused on the response of the intercellular junction to shearing forces on the apical surface of the endothelial cell. The intercellular junctions are also exposed directly to fluid shear stresses on their apposing surfaces that are induced by transmural flow (J_v) passing through the intercellular conduit. Tarbell et al. (94), using the dimensions of the adherens junction and typical values for J_v, have estimated these shear stresses to be of the order 25 dyne/cm², comparable to the shear stresses of blood flow on the apical surface. Changes in transmural pressure drive changes in this intercellular shear stress, not only because transmural pressure affects transmural flow (Equation 22), but also because transmural flow affects L_p, as demonstrated with BAEC monolayers in vitro by Tarbell et al. (94). In addition, transmural flow imposes shear stress on the basal surface of endothelial cells as fluid moves laterally in the subendothelial intima between the basal endothelial surface and the internal elastic lamina (IEL) toward a fenestration in the IEL. Tada & Tarbell (95) estimated this shear stress to be of the order 10 dyne/cm², again a level comparable to the shear stress of blood flow on the apical surface.

Leaky junctions have been associated with cell proliferation or turnover (mitosis) and cell death (apoptosis) as described earlier. Both of these processes are affected by fluid shear stress. A number of studies have demonstrated that increasing levels of steady shear stress reduce the rate of endothelial cell proliferation (78, 96). Increased cell turnover in regions of separated or recirculating flow has been implicated in the process of early atherogenesis. The fluid mechanical mechanism driving this increase has been attributed to elevated spatial gradients of shear stress in the separation zone (e.g., Reference 97) and to increased temporal gradients of shear stress in the same region (98).

Several studies have shown that a reduction in the rate of blood flow or shear stress induces an increase in the rate of endothelial cell apoptosis, whereas an increase in shear stress exerts the opposite effect (99, 100). Studies in HUVECs indicate that the apoptosis inhibitory effect of shear stress is mediated by upregulation of NO synthesis and that this mechanism is impaired in aged cells that do not display the antiapoptotic effect of shear stress (101). NO contributes to protection against endothelial cell death via S-nitrosylation of caspases. Freyberg et al. (102) have shown that irregular flow conditions, such as turbulence or unsteady laminar vortex flow, increase apoptosis rates and that an autocrine loop involving thrombospondin-1 and the alpha(v)beta(3) integrin/integrin-associated protein complex is the molecular coupling device between flow and apoptosis. Analysis of human atherosclerotic plaques from carotid arteries shows a systematic preferential occurrence of elevated apoptosis in the downstream regions of plaques, where low flow and low shear stress prevail in comparison to the upstream regions (103).

Taken together, these studies suggest that endothelial cell turnover rates and apoptosis rates, and by association the prevalence of leaky junctions, will be greater in regions of low mean shear stress (near separation and reattachment points) where spatial gradients and temporal oscillations of shear stress are colocalized (104).

These fluid mechanical features are often associated with the localization of atherosclerosis. It should be noted that hypoxia induces endothelial cell apoptosis as well (105, 106), implying that fluid mechanical conditions that produce local hypoxia will induce leaky junctions. As described in an earlier section, hypoxia is most likely to occur in regions of separated and secondary flow that are associated with atherosclerotic plaque formation.

Davies et al. (107) determined the effect of shear stress on pinocytosis (fluid-phase endocytosis) in BAECs by monitoring HRP uptake into cells. They found that continuous exposure to steady shear stresses (1–15 dyne/cm²) in laminar flow stimulated time- and shear stress level–dependent increases in pinocytotic rate that returned to control levels after several hours of application. A sudden removal of steady shear stress also resulted in a transient increase in pinocytotic rate. Oscillating laminar shear stress at 1 Hz did not influence pinocytotic rates, but changes over longer cycle time (15 min) did elevate pinocytotic rates. Elevated steady shear stress has also been shown to increase LDL endocytosis via increased expression of cell surface lipoprotein receptors (108). Kudo et al. (109) showed that after 48 h of exposure to steady shear stress, porcine aortic endothelial cells displayed an increased intracellular uptake of albumin that peaked at 10 dyne/cm² and then decreased with increasing shear stress out to 80 dyne/cm². None of these studies of vesicle rates measured the overall transport rates across the endothelium, and therefore, we can't be sure that an increase in pinocytotic rate can be equated with an increase in transendothelial transport. Further studies are required to clarify this point. The study by Davies et al. (107) does suggest that changes in shear stress might elevate vesicular transport transiently, and this would be consistent with the transient adaptation hypothesis of Fry & Friedman (93) mentioned previously.

When transmural pressure increases as a result of an increase in vascular pressure, there is an increase in transmural flow (J_v) associated with the increased driving force (Equation 22), but several studies have demonstrated that the L_p for the entire arterial wall actually decreases with increasing pressure in the range 50–180 mmHg (20, 110). These same studies show that when the artery is denuded of endothelium, L_p increases (by about twofold), but is no longer pressure dependent. The L_p of the denuded artery is dominated by the medial layer of the wall, and the observation of pressure-independent L_p is consistent with recent theoretical models of aortic media predicting that aortic volume and L_p are relatively insensitive to pressure (111). Huang et al. (112) proposed that the reduction in L_p of the intact artery with increasing pressure is a result of compaction of the arterial intima, specifically the extracellular basement matrix layer between the endothelial cells and the internal elastic lamina. Subsequent experiments in perfusion-fixed rat aortas provided electron microscopic evidence that the intima was compressed from a thickness of 0.62 to 0.12 μm as the pressure was increased from 0 to 150 mmHg (113). The deformable intima would also be expected to influence macromolecular transport, as the properties of the intimal matrix change with pressure, but this aspect has not yet been investigated.

The direct effects of hydrostatic pressure changes on endothelial cell function have received very little attention in the literature, although Schwartz et al. (114) showed that a sustained hydrostatic pressure of only 4-cm H₂O could stimulate a selective increase in the expression of integrin subunit alpha (v) within 4 h and increased cell proliferation within 24 h in HUVECs in vitro. Changes in pressure are more often associated with the induced changes in strain (stretch) of the endothelial cell layer that is supported by the elastic medial layer of the artery. Stretch effects on endothelial cells have been investigated quite extensively in vitro using cells supported on deformable substrates (e.g., References 115, 116). There have not been any reports of the effects of stretch on transendothelial transport, apparently because deformable substrates used to support endothelial cells in stretch experiments are not porous enough to be used in transport experiments and porous supports are not sufficiently compliant for stretch experiments. A highly porous and deformable material would be useful for studies of the effects of stretch on transport.

MACROMOLECULAR TRANSPORT AND ATHEROSCLEROSIS

The accumulation of lipoproteins in the arterial intima is a hallmark of atherosclerosis (117, 118). LDL is the most abundant atherogenic lipoprotein in plasma and high plasma levels of LDL are causally related to the development of atherosclerosis (119). The flux of LDL into the arterial wall depends on the plasma concentration and the permeability (P_e; Equation 24), which is similar for the aorta of humans, monkeys, rabbits, and pigeons—lying in the range 1.1 × 10⁻⁹–2.6 × 10⁻⁸ cm/s. In experimental animals, the focal sites of predilection for either spontaneous or dietary-induced atherosclerosis can be detected before plaques become visible macroscopically. Such lesion-prone areas are delineated by their in vivo uptake of the protein-binding azo dye, Evans Blue. The Evans Blue model has been studied extensively and characterized both morphologically and biochemically (120). The lesion-prone (blue) areas display increased endothelial permeability to and intimal accumulation of plasma proteins, including albumin (3.5-nm radius), fibrinogen (5.5-nm radius), and LDL (11-nm radius). Thus, a correlation between enhanced endothelial permeability to macromolecules and the localization atherosclerotic plaques is well established.

Focal Nature of Macromolecular Uptake

The focal nature of plaque formation is a striking feature of atheroscleosis (121), and studies in rabbits support the idea that LDL permeability is increased focally at susceptible branch sites of the aorta compared to adjacent atherosclerosis resistant nonbranch sites (7, 122, 123). Using horseradish peroxidase (HRP), Stemerman et al. (124) detected foci in the normal rabbit aorta with a luminal surface area of about 0.1 mm² that had an LDL permeability up to 47 times greater than surrounding areas and yet were covered by an apparently intact endothelium. The number of high permeability foci per unit area was ten times higher in the atherosclerosis-susceptible areas, such as the ostia of aortic branches, than in nearby resistant areas. Detailed analysis of the distribution of high LDL permeability sites around intercostal and coeliac arteries of the rabbit aorta showed that high permeability sites occurred most frequently at the distal and lateral edges of the orifices, which are the most susceptible sites for atherosclerosis (123).

The distal (downstream) predominance of LDL uptake around intercostal ostia in the rabbit model described above has also been observed for albumin in young animals, but attention has recently been drawn to the fact that this distribution pattern is reversed to one of upstream predominance in mature animals (125). These patterns of transport correlate with the distribution of lipid deposits in the human aortic wall. Lesions occur most frequently downstream of intercostal ostia in fetuses, neonates, and infants (126), but upstream in adult vessels (127). Forster & Weinberg (125) shed light on the mechanism of this distribution pattern by observing that treatment of aortas with the NOS inhibitor L-NMMA for only 25 min could reverse the uptake pattern in mature animals to the juvenile pattern (reverting from upstream to downstream predominance of uptake). Because age, hypercholesterolemia, and inhibition of NOS all induce the upstream pattern, Staughton et al. (128) have speculated that they all act via influences on NO synthesis.

In a subsequent study, an effect of blood flow pattern (and indirectly, fluid shear stress pattern) on albumin uptake around rabbit intercostal ostia was demonstrated by comparing the uptake pattern around normal ostia with the pattern around others that had the intercostal artery occluded to prevent flow into the side branch (128). In mature animals, the normal pattern of upstream predominance was converted to the immature downstream pattern around ostia that had been occluded for only 35 min. Because this altered pattern required only 35 min of altered hemodynamics, it seems unlikely to have been induced by significant remodeling of the endothelium. Because the altered uptake patterns induced by inhibiting NO synthesis and manipulating hemodynamics were similar, they suggest that the hemodynamic effect was mediated by altered NO release. This is consistent with several studies described previously that showed acute influences of altered fluid shear stress on NO release by the endothelium along with altered transport rates (73).

The fluid mechanics around the intercostal ostia are undoubtedly complex and have not been described in detail. However, pulsatile flow around the aorto-coeliac junction has been simulated, and the associated wall shear stress features have been compared to LDL uptake patterns in the rabbit (129). The downstream dominant uptake pattern characteristic of young rabbits was associated with high mean wall shear stress without reversal of shear stress direction around the downstream flow divider and low mean wall shear stress with reversal around the upstream region. This is the generic shear stress pattern one would expect to find around a bifurcation, as revealed in studies of other arterial bifurcations (e.g., References 130, 131). By occluding the intercostal artery, Staughton et al. (128) very likely reduced the significant differences in the wall shear stress patterns between the upstream and downstream sides of the junction. All of this suggests that differences in NO production rates between low mean (reversing) shear patterns and high mean (nonreversing) shear patterns may be important in explaining uptake patterns and disease patterns around arterial bifurcations. Further studies are required to clarify these issues.

The fluid phase, endothelium, and intima have generally been viewed as the most influential structures regulating uptake of macromolecules by the artery wall, and they have been discussed in detail in this review. But, as suggested by Caro et al. (3), the contribution of the medial layer should not be overlooked. The importance of the media was emphasized by Lever et al. (132), who showed that the pulmonary artery and the inferior vena cava, vessels not normally associated with atherosclerotic plaques, had much higher endothelial permeability to albumin and fibrinogen than did the atherosclerosis-prone aorta, carotid artery, and renal artery. Based on these observations, they suggested that the accumulation of plasma components during the development of atherosclerotic lesions is not determined simply by the rate at which macromolecules enter the tissue. The degree of accumulation of material in the intima may also be influenced by hindrance to outward passage of the components through the media. In support of this view, Lever & Jay (133) observed that the pulmonary arteries and veins have a significantly higher distribution volume (porosity) for albumin than do the arteries that are susceptible to disease. This higher porosity of the media of the vena cava and pulmonary artery may permit the easy drainage of macromolecules, preventing their accumulation in the tissue and thereby contributing to the low susceptibility of these vessels to disease.

Drainage or clearance of macromolecules by transport across the wall will also be influenced by the transmural water flux (J_v) that can flush macromolecules out of the wall by convection. This will be an important mechanism for molecules having a permeability coefficient (P_e—Equation 24) that is small relative to the volume flux (J_v; Equation 22). This condition is certainly satisfied by molecules the size of albumin or larger because albumin P_e for arteries is of order 0.25–1.0 × 10⁻⁷ cm/s and J_v for arteries is typically 1.0–5.0 × 10⁻⁶ cm/s (31). The importance of this flushing mechanism appears to have been demonstrated by Lever et al. (132) in experiments in which collars were placed around rabbit carotid arteries and macromolecular tracer uptake was measured. Steady state levels of fibrinogen and LDL in the media of vessels with collars were 20 to 50 times higher than in uncollared arteries. An appealing interpretation of these observations is that the collar greatly reduced J_v, and as a result, solute in the media beyond the high resistance intima was not flushed out. This is consistent with the discussion in Lever et al. (132) and a simple, one-dimensional, steady state, convective-diffusion model of medial transport with a boundary condition at the endothlelial (intimal) surface that matches the solute flux across the endothelium (Equation 11) to the combined convective/diffusive flux in the media. This simple model shows that the steady state medial uptake is inversely proportional to J_v.

MASS TRANSPORT MECHANISMS IN ATHEROGENESIS

The preceding sections have provided a review of the basic mechanisms of mass transport in arteries and the patterns of macromolecular uptake in relation to atherosclerosis. In this final section, a synthesis of these basic mass transport processes in relation to patterns of uptake is presented in the form of several plausible mechanisms of atherogenesis in which mass transport plays a central role. Although several of these mechanisms have been proposed in the literature previously, they have not been described in the context of general arterial mass transport phenomena.

Fluid-Phase Controlled Hypoxia Regulates Endothelial Permeability

As pointed out in the section on fluid-phase resistance to transport, high-molecular-weight species, such as LDL and albumin, are not limited by the fluid phase. Oxygen transport, however, may be fluid-phase limited in regions of low fluid-phase mass transfer rates (Sherwood number—Sh), such as the outer walls of bifurcations and the inner walls of curved vessels where enhanced LDL uptake and atherosclerotic lesions localize. Hypoxia in such regions has been confirmed by direct measurements in the carotid bifurcation (15), around vascular graft anastomoses (134), and in other vessels (135).

Hypoxia in the arterial wall has for many years been implicated in the development of atherosclerosis (136). Local hypoxia can affect the uptake of LDL and other macromolecules by the arterial wall through several mechanisms: (a) Hypoxia can break down the endothelial barrier and form interendothelial gaps leading to increased macromolecular transport (137, 138, 139). Hypoxia also induces endothelial cell apoptosis (105, 106, 140), which can increase LDL transport through leaky junctions. (b) Hypoxia can upregulate VEGF release by vascular cells and affect endothelial permeability. Many cell lines produce increased amounts of VEGF when subjected to hypoxic conditions, as do normal tissues exposed to hypoxia, functional anemia, or localized ischemia (141). VEGF is a multifunctional cytokine that acts as an important regulator of angiogenesis and as a potent vascular permeabilizing agent (142). VEGF is believed to play an important role in the hyperpermeability of microvessels in tumors, the leakage of proteins in diabetic retinopathy, and other vascular pathologies (79, 86, 143). Thus, a plausible scenario for the increase in lipid uptake is the following: Hypoxia upregulates the production of VEGF by cells within the vascular wall and VEGF in turn permeabilizes the endothelium, allowing increased transport of lipid into the wall. In support of this view, several recent studies have shown that VEGF is enriched in human atherosclerotic lesions (144, 145, 146).

Leaky Junctions Control Endothelial Permeability

The uptake of LDL is controlled by the endothelium, not the fluid phase, and leaky junctions, not tight junctions, would appear to constitute the principal pathway for transport of LDL across the endothelial layer. Leaky junctions are associated with cells in a state of turnover (mitosis) or death (apoptosis), and these processes are affected by local fluid mechanics. As described previously, elevated steady shear stress tends to suppress both mitosis and apoptosis, whereas low shear stress and separated or disturbed flow increase these processes. Therefore, it is expected that leaky junctions would be more prevalent in regions of low shear stress and separated flow than in regions of higher, unidirectional shear stress. These are precisely the regions where atherosclerotic plaques tend to be localized at the carotid bifurcation (131), in coronary arteries (147, 148, 149), and the aortic bifurcation (132, 150, 151).

Transient Intercellular Junction Remodeling Controls Endothelial Permeability

Friedman & Fry (93) and Friedman et al. (152) suggested that macromolecular uptake by arteries is enhanced during transient adaptation of the endothelium to changes in blood flow (shear stress). Such changes occur on a diurnal schedule as well as in response to many physiological stimuli. As described in detail in an earlier section of this review, virtually all transport pathways respond to changes in shear stress: intercellular junctions (tight junctions and adherens junctions) remodel, vesicle formation rates are altered, and rates of mitosis and apoptosis that modulate leaky junctions change. The timescales for these responses to shear stress vary from minutes for the phosphorylation of tight junction proteins (84) to days for the reestablishment of the adherens junction (91). Whether “changes” in the local hemodynamic environment are more influential than the “average” environment integrated over long periods of time in determining arterial wall uptake is not known. Most mass transport mechanisms in atherosclerosis, including the other three mentioned in this section, are based on the notion that the “average” environment controls the long-term behavior, even though dynamic processes, such as pulsatile flow (shear stress), contribute to the “average” environment.

Convective Clearance Alters Intimal/Medial Accumulation

For large macromolecules, such as LDL, that have a low endothelial permeability (P_e) relative to volume flux (J_v), an increase in J_v with fixed P_e will reduce the accumulation of solute beyond the endothelial layer (intima/media) by convectively clearing (flushing) out the region beyond the high resistance endothelial barrier. If we assume that a macromolecule crosses the endothelium primarily through leaky junctions as discussed above, and that volume flux (primarily water flux) is controlled principally by the intercellular junctions that have a much greater total area than the leaky junctions, then factors that affect J_v but not P_e can influence the accumulation of macromolecules within the wall. For example, as discussed previously, this could explain why collars around arteries led to an increase in macromolecule accumulation (132). This would be expected if the collar reduced J_v but did not alter P_e greatly, as seems plausible.

There are other examples that might be explained by this mechanism. Staughton et al. (128) observed that the albumin uptake pattern around intercostal ostia of rabbits could be transformed from one of upstream predominance to one of downstream predominance after only 35 min of flow alteration induced by tying off the intercostal artery. Alteration of the flow (wall shear stress) pattern is not likely to have influenced endothelial permeability to albumin within 35 min if leaky junctions (controlled by mitosis and apoptosis) provided the dominant transport pathway. On the other hand, it is known that an alteration in wall shear stress pattern can significantly affect hydraulic conductivity (L_p) within 30 min (73).

Most studies have shown that endothelial L_p, which is controlled by the intercellular junction, is higher when fluid shear stress is elevated. Therefore, it is expected that the convective clearance mechanism will contribute to lower macromolecular accumulation in regions of high shear stress. Because macromolecular permeability, if controlled by leaky junctions, is expected to be lower when the fluid shear stress is elevated (discussed previously), convective clearance is a mechanism that further reinforces low accumulation in high shear regions.

It is hoped that further study of these mass transport mechanisms of atherogenesis will lead to new understanding of the overall atherosclerotic disease process and may contribute to the development of therapeutic strategies based on regulation of mass transport phenomena.

ACKNOWLEDGMENTS

This work was supported by NIH NHLBI Grants HL35549 and HL57093 and NASA Grant NAG3-2746.

The Annual Review of Biomedical Engineering is online at http://bioeng.annualreviews.org