OUP user menu

★ editor's choice ★

Microvascular fluid exchange and the revised Starling principle

J. Rodney Levick, C. Charles Michel
DOI: http://dx.doi.org/10.1093/cvr/cvq062 198-210 First published online: 3 March 2010


Microvascular fluid exchange (flow Jv) underlies plasma/interstitial fluid (ISF) balance and oedematous swelling. The traditional form of Starling's principle has to be modified in light of insights into the role of ISF pressures and the recognition of the glycocalyx as the semipermeable layer of endothelium. Sum-of-forces evidence and direct observations show that microvascular absorption is transient in most tissues; slight filtration prevails in the steady state, even in venules. This is due in part to the inverse relation between filtration rate and ISF plasma protein concentration; ISF colloid osmotic pressure (COP) rises as Jv falls. In some specialized regions (e.g. kidney, intestinal mucosa), fluid absorption is sustained by local epithelial secretions, which flush interstitial plasma proteins into the lymphatic system. The low rate of filtration and lymph formation in most tissues can be explained by standing plasma protein gradients within the intercellular cleft of continuous capillaries (glycocalyx model) and around fenestrations. Narrow breaks in the junctional strands of the cleft create high local outward fluid velocities, which cause a disequilibrium between the subglycocalyx space COP and ISF COP. Recent experiments confirm that the effect of ISF COP on Jv is much less than predicted by the conventional Starling principle, in agreement with modern models. Using a two-pore system model, we also explore how relatively small increases in large pore numbers dramatically increase Jv during acute inflammation.

  • Starling principle
  • Glycocalyx
  • Fluid exchange

1. Introduction

The plasma, interstitial fluid (ISF), and lymph compartments are linked in series and, in the steady state, fluid flows continuously from one compartment to the next. Lymph drains back into the circulation chiefly at the major veins at the base of the neck. Accidental lymphatic fistulae in the neck indicate a total post-nodal lymph flow of up to 4 L/day in humans. Later work revealed that roughly half the fluid content of afferent lymph can be absorbed by lymph node microvessels,1,2 raising the fluid turnover estimate to ∼8 L/day.3 This is a considerable fluid turnover; since human plasma volume is only ∼3 L, the entire plasma volume (except the proteins) leaves the circulation approximately once every 9 h.

Substantial fluid movements between the plasma and interstitium account for the rapid swelling of acutely inflamed tissues (minutes), and for the oedematous swelling of venous thrombosis, cardiac failure, and lymphatic failure over hours to days. Conversely, haemodilution following an acute haemorrhage reveals a rapid absorption of ISF into the blood stream (∼0.5 L in 15–30 min). Acute fluid transfers are important medically, because plasma volume is a major determinant of the cardiac filling pressure and thus cardiac output (Starling's ‘law of the heart’).

The fundamental principle governing such fluid shifts was laid down by Starling4 in 1896. Starling4 showed that isotonic saline injected into the interstitial compartment of a dog hind limb appeared in the venous blood, which became haemodiluted; but when serum rather than saline was injected, the fluid was not absorbed. Starling therefore proposed that the walls of capillaries (and post-capillary venules) are semipermeable membranes. Consequently, fluid movement across them depends on the net imbalance between the osmotic absorption pressure of the plasma proteins [colloid osmotic pressure (COP)] and the capillary hydraulic pressure generated by the heart beat. Starling also recognized that since ISF has a substantial concentration of plasma proteins, microvascular semipermeability is imperfect; the endothelial barrier slowly ‘leaks’ plasma proteins into the interstitium. The degree of leakiness to a specific solute can be quantified by Staverman's osmotic reflection coefficient,5 σ, which ranges in value from 0 to 1; unity means perfect, 100% reflection, and thus no leakage of the specified solute. For a simple membrane separating well-stirred solutions of a single solute at two different concentrations, it follows from the principles of irreversible thermodynamics5,6 that: Embedded Image 1a where Jv is the volume filtration rate per unit endothelial area A, ΔP the difference in hydrostatic pressure, and ΔΠ the difference in osmotic pressure across the membrane.

Plasma and ISF each contain many solutes (species ‘n’). Therefore, to describe fluid movement across the intervening capillary wall, Eq. (1a) should be written as: Embedded Image 1b where ∑σnΔΠn represents the sum of the differences in osmotic pressure exerted across the vessel walls by all the solutes in plasma and ISF. In Eq. (1b), ΔP is the difference between local capillary blood pressure Pc (diminishing with axial distance) and ISF hydrostatic pressure Pi. In most microvascular beds, only the macromolecular solutes are present at significantly different concentrations, and for the small solutes, σ has a value of 0.1 or less. Under these conditions, Jv can be described approximately by Eq. (1a), where σΔΠ is the difference between the effective osmotic pressure exerted by the macromolecules (COP) in plasma (Πp) and ISF (Πi) (Figure 1A). This leads to the conventional expression Embedded Image 1c If σ has a value close to 1.0 and both Pi and Πi are close to zero, as in the following experimental studies, Eq. (1c) summarizes the classic observations of Landis7 and Pappenheimer and Soto-Rivera,8 whose investigations are generally regarded as having established Starling's hypothesis. By measuring the prevailing Pc by direct micropuncture in single capillaries in the frog mesentery and estimating the corresponding rates of filtration and absorption by a red cell tracking method, Landis was able to demonstrate a linear relation between Jv and Pc for a population of vessels. Using isolated perfused cat and dog hind limbs, Pappenheimer and Soto-Rivera ingeniously estimated the mean tissue Pc from the arterial and venous pressures and vascular resistances. They were able to vary mean Pc and demonstrated a linear correlation with rates of microvascular fluid filtration/absorption, estimated from changes in limb weight. Landis9 also showed, using micropipettes inserted into human nailfold skin capillaries at heart level, that blood pressure falls along a capillary, from >45–35 mmHg at the arterial end to ≥12–15 mmHg at the venular end (depending on skin temperature10,11). This gave credence to a simple picture of fluid exchange symmetry first suggested by Starling.12 The picture is that since Pc exceeds Πp (25–28 mmHg in humans) over the arterial half of the capillary bed at heart level, there is fluid filtration here; and this is balanced by absorption over the venous half, where Pc falls below Πp. The concept was developed by Landis and Pappenheimer13 and although they circumspectly noted that ‘with equal or greater licence (our italics) an average limb capillary and lymphatic can be drawn’, the caveat was largely ignored and the model became widely reproduced as an established fact (Figure 2A, left panel). It was not, however, based on a single tissue or species, because modern methods for measuring the important interstitial terms Pi and Πi in human skin were not then available.14 Over the past 25 years, it has become clear that the conventional model does not describe fluid exchange in the real microcirculation, because (i) both Pi and Πi change when Jv changes, and (ii) the fluid immediately downstream of the semipermeable membrane (the subglycocalyx fluid, see later) can have a markedly different composition to bulk ISF.

Figure 1

Comparison of traditional and revised views of the endothelial semipermeable membrane and the forces acting on it. (A) Traditional view of continuous endothelium as a semipermeable membrane. (B) The glycocalyx–cleft model identifies glycocalyx as a semipermeable layer. Its underside is subjected to the COP of fluid high inside the intercellular cleft rather than ISF, with important functional consequences.80,82 Symbols defined in main text. Grey shade denotes concentration of plasma protein.

Figure 2

Imbalance of the classical Starling pressures when all four terms are measured in the same tissue and species. (A) Left panel shows the traditional filtration–reabsorption model; interstitial terms are considered negligible. Right panel shows the imbalance when all four classic Starling forces are measured in human skin and subcutis at heart level.18 Black arrows indicate net force imbalance and hence the direction and the magnitude of fluid exchange are based on the four classic forces. Dot-dashed line (red) illustrates qualitatively the much smaller net filtration force predicted by the glycocalyx–cleft model. For symbols see text. (B) Sum of three classic Starling forces opposing microvascular blood pressure (Pc0 = σΠp − σΠi + Pi) plotted against the lowest microvascular blood pressure in different tissues (Pc). Each point is based on measurements in the same tissue and species except the intestine (square symbols), where Pc value from rats are plotted against values Pc0 derived from cats. Closed square is non-absorbing intestine; open square is during water absorption from gut lumen. Pc0 is capillary pressure at which the classic net filtration force would be zero. If venular pressure exceeds Pc0 (i.e. lies above the line of equality), venules and venous capillaries are in a state of filtration. Data from many laboratories.19

In this short review, we first consider data bearing on the conventional steady-state filtration–reabsorption model, then experiments highlighting the distinction between the linear, transient fluid exchange vs. pressure relation and the non-linear, steady-state relation. This is followed by a review of the altered perspective resulting from the recognition of the glycocalyx as the semipermeable layer coating the inside of the capillary wall. For a broad overview of endothelial physiology, see Levick.15

2. Traditional filtration–reabsorption model vs. Starling pressures measured in specific tissues: sum-of-forces evidence

Doubts about the validity of the filtration–absorption model surfaced quietly at first. In a study of rabbit mesenteric and omental capillaries, using the Landis red cell method to monitor fluid exchange, Zweifach and Intaglietta16 noted briefly ‘We have not to date … encountered vessels that consistently showed inward filtration’ (i.e. absorption). Ten years later, Levick and Michel10 documented how human skin capillary pressure increases with distance below heart level, albeit partially protected by postural vasoconstriction of the pre-capillary resistance vessels; and they pointed out that tissue fluid balance cannot be maintained by downstream reabsorption in skin more than ∼10 cm below heart level, because venous capillary Pc then exceeds Πp.10,17 In skin capillaries above the heart, Pc does not decrease in an analogous way, because the superficial veins collapse and Pc becomes independent of further elevation.9 The issue of downstream force balance in human skin at heart level was addressed by Bates et al.,18 who measured the interstitial terms using acutely inserted wicks. This enabled the sum of all four Starling pressures to be calculated. They found that due to the subatmospheric nature of Pi (−2 mmHg) and the substantial magnitude of Πi (15.7 mmHg), there was no net absorptive force in the cutaneous venous capillaries or venules at heart level (ΔP > σΔΠ) (Figure 2A, right panel). A review of the literature revealed the same situation in 14 different tissues/species, including skeletal muscle (∼40% of body mass) and the lung, the tissue with the lowest Pc (Figure 2B). It is important to note that since the Starling pressures differ between tissues and species, a correct sum of forces necessitates measurement of all four forces in the same tissue.19 The absence of a net absorptive force in downstream vessels in the steady state is in line with theoretical expectations, since a large ΔΠ only develops in tissues with large filtration rates (see later). On the basis of today's glycocalyx–cleft model, Πi is not the correct term to use in sum-of-force calculations (it should be the COP beneath the glycocalyx, Πg, see below). However, the inference, i.e. no downstream net absorptive force in the steady state, remains valid, because absorption would raise extravascular Π and thus reinforce the sum-of-forces argument.

3. Absorption is observed transiently but not in the steady state at capillary pressures below plasma COP

The traditional steady-state filtration–reabsorption model was tested directly by Michel and Phillips.20 They used a development of the Landis red cell method to measure the direction and rate of fluid exchange in perfused single capillaries of the frog mesentery under two different conditions. In one set of experiments, Pc was raised or lowered abruptly and fluid exchange was measured immediately after the change; no time was allowed for the extravascular forces to change (transient state). In a second set of experiments, Pc was maintained constant for at least 2 min (and usually longer) before measuring the filtration or absorption rate at the same value of Pc (steady state). As in Landis's study,7 there was a linear relation between Jv/A and Pc in the transient experiments, with transient fluid absorption when Pc was lowered below σΠp. In the steady state, however, a dramatically different relation emerged (Figure 3A). When Pc exceeded σΠp, filtration rate again increased linearly with Pc; but when Pc was lower than σΠp, no absorption occurred in the steady state—contrary to the traditional downstream reabsorption model or transient state. The observations have been confirmed in both frog and rat mesenteric vessels21 and analogous results were obtained in a study of filtration/absorption across confluent cultured endothelium in an Ussing-style chamber.22

Figure 3

Direct observation of direction of fluid exchange in capillaries in which Pc is maintained below Πp. (A) Comparison of fluid exchange immediately after raising or lowering Pc (transient state) and after Pc has been held constant for several minutes (steady state). Measurements made on single frog mesenteric capillaries perfused in situ with a Ringer solution plus serum albumin and the macromolecule Ficoll 70 (Πp = 32 cmH2O). When Pc was lowered from 30 to 10–15 cmH2O, a brisk uptake of fluid into the capillary was observed (e.g. green point on transient relation). If the vessel was now perfused with Pc held constant in this range (10–15 cmH2O), absorption attenuated and after several minutes could no longer be detected (red point on steady-state curve).20 Insets illustrate concentration of pericapillary plasma protein (dots) following transient water absorption (long thin arrows) and interstitial plasma protein reflection (thick curved arrow). (B) Explanatory sketches. Left panel shows how extravascular Starling forces change with time (transient state) when Pc is lowered below Πp, until a new steady state is reached. In intact tissues (cf. superfused mesentery), the absorption phase probably lowers Pi as well as raising pericapillary Πp, further cancelling the net absorptive term. Right panel shows how the intersection of the Starling and extravascular dilution relations determines the steady state. The curvilinear ISF relation23,24,100 describes the effect of filtration on ISF COP Πi (red line indicates corresponding Πg), while the linear, Starling's relation describes effect of extracapillary COP on filtration. Note that the latter relation is plotted contrary to convention, with the independent variable Πi on the y-axis and dependent variable Jv on the x-axis.

Thus, capillaries with a low blood pressure can absorb fluid transiently but not in the steady state (the lungs are an important example)—in keeping with the sum-of-forces evidence presented earlier. Why is this so generally the case? The reason is that an additional principle influences the steady state besides the Starling principle, namely the coupling of extravascular plasma protein concentration (and hence Πi) to capillary filtration rate. Furthermore, in some tissues, such as the lung and subcutaneous tissue, Pi increases non-linearly with ISF volume following increases in Jv/A.14

4. Inverse dependence of pericapillary COP on filtration rate restricts absorption to transient state in most tissues

As reviewed by Taylor and Granger,23 in the steady state, the concentration of ISF plasma protein, Ci (determining Πi), is not a fixed quantity but is inversely related to the capillary filtration rate; see the extravascular dilution curve of Figure 3B, right panel.24 This is because ISF is continually renewed and Ci is determined dynamically by the rate of solute influx, Js, relative to the rate of water influx Jv. Embedded Image 2

It has been known for over 120 years (see Cohnheim25) that raising the capillary filtration rate ‘dilutes’ the macromolecules of the ISF and lymph.3,26 Here, we encounter a tricky situation; Jv depends partly on extracapillary protein concentration4 [Eq. (1a–c)]; but extracapillary protein concentration depends partly on Jv [the extravascular dilution curve, Eq. (2)]. By dividing the standard convection–diffusion equation for macromolecular transport (Js) by Jv, Michel27 derived an expression for Ci and hence Πi in terms of Jv. This was then substituted into Eq. (1c) to predict the steady-state relation between Jv and Pc. The resulting, somewhat complex, expression [Eq. (3)] describes a curve that fits data for fluid exchange in the lung28 and human limbs29 and conforms closely to the steady-state results of Michel and Phillips,20 with no sustained reabsorption even at Pc ≪ σΠp (Figure 3A): Embedded Image 3 where Pe = Jv(1 − σ)/pdA and pd is the diffusional permeability of the capillary barrier to the macromolecule. Pe, the Péclet number, is the ratio of the velocity at which the macromolecule is washed through by convection (solvent drag) to its diffusive velocity within the endothelial barrier.

It is important to note that Eq. (3) does not predict Jv/A in terms of ΔP, since Jv/A is present on the right-hand side too, hidden within the Péclet number. The equation can be rearranged to express ΔP as a function of Jv/A (see Appendix, end of this article) and this is how it is solved.27

The reason that Eq. (3) predicts no absorption of fluid at values of ΔP less than σΔΠ is illustrated in Figure 3B (left panel). When Pc and Jv are reduced, Ci and Πi rise with time [Eq. (2)]. This progressively reduces the absorptive force Πp–Πi, until the system finally reverts to slight filtration, at which point a new steady state is reached.

Rather than solving the Starling and extravascular dilution equations mathematically, one can draw out the linear, transient Starling's relation [Jv vs. Πi for a given Pc, Eq. (1c)] and the interstitial dilution curve [Ci and hence Πi vs. Jv, Eq. (2)] on the same graph, as in Figure 3B. The intersection point is the steady-state value. The intercept is always a positive Jv value, even as Pc approaches zero.19

In view of the coupling of Πi to Jv, steady-state (but not transient) absorption in venous capillaries and venules is unlikely in most tissues. As Starling himself succinctly pointed out4 ‘With diminished capillary pressure there will be an osmotic absorption of salt solution from the extravascular fluid, until this becomes richer in proteids (Figure 3B, left panel); and the difference between its proteid osmotic pressure and that of the intravascular plasma is equal to the diminished capillary pressure’.

Perhaps, the most surprising finding of Michel and Phillips20 was the speed with which a steady state of fluid exchange was established. In other tissues, the time constant is considerably longer, being of the order of tens of minutes for microvessels in the skeletal muscles of large mammals.8 This may be because differences in the architecture of the pericapillary spaces allow protein concentration gradients to develop more rapidly in some tissues than others (see below). In addition, the time course may be influenced by aquaporins (water-only conducting pores in the endothelial cell membranes), which may contribute significantly to Lp in skeletal muscle microvessels3032 (see later).

5. Exceptions to the ‘no steady-state absorption’ rule: effect of local epithelial transport

There are physiologically important exceptions to the ‘no steady-state absorption’ rule. In the kidney, fluid is continuously taken up by the peritubular capillaries of the cortex and by the ascending vasa recta (AVR) of the medulla. In the intestine during water absorption, fluid is absorbed continuously by the mucosal capillaries. In lymph nodes, the capillaries continuously take up fluid from the inflowing pre-nodal lymph (see earlier). How do these vessels manage to break the rule? In lymph nodes, the ISF is continuously replaced by pre-nodal lymph of low protein concentration. In the kidney and the intestine, the ISF is renewed by an independent stream of protein-free fluid which is secreted by the neighbouring epithelia. The rapid renewal of the ISF breaks the inverse coupling between Jv and Πi [Eq. (2)] and is often accompanied by a rise in Pi. When this explanation was first proposed,27 it appeared to be consistent with data for the small intestine and renal cortex peritubular capillaries. In these tissues, the protein leaking into the ISF from the plasma is flushed into the local lymphatics by the epithelial secretions. For the renal AVR, however, the theory was less certain, because no lymphatic drainage from the medullary ISF has yet been demonstrated (see below). It should be noted that none of the microvessels in which steady-state absorption is maintained have the low continuous endothelium found in the skin, muscle, and connective tissue.

5.1 Intestinal mucosa

Although 70% of the blood flow to the rat ileum supplies the mucosa, direct microvascular pressure measurements showed that Pc in the fenestrated mucosal vessels lies well below Pc in the continuous capillaries of the parallel microcirculations of the circular and longitudinal smooth muscle.33 In cats, the intestinal lymph flow increases and its protein concentration decreases as fluid is absorbed from the small intestine lumen.34 In the non-absorbing ileum, lymph COP (ΠL) had a mean value of 10 mmHg, whereas during brisk fluid uptake, ΠL fell to 3 mmHg. The fall in Πi must be as least as great, because intestinal fluid absorption increases the local ISF volume, which reduces the fractional exclusion of serum albumin.35 The latter should amplify the reduction in Πi. Assuming that (i) Pc is approximately the same in the absorbing and non-absorbing states and (ii) similar values for Pc are present in cat and rat mucosal capillaries, the Starling pressures across the capillary walls change from favouring a low level of filtration when the gut is not absorbing fluid to a state of sustained fluid uptake when the gut is absorbing fluid (Figure 2B).

During the absorption of digested food, this simple picture may be complicated by osmotic gradients set up by the absorbed small solutes. At high rates of glucose uptake, the glucose concentration difference across the walls of the capillaries at the tips of intestinal villi may rise to 70 mM. Even if the osmotic reflection to glucose at the walls of these vessels is as low as 0.01, the effective osmotic pressure set up by this glucose gradient opposes and exceeds the Starling pressures by 15 mmHg.36 It is suggested that since the epithelial absorption of glucose and amino acids is confined to the outer third of the villus, fluid absorbed through epithelium here might flow through the villus interstitium to be taken up into the extensive capillary bed of the lower two-thirds of the villi36 and into the lacteals.

5.2 Renal cortex

The peritubular capillary Starling forces have been determined in rat and dog kidneys by several workers and mean values are shown in Table 1 for hydropenic rats and for rats with ISF volumes expanded by isotonic saline.37 In hydropenic rats, the Starling pressures favour fluid uptake, due to relatively low values of Pc and Πi and relatively high values of Pi and Πp. Local Πp is 20–30% higher than in systemic arterial blood, because the peritubular vessels are fed with plasma concentrated by glomerular filtration. Following ISF expansion by intravenous saline infusion, local Πp is reduced and Pc is increased but the net pressure favouring fluid uptake is maintained by a large increase in Pi and a further fall in Πi. Similar observations have been made in hydropenic and volume expanded dogs. The low concentration of interstitial plasma protein, and hence Πi, is maintained by the low macromolecular permeability of the fenestrated peritubular capillaries (σalbumin ≈ 0.99) and the fast production of protein-free ISF by the tubular epithelium, which flushes the relatively small amount of plasma protein entering the cortical ISF into the cortical lymphatics.

View this table:
Table 1

Starling pressures (mmHg) across the walls of peritubular capillaries and AVR in rat kidney

Cortical peritubular capillaries (σalbumin = 0.99)
 Fluid balanceΠpΠiPcPiσΔΠ − ΔP
 Hydropenia (mean ± SEM)24.8 ± 1.393.9 ± 0.613.2 ± 1.33.3 ± 0.610.0
 Volume expanded (mean ± SEM)16.6 ± 1.341.3 ± 0.1819.5 ± 2.7111.5 ± 0.787.1
Ascending vasa recta of the renal medulla (σalbumin = 0.7)
 HydropeniaΠpΠiPcPiσΔΠ − ΔP
 Papilla tip (mean ± SEM)26.0 ± 2.33.7 ± 0.267.8 ± 0.46.0 ± 0.313.8
 Papilla base (mean ± SEM)16.7 ± 1.33.7 ± 0.266.7 ± 0.486.0 ± 0.38.4
  • The column headed σΔΠ − ΔP lists values for the net Starling pressure favouring fluid absorption into the vessels. Data for cortical peritubular capillaries are based on four studies summarized by Ulfendahl and Wolgast.37 Data for ascending vasa recta of the renal medulla are from several laboratories reviewed by MacPhee and Michel43 and Michel.45

5.3 Renal medulla

Fluid that is absorbed by the collecting ducts from the nascent urine is cleared continuously from the medullary ISF by absorption into the AVR. Estimates of the Starling pressures across the walls of the AVR in the inner medulla of rats indicate high values of Πp and Pi and low values of Pc and Πi, consistent with high rates of fluid uptake (Table 1).

Until recently, the low value of Πi (implying good protein clearance) was puzzling for two reasons. Unlike the cortical peritubular capillaries, the AVR have a relatively high permeability to plasma protein; and unlike the cortex, lymphatic drainage has never been demonstrated convincingly in the renal medulla. In both dogs38 and rats,39 labelled plasma protein reached 85% of its steady-state concentration in the renal medullary ISF within 3 min of its injection into the circulation. Ultrastructural studies have shown that ferritin, catalase, and horseradish peroxidase can pass through the fenestrae of the AVR into the ISF.40 It was believed that pre-lymphatic channels might drain medullary ISF41 into the renal cortex but these have not been confirmed. Relatively recently, Tenstad et al.42 demonstrated that labelled albumin is cleared from the medullary ISF directly into the blood. To account for this, MacPhee and Michel43 suggested that proteins might be washed into the AVR (convective protein transport) during fluid absorption, as follows.

If convective transport of protein from ISF into AVR is to be effective, the AVR σ to plasma proteins has to be significantly <1. Measurements of σalbumin in single perfused AVR have given values between 0.59 and 0.78.43,44 With σ in this range, and protein-free fluid from the collecting ducts replacing the fluid passing into the AVR, calculations indicated that convective clearance into the AVR (driven by the Starling pressures, Table 1) can maintain a low ISF albumin concentration.43,45 More detailed modelling46,47 has provided additional support for the effectiveness of convective protein uptake in the medullary countercurrent exchange system.

Although Pc in the AVR is normally greater than the surrounding Pi, it is possible that Pi might exceed Pc if medullary ISF volume is rapidly increased. Because the hydraulic permeability of the fenestrated AVR is high (∼100 × 10−7 cm/s/cmH2O), relatively large volumes of fluid carrying a low concentration of protein can be driven into the AVR by a trans-endothelial hydrostatic pressure difference of just 2–3 cmH2O. Although a high Pi threatens to compress the AVR, these vessels were observed not to collapse until Pc fell more than 4 cmH2O below atmospheric pressure.48 The mechanical stability of the AVR is probably achieved by tethering of the endothelial cells to the basal lamina of neighbouring descending vasa recta and tubules by the fine processes described by Takahashi-Iwanaga.49

6. Low filtration force paradox

On the basis of the three lines of evidence reviewed above (sum of measured forces, direct observation of fluid exchange at low Pc, and theoretical considerations), tissue fluid balance in the muscle, skin, lung, and many other tissues is unlikely to be achieved through a near-balance of upstream filtration and downstream reabsorption in the steady state as traditionally supposed. The drainage of capillary filtrate by the lymphatic system is therefore the dominant factor responsible for interstitial volume homeostasis.50 This view, however, introduces a new problem. As pointed out by Aukland and Reed,51 the lymph flow predicted by Eq. (1a–c) from measurements of LpA and the four classic Starling forces is often an order of magnitude bigger than the observed lymph flow. To express this paradox another way, the true, globally averaged net filtration force calculated from the rate of generation of lymph (∼1 mmHg in many tissues) is much smaller than (PcPi) − σ(Πp − Πi), which is typically 5–10 mmHg (Figure 2B)—a discrepancy too large to dismiss as measurement error.

Several possible solutions have been proposed. In tissues with marked arteriolar vasomotion cycles (e.g. some skeletal muscles), the arteriolar contraction phase will lower Pc in the downstream capillary module,8 leading to transient fluid absorption (Figure 3, transient line), followed by reversion to filtration during the arteriolar relaxation phase, as first suggested by Chambers and Zweifach52 and developed by Intaglietta and Endrich.53 In this model, fluid balance is temporal rather than spatial. Not all tissues, however, show the regular, strong vasomotion pattern on which the temporal model is based.

More recently, exciting new insights into the complex pore structure of endothelium have helped to resolve the low filtration force paradox, by showing that the true net filtration force across endothelium depends not so much on Πi as on the COP of fluid just below the endothelial glycocalyx. In the next section, we will see that the plasma protein concentration here can be substantially less than that in the bulk ISF.

7. Pore exit microgradients and the glycocalyx–cleft model for continuous capillaries

7.1 Fenestrated capillaries

The classic Starling principle is symmetrical—the effect of increases in Πi is predicted to have the same effects on fluid exchange as decreases in Πp. This proved not to be the case, however, when tested in the fenestrated capillaries of synovium, the tissue that lines joint cavities. Synovial capillaries lie just a few micrometres below the highly porous surface of the tissue, with small clusters of fenestrae (1–2% of surface) orientated towards the joint cavity. Consequently, intra-articular infusions of albumin can be used to alter Πi, while measuring the resulting change in trans-synovial flow.54 When the response was compared with that brought about by changes in Πp following intravascular albumin perfusion, a marked asymmetry was observed. Changes in bulk Πi had approximately one-third as much effect on fluid exchange as changes in Πp. A two-dimensional model of synovial fluid exchange indicated that the explanation lay in the development of gradients of ISF albumin concentration (and hence Πi) around the exits from the fenestral clusters, supported by the poorly stirred nature of ISF.55 The ISF COP adjacent to the fenestrations was substantially lower than in the bulk ISF, because the ultrafiltrate emerging from the fenestrations diluted the immediately adjacent ISF over several micrometres. In other words, plasma ultrafiltration set up ‘microgradients’ of ISF protein concentration (steep gradients over micrometre distances). The generality of these findings has yet to be tested in other fenestrated capillary beds.

7.2 Continuous capillaries

7.2.1 Structure of intercellular pathway

Asymmetrical behaviour is even more extreme in continuous capillaries. Here, the principal fluid-conducting pathways are through the endothelial intercellular clefts. (In most microvessels, aquaporins in the endothelial cell membranes contribute <10% of the hydraulic conductance, except in blood–brain barrier and possibly skeletal muscle capillaries.56 Aquaporins have a role when interstitial osmolarity increases, e.g. renal descending vasa recta57 and swelling, ‘pumped’ exercised muscle.30) Ultrastructural studies show that the pathway through the intercellular cleft follows a long, narrow, tortuous route.5860 The cumulative perimeters of the cells over 1 cm2 endothelium (i.e. total length of the luminal entrance to the intercellular clefts in the surface plane) is a remarkable 12–20 m. Within the clefts, however, 90% of this length is sealed by the junctional strands (‘tight junctions’). Fluid is funnelled through breaks in the strands that occur at 2–4 µm intervals, each break being only 200–400 nm in length (Figure 4A). The clefts are 14–21 nm wide throughout; they do not narrow at the breaks in the junctional strands. Consequently, the breaks are too wide to filter out the plasma proteins (molecular diameter of serum albumin 7.1 nm). The protein-reflecting element or ‘small pore’ system has an effective diameter of ∼8 nm.23,6163 Based partly on these considerations, Curry and Michel64 proposed that the ultrafiltering pores are located in the ‘fibre-matrix’ of the endothelial luminal glycocalyx, which overlies the entrance to the clefts.

Figure 4

The two-dimensional glycocalyx–cleft model of capillary fluid exchange, and an experimental test of its predictions. (A) A mathematical model of the glycocalyx and intercellular cleft (top left).82 The semipermeable glycocalyx layer is modelled as a set of fine rods. Graph shows calculated gradient of plasma protein from interstitium to the ‘protected’ subglycocalyx space, in a continuous capillary when Pc > Πp. The subglycocalyx space is protected from equilibration with the bulk ISF by the convergent stream of ultrafiltrate passing through the narrow orifice formed by breaks in the intercellular junctional strands. (B) Fluid exchange in single rat mesenteric venules at controlled microvascular pressure, measured by the modified Landis red cell method (transient state).60 The lumen was perfused with albumin solution and the exterior was superfused with saline (‘no tissue albumin’, open symbols; Πp − Πi = Πp) or the same albumin solution as in the lumen (‘tissue albumin’, filled symbols; Πp − Πi = 0). Short dashes show the expected increase in filtration rate for the latter according to the classic Starling principle. The much smaller observed response was as predicted by the glycocalyx–cleft model. Steady-state results (data not shown) approximated to Eq. (3).

The glycocalyx is a complex, quasi-periodic network of glycosaminoglycan chains (syndecan-1, glypican, and hyaluronan) and sialoglycoproteins, extending 60–570 nm from the endothelial luminal membrane and coating the inner surface of capillaries, including the entrance to the intercellular cleft and most fenestrations.6568 Evidence that the glycocalyx is a major determinate of capillary permeability includes large increases in Lp when glycocalyx-binding proteins are removed from the capillary perfusate (the ‘protein’ effect6972) or when glycocalyx components are removed by enzymatic digestion,73,74 and direct observations.75 Enzyme digestion studies also indicate that the glycocalyx mediates shear-induced changes in Lpd.76 In addition to the glycocalyx–intercellular cleft pathway for water and small solutes, a separate, parallel ‘large pore system’ transports plasma proteins slowly into the interstitial compartment. This system comprises, controversially, large (≥50 nm wide) pores and/or vesicular transport.7779

7.2.2 COP difference across the intercellular pathway

Michel80 and Weinbaum81 independently recognized that since the pathway for macromolecules for continuous endothelium lies in parallel with the main fluid-conducting pathway (intercellular clefts), the COP difference across the vascular barrier could no longer be calculated as Πp minus Πi. The COP difference that determines fluid exchange is that across the semipermeable glycocalyx. The fluid at the abluminal side of the glycocalyx is separated from the pericapillary ISF by the tortuous path through the intercellular clefts (Figure 4A) and since there can be protein concentration gradients along this path, Eq. (1c) should be written as: Embedded Image 4 where Πg is the COP of the ultrafiltrate on the underside of the glycocalyx. Πg can be very low, for two reasons; σ is high, and the outward flow of the ultrafiltrate prevents protein diffusion equilibrium between the subglycocalyx fluid and the pericapillary ISF. The diffusion disequilibrium is greatly exacerbated by cleft structure. As the ultrafiltrate converges to pass through the narrow breaks in the intercellular junctional strands, the fluid velocity increases at least 10-fold (Figure 4A), impeding the upstream diffusion of ISF plasma proteins.60,8082 Rough calculations indicate that even the low fluid filtration rate generated by a pressure difference of 1 cmH2O suffices to prevent protein diffusion equilibrium between the subglycocalyx fluid and pericapillary ISF at the cleft exit.80,81

The above concept was developed into a two-dimensional mathematical model by Weinbaum and coworkers.60,82 The model predicts intercellular cleft flow and plasma protein concentration in the subglycocalyx space for a given bulk ISF composition. Depending on the magnitude of the filtration pressure Pc, the model predicts big differences between subglycocalyx plasma protein concentration (hence Πg) and bulk ISF protein concentration (hence Πi); the ‘race’ between upstream diffusion and downstream washout, particularly at the junctional breaks, can result in a subglycocalyx protein concentration that is ∼10% of Ci (Figure 4A). Under these circumstances, Πg is so low that the effective osmotic pressure difference opposing fluid filtration approximates to Πp rather than the classical Starling term Πp − Πi (Figure 2B). This reduces basal Jv to similar levels as reported lymph drainage rates, and thus helps explain the low filtration force paradox.

7.2.3 Experimental tests of two-dimensional glycocalyx–cleft model

The Weinbaum model predicts that changes in Πi around a filtering capillary should have much less effect on Jv than predicted by the classic Starling principle [Eq. (1c)]. The conflicting predictions were tested by Curry, Adamson and coworkers60,83 with dramatic results (Figure 4B). When interstitial albumin concentration around rapidly filtering frog mesenteric capillaries was raised by superfusion to equal that perfused through the capillary lumen (Πp − Πi = 0), there was almost no change in the filtration rate—a finding incompatible with the classic Starling principle but in line with the Weinbaum model.83 In a definitive study on rat mesenteric venules,60 the increase in Jv caused by raising Πi from zero to plasma level was ∼20% of that predicted by the classic Starling principle. Measurements of filtration across confluent cultured endothelium in a Ussing-style chamber have likewise supported the glycocalyx–cleft model.22

7.2.4 Simplified one-dimensional model, time constants, and aquaporins

To facilitate computation and examine events during microvascular fluid uptake, Weinbaum and coworkers84 developed a simplified one-dimensional model, arguing that this more readily handles the differences in scale between the gradients inside the intercellular cleft and in the surrounding tissues. The one-dimensional model has been used to examine the transition from non-steady-state absorption to a low level of filtration. Zhang et al.21 have argued that cuffs of pericytes encircling microvessels accelerate the transition from transient to steady-state fluid exchange by forming ‘trapped microdomains’ of fluid outside the vessels. They suggest that this phenomenon accounts for the rapid development of the steady state reported by Michel and Phillips20 and confirmed in their own laboratory.21 It is possible that trapped microdomains are not restricted to spaces beneath pericytes. Different tissues have different cellular architectures surrounding the microvessels, and this could contribute to the wide differences in the time constants of transition from transient to steady state.

The time constants could also be lengthened considerably if the aquaporin pathway contributes substantially to the microvascular Lp. Although Michel and Curry56 concluded that aquaporins represent ≤10% of the Lp of most microvessels, Watson and Wolf3032 have argued that their contribution to Lp may be 40–50% in cat skeletal muscle capillaries. The osmotic pressures driving water through the aquaporin pathway are those of the capillary plasma and mean ISF. If aquaporins were responsible for 50% of Lp, once fluid absorption has started following a fall in Pc, it should continue until the reduction in ISF volume has concentrated the ISF proteins sufficiently for the consequent rise in Πi to reduce ΔΠ below ΔP. This could take an hour or more, as calculated using typical values for capillary Lp and ISF volume in skeletal muscle. Before regarding this as an explanation of the slow development of steady-state fluid exchange in skeletal muscle, a second consequence of a large aquaporin component of Lp should be considered. Since the osmotic pressure difference across the aquaporin pathway is Πp minus the mean ISF Πi (not Πg), relatively high rates of fluid filtration should occur when Pc lies within its usual range (Figure 2A, right panel). One would therefore expect a brisk production of lymph in resting skeletal muscle—whereas resting lymph flow in skeletal muscle is so low that it is usually necessary to massage or passively exercise the muscle to obtain samples. Further investigation of the contribution of aquaporins and vasomotion to fluid exchange in skeletal muscle is required to resolve this question.

8. Clinical importance: increased endothelial permeability and tissue swelling

Microvascular fluid exchange underlies several major pathophysiological states. Starling himself was familiar with the transient absorption of ISF and attendant haemodilution following an acute hypotensive episode. Around 0.5 L of ISF can be absorbed into the human circulation within 15–30 min, supporting the cardiac output during hypotensive stress. Conversely, excessive microvascular filtration leads to interstitial oedema—a potentially fatal condition when it affects the lungs. Oedema can result from abnormal Starling's forces, increased endothelial permeability (e.g. inflammation), or impaired lymphatic drainage.50 The biochemical processes mediating increases in endothelial permeability were reviewed recently by Mehta and Malik.85 Here, we consider the impact of relatively small changes in endothelial membrane properties on fluid exchange.

Substantial pathology can result from relatively small reductions in vascular barrier properties. Rippe and Haraldsson86 pointed out that Eq. (3) describes fluid exchange through a homoporous membrane (one where all the fluid-conducting channels have the same hydraulic resistance and molecular-sieving properties); whereas endothelium has at least two fluid-conducting pathways, the ‘small pore’ and the ‘large pore’ pathways, as well as an aquaporin pathway in continuous capillaries. In normal healthy vessels, the small pores are estimated to contribute at least 95% and the large pores no more than 5% to the Lp of capillaries in skeletal muscle.63,87 Using these values and Eq. (3), we have constructed curves for steady-state fluid exchange (Figure 5A); see Appendix. For models that also incorporate the low-conductance aquaporin pathway, see Wolf31 and Rippe et al.88 We have assumed that the small pore radius is 4 nm and the large pore radius is 22.5 nm. Since the large pores contribute only 5% to the LP, this is equivalent to 19000 small pores for every large pore. In Figure 5A, we have plotted the separate relations between Jv/A and ΔP for the small and large pore systems, and the summated effect. Increased permeability can be thought of most simply as an increase in the number of large pores. Figure 5B shows the effect of a 10-fold increase in the number of large pores, with no change in the small pore population. The resulting increase in permeability resembles that seen during mild inflammation.89 The increase in overall LP is just over 50% and the overall σ to serum albumin falls only from 0.93 to 0.66. If the mean microvascular pressure were unchanged at just above 20 cmH2O, the addition of nine large pores per 19 000 small pores would increase the fluid filtration rate five times. There would also be a noticeable increase in ISF protein concentration. The effect of this increase on Jv is greatly magnified by a rise in Pc, which results from the increased blood flow and vasodilatation of inflammation. If Pc rose to 40 cmH2O (a relative modest increase), the net filtration of fluid into the tissue would increase 17-fold (arrow, Figure 5B). Very much larger increases in LP and reductions in σ occur during the first phase of acute inflammation.9092 Changes of this magnitude are usually short-lived, however.

Figure 5

Steady-state fluid exchange simulated for a post-capillary venule, with the fluid-conducting pathways modelled as parallel small pore and large pore populations; effect of inflammation. (A) Basal low permeability state; 95% of the hydraulic conductance is represented by the small pores (radius = 4 nm; blue curve) and 5% by large pores (radius = 22.5 nm; red curve). The black solid curve shows the total fluid exchange (sum of red and blue lines) at varying values of Pc (see Appendix, end of this article). The vessel is perfused with a Ringer solution containing serum albumin (Πp = 25 cmH2O). Pi is assumed to be constant and aquaporin pathway negligible (≤10% of total conductance). B. Steady-state fluid exchange when permeability has been increased in the same vessel. The red curve represents flow through the large pore system after inflammation has increased the number of large pores 10-fold. The small pore population is unchanged. The dashed lines are extrapolations of the linear parts of the steady-state summed relations to the pressure axis, where their intersection gives the value of the effective COP opposing fluid filtration (reduced in inflammation). Vertical arrows show typical values of microvascular pressure under basal conditions (A) and in mild inflammation (B). The rise in pressure contributes importantly to the dramatic 17-fold increase in filtration rate.

9. Advances in interstitial and lymphatic physiology

Space confines us to merely sketching advances in understanding the regulation of Pi and lymphatics. During acute inflammation Pi may fall initially, independently of microvascular exchange, and so amplify the initial rate of local oedema formation.93 It is suggested that connective tissues are held under mild compression by collagen fibrils that are β1-integrin bonded to fibroblasts and oppose the inherent tendency of interstitium to expand (glycosaminoglycan swelling). Fibril detachment serves to reduce Pi to more subatmospheric values. Supporting evidence includes a fall in Pi following treatment of the tissues with β1-integrin antibodies.

The active role of lymphatics in clearing fluid away from peripheral tissues has been supported by investigations ranging from ion channel electrophysiology to human arm studies,94,95 and lymphatic pump failure has been demonstrated in human breast cancer-related lymphoedema.96 Advances in understanding the molecular and genetic control of lymphangiogenesis have underpinned insights into the defects that underlie some hereditary lymphoedemas.97,98

10. Summary and future directions

The COPs influencing filtration across both fenestrated and continuous capillaries are exerted across the endothelial glycocalyx; the osmotic pressure of the ISF does not directly determine transendothelial fluid exchange. There is substantial evidence that with important exceptions such as the renal cortex and medulla, downstream microvessels are not in a state of sustained fluid absorption as traditionally depicted. Although doggedly persistent in textbooks and teaching, the traditional view of filtration–reabsorption balance has little justification in the microcirculation of most tissues. Tissue fluid balance thus depends critically on lymphatic function in most tissues. In making these forceful statements, we are mindful of William Harvey's remark in his classic, De Motu Cordis (1628): ‘I tremble lest I have mankind for my enemies, so much has wont and custom become second nature. Doctrine once sown strikes deep its root, and respect for antiquity influences all men’.99

We would emphasize that despite the above advances, much remain unknown. The time course of fluid absorption into the circulation requires more detailed investigation in a wider range of tissues; and the corresponding microstructure of the transendothelial pathway needs to be better defined, including estimates of the aquaporin contribution to the overall microvascular LP, which appears to vary considerably.32 The structure and composition of the glycocalyx over fenestrations and intercellular clefts should be explored. Understanding the regulation of the glycocalyx synthesis and turnover could lead to novel therapeutic strategies to reduce pathologically increased permeability. It also worth emphasizing that little is known of the nature of fenestral diaphragms; and that the classical problem of the transport of macromolecules through endothelia (large pores or vesicles) remains unresolved 50 years after it first became a controversy. Genetic knock-out mice offer a promising approach to some of these problems.79,98


We apologize to the many authors who have contributed importantly to this field and whose work has not been cited due to space limitations.

Conflict of interest: none declared.

Appendix: two-pore model used to generate Figure 5

The method was that described by Rippe and Haraldssson,63,87 where the capillary wall is modelled as an impermeable membrane of thickness Δx penetrated by two sets of cylindrical pores—small pores of radii 4 nm and large pores of radii 22.5 nm. Lp was chosen to be 2 × 10−7 cm s−1 H2O−1 (typical of a rat mesenteric venule), 95% of which was accounted for by small pores and 5% by large pores. Thus, for the small pores Lp(S) = 1.9 × 10−7 cm s−1 cmH2O−1 and for the large pores Lp(L) = 0.1 × 10−7 cm s−1 cmH2O−1 and the overall hydraulic conductance is Embedded Image A1 Pore theory (e.g. Curry100) defines the Lp of a population of pores of constant radius (r) in terms of Poiseuille's law, i.e. Embedded Image A2 where N is the number of pores per unit area of membrane and η is the fluid viscosity within the pores. Thus, by assigning values for Lp(S) and Lp(L) and pore radii, we have determined the values of NSx and NLx, the ratios of the numbers of small and large pores to their length.

The partition coefficient, λ, and the reflection coefficient, σ, for each set of pores to a particular solute depend on the ratio of the molecular radius (a) to the pore radius. They have been calculated from the following expressions: Embedded Image A3 and Embedded Image A4

Finally, the diffusional permeability, p, of the solute for each set of pores is related to the diffusion coefficient of the solute in aqueous solution, D, and a function dependent on (a/r): Embedded Image A5 where Embedded Image A6

Equations (A1)–(A6) were used to estimate the values for Lp, σ and p for serum albumin at each population of pores, assuming that a = 3.6 nm for albumin. Equation (3) was re-arranged as Embedded Image A7

with the relevant values of Lp, σ, and p, equation (A7) was used to determine the values of ΔP for a range of values of JV/A for the small pores and the large pores. The values of JV/A at the same ΔP for the small pores and the large pores were added to give the overall JV/A for unit area of vessel wall (Figure 5A). The exercise was repeated for Figure 5B. Here, the contribution of Lp(S) remained constant at 1.9 × 10−7 cm s−1 H2O−1 and Lp(L) was increased 10-fold, raising the overall Lp only modestly, from 2.0 to (1.9 + 1.0) = 2.9 × 10−7 cm s−1 H2O−1.


  • This article is part of the Spotlight Issue on: Microvascular Permeability


View Abstract