Capillary Exchange, Lymph and Edema

Adults born with almost no albumin in their plasma usually have only mild edema, and a few have none. (In newborns and young children the condition can be serious, with fluid retention.) Albumin supplies most of the oncotic pressure that holds water in the blood, and the textbook balance of capillary forces predicts that losing it should flood the tissues. Why it does not is this lesson’s question.

Blood vessels, blood and lymph gave the first look: Starling’s balance, Landis’s frog capillaries and the arterial-end, venous-end model. Here you take the balance apart, add the leak, and follow the fluid home.

Three kinds of wall, several routes across

Capillary walls differ in how their endothelial cells join (epithelia).

• Continuous capillaries (skeletal muscle, skin, lung; the tightest kind in the brain) have an unbroken lining. Cells meet at narrow intercellular clefts and, outside the brain, carry many caveolae, small membrane pits and vesicles (endocytosis).
• Fenestrated capillaries (intestinal villi, renal peritubular capillaries, endocrine glands) add fenestrae, pores through the thinnest parts of the cell. In rat tissues a thin diaphragm across each fenestra is labelled by the protein PV-1 (Stan, Kubitza and Palade, 1999). Glomerular fenestrae have no diaphragms (glomerular filtration).
• Sinusoidal capillaries (liver, spleen, bone marrow) have wide openings, large open fenestrae in the liver and gaps between cells elsewhere, and an incomplete or absent basement membrane. They hold back even large molecules little.

Over all of them lies the glycocalyx, a luminal coat of proteoglycans and glycoproteins whose fibres cover the cleft entrances and fenestrae.

The routes follow from the wall. Oxygen, carbon dioxide and other lipid-soluble molecules diffuse through the endothelial membranes over the whole surface (membrane transport). Outside the brain, small water-soluble solutes such as glucose, ions and urea diffuse through clefts and fenestrae; brain capillaries close this route, and glucose crosses them on transporters. For them, diffusion moves far more across the wall than filtering water carries, so the forces below govern water and protein, not nutrient supply.

Plasma proteins take a separate, sparser route. In the two-pore model, many small pores, thought to be the spaces between glycocalyx fibres over clefts and fenestrae, pass water and small solutes, and rare large pores pass macromolecules. What the large pores are is not settled. Bruns and Palade (1968) found intravenous ferritin inside endothelial vesicles of rat diaphragm-muscle capillaries, not in junctions, and concluded that vesicles are the large pores; a 2021 review lists rare channels or possibly vesicles.

Starling’s balance, written with a leak term

Starling showed in 1896, in perfused hind limbs of anaesthetized dogs, that saline injected into the tissues was absorbed into the blood while plasma was not, and argued that capillary pressure and plasma oncotic pressure set fluid exchange. The form used today adds a term he never wrote:

Jv = Lp S [(Pc - Pi) - sigma (pi_p - pi_g)]

• Jv: volume flow across the wall, mL per min, positive outward (filtration).
• Lp: hydraulic conductivity, water flow per unit area per unit pressure (cm per s per cmH2O); S is wall area. Their product, the filtration coefficient Kf, is given for whole tissues per 100 g.
• Pc: mean capillary hydrostatic pressure, placed between arterial and venous pressure by the ratio of pre- to postcapillary resistance (resistance, compliance and local flow), so a rise in venous pressure moves it more than an equal rise in arterial pressure.
• Pi: interstitial hydrostatic pressure, slightly below atmospheric in loose tissues such as skin, intestine and lung, though positive in some organs such as liver and kidney; Noddeland found a mean of about -1.4 mmHg in subcutaneous tissue at heart level in 10 people. A rise in Pi opposes filtration.
• pi_p: plasma oncotic pressure, about 26 mmHg in adults by one review. Albumin supplies most of it, about 65 to 80 percent by different reviews, and globulins supply the rest. Oncotic pressure rises more steeply than linearly with protein concentration, so the low-albumin numbers in this lesson are round illustrative values, not a clinical conversion.
• pi_g: oncotic pressure of the fluid just beneath the glycocalyx, on the tissue side of the sieve.
• sigma: the reflection coefficient, the fraction of a solute’s full osmotic pressure it exerts across this wall: 1 if the wall passes water and no solute, 0 if it hinders the solute no more than water. For albumin it is high, about 0.8 to 0.96 in the single vessels and tissues measured, varies with the tissue, and approaches 0 in sinusoids.

Staverman defined the reflection coefficient, and Kedem and Katchalsky gave the flow equation that uses it. Putting the subglycocalyx oncotic pressure into it came later, from Michel’s work on the revised principle. Starling gave the principle.

Two consequences follow. A lower sigma means the same protein gradient opposes filtration less, so a leakier wall passes more water as well as more protein; in a liver sinusoid oncotic pressure barely restrains filtration. And because the wall leaks protein, there is no true Starling equilibrium: if filtration stopped, protein would diffuse out until the oncotic difference vanished.

Most capillaries filter all along their length

Woodcock and Michel’s 2021 review states that the arterial-end, venous-end model has no reproducible experimental evidence behind it, and Levick concluded in 1991 that sustained downstream reabsorption is difficult to uphold.

The revised picture puts the sieve at the glycocalyx, like a coffee filter laid over the mouth of a narrow channel. Filtrate passing it is nearly protein-free and flows out through the cleft, washing the small subglycocalyx space. Protein reaching the tissue by large pores must diffuse back against that flow, so pi_g is lower than average interstitial oncotic pressure and depends on Jv. Reviews by Levick and Michel (2010) and Michel, Woodcock and Curry (2020) draw four consequences:

• The hockey-stick curve. Steady-state filtration plotted against Pc is steep above an inflection just below plasma oncotic pressure, and shallow below it.
• The no-absorption rule. The oncotic difference opposes filtration but does not reverse it in the steady state.
• Transient absorption. A sudden fall in Pc gives absorption that concentrates protein beneath the glycocalyx until it stops; after blood loss this is autotransfusion (reflexes are in blood pressure regulation).
• Steady absorption only where an independent stream flushes the tissue with protein-poor fluid: intestinal mucosa (absorption and the liver) and renal peritubular capillaries, where an epithelium supplies nearly protein-free fluid, and lymph nodes, where incoming lymph, lower in protein than plasma, does the flushing.

The modern test. If the glycocalyx is the sieve, stripping part of it should raise Lp and lower sigma in the same vessel. Adamson (1990) found that a minute of pronase, a broad protease, raised Lp of single perfused frog mesenteric capillaries from 2.0 to 4.9 x 10^-7 cm per s per cmH2O with clefts unchanged. Betteridge, Arkill and colleagues (2017) cannulated single mesenteric microvessels in anaesthetized rats and measured Lp by Landis-Michel occlusion (the slope of filtration against pressure), the albumin reflection coefficient, and glycocalyx depth. Lp and sigma were measured in each vessel before and after neuraminidase, which removes sialic acids (n = 7 pairs): Lp rose 1.9-fold, from 1.39 to 2.68 x 10^-7 cm per s per cmH2O, and sigma for albumin fell from 0.96 to 0.87. Depth by electron microscopy was 78 nm in vehicle-treated vessels and 40 nm in neuraminidase-treated vessels, a comparison between different vessels.

That ties sialic acids to water and albumin permeability in rat mesentery. It does not show the effect in human vessels or disease, neuraminidase may act beyond the glycocalyx, and depth estimates in one rat study ranged from 0.17 to 3.02 micrometres with label and method. Dull and Hahn (2022) argue that junctions, not the glycocalyx, limit protein permeability; the debate is open.

Worked problem: one muscle, three states

Made-up round numbers for adult human skeletal muscle at rest, steady state unless stated: Kf = 0.005 mL per min per mmHg per 100 g (illustrative, of the order estimated in human limbs), 10 kg of muscle (100 units of 100 g), Pc = 18 mmHg, Pi = -1 mmHg, pi_p = 26 mmHg, pi_g = 6 mmHg, sigma = 0.9. Assume one lumped Pc and constant Kf and sigma.

A, rest. Hydrostatic difference = 18 - (-1) = 19 mmHg. Oncotic opposition = 0.9 x (26 - 6) = 18 mmHg. Net = +1 mmHg, outward. Jv = 0.005 x 1 x 100 = 0.5 mL per min, or 720 mL per day, returned as lymph.

B, venous pressure rises so Pc = 28 mmHg (as in heart failure; cardiac output and venous return). At first net = 29 - 18 = 11 mmHg and Jv = 5.5 mL per min. Over hours, in this example, Pi rises to +2 mmHg and pi_g falls to 2 mmHg: net = (28 - 2) - 0.9 x (26 - 2) = 4.4 mmHg, Jv = 2.2 mL per min, 60 percent less. If lymph can carry at most 1.5 mL per min, fluid accumulates at 0.7 mL per min, about 1.0 L per day: edema.

C, blood loss lowers Pc to 10 mmHg. Net = 11 - 18 = -7 mmHg, absorption of 3.5 mL per min at first. It stops when 0.9 x (26 - pi_g) = 11, at pi_g = 13.8 mmHg, and the vessel sits on the shallow arm of the hockey stick.

Mean capillary pressure Pc 18 mmHg Plasma albumin 4 g per dL (π_p = 26 mmHg) Reflection coefficient σ 0.90 (0.9 or more in most tissues)

Lymph drainage

Example capacity 1.5 mL per min Blocked (invented 0.3 mL per min)

Net pressure

+1.00 mmHg, outward

Steady filtration Jv

0.50 mL per min · 0.72 L per day

Lymph flow

0.50 mL per min · 0.72 L per day

Accumulation in the tissue

0.00 mL per min · 0.00 L per day

In this sketch: steady state, adult human skeletal muscle at rest, one lumped Pc. All numbers are made up for the sketch, as in the lesson’s worked problem: Kf = 0.005 mL per min per mmHg per 100 g (illustrative) for 10 kg of muscle, so Jv = 0.5 × net; Pi = −1 mmHg; π_g = 6 mmHg; π_p = 2 + 6 × albumin, the lesson’s 6 mmHg per g per dL (26 mmHg at 4 g per dL), a linear simplification; lymph capacity of 1.5 mL per min is the lesson’s example, and the blocked capacity of 0.3 mL per min is invented here. Values of σ below 0.9 and the two lymph states are sketch settings. Safety factors (Pi rising, π_g falling) are not modelled, so accumulation here is an upper estimate. If net is negative the sketch shows transient absorption and sets steady filtration to zero (a crude stand-in for the no-absorption rule; the hockey stick is shallow below its inflection, not zero).

Rule to keep: edema forms when filtration exceeds lymph drainage, so a higher Pc, lower plasma protein or lower σ raises filtration, and blocked lymph lowers what can be drained.

Lymph: buttons, lymphangions and the thoracic duct

Filtrate enters blind-ended initial lymphatics, whose overlapping cells meet at discontinuous button junctions, with flap-like openings between the buttons acting as the primary valves. Collecting lymphatics have secondary valves, and each segment between two is a lymphangion, a contractile pump. Lymph moves by this intrinsic pumping plus extrinsic squeezing from muscle and breathing. In isolated rat mesenteric lymphatics and thoracic ducts, imposed flow lowered contraction frequency and amplitude (Gashev, Davis and Zawieja, 2002). Skeletal-muscle lymphatics are reported to lack smooth muscle, so muscle movement probably drives flow there.

Recent reviews estimate that about 8 L of fluid a day filters from an adult’s blood capillaries and put lymph transport at 8 to 12 L a day, with roughly half thought to be absorbed into blood within lymph nodes, an estimate resting mainly on animal studies (lymphoid organs and immune traffic); the familiar 24 L filtered and 20.4 L reabsorbed come from the reabsorption model. The rest enters the veins by the thoracic duct, at the junction of the left subclavian and left internal jugular veins, and by the right lymphatic duct on the right, carrying back the plasma protein that leaked.

Long-chain fats leave intestinal cells as chylomicrons, enter lacteals and reach the blood through the thoracic duct, bypassing the liver at first; medium-chain fatty acids go mostly by the portal vein. In mice, endothelial deletion of Nrp1 and Flt1 raised VEGF-A signalling through VEGFR2, zippered lacteal junctions, blocked chylomicron uptake and gave resistance to diet-induced obesity (Zhang and colleagues, 2018).

Four ways to make edema, and what holds it back

Edema is interstitial fluid accumulation when filtration exceeds lymph drainage. Four routes lead there: raised Pc (Part B), low plasma protein, leaky microvessels, and blocked lymph. In lymphedema, chronic impaired drainage, the swelling includes fat deposition; in one family, Milroy disease cosegregated with a VEGFR3 mutation (Irrthum and colleagues, 2000).

Where does an inflammatory leak happen? Guido Majno and George Palade (1961) injected histamine or serotonin under the scrotal skin of rats, carried a colloidal mercuric sulfide tracer in the blood, and fixed the cremaster muscle 1 to 12 minutes later for electron microscopy. If leak occurred where exchange normally happens, capillaries should open. Instead, openings about 0.1 to 0.8 micrometres wide appeared between endothelial cells in vessels of 7 to 8 micrometres or larger, venular-sized, while capillary-sized vessels (3 to 5 micrometres) looked intact, and the tracer passed the gaps between endothelial cells but was held back by the intact basement membrane. That located this leak in venules, between cells, and ruled out capillaries as its site in this tissue. It did not show how the gaps form, that every organ and mediator behaves this way, or anything about baseline protein leak (innate immunity).

Three safety factors limit swelling: Pi rises as fluid collects, washout lowers tissue and subglycocalyx protein, and lymph flow rises.

Low plasma protein, as mechanism. Suppose albumin falls from 4 to 2 g per dL and, as a round illustrative value, pi_p falls from 26 to 14 mmHg. With pi_g fixed at 6 mmHg, Part A’s net rises from 1 to 19 - 0.9 x 8 = 11.8 mmHg, a large rise in filtration that would cause severe edema if nothing else adjusted. Yet in congenital analbuminemia edema is common but usually mild, and a few people have none. Washout lowers pi_g: at 2 mmHg, net = 19 - 0.9 x 12 = 8.2 mmHg. Pi and lymph flow rise, and globulins may partly stand in for albumin. Directions: pi_p down, Jv up (most at first), lymph flow up, interstitial protein down, Pi up. For nephrotic protein loss, the underfill hypothesis (low oncotic pressure drains the plasma) is debated. By the no-absorption rule, raising plasma oncotic pressure can slow steady filtration but not reverse it.

Design problem

An anaesthetized dog has a cannula in a prenodal lymphatic draining the hind paw, so lymph reports that tissue’s capillaries before a node alters it. Plasma protein is 60 g per L. You can raise limb venous pressure, infuse histamine into the femoral artery, or lower plasma protein by plasmapheresis. Predict lymph flow, the lymph-to-plasma protein ratio (L/P) and protein flux for each, and interpret these invented results: baseline 0.05 mL per min, L/P 0.5; raised venous pressure 0.25 mL per min, L/P 0.25; histamine 0.25 mL per min, L/P 0.7.

One way to do it

1. Protein flux = lymph flow x lymph protein. Baseline 0.05 x 30 mg per mL = 1.5 mg per min; venous pressure 0.25 x 15 = 3.75 mg per min; histamine 0.25 x 42 = 10.5 mg per min, seven times baseline.
2. Pressure. Water flow rises more than protein flow, so L/P falls (sieving, the washout safety factor).
3. Permeability. Histamine opens venular gaps and sigma falls, so L/P stays high or rises and protein flux climbs. Histamine also dilates arterioles and raises Pc, so the rise in lymph flow alone does not separate the two causes; the high L/P is what points to permeability.
4. Low plasma protein. Lymph flow rises while lymph protein falls with plasma protein, so L/P may change little (a simplification).
5. Controls. Reach a new steady state at each step, keep the cannula prenodal, compare the other limb, allow for anaesthesia, measure Pi if you can. A dog’s paw does not stand for every tissue or for people.