Você está na página 1de 27

MICROVASCULAR RESEARCH 4‘%,85-111 (1992)

A Three-Dimensional Junction-Pore-Matrix Model for


Capillary Permeability

S. WEINBAUM,*” R. TsAY,t AND F. E. CURRYZ~?


Departments of *Mechanical Engineering and tChemica1 Engineering, The City College of the City
University of New York, New York, New York 10031; and $Department of Human
Physiology, School of Medicine, University of California at Davis,
Davis, California 95616

Received February 14, 1991

A three-dimensional model is presented for the hydraulic conductivity and diffusive perme-
ability of capillary endothelial clefts with a junctional strand with discrete pores and a fiber
matrix in its wide parts. The model attempts to provide new insight into long-standing issues
concerning the relative importance of open junction discontinuities, restricted slit regions,
and matrix components in determining the permeability and selectivity of the capillary wall.
The predictions drawn from the model are used to formulate new experiments to test two
hypotheses concerning the molecular organization of the junction strand and the location
of matrix structures in the wide part of the cleft. Using the three-dimensional theoretical
approach recently developed by Tsay, Weinbaum, and Pfeffer (Chem. Eng. Comm. %2,67-
102, 1989), the model first explores the behavior of three different molecular models for
the junctional strand discontinuities: (i) a more frequent circular pore of 5.5-nm radius
formed by isolated missing junction proteins; (ii) a restricted rectangular slit of four to eight
missing proteins and 8-nm gap height; and (iii) larger more infrequent breaks of four to
eight missing proteins with a gap height of 22 nm, equal to the width of the wide part of
the cleft. For the circular and 8-nm gap height pores the primary molecular sieve can be
located at the level of the junction strand, whereas for the 22-nm gap height pores, matrix
components must be present in at least some portion of the cleft to provide the molecular
filter. The water flow through the cross-bridging fibers in the wide part of the cleft is described
either by a new exact three-dimensional theory (Tsay and Weinbaum, J. Fluid Mech. 226,
125-148, 1991) for an ordered periodic array or by a new approximate theory for a random
array of perpendicular fibers. Both this theory and the new approximate theory for diffusion
presented herein take into account for the first time the interaction between the fibers and
plasmalemma boundaries. The principal predictions of the model are that (i) infrequent
larger breaks are most likely required to account for small solute permeability; (ii) these
larger breaks must be accompanied by a sieving matrix, but this matrix probably occupies
only a small portion of the depth of the cleft and/or its entrance at the luminal surface;
(iii) neither junctional pore, restricted slit, or fiber matrix models can by themselves satisfy
permeability and selectivity data; and (iv) one-dimensional models are a poor description
of a cleft with infrequent larger breaks since the solute will be confined to small wakelike
regions on the downstream side of the junction strand discontinuities and thus not fill the
wide part of the cleft. 0 1~2 Academic PKSS, I~C.

’ To whom correspondence and reprint requests should be addressed.


* Present address: Institute of Biomedical Engineering, National Yang-Ming Medical College,
Taipei, Taiwan 11221.

85

w26-286292 $5.00
Copyright 0 1992 by Academic Press, Inc.
All rights of reproduction in any form reserved.
Printed in U.S.A.
86 WEINBAUM, TSAY, AND CURRY

width not
6-611 speoiffed

w
17-22nm

1,
J

Ilf 1 ,
FIG. 1. Two one-dimensional models of intercellular pathway. (a) Constricted channel model. This
model proposed that the junctional cleft is 17-22 nm wide except for one or more constrictions. The
constriction is assumed to be the main molecular filter and its width, 6-8 nm, is chosen to fit reflection
coefficient data. (b) Fiber matrix model. The fiber matrix is assumed to be contained in the wide part
of the channel and acts as the molecular filter. Both models assume that the permeability is proportional
to the fractional length of the tight junction which is effectively open.

INTRODUCTION
The reasons for developing the theoretical model described in this paper were,
first, to more rigorously evaluate two alternative hypotheses describing the size-
selective structures that determine capillary permeability and, second, to use the
theory to conceptualize new experiments to test these two hypotheses. One hy-
pothesis proposes that the molecular filter lies within the small discontinuities in
the strand of molecules forming the junctional complex between adjacent en-
dothelial cells. These discontinuities also determine the length of the junction
effectively open for the diffusion of small hydrophilic solutes. The second hy-
pothesis also proposes that the breaks in the junctional strand determine small-
solute permeability, but assumes that the primary molecular filter is a fibrous
network associated with the cell surface and the wide part of the junctional cleft.
To quantitatively test the feasibility of these two hypotheses we have developed
specific molecular models for the structure of the junctional pore and the fiber
matrix. The prediction of the models are evaluated using data on the hydraulic
conductivity, permeability, and selectivity of microvessels with continuous en-
dothelium and the ultrastructure of microvessels of known permeability properties.
The starting point for our analysis is the model in Fig. la, which is based on
published studies of the ultrastructure of the cleft between adjacent endothelial
cells, most of which were completed prior to 1980 using electron microscopy on
conventional thin sections of 40-50 nm thickness, and protein tracer molecules,
which could be used as size-specific molecular probes of the pathways through
the cleft (Perl, 1971; Casley-Smith et al., 1975; Wissig, 1979; Perry, 1980; Crone
and Levitt, 1984). The molecular filter was assumed to be associated with slitlike
openings in the junctional complex, where the adjacent membranes appear to be
very close, but not fused. This size-selective pore could also be formed by a
tortuous pathway across a multistranded array in which these slitlike openings
could appear in different planes of section. These views lead several authors to
A MODEL FOR CAPILLARY PERMEABILITY 87

propose a simple one-dimensional constricted channel model, as shown in Fig.


la, in which the permeability was proportional to the fractional length of the
junction strand, which was effectively open for the diffusion of small solutes, and
the size-selective structure was a restrictive slit of 6- to 8-nm gap height, connecting
the wider regions of the cleft whose width was estimated to be 17-22 nm by
various investigators. In muscle capillaries the measured hydraulic conductivity
and permeability coefficient for small solutes could be accounted for if 10% of
the junctional strand was open, while in frog mesenteric capillaries close to 90%
of the junctional strand needed to be open to account for these permeability
properties. The restricted cleft height was the equivalent slit dimension that would
satisfy the measured reflection coefficients for solutes of 1.5 to 3.5-nm radius
(Curry, 1986; Crone and Levitt, 1984).
Over the past decade there have been two significant challenges to important
details of this model. First, three-dimensional reconstructions of serial sections of
capillaries in rat heart muscle indicated that the discontinuous breaks in the
junctional strand were only of the order of a single conventional thin-section
thickness (40 nm) and thus was much shorter than the breaks proposed by earlier
investigators, who had based their estimates on random thin sections. Further-
more, the 40-nm breaks observed by Bundgaard (1984) were very infrequent
(approximately one per micrometer of junction length). These discontinuities thus
represented a fractional length of open junction that was considerably shorter
than the length of the discontinuities predicted by the one-dimensional models.
The 3-D reconstructions also showed that the pattern of the junctional complexes
representing the points of contact between adjacent cells was very similar to the
pattern of ridges and groves seen in freeze-fracture electron micrographs of the
junctional complex. This was the result expected if the junctional complex seen
in conventional EM corresponds to the arrays of intramembranous particles seen
in freeze-fracture as is also suggested by Wissig (1979). Bundgaard pointed out
that discontinuities in the junctional strand shorter than 40 nm could not be
identified in conventional thin sections. After reexamining a small sample of
ultrathin sections (12-15 nm) from a previous study, he suggested that discon-
tinuities as small as 12 nm in the strand might represent a population of smaller
pores. We also evaluate this hypothesis in more detail in this paper.
The second challenge to the model in Fig. la is the observation that the perme-
ability and selectivity of a variety of continuous capillaries can be accounted for
if a matrix of fibrous molecules covered the endothelial cell surface and filled all
or part of the cleft as shown in Fig. lb. The fiber matrix hypothesis placed the
primary molecular filter in the matrix and not at the junctional complex. Reviews
evaluating this hypothesis have been published elsewhere (Michel, 1984, 1988;
Curry, 1986; Crone and Levitt, 1984; Tsay et al., 1989). Further experimental
support for this concept is the recent demonstration that enzymatic removal of
the endothelial glycocalyx in frog mesenteric capillaries decreases the hydraulic
resistance by roughly 50% (Adamson, 1990). Since the matrix does not constitute
a significant barrier to small solutes, the predicted permeability to small solutes,
assuming the cleft is open, is much too large. To account for this shortcoming,
it is necessary to postulate that small solute permeability is regulated by pores or
breaks in the junctional strand. One fundamental difference between the proposed
junctional strand-pore-fiber matrix model considered herein and the straight junc-
tional strand-pore model considered previously in Fig. la is that the structure of
88 WEINBAUM, TSAY, AND CURRY

‘missing junction partihes

f3nm
b)
%ght regibn ?ong rectangular
44423nm

...........................
.........................
........................
FIG. 2. Three schematic diagrams of junction protein arrangements. These protein arrangements
correspond to (a) small gaps suggested by Bungaard’s (1984) ultrathin sections (the zigzag protein
pattern was first proposed in Tsay et al. (1989)); (b) 1on g rectangular junction pores, similar to those
in the one-dimensional constricted channel model; and (c) large pores formed by discontinuities of
44-88 nm in the junctional strands.

the junctional strand discontinuities does not have to be restricted to dimensions


required of a molecular sieve if a matrix is present.
The development of the present models was also driven by the need to obtain
more precise molecular models of the junctional complex. Either the discontin-
uities are so small that they are not observed in 40-nm sections or they are larger,
but infrequent, so that most of the junctional strand is a tight barrier impenetrable
to water and solutes. Figure 2 summarizes several arrangements of the junction
proteins considered in this manuscript. The interlocking structure in Fig. 2a was
suggested by Tsay et al. (1989) (TWP) as a way in which size-selective pores could
be formed by individual missing proteins. This interlocking arrangement was
suggested by the observation in Firth et al. (1983) that the average spacing between
junctional proteins in each membrane of the guinea pig placental capillaries was
22 nm, just twice the diameter of the individual proteins (11 nm). An important
consequence of the model in TWP was that the predicted spacing between the
pores, required to account for measured value of L, in frog mesenteric capillaries,
(-100 nm) was too small to neglect the three-dimensional interaction of the
velocity fields between pores and a three-dimensional model for the junctional
strand was developed.
The alternate possibility that the discontinuities in the junctional strand are
larger than a single missing protein, and widely spaced, is usually represented as
shown in Fig. 2b. The rectangular slit length is shown as 44-88 nm, representing
four to eight missing proteins, and the slit height is shown as 6-8 nm in order
that the slit can function as the primary molecular filter. However, it is difficult
to explain how this slit height is maintained. One expects that, if junction proteins
were absent in these longer discontinuities, the membrane forces that determine
the equilibrium spacing in the wide part of the cleft would also be operative in
A MODEL FOR CAPILLARY PERMEABILITY 89

Junction

FIG. 3. Sketch of the three-dimensional junction-pore-matrix model of the intercellular cleft.


Junctional strand with periodic pores lies parallel to luminal front. Lz is the depth of pores in the
junctional strand and 15, and 15, are the depths between the junctional strand and the luminal and
abluminal fronts. Within the wide portion of cleft, cross-bridging fibers are represented by either a
periodic square array or a random array of cylindrical posts. The radius of these posts is a and the
gap spacing between posts for periodical array is A. The distance between two adjacent pores in the
junctional strand is 20.

the region where these breaks appear. Thus, if the membrane-bending resistance
over these longer breaks was small, it is anticipated that their gap height would
be nearly the same as the background 17- to 22-nm height of the wide part of
the channel, as shown in Fig. 2c. The larger breaks observed in Bundgaard (1984),
whose lengths are typically single conventional FM sections, support this conjec-
ture concerning the gap height of the larger breaks. If such pores are present
they cannot be the primary size-selective structure and a fiber matrix in the wide
part of the cleft is also required to provide the molecular sieve.
The theoretical model described in this paper is an extension of the 3-D model
in TWP. The simplest form of the model is shown in Fig. 3, where the interlocking
junction strands were treated as an impermeable barrier with periodically dis-
tributed discrete circular pores representing the individual missing proteins as
shown in Fig. 2a. In order to represent the fiber matrix components in the wide
part of the cleft, a two-dimensional square array of slender cross-bridging fibers
of radius a and height 2B was introduced with a periodic spacing 2w and gap
spacing A as sketched in the left hand diagram in Fig. 3. The fibers are shown
in an ordered array to account for the fact that part of the interaction of albumin
with the matrix appears to involve ordering of the fibers as described by Michel
(1988).
In the present paper we shall explore three important extensions of the model
in TWP. First, the pore in TWP was assumed to be formed by individual missing
proteins in the junction protein strand whose cross-sectional shape was a circular
hole of radius rp = 5.5 nm. This value of pore radius was selected to fit the
osmotic reflection coefficient data for intermediate-size solutes smaller than al-
90 WEINBAUM, TSAY, AND CURRY

burnin (effective hydraulic diam 7 nm). In this study the various rectangular
junction pore geometries shown in Fig. 2 will also be examined and the perme-
ability properties of small clusters of up to eight missing proteins will be analyzed.
Second, a much more rigorous hydrodynamic theory has been developed for the
fiber matrix, which is valid for an ordered array of cross-bridging fibers of arbitrary
aspect ratio and spacing (Tsay and Weinbaum, 1991). This new theory will be
used to (i) describe the additional hydrodynamic resistance due to an ordered
matrix in the wide part of the cleft and (ii) to develop a simple model for randomly
distributed cross-bridging fibers. Third, the model in TWP considered only filtra-
tion, In the present study an equivalent theory for the junctional strand-pore-
matrix model has also been developed for diffusion. Thus, it will be possible to
examine the effect of pore geometry and fiber matrix components on the solute
permeability coefficient P.

METHODS
1. Hydrodynamic Interaction between Junction Pores
The model for filtration through the idealized intercellular cleft geometry
sketched in Fig. 3 combines two different types of channel flows; the three-
dimensional hydrodynamic interaction between the pores in the junction strand
and the flow through the fiber array. The expression for L, for the entire cleft
can be written as the sum of three resistances in series,
L, = (R, + R2 + R&l, (1)
where RI and R3 describe the mixing interaction of the Stokes jets within the
wide parts of the channel and Rz the resistance within the junction pore itself,
The expressions for these resistances are given for a circular pore of radius ri, in
TWP as
3~
Ri = -[B2Lj,
& + I z tanh (X,L,) sin* (h,d)
2 & A3, 1, i = 1,3 (2)

A, = F, d=g (3)

R = 16pL2D
2
7rri Lj, ’
(4)
where p is the fluid viscosity, all the geometric lengths are defined in Fig. 3, and
Lj, is the total cleft length (total cell perimeter) per unit endothelial surface area.
For a pore of rectangular cross-section of height 26 and width 2d, Eq. (4) for R2
is replaced by

R 2

(5)
Gr, y = b/d.

The foregoing results are derived by matching the volumetric flow and average
A MODEL FOR CAPILLARY PERMEABILITY 91

pressure entering and leaving the junction pore with the entrance and exit con-
ditions in regions 3 and 1. The nonuniformity of the pressure at the pore entrance
and exit obtained from the Hele-Shaw solutions in regions 1 and 3 is small and
gives results that are nearly indistinguishable from those obtained by requiring
the pressure to be uniform across the entrance and exit of the pore.

2. Hydrodynamic Resistance of Fiber Matrix


The effect of the hydrodynamic interaction with the cross-bridging fibers in the
wide portion of channel can be treated by replacing the actual fluid viscosity ,U
in Eq. (2) by an effective viscosity peff. peff is defined by V (P) = - heff (w/B’,
where ( ) denotes an average value over a region that is small compared to the
depth of the cleft L. For the periodic fiber arrangement shown in Fig. 3 the
average is taken over a single periodic unit containing one fiber in the array. The
effective viscosity peff can be written as pf, where f is an hydrodynamic interaction
function that depends on the fiber configuration, the fiber volume fraction S, and
the aspect ratio B/a of the fibers.
The boundary value problem for determining peff requires that the no-slip
conditions be satisfied on the surface of each fiber or on the channel walls. In
Tsay and Weinbaum (1991) an exact infinite series solution to the full boundary
value problem for the doubly periodic fiber array is constructed using the fun-
damental singularities for a single circular post developed in Lee and Fung (1969)
and a quasi-periodic function that is a linear combination of the Weierstrass 4
function and its derivatives to describe periodicity conditions for each fiber unit
cell. This highly accurate solution is far too complicated to be used with the same
ease as a Carmen-Kozeny equation for an infinite medium. However, this solution
provides important new insight into the velocity field surrounding the fibers and
is used to justify the validity of the Brinkman approximation introduced in the
next section. The important observation is that for long slender fibers the viscous
boundary layers surrounding the fibers can be much larger than their radius and
when the fiber boundary layers overlap there will be a large increase in drag or
peff (decrease in Lp). In contrast, if these boundary layers remain distinctly sep-
arated there will be very little increase in drag beyond what is caused by the walls
of the channel. One thus anticipates that large increases in drag will start to occur
when A/B < 1.

3. Brinkman Approximation for a Bounded Fiber Matrix


Based on the foregoing discussion, the fiber matrix will be an important con-
sideration in determining the filtration resistance if A is less than 10 nm since B
is approximately 10 nm. In this regime the two-term approximate solution in Lee
(1969) is not accurate and we seek a simpler approximation to the exact infinite
series solution in Tsay and Weinbaum (1991) for a bounded porous media using
a Brinkman-type approximation. In the Brinkman approach the resistance due to
the fibers is taken into account by adding a Darcy-type resistance term in the
governing equation for the channel flow:

VP=j+LV%.
P
92 WEINBAUM, TSAY, AND CURRY

Here p is the fluid viscosity and I& is a Darcy permeability coefficient, which
describes the flow through an infinite matrix that has the same fiber geometry as
the interior of the bounded flow under consideration. In the present case the
unbounded flow geometry corresponds to flow perpendicular to a periodic array
of infinitely long fibers. The exact solution to this two-dimensional unbounded
flow problem is given in Sangani and Acrivos (1982). Tsay and Weinbaum (1991)
showed that the following simple approximation is accurate to within 10% for
s < 0.7:
Kp = 0.0572 a2 (A/a)2.377 (7)
The solution to the Brinkman equation that satisfies the no-slip conditions on
the channel walls is

is _ - Kp dP 1 _ cash (2/I&)
P dx ( cash (Pimp) 1*
By averaging this solution for the velocity across the channel height we can define
an effective Darcy law permeability by

K p,eff (9)
= -CL
-dx
\ I
which describes the effect of the channel walls in increasing Kp.
The above expression for Kp,eff can be directly related to the channel friction
coefficient f using the expression Kp,eff = B2/3f for Poiseuille flow in a channel.
After some algebra f can be written using Eq. (9) in the compact form

f= p3
3@ - tanh /3)’
where p = B/fip. The effective hydraulic permeability of a uniform channel
with a fiber matrix is related to K,,eff by

where L, is the depth of the channel with matrix components and Lj, is the total
cleft length.
The accuracy of the Brinkman approximation and the various approximate
solutions described in this and the previous subsection are shown in Fig. 4, where
the solutions for the increase in the channel drag correction factor f as a function
of the solidity fraction S are plotted for three different aspect ratio fibers B/a =
0.5, 5, and 20. B/a = 20 is representative of a glycoprotein side chain spanning
the wide part of the cleft, B/a = 5 describes the protein core of the proteoglycan,
and B/a = 0.5 is typical of the septal posts in lung alveoli for which the theory
in Lee and Fung (1969) and Lee (1969) was developed. Four sets of curves are
shown. The solid set is the exact infinite series solution in Tsay and Weinbaum
(Ml), the dashed set is the two-dimensional solution in Sangani and Acrivos
(1982) for unbounded infinitely long fibers, the dash-dot set is the two-term
A MODEL FOR CAPILLARY PERMEABILITY 93

:E-3 lE-2 lE-1 1


S

FIG 4. A comparison between the drag coefficients, f, obtained by the three-dimensional exact
solution (-, Tsay and Weinbaum, 1991) the Brinkman approximation (. . . , eq.(lO), the two-term
asymptotic solution (-..-, Lee, 1%9), and the 20 solution (- - -, Sangani and Acrivos, 1982). The long
dashed curve with symbols corresponds to the fiber array when its gap spacing equals B/2.

asymptotic solution previously used in TWP (Lee, 1969), and the dotted set in
the new Brinkman approximation described by Eq. (10). The difficulty with the
two-term asymptotic approximation used in TWP is clearly evident in Fig. 5. This
asymptotic solution accurately describes the additional hydraulic resistance when
the viscous boundary layers surrounding the fibers do not overlap and breaks
down entirely for high aspect ratio fibers when A/B < 1. From the geometry of
the fiber array in Fig. 3, A and S are related by

(12)

The value of S at which A/B = 1 is readily obtained by setting A = B in (12)


and solving for S. For B/a = 0.5, which is the case treated by (Lee, 1969) in his

1.00

l.OE-2

FIG. 5. The ratio of D,,,&D,, for square array of cross-bridging fibers vs the effective fiber solid
fraction S,.
94 WEINBAUM, TSAY, AND CURRY

model for the septal posts in the lung, A/B = 1 when S = 0.5. Thus, for
B/a = 0.5 the two-term approximation is valid for nearly all values of S except
for the dense septal post limit. However, for B/a = 20 the two-term approximation
breaks down for S > 0.0065 or A < 12.0 nm if 2B is 22 nm. The corresponding
values for B/a = 5 are S > 0.063 or A < 15.5 nm. This approximation, therefore,
cannot be used for fiber spacings where the fiber matrix is the primary molecular
filter for molecules smaller than albumin, effective diameter 7.1 nm. In contrast,
the results clearly show that the Brinkman approximate solution is a remarkably
accurate approximation for all values of S when B/a > 5, the aspect ratios of
interest in the present application.
The excellent approximation provided by the Brinkman approximation when
B/u > 5 suggests a very simple solution for a random matrix of cross-bridging
perpendicular fibers in a channel. Instead of using the Sangani and Acrivos solution
for K,, for a square periodic array, we introduce instead in the Brinkman equation
a random cell Carman-Kozeny approximation for the two-dimensional perpen-
dicular array. The Carman-Kozeny approximation is given by

K = (1 - 9”. g )
P
s*
03)
( 1

where C is a fiber density correction factor. When the fibers are circular cylinders
perpendicular to the flow, Happel (1959) used a periodic unit cell model with
vanishing shear at the edge of the periodic unit to obtain the following approximate
expression for C:

2
s [In($) - i: J z1il-l.
c = (1 - S13

(14)

The randomness in the distribution of fibers can now be taken into account using
a stochastic model developed by Yu and Soong (1975). In this model N fibers
are randomly inserted into M subregions and a nonuniform fiber density distri-
bution is generated. Each subregion has a different value of S given as

S(k)
1 = #)1 !?! s
0N ’

when it i(k) is the number of fibers in subregion i. By assuming that there is at


least one fiber in each subregion, a different value of Kp is obtained for each
region. The average value of K,,, which we denote by KPr is calculated as:

(N - M)!
KPr = i i . .. i e)
n,=l n*=l
mu=’ fi (,#d _ 1)!M(-‘)’

i=l (16)
n1 + 122+ ..- + nM = N.

The mean KPr is used to replace Kp in Eq. (6). In the results presented in Fig.
8a and 8b we have chosen N and M as 25 and 10, respectively.
A MODEL FOR CAPILLARY PERMEABILITY 95

4. Diffusive Permeability
In the same way that a separate theory was required to describe the Hele-
Shaw flow through the junction strand with its pores and the effective hydraulic
resistance of the cross-bridging fibers in the wide part of the cleft, separate the-
oretical models are required to describe (i) the basic diffusive resistance of the
cleft with its junction strand and (ii) the added resistance due to fiber matrix
components in the wide part of the cleft for solutes of finite size. The theoretical
model for the lateral spreading of solute and the diffusive interaction between
pores closely parallels the analysis developed in TWP for the derivation of Eqs.
(l)-(4) in the present study, since the governing equation and boundary conditions
for the pressure field in the Hele-Shaw flow through the junction strand are
closely related to the boundary value problem for pure diffusion of a solute through
this same geometry. The details of this solution are therefore omitted and only
the final results are given below. The solution of problem (ii) for the diffusion
of a finite-size solute through a bounded periodic fiber array is much more difficult
since the effective diffusion coefficient D.rw,effin the wide part of the cleft in regions
1 and 3 in Fig. 3 varies spatially and depends on the solution to a complicated
hydrodynamic boundary value problem in which the no-slip conditions are satisfied
on both the solute and channel-fiber surfaces. However, a simplified approximate
expression for Diw,eff in regions 1 and 3 based on the Brinkman approximation
will be proposed, which should provide reasonable accuracy provided the solute
diameter 2r, does not approach the fiber gap spacing A. Current statistical theories
for diffusion in a fiber matrix take account of the steric exclusion of the particles
and introduce a collision-mean-free path to take account of the interaction of the
solute with the matrix fibers using a statistical model for the fiber density distri-
bution, see Ogston et al. (1973). However, this statistical theory does not consider
the hydrodynamic interaction of the solute molecule with the disturbance velocity
field of the matrix. Therefore, the statistical model would be anticipated to un-
derestimate the diffusive resistance of the periodic fiber array considered in Fig. 3.
We consider first the diffusion of the solute through a cleft with a junction
strand with periodic circular or rectangular pores, but no fiber matrix components
in the wide part. For this simplified junction-pore model the diffusion coefficient
in the wide part Di, can be treated as a constant. The finite size of the solute
can be taken into account by considering the additional hydraulic resistance due
to the channel walls using the solutions in Ganatos et al. (1981) for the motion
of a finite sphere for this geometry as described in Curry (1984). In place of Eqs.
(4) and (5) one has for circular or rectangular pores, in that order.
R=2L, D-1-
2
()O Ljt
nri 02
(17)

R=2L, Dl
2
()O
2bd Ljt z’ (18)
where D2, the restricted diffusion coefficient, depends on the cross-sectional ge-
ometry of the pore. Expressions for Dz are given in Curry (1984) for both a
circular pore and a pore that is a slit with parallel walls. The latter is a good
approximation for the rectangular pores considered herein since b < d. In contrast
to the hydraulic resistance, which varies as the fourth power of the pore radius
96 WEINBAUM, TSAY, AND CURRY

or the third power of the slit height, the diffusional resistance of the junction
strand is proportional to the open pore area and thus the junction strand constitutes
a smaller fraction of the total resistance of the cleft.
The solution to the boundary value problem for the concentration field in regions
1 and 3 in Fig. 3 leads to the following expression for the diffusional resistance
in these regions:

where
m tanh (h,LJsin*(h,d)
An3 1

i = 1,3 (19

‘TT(rp - rs)2
A, = F, d=
4 (B - rs)
Here Di, is the restricted diffusion coefficient for a finite sphere in a channel
obtained from Ganatos et al. (1981). Note that (19) is of the same form as (2)
except for multiplicative parameters. The permeability P is related to the diffu-
sional resistances in (17)-(19) by
P = [R, + R2 + R3]-‘. (20)
We consider next problem (ii), the added resistance due to matrix components
in the wide part of the cleft. Expression (19) could be easily modified to take
account of cross-bridging fiber components in regions 1 and 3 if one could find
a suitable expression for Diw,effto replace Di, in these regions. A rigorous treatment
of the additional attenuation of the velocity disturbance from the sphere due to
the presence of the fibers would require a solution that satisfies zero-slip conditions
on the surface of the sphere and all the surrounding fibers. A reasonable ap-
proximation to this shielding behavior for the high aspect ratio fibers of current
interest is suggested by the excellent approximation provided by the Brinkman
equation (6) shown in Fig. 5, when B/a > 5. An exact solution of (6) that satisfies
the no-slip conditions of the channel walls and the solute (sphere) surface is beyond
the scope of the present investigation, but a very good approximation to this
much more difficult problem can be obtained by combining the solution in Ganatos
et al. (1980) for a sphere in a channel and the known solution for a sphere in an
unbounded Brinkman medium.
The solution to Eq. (6) for an unbounded sphere is given in the original theory
of Brinkman (1947). This solution leads to the expression for the drag on the
sphere,
1 r*
F,=6rpVr, l+L+&, (21)
( l&7 p)
where KP for our periodic fiber array is given by equation (7), r, is the sphere
radius, and the term in parentheses is the Brinkman correction to the Stokes drag
on a sphere. Result (21) neglects the walls of the channel, but Fig. 4 shows that
for A < B (see Eq. (12)) the walls of the channel provide only a modest correction
to the unbounded solution of Sangani and Acrivos (1982), whereas for A % B
one should approach the exact limiting solution of Ganatos et al. (1981) in the
same way that the Brinkman approximation (10) approached the solution for
A MODEL FOR CAPILLARY PERMEABILITY 97

Poiseuille flow in a channel when the sphere was not present in the dilute fiber
limit. Therefore, using the Stokes-Einstein relation, we propose the simple ap-
proximate relation for the solute diffusion coefficient &, in the available space
between fibers,
1r2 -’
1 +L+3+ i = 1,3, (22)
I& pI
where Di, is the average diffusion coefficient integrated across the channel height
in the absence of fibers obtained from the solutions in Ganatos et al. (1981).
Equation (22) approaches the correct limiting behavior for both dilute fiber arrays,
where the walls of the channel provide the primary interaction, and more dense
fiber arrays (A < B), where wall effects are not important.
The above expression for Dim still does not include the steric exclusion of the
solute by the fibers and the channel walls and the spatial variation of the solute
concentration in the available space of the matrix. The latter effects can be treated
by solving the solute diffusion equation for the periodic fiber array in Fig. 3 in
which a zero-flux boundary condition is applied at the exclusion radius, r, = a +
r, of the fibers. This last boundary value problem is analogous to the boundary
value problem for the potential pressure field for the Hele-Shaw flow past a
periodic fiber array. The solution to the latter problem in Tsay and Weinbaum
(1991) can be readily applied to derive the expression for Diw,eff,

(23)
where b1 is the coefficient of the leading term of the doubly periodic Wierstrasse
expansion series that is used to describe the disturbance produced by each fiber
in the array. Although (23) contains only the coefficient of the leading term, it
is an exact expression since the value of b1 is determined by solving an infinite
matrix equation that contains the coefficients b, of all the higher order terms.
Once can show that this solution for bl depends only on the effective fiber solid
fraction S, = S(l + r-,/a)‘. The coefficient of Dim is, therefore, only a function
of S, and is plotted in Fig. 5. For a pure fiber matrix model in which there is no
junction strand, P is related to Diw,effby

P = 2BLj, Di, eff, (24)


L ’
whereas if a junction strand is present, Di, in Eq. (17) is replaced by the expression
for Diw,effin Eq. (23). A comparison of the hydrodynamic theory and the stochastic
theory is given in the appendix.

RESULTS AND ANALYSIS


The results for L, and P are presented in three sections: (1) the cleft with only
a junction strand with pores and no fiber matrix components in its wide parts,
(2) a cleft with fiber matrix components distributed uniformly through the entire
depth and no junction strand, and (3) a combined model with both junction strand
and fiber matrix components. In addition, three types of junction pores are ex-
98 WEINBAUM, TSAY, AND CUBBY

2b=22 2d=4
= pb_‘-585 2d=44
.
b4OOkl
2B=22nm
L2=l lnm

10 100 1000 lE4


2D h-4
FIG. 6. Solutions for the hydraulic conductivity for clefts with pores in the junctional strand whose
cross-sectional shape is one of three basic types are shown. Results are shown for cylindrical pores
of 5.5nm radius (-..-), rectangular pores 8 nm high and 44 nm long (-), and pores 22 nm high
and either 44 nm long (...) or 88 nm long (--).

amined: (i) a small circular pore representing individual missing proteins in an


otherwise impermeable junction strand, (ii) a longer rectangular pore with four
missing proteins and an 8-nm slit height (for these rectangular openings the junc-
tion pores would serve as the primary molecular filter since this slit height provides
an optimum fit for the reflection coefficient data for all the intermediate size
solutes between 1.5 and 3.5nm radius, see Fig. 4 in Curry (1986)), and (iii) a
wider rectangular pore with four missing proteins and a 22-nm slit height. In the
latter case the junction strand would contribute to L, and P but a fiber matrix
would need to be present, at least in some portion of the cleft or at its entrance,
since this large pore could not possibly function as the primary molecular sieve.

A exp. results
. 2b=22 2d=44 2D=1220.
-- 2b=22 2d=88 2D=1650’

FIG. 7. Solutions for the diffusive permeability for clefts with pores in the junctional strand having
geometry as in Fig. 6. The spacing between pores (20) is determined from the hydraulic conductivity
data for frog mesentery 5.9 x lo-’ cm/set/cm-HzO. The solid triangles are directly measured values
for the solute permeability.
A MODEL FOR CAPILLARY PERMEABILITY 99

1. Cleft with Junction Strand Only


In Figs. 6 and 7 we have plotted, respectively, the solutions given by Eqs. (1)
to (5) for the hydraulic conductivity L, and Eqs. (17) to (24) for the diffusive
permeability P for a cleft with a junction strand with pores whose cross-sectional
shape is one of the three basic types summarized above. The values for the cleft
depth, L = 400 nm, channel height, 22 nm, and the total cleft length per unit
area, Lj, = 2000 cm/cm2, are based on the measurements for single perfused
vessels in frog mesentery by Clough and Michel, (1988). These data were selected
because morphometric electron microscopic studies were performed on the same
vessels for which the measurements of hydraulic permeability and osmotic re-
flection coefficient for serum albumin were obtained. The measured average value
of L, was 5.9 x 10m7 cm/set/cm H20.
The lowest curve in Fig. 6 is the solution for a circular pore of 5.5 nm radius.
This value for rP is estimated from the measured osmotic reflection coefficient for
albumin u = 0.83 using centerline pore theory (Curry, 1984). The second curve
in Fig. 6 is the solution for a rectangular slit of 8 nm height and 44 nm width.
One notes that to achieve a value of L, of 5.9 x 10m7 cm/set/cm HzO, the
spacing between pores, 20, will be 90 nm for the 5.5-nm radius pore and 370
nm for the rectangular slit. The fractional length of open junction for the circular
pore and the rectangular slits is roughly 12%. Results for the 22-nm-high rectan-
gular pore are shown by the upper two curves in Fig. 6. For a pore 22 x 44 nm
a spacing of 1.22 pm would account for the measured L,. An important aspect
of this model is that L,‘s of the order of 20 x 1O-7 cm/set/cm H20, measured
after albumin is removed from the perfusate, can be accounted for if the pore
spacing is reduced and if the resistance of the fiber matrix is also lowered by the
removal of albumin. For example, with a 44-nm-wide pore a value of L, close
to 20 x 10e7 cm/set/cm H20 can be achieved with a pore spacing of 480 nm.
In contrast, with the 5.5~nm-radius pore these large values of L, are not possible
for any pore spacing. The horizontal line in the left-hand margin of Fig. 6 is the
value of L, for a cleft without a junctional strand.
In Fig. 7 the corresponding solutions for the solute permeability P have been
plotted as a function of solute radius for the four different pores shown in Fig.
6. In obtaining the predicted curves, the averge periodic spacing 20 between
pores is first determined from Fig. 6 by requiring L, to be equal to the average
measured value for frog mesentery 5.9 x 10e7 cm/set/cm H20 cited above. The
solid triangular symbols in Fig. 7 are directly measured values for the solute
permeability that have been obtained for a spectrum of solutes up to the size of
albumin, 3.5 nm radius, in studies using single perfused frog mesentery capillaries.
The continuous theoretical curves for the solute permeability P use the Stokes-
Einstein relation to determine the solute diffusion coefficient in an unbounded
free solution.
In Fig. 7 the predicted values of P for 5.5-nm-radius pores closely fit the
measured data for very small solutes and also provide a reasonable approximation
for albumin, but deviate significantly from the measured values (by a factor of
2- to lo-fold) for all intermediate size solutes. The combined results for the 5.5-
nm pore thus closely satisfy the measured values for u and L, and small solute
permeability. The deviation of the predicted curves in Fig. 7 from the measured
values of P for a 5.5-nm pore is an indication of the inability of the junction-
100 WEINBAUM, TSAY, AND CURRY

circular pore model to describe the full range of permeability data on the perme-
ability and selectivity of the capillary wall in frog mesentery. The reason is that
although the pores in the junctional strand are the major resistance to water, the
pores being quite short (11 nm), offer relatively little resistance to the diffusion
of solutes until molecules the size of albumin are approached. This is not the
pattern displayed by the experimental data in which permeability decreases quite
significantly with successive increases in molecular size over the entire range of
solute sizes.
Figure 7 also shows the same computations for S-nm-high rectangular junction
pores that are 44 nm wide. These values correspond to periodic junction discon-
tinuities, which would be created by four missing proteins assuming each protein
is 11 nm wide. The 44-nm-wide break is typical of a single standard section
thickness. The predicted curves for P in Fig. 7, which satisfy a value of L, of
5.9 x 10e7 cm/set/cm HzO, show that the 44-nm-wide rectangular slit model
with an 8-nm gap height that satisfies the requirements for the osmotic sieve, fits
the measured data for small solutes up to the size of sucrose (0.48 nm radius),
but substantially overestimates the measured data for P for all the larger solutes,
including albumin, which could be better matched by the 5.5nm circular pore.
The values of the small-solute permeability coefficients calculated in Fig. 7 for
the 22-nm-high rectangular pores, which account for the measured L, are about
one half the experimental values. This reflects the fact that, per unit of open
junctional strand, the hydraulic conductivity of these junctional pores is larger
than for the more restrictive pores. For the larger solutes the predicted values of
P are too large because the gap height of the 22-nm-high rectangular pore is too
large to offer significant resistance to solutes between 0.8 and 3.5 nm radius. It
is unlikely that this junction-pore model could exist by itself and fiber matrix
components would have to play an important role to account for the osmotic
sieving properties of the cleft. The importance of this model is that it is flexible
enough to enable an additional selective barrier to be inserted. In the next section
the properties of a cleft filled with fiber matrix alone is considered first.

2. Cleft with Fiber Matrix Only


All the results in the preceding section are for a cleft in which there is no fiber
matrix present in its wide portions. In Fig. 8a we have plotted the various solutions
for L, for a cross-bridging proteoglycan matrix in a uniform cleft without junctional
strands with cross-bridging fibers of aspect ratio B/a = 18.3. The cleft height and
depth again correspond to the measurements of frog mesentery interendothelial
clefts in (Clough and Michel, 1988). Starting from the left the curves in this figure
represent the two-term asymptotic approximation in TWP, the square periodic
array solution given by either the infinite series solution in Tsay and Weinbaum
(1991) or the Brinkman approximation for L, obtained from equations (9)-(ll),
the Brinkman approximation for the statistically random perpendicular array based
on the value of Z$ described by Eqs. (15) and (16), and the original random
array solution for an infinite medium used by Curry and Michel, (1980) where
the Carman-Kozeny parameter is C = 5.
One notes in Fig. 8a that there is a relatively small difference between the
solutions for the square and random arrays at the fiber densities of interest, S <
0.02 or A > 6.3 nm, and that the two-term solution used in TWP, as suggested
by the results in Fig. 4, deteriorates rapidly for S > 0.005. One also observes
A MODEL FOR CAPILLARY PERMEABILITY 101

4
I I
1 .bE-3 1 .OE-2 0.1
Solidity Ratio S

b
100.0
In
0

; 10.0

IIT
ma,
3-r: 1.0
ZE
6.5
CL
P 0.1
3 without junction strand
isi
1 .OE-2
0.0 1.0 2.0 3.0 4.0 5.0

SOLUTE RADIUS (nm)


FIG. 8. (a) Solutions for hydraulic conductivity for ciefts with fiber matrix present only. The figure
shows various solutions for the hydraulic conductivity I!,, for a cleft with cross-bridging fibers of B/a
= 18.3. From the left, the curves represent two-term asymptotic approximation (TWP, 1989) (.a*),
square array solution (Tsay and Weinbaum, 1991) (-), the Brinkman approximation for two-dimen-
sional random array (-..-), and the Carman-Kozney solution with C = 5 (Curry and Michel, 1980)
(- - -). (b) Solutions for the solute permeability P as a function of the solute radius r,. The solid curve
is for a = 0.6 nm, dashed curve for a = 2 nm, and dotted curve for a = 5. The top three curves
represent solutions for clefts with random cross-bridging fibers and the bottom ones are for periodic
fiber arrays.

that the measured value of L, of 5.9 x lo-’ cm/set/cm Hz0 can be achieved
by a fiber matrix in a cleft without junctional strands with a solid volume fraction
that lies between 0.017 and 0.025 for the square periodic and random arrays,
respectively. A fiber matrix with these values of S, which fills the entire cleft,
reduces the hydraulic conductivity of the cleft by more than sevenfold if the
resistance of the junction strand is neglected. When arranged in square array, a
matrix of 0.6-nm-radius fibers has an open spacing between fibers A of 7 nm when
S = 0.017. This is close to the spacing required if the fiber matrix is to serve as
the primary molecular sieve. It follows that if there is any additional resistance
102 WEINBAUM, TSAY, AND CUBBY

to water movement through the cleft from the junctional strand, there must be
reduced resistance to water flow from the matrix to satisfy the measured value
of L,. Therefore, if one requires that the matrix continue to be the primary filter,
the matrix must occupy less than the total cleft depth.
The solutions for the solute permeability P as a function of the solute radius
r, for a cleft with either periodic or random cross-bridging fibers are shown in
Fig. 8b. These calculations are based on Eqs. (22) and (23) for the ordered square
array. For the random array the statistical theory in Ogston et al. (1973) has been
used to calculate the steric exclusion of the fibers, while the hydrodynamic pre-
dictions to determine the relation between L, and S are based on Eqs. (11) and
(16). Curves for three fiber radii are presented and for each curve the fiber solid
fraction S has been selected to satisfy the measured value of L,, 5.9 x lop7
cm/set/cm H20, for frog mesentery. The results show that the fiber matrix offers
little resistance to small solutes as indicated by the fact that the predicted perme-
ability coefficient of the fiber-filled cleft for solutes <.48 nm radius is close to
that calculated for an open cleft. The present higher estimates for small solute
permeability compared with previous fiber matrix calculations for frog mesenteric
capillaries (Curry and Michel, 1980; Curry, 1986; Michel, 1988) reflect the larger
value for cleft width and reduced cleft depth used in the present model. To account
for small-solute permeability it is, therefore, necessary to assume that junctional
strands act to reduce the effective length of open junction. Other combinations
of fiber radius and fractional fiber density describe a relation between permeability
and solute size for molecules up to 2 nm in radius that is similar to that for the
matrix with 0.6-nm fiber radius, but these combinations fail to describe the large
solute permeabilities due to the wider spacing between fibers. The calculations
in Figs. 6-8 indicate that neither the junction pore or fiber matrix model by itself
can account for all the data.

3. Combined Model for Cleft with Junction Strands and Fiber Matrix
In Fig. 9a the solutions for the junction pore interaction given by Eqs. (l)-(5)
are combined with the solution for pen for the fibers in the wide part of the
channel. Three sets of curves are shown, a junction strand with a circular cylindrical
pore of 5.5 nm radius and two rectangular slit pores of 44 nm width, one with a
gap height of 8 nm and the other an unconstricted height of 22 nm. In each case
there are fibers of 0.6 nm radius and a spacing A = 7 nm (S = 0.017) filling the
wide portion of the cleft. As expected from our results presented above, there
is a limited range of parameters that are consistent with the experimental data
when the matrix fills the whole cleft. The predictions in Fig. 9a indicate that the
measured value of L, = 5.9 x 10m7 cm/set/cm Hz0 cannot be achieved by a
junction strand with 5.5-nm-radius circular pores even if every other protein is
missing (20 = 22 nm). If we consider the latter junction strand as the most
permeable possible for the circular pore geometry, the model predicts that the
maximum value of S for which Clough and Michel’s measured value of L, can
be achieved is S = 0.01 (A = 10 nm) if the matrix fills the entire wide part of
the cleft. For the rectangular slit geometry with 8 nm gap height and 44 nm width,
with A = 7 nm, L, approaches a limiting value close to 5.5 x 10M7 cm/set/cm
HZ0 when there is only a single protein separating junctional pores (20 = 55
nm). For the 22 x 44-nm rectangular pore this limiting value of L, is 6 x 1O-7
cm/set/cm H20. Based on these predictions, the fiber matrix could not serve as
A MODEL FOR CAPILLARY PERMEABILITY 103

‘.. 2b=22 2d=44 nm


-- 2b=6 2d=44 nm

a=0.6nm. S=.O17

5 10 100 1000 lE4


2D (nm)

A experimental results

A
i?i
l.OE-2 ....‘...“,““..‘.‘.‘.,
0.0 1.0 2.0 3.0 4.0 5.0

SOLUTE RADIUS (nm)


FIG. 9. Solutions for the hydraulic conductivity and the diffusive permeability for clefts with
junction strands and fiber matrix are presented in (a) and (b). The set of curves shows junction-
pore-matrix model with fiber density, S = 0.017. The three curves shown in (a) are for junction
strands with (i) circular pores for 5.5nm radius; (ii) rectangular slit pores of 8 x 44 nm, and (iii)
rectangular pores of 22 x 44 nm. In (b), results are presented only for the cases in (a) in which the
measured value of L, can be achieved and in which the fiber spacing A is of the dimensions for an
adequate molecular sieve for albumin.

the primary molecular sieve if the matrix is to occupy all or nearly all of the wide
part of the cleft with either the 5.5nm radius or S-nm gap height rectangular
pores. However, matrix components with primary sieving properties filling the
entire cleft are just on the borderline of possibility for the larger 22-nm gap height
pore.
An important property of the combined junction-pore-matrix model is that
for all pore spacings 20 > 200 nm, the presence of a fiber matrix reduces the
L, relative to that of a cleft without matrix by approximately the same extent for
all spacings between the junctional pores for each matrix geometry. This reflects
the fact that all the L, curves in Figs. 6 and 9a are parallel in this region. For
104 WEINBAUM, TSAY, AND CURRY

example, the addition of a matrix formed by 0.6-nm fibers (S = 0.017) decreases


the L, of a cleft with junctional pores (2d = 44 nm, 2b = 22 nm) spaced 200
nm apart from 24 x 10d7 cm/set/cm Hz0 without matrix (see Fig. 6) to 4.0 x
10e7 cm/set/cm HZ0 (a factor of 6) and decreases the L, of a cleft with the same
pores spaced a distance of 600 nm apart from 10 x 10e7 cm/set/cm HZ0 (see
Fig. 6) to 1.7 x 10m7 cm/set/cm HZ0 (also a factor of 6).
Solutions for P are shown in Fig. 9b only for the single case in which the
measured value of L, can be achieved. This corresponds to the 22-nm gap height
rectangular pore, where the fiber spacing is just adequate for the matrix to be
the primary molecular sieve when the pore spacing 20 = 55 nm.

DISCUSSION
The calculations described in this paper provide new detailed theoretical pre-
dictions to evaluate some of the principal questions in current studies of the
structure-function relationship for water and solute flow through the interen-
dothelial clefts. While none of the present models provide an entirely satisfactory
description of the available experimental data, two fundamental new observations
can be drawn from the calculations. One is that widely separated small breaks in
the junctional strand represented by the 22 x 44-nm rectangular pore are con-
sistent with the measured value of L, in Clough and Michel, 5.9 x low7 cm/set/cm
HzO. The solutions in Fig. 6 for this rectangular pore show that the spacing can
be as large as 20 = 1220 nm if no matrix is present. However, to account for
the measured small-solute permeabilities and the selectivity of the microvessel
wall to large molecules, these gaps are too far apart and a fiber matrix is required
in some portion of the cleft to further reduce the hydraulic conductivity and
provide the molecular sieve. There are a variety of combinations of pore spacing,
matrix density, and depth of matrix penetration into the cleft that need to be
investigated in future extensions of the present model. The second observation
is that the small sieving pores in the junctional strand (the 5.5-nm-radius circular
pores spaced 90 nm apart or the 8 x 44-nm rectangular pores spaced 370 nm
apart), with no additional resistance due to a matrix, can account for the measured
L,, the small solute permeability, and also the measured selectivity properties.
These small-pore models, however, do not properly describe the data for inter-
mediate size solute permeability.
The above calculations also indicate two experimental approaches for future
studies. One is that it should be possible to determine the site of the principal
molecular filter in the cleft based on the frequency of discontinuities in the junc-
tional cleft. Specifically, if small (5.5-nm radius) junctional pores are the principal
molecular filter, then they must be present at a very high frequency in the junc-
tional strand (approximately 1 protein in 8 missing 20 = 90 nm). On the other
hand, if the fiber matrix is the principal molecular filter, larger discontinuities
spaced approximately several hundred to lo3 nm apart are required to account
for the small solute permeability. Preliminary experimental data that address these
considerations are described at the end of the discussion.
Before we elaborate on the experiments mentioned above, we first discuss the
approximations and limitations of the model and the analysis. The approximations
fall under two headings: approximations related to the hydrodynamic modeling
A MODEL FOR CAPILLARY PERMEABILITY 105

and approximations related to the assumed geometry of the cleft and its junctional
structures. These are discussed separately.

Approximations in the Hydrodynamic and Diffusion Models


The principal motivation for the development of the three-dimensional junc-
tional model is to account for the lateral spreading of water and solute proximal
and distal to the junction strand pore in the wide portions of the cleft. This
spreading has been ignored in all previous one-dimensional models (Fig. 1). The
main approximation in the junction-pore model is that the junction strand array
can be replaced by an idealized single equivalent strand with periodically spaced
pores whose average spacing is the same as in the junction strand array. As
expected the largest deviations between the 1-D and 3-D models for the wider
rectangular pores arise whenever the wide portions of the cleft are the major
contribution to the total resistance. This appears to be nearly always the case for
solute diffusion. The agreement between the 1-D and 3-D models is particularly
poor when the spacing between pores is large. For example, for the 22 nm x 44
nm rectangular pore that satisfies the measured L, in Fig. 6 (20 = 1220 nm),
the junctional region would account for only 11% of the resistance to a small
solute, whereas the junctional region would account for 44% of the diffusive
resistance in the 1-D model. If one were to plot solute flux lines (these lines are
everywhere perpendicular to isoconcentration lines), one would observe that when
the pores are far apart these lines would be confined to small wakelike regions
surrounding each pore exit and thus not fill the wide part of the cleft. When this
occurs 1-D theory grossly underpredicts the junctional resistance of the cleft since
it neglects the finite distance for the lateral spreading of the solute to fill the wide
portion of the cleft on the downstream side of the junction strand barrier.
The behavior, just described, is also observed for filtration, where the fluid
streamlines leaving the pore exit replace the solute flux lines. The principal dif-
ference is that in the case of filtration the junctional strand offers a more significant
fraction of the total cleft resistance when the pore dimensions are sufficiently
small for the junction strand to function as the primary size selective structure
(5.5nm pores and S-nm-high rectangular pores). In this case the model for the
pores is more important than the lateral spreading of the water downstream of
the junction strand. In contrast, for the larger 22 x 44-nm rectangular pore one
finds that the junction strand provides 15% of the total hydraulic resistance when
20 = 1226 nm. This is less than one-third the value predicted by the 1-D model.
The principal approximation introduced in the treatment of the ordered fiber
array is that the fibers can be treated as rigid, periodic structures arranged at
right angles to the direction of flow. The solution for L, for this geometry is close
to exact. The important insight gained from the hydrodynamic analysis of water
flow through this mathematically idealized geometry is that the increased viscous
drag is due primarily to the additional viscous dissipation that occurs when the
viscous layers surrounding each fiber overlap. This behavior is confirmed by the
detailed theoretical analysis of the three-dimensional velocity profiles for water
flow around each fiber in Tsay and Weinbaum (1991), which show that the thick-
ness of the viscous layers is of order B. Thus, if B/A > 1 the region of viscous
overlay will fill the entire region between adjacent fibers. This increase in resistance
can occur even if the fiber density is low (B/a + 1). As seen in Fig. ga, for values
106 WEINBAUM, TSAY, AND CURRY

of fiber density ~0.1, the new model predicts a much larger resistance than the
Carman-Kozeny equation. For example, the fiber volume density S = 0.017
predicted by the present model to achieve an L, of 5.9 x 1O-7 cm/set/cm Hz0
is significantly less than the value of S = 0.05 obtained in the original calculations
of Curry and Michel (1980). The latter value was based on a Carman-Kozeny
equation in which the density correction factor C in Eq. (13) was assigned a
constant value of 5, representative of the dense fiber limit. As observed in Fig.
8a, the new model approaches the predicted curve for C = 5 in Curry and Michel
(1980) for S > 0.1, but the result deviates significantly from the predictions of
the Carman-Kozeny relation at lower value of S. As shown in Levick (1987) the
Kozeny factor C needs to be increased from 5 to more than 20 when S is decreased
from 0.1 to 0.01 if it is to fit the experimental data.
The principal approximation used in the development of the new model for
diffusion in the ordered array is that the decrease in effective diffusion coefficient
in the matrix can be accounted for by the increased drag on the solute, (Eqs.
(21) and (22)) and the increased exclusion of the solute imposed by the hydro-
dynamic flow between the fibers, (Eq. (23)). Equation (23) includes not only the
excluded volume effect due to the presence of the fibers but also takes into account
the nonuniform concentration distribution of the solute that arises from the in-
teraction between the fibers. The approach is quite different from that used in
the stochastic description of the reduction in diffusion within a matrix introduced
by Ogston et al. (1973). The latter approach describes the reduction in the diffusion
coefficient in terms of the probability of collision between solute and fibers and
the classical excluded volume effect of the fibers. The predictions of the Ogston
approach are compared with the new hydrodynamic formulation in Appendix 1
for the case S = 0.017 or A = 7 nm. It is shown for small and large solutes that
the reductions in diffusion coefficient using the new hydrodynamic formulation
are quite close to those of the Ogston model for the case in which the fibers are
in an ordered array, even though the relative importance of the exclusion and
drag/collision terms differ significantly for the two approaches.
While the present theory for solute diffusion includes, for the first time, the
far-field hydrodynamic interaction of the solute molecule with the surrounding
fibers, it does not accurately describe the near-field interaction with the closest
neighboring fibers. This near field interaction has not previously been studied,
since the undisturbed velocity field in a bounded matrix in the absence of the
solute molecule was not known. The new solutions in Tsay and Weinbaum (1991)
for the velocity field in the fiber array, when combined with the Faxen reciprocal
theorems for Stokes flow, should enable us to examine problems of this type in
the future.

Approximations Related to Assumed Cleft and Pore Geometry


The data for cleft depth, cleft height, and effective cleft length per unit area
in Clough and Michel (1988) are mean values based on measurements in single
perfused microvessels in which vessel hydraulic conductivity was also measured.
While this is arguably the best data available to correlate structure and function,
the shortcomings of the data should be noted. The length of junction per unit
area 2000 cm/cm’ is similar to that estimated by Bundgaard and Frokjaer-Jensen
(1984) in frog mesentery and Bundgaard (1984) in the rat heart. The cleft height
(22 nm) is larger than that reported by Mason et al. (1979) and Bundgaard and
A MODEL FOR CAPILLARY PERMEABILITY 107

Frokjaer-Jensen (1984) (17 nm) and the cleft depth (0.395 pm) is shorter than
the value of 0.7 pm described in these earlier publications. There is no simple
explanation for these discrepancies. In addition, there was considerable variation
in the measured value for cleft depth both within a given vessel, and between
vessels. For example, the value of 0.395 pm for the cleft depth is calculated from
a population that varied from 0.104 pm to 1.7 pm.
Measured values of L, in Clough and Michel (1988) also varied over a seven-
fold range from 1.84 to 12.5 x lo-’ cm/set/cm H20. Clough and Michel found
a reasonably good correlation between L, and the inversion of the harmonic mean
cleft depth (Z/L). This correlation would not be expected if the pores in the
junction strand are the primary resistance to water flow. The Z/L correlation
suggests that the resistance resides in the wide part of the cleft for frog mesentery.
Thus, although it was noted above that the calculations for L, based on mean
values for cleft geometry might be compatible with a junctional pore model with
5.5-nm pores in the strand, this model is not compatible with observed relation
between L, and Z/L because the predictions of our 3-D model for the junction
pores with rP = 5.5 nm and 20 = 90 nm show that only 18% of the total
resistance to flow resides in the wide part of the cleft. Thus the variation in the
depth of the wide portion of the cleft can not be explained by this small circular
pore model for the junction pores. In contrast, our predictions for the 22 x 44-
nm rectangular pore with 20 = 1220 nm that satisfy the measured value of L,
require that only 15% of the total hydraulic resistance resides in the junction
strand. This model is, therefore, more consistent with the Z/L correlation.
Another area of uncertainty concerns the geometry of breaks in the junctional
strand. There is no definitive data on the geometry of the junctional strands. The
size of individual missing proteins is at best an educated guess, constrained as
much by the measured selectivity of the vessel wall as any precise knowledge of
the channel geometry. There is also no definitive data on the effective depth of
the strand. In one-dimensional models (see Fig. 1) the depth of the channel in
the region of the junction strand has been assumed to be as much as 10% of the
total cleft depth, while in the present model this depth has been related to the
diameter of the missing proteins, 11 nm or <3% of cleft depth. This assumption
reduces the relative contribution of the junctional strand to the total resistance.
An important assumption in the fiber matrix model is that the fibers are dis-
tributed throughout the cleft depth. There is no clear evidence that fibers are
present in the cleft. The fibers described here are conjectured to be either gly-
coprotein side chains of typically 0.6-nm radius (Curry and Michel, 1980; Curry,
1986) or core proteins of 2- to 2.5-nm radius, whose length is equal to the gap
height of the wide part of the cleft. For bridging molecules within the cleft, the
most likely fiber orientation is perpendicular to the wall. The evidence to support
this suggestion is from studies of proteoglycan molecules similar to those described
above, which have been found to perpendicularly span the space between neigh-
boring collagen fibrils in subendothelial connective tissue matrix using rapid freeze-
etching techniques (Frank and Fogelman, 1989). It is noted that conventional EM
techniques do not show bridging fibers since these structures do not withstand
chemical fixation. Rapid freeze-etching EM techniques are thus likely to be es-
sential tools for investigating the structure of fiber matrices in the junction and
on the cell surface in the future. It is noted that there is evidence that fibers
associated with the endothelial cell surface glycocalyx account for 50% of the total
108 WEINBAUM, TSAY, AND CURRY

hydraulic resistance in frog mesentery , Adamson (1990). This surface glycocalyx


is not included in the present model. At the end of the discussion we shall estimate
the resistance of such a layer.
We also note that the model only indirectly accounts for the presence of ad-
sorbed proteins that are assumed to be part of the matrix. Michel (1984) has
hypothesized that one of the important actions of albumin may be to order the
matrix. This ordering is suggested to arise from the electrostatic interaction be-
tween positively charged arginine groups on albumin and the negative charges on
the proteoglycans. Thus the ordered array described in the model may be rep-
resentative of the matrix structure in the presence of albumin. The additional
resistance due to the albumin molecules, themselves, has been neglected in the
model. The diffusion model also ignores the effect of charge on the channel walls
and/or fibers. Charge effects reduce the permeability of negatively charged in-
termediate-size solutes to values that are l/3 to l/5 of the uncharged solutes of
the same size (Adamson et al., 1987; Curry et al., 1989). These effects are roughly
of the same order of magnitude as the deviation between theory and experiment
for the 1-Znm-radius solutes shown in Fig. 7 for the 22 x 44-nm pore.

Structure-Function Correlation: Future Studies


In spite of the limitations and the uncertainties in the model there are important
predictions to be investigated in new experiments. These are discussed below. If
small breaks in the junction strand typical of a single EM thin section separated
by distances of a few hundred to 1000 nm modulate small-solute permeability and
the effective area for flow, it should be possible to determine their size and
frequency from the lateral spreading of electron-dense tracers in the wide part of
the cleft distal to the junctional strand. The spreading of the tracer can be described
as a wake. Clough and Michel (1988) showed that, for the microvessels used to
obtain the ultrastructure data used throughout this paper, the electron-dense tracer
lanthanum was present in 50% of the random thin sections examined in EM.
They interpreted these results as indicating that 50% of the junction strand was
open to diffusion for small solutes and water flow. If, however, there is lateral
spreading due to convection and diffusion on the distal side of the junction strand,
the wake of a single break will occupy a number of section thicknesses, and far
fewer breaks are required to explain the data. The number of breaks and their
effective size could be determined from the three-dimensional information that
would be obtained from a more detailed study of the wake structure. The present
arguments provide a strong rationale for an extended study combining the ap-
proach of Clough and Michel and the theoretical framework outlined in this paper.
Preliminary studies of this nature have been started (Adamson and Michel, 1991).
The shape of the solute wake of the junction strand will also be influenced by
the filtration flow if convection and diffusion are of comparable importance.
Adamson (1990) has recently shown that the treatment of the endothelial cell
surface with proteolytic enzymes to remove all or a part of the endothelial cell
glycocalyx reduces the hydraulic conductivity of the microvessel wall by roughly
one half for frog mesentery. Gross ultrastructural observations suggests that the
junction strand and cleft geometry are not altered. However, there is no quan-
titative data to show that the frequency of the discontinuities in the junction strand
has not been changed. Assuming that a significant portion of the resistance to
water flow resides in a fiber matrix extending into the microvessel lumen proximal
A MODEL FOR CAPILLARY PERMEABILITY 109

to the endothelial cleft, a rough estimate of the thickness of this matrix layer can
be deduced from the results in Figs. 6 and 9a for the 22 x 44 nm pore. As seen
in the Results section, L, for this rectangular pore is increased by approximately
a factor of six (in the region where the L, curves are parallel) when a selective
matrix (S = 0.017, fiber radius = 0.6 nm, A = 7 nm) is removed from the entire
depth of the cleft, suggesting that the matrix components have roughly six times
the resistance of the cleft with its junction strand. Therefore, a sieving matrix
with A equal to 7 nm occupying about one-sixth the cleft depth, or roughly 60-
70 nm, would offer a resistance that is comparable to a cleft with a junctional
strand only. From Fig. 9a the L, of a cleft with a junctional strand composed of
pores 22 x 44 nm spaced 600 nm apart is 10 x lo-’ cm/set/cm H20. Since the
resistance of this cleft would be doubled if there is a fiber matrix, 60-70 nm thick,
near the cleft entrance, L, would be reduced to 5 x lo-’ cm/set/cm HZ0 if this
thin fiber layer were present. A combined model that includes 22 x 44-nm
rectangular pores and a fiber matrix that occupies only a small fraction of the
cleft depth is thus compatible with Adamson’s measurements for L,. Note that
the above result would not be predicted using the Carman-Kozeny equation
because the predicted resistance of the thin matrix layer would be too small to
reduce L, sufficiently.
The foregoing hypothesis could be more critically explored if similar experiments
were performed to also assess the changes in solute permeability and selectivity
that would result from the removal of the surface glycocalyx for different size
solutes. To interpret these results the present theory would have to be modified
by including a fourth region at the entrance to the cleft in Fig. 2 in which a thin
region of ordered matrix elements would be present. This layer would be treated
using the same Brinkman approach that has been described in the present study.
In this manner it should be possible to evaluate the relative contributions of the
junctional strand and a thin matrix layer to the diffusive resistance and molecular
selectivity.

APPENDIX
The diffusive permeability coefficient of the cleft, neglecting channel wall effects,
is determined by both the solute diffusion coefficient in the cleft (D/Do) and the
partition coefficient for the solute in the matrix @. The reduction in permeability
coefficient due to the presence of the matrix, P/PO, is proportional to the product
(D/D0 . @ (Curry, 1984). The Ogston model for a random array predicts
P/PO = [exp - fi (1 + r,/a)] * [exp - S &,/a + rfla”)]. (Al)
The first term on the right describes the reduction in diffusion coefficient due
to increased collisions of the solute with the fibers. The second term on the right
describes exclusion of the solute from the matrix (Curry, 1984).
The relation for a random array overestimates P/P,, when the matrix is ordered
as in the hydrodynamic model. For an ordered array the exponential terms on
the right side of Eq. (Al) are replaced by linear relations. For an ordered array
Curry (1985) used the term fi . (1 - 2r,/$a) to describe the reduction in
diffusion coefficient. Michel(l984) used the term (1 - S,) to describe the partition
coefficient.
110 WEINBAUM, TSAY, AND CURRY

In the hydrodynamic model, the corresponding equation for P/P,, with channel
wall effects neglected, is given by the product of text Eqs. (21) and (23):
p/p, = [Cl + r,/l&) - ~~/3QJ)l-’ [(I - b%l(1 + h&)1. WI
Values calculated for the Ogston model and the hydrodynamic models using Eqs.
(Al) and (A2), respectively, are compared below.
TABLE Al

Ogston S = 0.017 Hydrodynamic S = 0.017

r, (nm) Diffusion Exclusion P/PO Diffusion Exclusion P/PO


0.2 0.951 0.970 0.922 0.928 0.94 0.870
0.5 0.878 0.943 0.828 0.832 0.89 0.740
1.0 0.755 0.880 0.664 0.701 0.78 0.547
2.0 0.510 0.682 0.341 0.513 0.52 0.207
3.0 0.264 0.390 0.102 0.070 0.22 0.085
3.5 0.129 0.190 0.031 - -

Note. In Equations Al and A2, call term one on the right “Diffusion” and term two on the right
“Exclusion.”

ACKNOWLEDGMENTS
The first two authors acknowledge NIH Grant HL 19454 and NSF Grant CBT-8803116 for the
support of those aspects of this research related to endothelial filtration in arteries. Research related
to microvessels is supported by NIH Grant HL44485.

REFERENCES
ADAMSON, R. H. (1990). Permeability of frog mesenteric capillaries after partial pronase digestion of
the endothelial glycocalyx. J. Physiof. 428, 1-13.
ADAMSON, R. H., HUXLEY, V. H., AND CURRY, F. E. (1987). Single capillary permeability to proteins
having similar charge but different size. Am. J. Physiol. 254, H304-H312.
ADAMSON, R. H., AND MICHEL, C. C. (1991). Ultrastructure of pathways through intercellular clefts
of frog mesenteric capillaries. J. Physiol. 435, 26.
BRINKMAN, H. C. (1947). A calculation of the viscous force exerted by a flowing fluid on a dense
swarm of particles. Appl. Sci. Res. Sect. A l(l), 27-34.
BUNDGAARD, M. (1984). The three-dimensional organization of tight junctions in a capillary endo-
thelium revealed by serial section electron microscopy. J. Ultrustruct. Res. 88, 1-17.
BUNDGAARD, M., AND FROKIAER-JENSEN, J. (1982). Functional aspects of the ultrastructure of terminal
blood vessels: A quantitative study of consecutive segments of the frog mesenteric microvasculature
Microvasc. Res. 23, l-30.
CASELY-SMITH, J. R., GREEN, J. L., AND WADEY, P. J. (1975). The quantitative morphology of skeletal
muscle capillaries in relation to permeability. Microvusc. Res. 10, 43-64.
CLOUGH, G., AND MICHEL, C. C. (1988). Quantitative comparisons of hydraulic permeability and
endothelial intercellular cleft dimensions in single frog capillaries. J. Physiol. 405, 563-576.
CRONE, C., AND LEVITT, D. (1984). Capillary permeability to small solutes. In “Handbook of Phys-
iology” (E. M. Renkin and C. C. Michel, Eds.), Sect. 2, Vol. 4, pp. 411-466. American Physiological
Society, Bethesda, MD.
CURRY, F. E. (1985). Effect of albumin on the structure of the molecular filter at the capillary wall.
Fed. Proc. 44, 2601-2613.
CURRY, F. E. (1984). Transcapillary exchange. In “Handbook of Physiology” (E. M. Renkin and
C. C. Michel, Eds.), Sect. 2, Vol. 4, pp. 309-374. American Physiological Society, Bethesda, MD.
A MODEL FOR CAPILLARY PERMEABILITY 111

CURRY,F. E. (1986). Determinants of capillary permeability: A review of mechanisms based on single


capillary studies in the frog. Circ. Rex 59, 367-380.
CURRY, F. E., AND MICHEL, C. C. (1980). A fiber matrix theory of capillary permeability. Microvasc.
Res. 20, 96-99.
CURRY, F. E., RUTLEDGE, J. C., AND LENZ, J. F. (1989). Modulation of microvessel wall charge by
plasma glycoprotein or orosomucoid. Am. J. Physiol. 257, H1356-H1359.
FIRTH, J. A., BAUMAN, K. F., AND SIBLEY, C. P. (1983). The intercellular junctions of guinea-pig
placental capillaries: A possible structure basis for endothelial solute permeability. J. Uftrastruct.
Res. 85, 45-57.
FRANK, J. S., AND FOGELMAN, A. M. (1989). Ultrastructure of the intima in WHHL and cholesterol-
fed rabbit aortas prepared by ultra-rapid freezing and freeze-etching. J. Lipid Res. 30, 967-978.
GANATOS, P., WEINBAUM, S., AND PFEFFER, R. (1980). A strong interaction theory for the creeping
motion of a sphere between plane parallel boundaries. 2. Parallel motion. J. Fluid Mech. 99(4),
755-783.
GANATOS, P., WEINBAUM, S., FISCHBARG, J., AND LIEBOVITH, L. (1981). A hydrodynamic theory for
determining the membrane coefficients for the passage of spherical molecules through an intercellular
cleft. Adv. Bioeng. 3, 193-196.
HAPPEL, J. (1959). Viscous flow relative to arrays of cylinders. AIChE J. 5, 174-177.
HUXLEY, V. H., AND CURRY,F. E. (1985). Albumin modulation of capillary permeability: Test of an
adsorption mechanism. Am. J. Physiol. 248, H264-H273.
LEE, J. S. (1969). Slow viscous flow in a lung alveoli model. J. Biomech. 2, 187-198.
LEE, J. S., AND FUNG, Y. C. (1969). Stokes flow around a circular cylindrical post confined between
two parallel plates. J. Fluid Mech. 37, 657-672.
LEVICK, J. R. (1987). Flow through interstitium and other fibrous matrices. Q. J. Exp. Physiol. 72,
409-438.
MASON, J. C., CURRY, F. E., WHITE, I. F., AND MICHEL, C. C. (1979). The ultrastructure of frog
mesenteric capillaries of known filtration coefficient. Q. J. Exp. Physiol. 64, 217-233.
MICHEL, C. C. (1984). Fluid movements through capillary walls. In “Handbook of Physiology” (E. M.
Renkin and C. C. Michel, Eds.), Sect. 2, Vol. 4, pp. 375-410. American Physiological Society,
Bethesda, MD.
MICHEL, C. C. (1988). Capillary permeability and how it may change. J. Physiol. 404,
l-29.
OGSTON, A. G., PRESTON, B. N., AND WELLS, J. D. (1973). On the transport of compact particles
through solutions of chain polymers. Proc. R. Sot. London, Ser. A. 333, 297-316.
PERL, W. (1971). Modified filtration-permeability model of transcapillary transport-A solution of the
Pappenheimer pore puzzle. Microvasc. Res. 3, 233-251.
PERRY, M. A. (1980). Capillary filtration and permeability coefficients calculated from measurements
of interendothelial cell junctions in rabbit lung and skeletal muscle. Microvasc. Res. 19, 142-157.
SANGANI, A. S., AND ACRIVOS, A. (1982). Slow flow past periodic arrays of cylinders with application
to heat transfer. Int. 1. Multiphase Flow 8, 193-206.
SILBERBERG, A. (1987). Passage of macromolecules and solvent through clefts between endothelial
cells. In “Microcirculation-An update” (M. Tsuchiya et al., Eds.), Vol. 1, pp. 153-154. Elsevier,
Amsterdam.
TSAY, R., WEINBAUM, S., AND PFEFFER, R. (1989). A new model for capillary filtration based on recent
electron microscopic studies of endothelial junctions. Chem. Eng. Commun. 82, 67-102.
TSAY, R., AND WEINBAUM, S. (1991). Viscous flow in a channel with periodic cross-bridging fibers of
arbitrary aspect ratio and spacing. J. Fluid Mech. 226, 125-148.
WARD, B. J., BAUMAN, K. F., AND FIRTH, J. A. (1988). Interendothelial junctions of cardiac capillaries
in rats: Their structure and permeabilities. Cell Tissue Res. 252, 57-66.
WISSIG, S. L. (1979). Identification of the small pore in muscle capillaries. Acta Physiol. &and. Suppl.
463,33-44.
Yu, C. P., AND SOONG, T. T. (1975). A random cell model for pressure drop prediction in fibrous
filters. J. Appl. Mech. Trans. ASME, June, 301-304.

Você também pode gostar