BME 2105 Introduction to Biomedical Engineering

Molecular and Cellular Fundamentals

Lecture 8 - Diffusion

Diffusion dominates at molecular and cellular scales:

Reynold’s number
vL
Re =

v=average velocity
L=characteristic length
=media density
=dynamic viscosity (1 poise = 1g*cm-1*sec-1; Pascal second = 1 kg*m-1*s-1)
For a fish (10 cm long) swimming in water (100 cm/sec) Re=105
For a bacterium (10-4 cm long) in water (10-3 cm/sec) Re=10-5
So, at molecular and cellular scales, turbulence does not significantly assist diffusion
(conduction) in transport. Similarly, convection and radiation are not too important.

1D Fick’s law

Break the space up into slabs of width L, sides Y and Z, each centered on x=0,L,2L,3L,....
For a given time interval t, all the particles in the slab move either left or right, with equal
probability (free diffusion). Choose t such that we only need to examine one crossing, and
no particles cross over two boundaries. In the diagram, more particles move from x=0 to
x=L than from x=L to x=0, simply because there are more particles in x=0 than in x=Lf.
Mathematically, for a situation in which N(x) = number of particles in a box of size L * Y *
Z = c(x) * L * Y * Z
Total transfer over the given time interval across any given barrier from x to x+L is

flux =
1
(N (x ) − N (x + L ))
2 * t
note that as L→0,

(N(x ) − N(x + L)) → −L dN(x)

dx
(go right to Fick’s law unless interested)
Substituting back in,
 d  1 L2 dc( x)
* * L *  − N ( x) = LYZc ( x) = −
flux 1 1
flux per area = = j=
Y *Z Y * Z * t 2  dx  t 2 dx
rewrite, using the definition we had from the discrete case, as
dC( x ) L2
j = −D * ;D = ; Fick’s Law
dx 2t
Continuity equation:
C( x, t ) j
=−
t x
which is simply a statement that for any region, the change in stuff as a function of
time is the balance of stuff flowing in against that flowing out. Together, we get
dC( x, t ) d 2 C( x , t )
= D* ; Diffusion equation
dt dx 2
or more simply
Ct = D * C xx ; Diffusion equation
This can be extended into higher dimensions. The Diffusion equation becomes
Ct = D 2 C

Impulse response
The most fundamental solution to this type of equation is the Gaussian distribution, which
described the evolution of a delta function input with time.

1  x 2  P( x ) =
1
 2
( )
exp − ( x − x 0 ) 2 /( 2 2 ) ; Gaussian
C( x , t ) = exp −  note that
4Dt  4Dt  
 
 exp(− y )dy = 
2
−
Things to point out:
2 goes as 2Dt, similar to that from the discrete case
Once again, for simple diffusion, D is a constant that can be measured, not dependent on
L or t, but a combination in which the details vanish.
So, why two different ways at getting at the same equation? Both are approximations of
reality. For example, the Gaussian solution to the diffusion equation does imply
that some particles move in excess of the speed of light. On the other hand, the
 idea that particles exhibit a set jumping point in lockstep t behavior, and a
detailed look at how the jumps actually happen, are unrealistic.
All spatial directions are independent. So, by separation of variables:
C( x, y, z, t ) = C( x, t ) * C( y, t ) * C(z, t )
That is, each direction contributes either a Gaussian or a factor of 1 since integration over the
Gaussian=1. So, the general form is
1  r2  1  r 2 
 i 
exp −  ; n-degrees of freedom
n/2 
C( r , t ) = exp − =
 
(4Dt) i=1..n  4Dt  (4Dt) n / 2  
 4Dt 
So, is diffusion down an axon truly a 1D system? To be accurate, no, it is a three-dimensional
problem. However, if you assume a single concentration at any cross section, 1D diffusion
applies. That is, a particular dimension can be separated if:
• Symmetry imposes the condition that there is no net flux in that dimension, or
• The system is such that diffusion is not allowed in that dimension
Subtle, but important difference.

Estimation of the diffusion coefficient:

See Nelson section 4.1.4 for more complete discussion. It is important to see how the discrete
model and continuous model interact to get these relations.

The Einstein relation for diffusion is:

D = k B T ;=mobility=v/f
kB = Boltzmann constant = 1.38 x 10-23 J/K
T = temperature
Where f is the force needed to maintain a steady velocity of v.
For a sphere of radius, R, moving through a medium with dynamic viscosity, :
1
= Stokes formula for creeping flow
6R

Thus, the Stokes-Einstein relation gives us:

k BT
D=
6R
Good predictor for things with spherical symmetry, moderate success for things of lower
dimensional symmetry. The problem is that there is no analytical solution to Stokes flow in 2
dimensions.
• As stated before, small particles in water, D ~ 1 m2/msec
• For proteins, MW/A0=(4/3)a3; ~1.3 g/cm3 for proteins, A0=Avogadros #.
 o Myoglobin, MW 17kDa, water=0.764 cp=7.64E-4 Ns/m -> a=1.7 nm, D=.16
m2/ms
• molecules of gas, D ~ 10 m2/s
• freely diffusing molecule in cell membrane, D ~ 1 m2/sec

What are some implications

For 1D diffusion of a small molecule in water (D~1 m2/sec) across characteristic distances:
cell of diameter 10m; t=50 msec.
large cell of diameter 100 m; t=5 sec
axon of length 10mm; t=50,000 sec = 14 hours
So, diffusion is great for most cell-level transport distances, and is energy-free, but not good for
longer distances. In those cases, some other transport mechanism is needed.

Steady State equations:

In terms of diffusion-based transport, this is an important counterpart to the transient solutions
we just worked through..
Again, the diffusion equation can be generalized as:
Ct = D2C
And in steady state, this can be rewritten as:
C t = D 2C = 0

1D diffusion through a region of defined diffusion coefficient – steady

state equation
The starting point for this discussion is to consider transport through a single slab. For the
moment, assume this slab represents a region of fluid or other structure which is unstirred,
nonmoving, and only diffusion can transport molecules. For simplicity, assume that the
regions next to these are extremely well-stirred, such that the concentration in each can be
defined as a single, constant number. We will know that this configuration is not possible,
but for sake of analysis of the middle layer, assume that this is true.
In rectangular coordinates, the diffusion equation, in short, says that the concentration profile
must be piecewise linear. For the following situation, in which the region between A and B is
composed of a medium of diffusion coefficient DAB. Forget how we get Ca and Cb
experimentally, just focus on what’s going on between A and B
 L
L
Ca
Ca Cb
C(x) Cb

x
A B A B
 Ca xA
 Cb − Ca
C( x ) = Ca + (x − A ) A  x  B
 L
 Cb xB
Note that the flux in region between A and B is
 Cb − Ca   Ca − Cb 
j = −D AB   = DAB  
 L   L 
Apparent paradox: the earlier discussion said that transport happens on t-1/2 time, but
the steady state solution seems to contradict this; it is, in fact constant with time. If we
follow a specific molecule, we do see that <x2> goes as Dt. However, bulk transport
independent of time (i.e., steady state)

Electrical Analogy
Electrical circuit: I=1/R * V
diffusion: J=DA/L*C; R=L/(DA)
more convenient to think in terms of conductance per area, or conductivity, k=D/L

These act as resistors, conductors, can be assembled as you encountered

in electrical engineering:
For two slabs (a) and (b) in series (different notation than above), the flux resultant from a
concentration difference C is:
1 1
J = Ja = J b = Ca = Cb
Ra Rb
C = Ca + Cb
1 1
J= * C = C; R ser = R a + R b
(R a + R b ) R ser

(or, for conductances)

 J = J a = J b = k a A a Ca = k b A b Cb
C = Ca + Cb
1 1 1 1
J= * C = k ser A a = b C; = +
1 1 k ser k a k b
+
k a Aa k bA b

And for two slabs (a) and (b) in parallel

C = Ca = Cb
J = Ja + J b
J = (k a Aa + k b A b ) * C = k par Aa = b C; k par = k a + k b
(or, for resistances)
C = Ca = Cb
J = Ja + J b
1 1 1 1 1 1
J=( + ) * C = C; = +
Ra Rb R para R para R a R b
Each representation is equally valid, and each will be useful in different contexts.
Finally, this electrical analogy should drive home the point that conservation of mass must hold.
In a slab geometry, flux is constant at all points. Thus, with no gradient in the regions outside
this single slab, no transport could take place. We hand-wave this by invoking the well-
stirred condition, which relies on inertial and forced flow to dominate. In fact, in a small
region near the slab, these additional factors cannot function, and there is a deviation from a
constant concentration right near the slab.

Other steady state solutions

Diffusion to adsorbing sphere
Consider an absorbing sphere of radius s in a bath of far field concentration C0. This means
C(r=s)=0, C()=C0
Steady state solution,
0 = D * C xx or D 2 C = 0
For spherical symmetry
1 d  2 dC(r ) 
r =0
r 2 dr  dr 
so,
dC(r ) dC(r )
r2 = C1 then = r − 2 C1 and thus C(r ) = C 2 − r −1C1
dr dr
now, since C() = C0, then C2=C0
since C(s)=0, C1=sC0
and thus,
 s
C(r ) = C 0 (1 − )
r
also note that the flux per unit area is
s
j(r ) = −DC(r ) = −DC 0
r2
which if integrated over a shell of radius r (and surface area 4r2) is constant! This is
the equivalent as stating that flow through many slabs in series is constant, and
dictated by the size of the sphere!
J = − D4C 0 s
note that you can think of this as J=C/R where R=1/(4Ds)

Diffusion to a disk adsorber

Consider an adsorbing disk of radius s in a semi-infinite bath of far field concentration C0. This
means C(r=s)=0
Steady state solution (the solution is ugly),
0 = D * C xx or D 2 C = 0
From cylindrical symmetry, the transport through the disk is:
J = 4DsC 0
which can be thought of as J=C/R, where R=1/(4Ds)

What else influences diffusion through layers?

Partition coefficient
[x]
Remember Henry’s Law:  =
Px
Partition coefficient is similar:  = [ x ]solvent1 /[ x ]solvent 2
Usually, this will be in reference to water, so:  = [ x ]solvent /[ x ] water = C solvent / C water
This also applies to materials, such as polymers in which a drug is dissolved. In that case, the
material is equivalent to the solvent.
The partition coefficient is a function of hydrophobicity/hydophilicity, among other things
What does this do to transport?

J = A (C A − Cb ) = A C

D D
L L
So, a change in partition coefficient increases or decreases flux, independent of diffusion
coefficient!
Note that the change in concentration is measured as if the material/solvent was immersed in
water, and the concentrations were measured in the water right next to the surface of the
material/region. This is the typical situation, in which a barrier is separating two aqueous
buffers. It is relatively easy to measure the concentration in the buffers, but the concentration
internal to the barrier must be inferred.
Diffusion under drift, against a barrier.
What happens when I apply a force to the particles being followed, but not the media?
Additionally, impose the condition that at some position there is a barrier to further diffusion
of particles. This is the case of centrifugation in a test tube, for example. The flux equation
becomes:
dC( x )
j( x ) = −D + vC ( x )
dx
To impose that j(x)=0 at a barrier further implies that j(x)=0 throughout the space for steady state
to be true; what goes in, must go out. The flux equation becomes:
dC( x ) dC( x ) v
D = vC ( x ) then gives us = C( x )
dx dx D
and thus
v 
C( x ) = C0 exp x 
D 
Not too informative in itself, let’s give it a context.

Sedimentation
Consider a tall container of fluid in a gravitational field, and we
follow a set of particles that are of a different density than
the liquid.
The force inducing drift is buoyancy (density and gravity m
working together), and the bottom of the glass or top of the
fluid are the no-flux boundaries of this system, depending
on buoyancy; denser items sink, less dense items float. V,m
g is gravity, acting downwards, but let’s leave the directionality
to the force calculation (g > 0, and not a vector itself).
z=0
F=g*(Vm-m)= - mnetg; where m=density of media, V=volume of particle, m=mass of particle,
mnet=m-Vm. m > V*m, F < 0, downwards
From this force, we get a drift velocity of v=-*g*mnet, where  is mobility. Using the Einstein
relation, we get:
v = - D*mnet*g/(kBT)
Put this in, and for sake of thinking in vertical coordinates, put in terms of z:
 m gz   z 
c(z) = C 0 exp − net  = C 0 exp − 
 k BT   z* 

Put some numbers into it:

Myoglobin, 17kDa. Protein has a general density of about 1.3 g/ml. This protein has a rough
diameter of 4 nm. The term z* works out to be about 42 m.
Thus, in a typical 4 cm tube, at equilibrium in a 1 X gravitational field, the ratio of concentration
at top compared to bottom is 99.9%. Never settles out!
However, centrifuge it long enough at 10,000 X g, z* is about 4 mm, you’ll get it out of solution.

Binding interactions:
Returning to the fundamental binding interaction:
k
⎯⎯→
1
A+B C
⎯
⎯⎯
k −1
Many biological interactions are diffusion limited. That is, when A encounters B, they interact,
so the reaction rate is limited by how often these molecules interact and that is limited by
diffusion.
This gives a typical forward rate reaction of k1~1 x 106 M-1s-1
This leads to a measure of how quickly the complex dissociates. Looking at only the
dissociation reaction, thus ignoring the forward reaction, we note that
C(t)=C(0)*exp(-k-1*t)
So, we can take the value 1/k-1 as a measure of how long the complex endures.
How to get k-1? Remember that

KD =
A* B = k −1
C  k1
Thus,
k −1 = K D * k1
Note: this is derived from the association reaction above, but is relating to the reverse reaction.
Be careful on how k1 and k-1 are defined.
For a single dissociation reaction with Km=25 mM=k-1(dissoc)/k1(assoc)
k-1=25mM*1E6(M-1s-1)=2.5 x 104 s-1
 = 1/k-1 = 4 x 10-5 sec = 40 sec.
Examples:
Oxygen delivery to tissues
The partial pressure of oxygen in tissues at rest is 40 mmHg, and
blood
tissue z
can drop to 20 mmHg with exercise. Here, we examine a
converse problem, how deep into tissues oxygen can be
delivered under these minimally oxygenated, under
equilibrium conditions. PO2
Most exchange between oxygen in red blood cells and tissues
happens in capillaries, the small-diameter, final branches of z
the vasculature. For simplicity, here this system is modeled
0 d
as a 1-D slab diffusion geometry; certainly, a cylindrical system centered on the capillaries
would better represent real tissues.
Assume a layer of well-stirred, oxygenated blood at PO2,blood of 100 mmHg overlying a slab that
represents living tissue. Metabolism in this tissue exerts an oxygen demand, which must be
furnished by transport from the blood.
We’re trying to find the maximum depth, d, at which oxygen demand can be supplied in this
configuration. For simplicity, we’ll assume that if we’re looking at a specific depth d, all of
the oxygen demand required by the slab of tissue between the blood and this depth is
localized to a plane at this depth:
Total demand (d) = A*d*jO2; jO2 = demand per volume of tissue, A = cross sectional area.
Since no demand is satisfied between the blood-tissue interface and that depth, there are no
sources-sinks of O2 until depth d, and the profile must thus be linear (see figure).
The greatest transport will be when PO2 at the depth d is 0 (why?).
From these ideas, we can readily derive the following relation for the maximum depth of
oxygenation:
DO 2 * k O 2 * PO 2, blood
d=
jO 2
get this from:
j = demand per area = (DO2/d)kO2PO2,tissue = d*jO2, Solve for d.
Some useful numbers:
• kO2 = solubility of oxygen in aqueous solution (tissues) = 1.4 x 10-6 M/mmHg
• jO2 = oxygen requirement of tissues = 1 x 10-8 mole/(sec*cm3)
• DO2 = diffusion coefficient for oxygen in tissues = 1 x 10-5 cm2/sec
So, for PO2,blood=100 mm Hg and assuming that the corresponding concentration is representative
of the concentration in tissue immediately adjacent to blood,
1E − 5cm2 / sec*1.4E − 6mole /( L * mmHg) * 100mmHg  cm3 
d= 1E − 3 = .0118cm = 118m
j1E − 8mole /(sec* cm3 )  L 

not a bad approximation, since the distance between capillaries tends to be in the 100 m range,
and this must supply times of high demand. Also note that it is not good for PO2 in tissues at
 any depth to equal 0; if this is the concentration outside a cell, it would be drawing whatever
O2 there is inside the cell out.

Chemotaxis
Cells are very sensitive to gradients of certain chemicals; a difference of 2-5% of some receptor
ligands in the local cell environment can direct cell migration. As suggested in the diagram,
consider the sensing by one cell (A) of a chemical being secreted by cell (B). Assume the
following:
• Cell A measures 50 m from the end closest 50 m
to cell B to the end farthest from cell B r
L
• Cell B measures 5 m in radius C B

• Assume steady state condition. CS

• Spherical geometry of this system, with cell B

cellA cell B
at the origin.
• Concentration is thus a function of r only; ignore geometry of cell A
• The chemical compound in question exhibits a diffusion coefficient of 1 m2/msec
• There is a bulk concentration CB of this compound (C(r=)=CB)
• There is a concentration of Cs of this compound at the surface of cell B (C(r=s)=Cs)
In the discussion on diffusion, we found the steady state solution for concentration of a perfectly
absorbing sphere, starting with the spherical geometry version of Fick’s law:
 1 d  2 dC(r ) 
C t = −D 2 C(r ) = −D  r 
2 dr  dr 
r
What is the steady-state solution to this system?
Set Ct to zero. This also implies that the terms D and the 1/r2 don’t mean anything.
So,
d  2 dC (r ) 
r =0
dr  dr 
Integrate once,
dC (r ) 1
= 2 * C1
dr r
Integrate again,
1 1
C (r ) = −   + C2
C1  r 
Now, apply boundary conditions (C(r=s)=Cs and C(r=)=CB), we get to

C(r ) = (CS − CB ) + CB
s
r
It is not dependent on diffusion coefficient; the diffusion coefficient influences
overall flux, but not concentration profile in any of the systems we studied.
Now, assuming that CB=0, and that cell A must experience a 5% difference in concentration of
the chemical across it’s 50 m extent to respond to cell B, describe as quantitatively as
possible the dependence of this response on the distance between cells A and B, denoted as L
in the diagram.
Setting CB=0,
C(r ) r + r
=
C(r + r ) r
Where r is the distance across cell A. r->L. As L gets bigger, this ratio drops
to 1, so the change across a 50 m distance drops. Specifically, for r=50m,
the cutoff for a 5% difference is 1.05=(L+50m)/L=> L=1000m=1mm
(5 points)

Diffusion across a synapse

The neuromuscular junction is defined as the interface
between a motor neuron and muscle cell, and is
depicted graphically to the right. The synaptic
cleft is a roughly 20 nm gap between the neuron
membrane and the postsynaptic surface of the
muscle cell. A typical signal transmission event
starts with the arrival of an electric signal at the
end of the neuron, which sequentially initiates: 1)
fusion of storage vesicles containing acetylcholine
with the presynaptic membrane, 2) release of
acetylcholine into the 20-nm thick synaptic cleft, 3) activation of acetylcholine receptors on
the other cell, which leads to activation of signaling systems in the second cell. The diffusion
coefficient for acetylcholine in the synaptic cleft is 0.9 m2/msec. Assume that the vesicle
fusion process produces a uniform distribution of acetylcholine along the neuron face in the
synaptic cleft.
2a) Modeling acetylcholine as a sphere and the extracellular fluid in the synaptic cleft as water,
what is the radius of acetylcholine?
The Stokes-Einstein relation, which applies for a spherical molecule, is written
as D=kBT/(6R). D yields R=2.4E-10m
2b) How long does it take acetylcholine to diffuse across the cleft? That is, assuming a Gaussian
solution accurately describes diffusion, when does =sqrt(2)=20nm
Given the symmetry along the synaptic cleft, this is a 1D solution. =sqrt(2Dt).
D=0.9 m2/msec, =20 nm -> 2E-4 msec
2c) Synaptic transmission normally takes 1 msec. Is the rate of diffusion of acetylcholine the
main determinant of the transmission delay?
no