You are on page 1of 5

Interpreting pulsedgradient spinecho diffusion experiments with

permeable membranes
E. von Meerwall and R. D. Ferguson

Citation: J. Chem. Phys. 74, 6956 (1981); doi: 10.1063/1.441059
View online: http://dx.doi.org/10.1063/1.441059
View Table of Contents: http://jcp.aip.org/resource/1/JCPSA6/v74/i12
Published by the AIP Publishing LLC.

Additional information on J. Chem. Phys.
Journal Homepage: http://jcp.aip.org/
Journal Information: http://jcp.aip.org/about/about_the_journal
Top downloads: http://jcp.aip.org/features/most_downloaded
Information for Authors: http://jcp.aip.org/authors
Downloaded 13 Sep 2013 to 192.12.88.151. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jcp.aip.org/about/rights_and_permissions
Interpreting pulsed-gradient spin-echo diffusion
experiments with permeable membranes
E. von Meerwall and R. D. Ferguson
a
)
Physics Department, The University of Akron, Akron, Ohio 44325
(Received 14 November 1980; accepted 26 February 1981)
We have developed an approximate expression for the spin-echo attenuation as function of the diffusion time
in a pulsed-gradient experiment on specimens containing regularly spaced barriers of arbitrary permeability.
The expression is a modification of the exact solution for impermeable barriers, and is simple enough to
permit use in curve-fitting programs.
INTRODUCTION
In the pulsed-gradient spin-echo method
1

2
of mea-
suring diffusion, the time over which diffusion is ob-
served is a well-defined quantity. This permits re-
stricted diffusion to be characterized; in particular,
the existence of regularly spaced barriers to diffusion
can be readily observed and measured. Tanner
2
has
derived expressions for the spin-echo attenuation as a
function of diffusion time for certain simple barrier
geometries, permitting the interpretation of correspond-
ing experiments.
In most important biological systems these barriers
are partially permeable. Exact solutions to this prob-
lem must take into account the possibility that diffusing
particles encounter and penetrate several such barriers.
The mathematical problem is formidable. Tanner
3
has
numerically calculated solutions for the resulting eigen-
value problem; one such solution consumes over one
minute on a very fast computer. Aside from certain un-
physical behavior and what appear to be possible numer-
ical instabilities, the required calculational effort alone
makes these solutions unsuitable for curve fitting to ex-
tract information from experimental data. On the other
hand, Tanner has also demonstrated
3
the failure of all
proposed approximations according to which the time-
dependence of the observed diffusion coefficient is ap-
proximated by a superposition of results for no barriers
and for impermeable barriers.
In a recent paper, Zientara and Freed
4
have demon-
strated a numerical solution based on the stochastic
Liouville equation, whose applicability includes the case
considered here, but extends to arbitrary sample in-
homogeneity (membranes of finite thickness, regions
with different solubilities, alternate diffusion paths) and
I
where e =yg6a, and d=(e/1T) I cosO'I.
For impermeable barriers, A = 0, B = 1. The sum may be
truncated after m"'" 1 0 + 2d terms to give a worst-case
a) Present address: Department of Physics. Cornell University.
Ithaca. N. Y. 14853.
can be adapted to various rf and gradient pulse sequences.
While the high-speed storage requirements for imple-
mentation of this method are quite modest and its speed
is sufficient to permit the use of a minicomputer, 4 the
iterative nature of the method best suits it for simula-
tion rather than curve fitting. Moreover, for arbitrarily
complex models of the substance under investigation, the
time-dependent diffusion constant D(t) may not enter the
theory linearlYj hence a fitting procedure may not pro-
duce unique fitted parameters. In such cases, compari-
son of suitably adjusted models with data primarily has
the character of hypothesis testing, and extraction of
parameters, where possible at all, becomes secondary.
For the simplest case, an infinite number of thin,
equally spaced plane parallel barriers, a solution linear
in D(t) eXists
2
j we propose an approximation for barriers
of finite permeability based on it. The approximation is
generally adequate, has no unphysical behavior or nu-
merical difficulties, and is evaluated rapidly enough to
permit its use in a (nonlinear) curve-fitting computer
program for direct extraction of parameters from data.
A preliminary report of this work has been given.
5
DEVElOPMENT
The pulsed-gradient spin-echo experiment involves
gradient pulses of magnitude g and duration 6, separated
by a diffusion time t. The gyromagnetic ratio of the dif-
fusing nuclei under observation is y, the intrabarrier
diffusion constant is Do, and the barrier spacing is a.
The barriers are assumed to be plane, with their nor-
mal inclined at an angle 0' with respect to the direction
of the principal axis of the field gradient. The relative
magnitude of the spin echo R(t) =A(t)/A(O) for imperme-
able barriers may be written, following Tanner2:
(1)
I
accuracy of five significant figures in R(t).
In the case of membranes of permeability p (i. e., re-
duced permeability P =ap/D
o
), any expression for R(t)
must satisfy certain requirements. If R(t) for P> 0 is
called Rp(t), then:
(a) the slope of a plot of lnRp vs t in the limit of small
6956 J. Chern. Phys. 74(12), 15June 1981 0021-9606/81/126956-04$01.00 1981 American I nstitute of Physics
Downloaded 13 Sep 2013 to 192.12.88.151. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jcp.aip.org/about/rights_and_permissions
E. von Meerwall and R. D. Ferguson: Spin-echo diffusion experiments 6957
t must be independent of P, hence Rp(t- 0) must agree
with Eq. (1);
(b) the slope of this plot at large t must be smaller
than the initial slope by the factor D.,/D
o
, where
3
Doc =Do/
(1 + lip); the value of D .. must not depend on e;
(c) the transition from initial to final slope should oc-
cur at a time when most diffusing nuclei have encountered
at least one barrier. Therefore 1 the t = 0 intercept of the
asymptotic behavior Rp(t - co) should be near or above the
limiting value of R(t - co) for impermeable barriers;
(d) Rp(t) must reduce to a single exponential for P _ co,
and to Eq. (l)forP=O.
These four requirements do not address the behavior of
Rp(t) in the intermediate regime where the particles
originating in one intrabarrier region are still concen-
trated in a few neighboring regions. Nevertheless, a
useful approximation to the exact Rp(t) may be obtained
by modifying Eq. (1) to produce the required behavior
at very short and very long times. Contrary to other
suggested approximations, this strategy at least accounts
for encounters with the innermost barriers.
Inspection of Eq. (1) shows that requirement (b) can
be met by multiplying it with an exponential function
with the required final slope. It is convenient to com-
bine this with the first exponential which is otherwise
used only when a* O. This results in
(2)
This change, however, also raises the initial slope,
which according to requirement (a) should have re-
mained unaffected. To correct for this the second fac-
tor in Eq. (1) is made to decrease with time at ~ cor-
respondingly lower rate until its saturation behavior
sets in. A detailed comparison with the form of the
exact solution [e. g., Eq. (5) in Ref. 3] shows that the
most appropriate way to do this is by multiplying the
argument of the second exponential by the factor
B=l/(l+P) . (3)
Because of the form of Eq. (1) and its modifications
Eqs. (2) and (3), reqUirements (c) and (d) are also met.
Equations (1), (2), and (3) together constitute the pro-
posed approximation for Rp (t) It is exact at small
times, and at large times has the required slope and
approximately the intercept of the exact solution. At
intermediate times, it represents a first-order approxi-
mation to the exact solution. The physical significance
of the approximation lies in its replacement of all bar-
riers beyond the innermost ones by a continuum.
The barrier geometries commonly encountered in
nature are confined to a few general types, of which
the plane and the spherical are the most important.
Rather than dealing with the relatively intractable solu-
tion:! for spherical geometry, it seems preferable to
decompose the sphere into annual segments to permit
the use of Eq. (1), with or without modification for bar-
rier permeability. In practice, it appears to be suffi-
cient to use only three segments, with a"'< 1.9r, 0.9r.
and O. 4r, where r is the radius of the sphere. Equa-
tion (1) is evaluated with a = 0 for each of these, and
the results are summed with coefficients h j-, and h
respectively, representing the base area fractions of
the annuli. For cylindrical geometries one may use
the same three values of a, but the area fractions of
the (now linear) segments are t each; here a is the
complement of the angle between the cylinder axes and
the sampled diffusion direction.
For simulating a distribution of barrier spacings, we
follow Tanner's practice:! of evaluating R(t) with spac-
ings a/k, a, and ak, and summing with coefficients i,
~ , and t, respectively. For spherical or cylindrical
geometry, this results in nine evaluations of Eq. (1)
for a complete calculation of Rp at one t value. Con-
ceivably, distributions in P might be simulated analo-
gously.
It should be remembered that Eq. (1) predicts:! a non-
monotonic decrease of R(t) with G
2
1)2 for large values of
the reduced time T = Dot la
2
This effect is ameliorated
in the presence of a distribution of barrier spacings,
and by our approximation for spherical geometry (it is
absent in the exact solution for that case
2
); introducing
finite permeability does not eliminate tbis effect. In
order to avoid encountering any resulting difficulties in
interpreting experiments, it is only necessary to choose
8 so that R(t) O. 1 at all times of interest in the experi-
ment. Insistence on this relatively nonrestrictive prac-
tice also simplifies the preliminary estimation of Do,
a, and p from the data, as is necessary in preparation
for curve fitting.
RESULTS AND DISCUSSION
In the following, we concentrate on the properties of
Eqs. (1), (2), and (3), i.e., plane geometry without dis-
tribution of spacings, at o! = 0, to permit direct com-
parison with the exact solution as published.
3
Figure 1 shows R(t) versus reduced time T =Dot/ a
2
, at
the same values of P and 8 used in Fig. 4 of Ref. 3; the
comparison shows good semiquantitative agreement. A
more sensitive standard of comparison is the effective
diffusion coefficient itself as function of time. USing
Tanner's definition3 of D(t), an expression for D(t)/D
o
was derived, based on Eqs. (1), (2), and (3); the re-
sults are shown in Fig. 2 for direct comparison with
Fig. 5 of Ref. 3.
The results for a given P are nearly independent of
8 for 8!S 2. 5. The required limits at large and small
times are reached at approximately the correct times.
Importantly, D(t)/D
o
at all values of 8 monotonically de-
creases with time, never falling below the asymptotic
limit ~ / D o . The exact solution,3 however, for e> 1
apparently does substantially decrease below D ~ / D o ; the
origin of this unphysical behavior is not well under-
stood.
3
Any e dependence of D(t)/D
o
in our approxima-
tion is confined to intermediate times. There D exhib-
its a progressively earlier and more rapid (monotonic)
decrease from Do to D ~ as e is increased. However. at
these intermediate times our approximation is least
trustworthy. While it is not clear whether D(t)ID
o
, cor-
rectly calculated for an infinite number of physically
reasonable barriers, should have any 8 dependence at
J. Chern. Phys., Vol. 74, No. 12, 15 June 1981
Downloaded 13 Sep 2013 to 192.12.88.151. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jcp.aip.org/about/rights_and_permissions
6958 E. von Meerwall and R. D. Ferguson: Spin-echo diffusion experiments
SPIN-ECHO HEIGHT vs TIME FOR e - 2.5
0.3
A(2'
A(O)
0.1
0.03
P-0.001
P-0.5
0.0 1
o 0.5 1.0 1.5 2.0 2.5
REDUCED TIME
FIG. 1. Spin-echo attenuation as function of reduced time,
for several values of the reduced barrier permeability; the
parameter () (see text) equals 2.5. This figure should be com-
pared with Fig. 4 of Ref. 3.
all, our rudimentary treatment of barriers produces a
(7 dependence which at least is in rough agreement with
the exact calculation at intermediate times. We suspect
that once the problem is better understood, any surviving
(7 dependence will have the qualitative features apparent
in Fig. 2. To our knowledge, a dependence of D(t) on (7
has not been observed. 6
Figure 2 shows thatthe time for which D is halfway
between Do and D", becomes shorter as P decreases, an
effect which is only differentially affected by the e depen-
dence. Thus our approximation, perhaps artificially,
implies
3
that the restrictive effects of the barriers are
manifested in proportion to the fraction of particles
which are reflected (rather than transmitted) by the bar-
riers. While this is contrary to the interpretation
evoked by the exact solution [which has an opposite P
TABLE I. Interpretation of restricted diffusion data.
Substance Do (10-
6
cm
2
/s) r(/lm) p (cm/s)
Red blood cells, 5.3 2.3 0.014
packed 4.60.8 2.80.9 0.030.02
Water and 23.8 0.2
alumina powder
a
26 5 6.52.8 O. 25 0.1
Yeast 6.8 4.1 "'0
7.10.3 3.20.4 SO.0005
Apple, yellow, 22 42 ""0
ripe 354 37 9 0.0080.003
0.8
D(t)
D (0)6
0.4 \
-8:s2.S
\
0.2 --8-10
\
\
\
,
P-oo
P-2
P-o.S
P-O.001
"-

0.001 0.01 0.1 10
REDUCED TIME
FIG. 2. Measured diffusion coefficient normalized to intra-
barrier diffusion coefficient as function of reduced time, for
various reduced barrier permeabilities. Solid lines apply for
values of () up to () '" 2.5, dashed lines for ()= 10. Symbols in-
dicate points where D (t) = (Do + D.,)/2. This figure is analogous
to Fig. 5 of Ref. 3.
dependence of the time of mean D(t)], this feature of
our approximation agrees with an earlier suggestions
and appears physically more reasonable.
We have incorporated our proposed approximation,
Eqs. (1), (2), and (3) with the modification for spherical
geometry, into a computer program
7
as one of ten model
equations to permit fitting to data from various types of
field-gradient spin-echo diffusion experiments. The ad-
justable parameters in this case are Do, the sphere radi-
us r, and the barrier permeability p. Since our experi-
ence shows that fitting more than three parameters to a
single echo attenuation experiment is unproductive, the
radius distribution factor k (see above) is set to a sensi-
ble constant value (k"" 1. 3), or the sum over the radius
distribution is omitted entirely.
The approximation produces excellent fits to our ex-
perimental data in appropriate samples. Fitting the
same data instead with a sum of two exponentials sub-
stantially worsens the chi-square goodness of the fit,
tending to falsify the intial and final slopes, so that these
cannot be taken to represent Do and D." respectively.
Reference
6 (Fig. 7)
6 (Our analysis)
6 (Fig. 6)
8 (Our analysis)
8 (Fig. 4, all data)
8 (Our analysis)
2 (p. 134)
Present work
Pine, aged
276 47 10 0.0020.0005 Present work
water saturated
aGeometry has been reinterpreted in terms of equivalent spacing and permeability. 6
J. Chern. Phys., Vol. 74, No. 12, 15 June 1981
Downloaded 13 Sep 2013 to 192.12.88.151. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jcp.aip.org/about/rights_and_permissions
E. von Meerwall and R. D. Ferguson: Spin-echo diffusion experiments
6959
For example, whereas our approximation extracts the
proper Do from data in water-saturated aged pine wood
or wet sandstone (Do"" 2.4x 10-
5
cmz/s), the fast decay
in the ill-fitting double exponential suggests values two
to four times as large in both cases. With some rein-
terpretation, the theory (and hence our approximation)
also adequately describes fluid diffusion in the interstices
of aggregates of impermeable particles. 4.6
Applying the approximation to published data [R(t) or
D(t) vs t] usually produces good agreement with the
original interpretations
2

8
; Table I shows such a com-
parison. However, for no published data known to us is
the validity of our approximation vis-a-vis the exact
solution put to a critical test; their differences are less
than the experimental uncertainties.
We find that for curve fitting it is quite important to
supply data from very small to very large diffusion
times. This only incidentally deemphasizes any inade-
quacies of the approximation at intermediate times and
minimizes uncertainties in the fitted parameters; its
most important role is in establishing the fact (not
merely the degree) of permeability, since diffusion be-
tween permeable barriers is easily miSinterpreted as
diffusion between impermeable barriers with irregular
spacing (k * 1)2 except at sufficiently large diffusion
times. In fact, we suggest that some of the data cited
in Table I (yeast, apple) may be better explained by a
finite, small cell wall permeability than with the origi-
nally postulated impermeability.
CONCLUDING REMARKS
We recommend that the present approximation be used
to interpret pulsed-gradient spin-echo diffusion data in
systems containing regularly spaced permeable barriers
whenever data R(t) vs t over wide ranges of t are avail-
able. In practical physical systems disagreements be-
tween theory and data, or misleading fitted parameters,
will more likely result from incompletely known distri-
butions of barrier spacing or permeability than from
failures of this approximation. Properly optimized for
computation, the formulas appear to offer a Sizeable re-
duction in calculation time over the exact calculations
(at least three orders of magnitude) and the numerical
solution,4 making possible the use of nonlinear curve-
fitting algorithms for direct interpretation of data.
Thus, use of valid approximation for R,(t) will survive
the clarification of the unphysical behavior in the exact
calculation, since it is unlikely that any exact solution
of eigenvalue problems of this kind
3
will have a simple
form.
ACKNOWLEDGMENTS
The authors wish to ackn!lwledge numerous helpful
discussions with Professor J. W. Doane and Dr. N. Vaz
of Kent State University, and members of the Depart-
ment of Mathematics at The University of Akron. The
Computer Center of The University of Akron was helpful
during the implementation of the curve-fitting algorithm.
This work forms a peripheral part of work supported by
the Faculty Research Committee of The University of
Akron.
IE. O. stejskal and J. E. Tanner, J. Chern. Phys. 42, 288
(1965).
2J E. Tanner, Ph.D. thesis (University of WisconSin, 1966).
3J. E. Tanner, J. Chern. Phys. 69, 1748 (1978).
4G. P. Zientara and J. H. Freed. J. Chern. Phys. 72, 1285
(1980).
5R. D. Ferguson and E. von Meerwall, Bull. Am. Phys. Soc.
(in press).
SR. L. Cooper, D. B. Chang, A. C. Young, C. J. Martin,
and B. Ancker-Johnson, Biophys. J. 14, 161 (1974).
7E. von Meerwall and R. D. Ferguson, Cornp. Phys. Cornrnun.
21, 421 (1981).
8J. E. Tanner and E. O. stejskal, J. Chern. Phys. 49, 1768
(1968).
J. Chem. Pl;lys., Vol. 74, No. 12, 15 June 1981
Downloaded 13 Sep 2013 to 192.12.88.151. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jcp.aip.org/about/rights_and_permissions

You might also like