37. We thank B. Spiller for crystallography advice, E.
Abbate for construction of several HPV18 mutants,
and D. S. King for mass spectrometry. We are
grateful to J. Holton and other members of the
Alber lab for helpful discussions; T. Alber, J. Berger,
R. Stevens, and R. Tjian for suggestions and critical
reading of the manuscript; L. W. Hung and T.
Earnest of the Advanced Light Source (ALS) for
sharing counsel and resources; and H. Bellamy of
the Stanford Synchrotron Radiation Laboratory
(SSRL). The ALS facility is funded by the Office of
Biological and Environmental Research of the U.S.
Department of Energy (U.S. DOE), with contributions
from Ernest Orlando Lawrence Berkeley National
Laboratory (LBNL), Amgen, Roche, University
of California at Berkeley, and Lawrence Livermore
National Laboratory. The SSRL is funded by
the U.S. DOE Office of Basic Energy Science. This
work was supported by NIH grants CA30490 and
CA42414 to M.R.B. X-ray coordinates are available
from the authors until the data are deposited in
the Protein Data Bank.
18 February 1999; accepted 19 April 1999
The Fourth Dimension of Life:
Fractal Geometry and
Allometric Scaling of Organisms
Geoffrey B. West,1,2
* James H. Brown,2,3
Brian J. Enquist2,3
Fractal-like networks effectively endow life with an additional fourth spatial
dimension. This is the origin of quarter-power scaling that is so pervasive in
biology. Organisms have evolved hierarchical branching networks that terminate
in size-invariant units, such as capillaries, leaves, mitochondria, and oxidase
molecules. Natural selection has tended to maximize both metabolic capacity,
by maximizing the scaling of exchange surface areas, and internal efficiency, by
minimizing the scaling of transport distances and times. These design principles
are independent of detailed dynamics and explicit models and should apply to
virtually all organisms.
Evolution by natural selection is one of the
few universal principles in biology. It has
shaped the structural and functional design of
organisms in two important ways. First, it has
tended to maximize metabolic capacity, because
metabolism produces the energy and
materials required to sustain and reproduce
life; this has been achieved by increasing
surface areas where resources are exchanged
with the environment. Second, it has tended
to maximize internal efficiency by reducing
distances over which materials are transported
and hence the time required for transport.
A further consequence of evolution is the incredible
diversity of body sizes, which range
over 21 orders of magnitude, from 10 13
g
(microbes) to 108
g (whales). A fundamental
problem, therefore, is how exchange surfaces
and transport distances change, or scale, with
body size. In particular, a longstanding question
has been why metabolic rate scales as the 3/4power
of body mass, M (1).
Biological scaling can be described by the
allometric equation Y Y0 Mb
, where Y is a
variable such as metabolic rate or life span,
Y0 is a normalization constant, and b is a
scaling exponent (1). Whereas Y0 varies with
the trait and type of organism, b characteristically
takes on a limited number of values,
all of which are simple multiples of 1/4. For
example, diameters of tree trunks and aortas
scale as M3/8
rates of cellular metabolism and
heartbeat as M 1/4
, blood circulation time
and life span as M1/4
, and whole-organism
metabolic rate as M3/4
. The question has been
why these exponents are multiples of 1/4
rather than 1/3 as expected on the basis of
conventional Euclidean geometric scaling.
Recently, we presented a model which
suggested that the explanation could be found
in the fractal-like architecture of the hierarchical
branching vascular networks that distribute
resources within organisms (2). The
model accurately predicts scaling exponents
that have been measured for many structural
and functional features of mammalian and
plant vascular systems. It is not clear, however,
how this model can account for the
ubiquitous 3/4-power scaling of metabolic
rate in diverse kinds of organisms with their
wide variety of network designs, and especially
in unicellular algae and protists, which
have no obvious branched anatomy. Here we
present a more general model, based on the
geometry rather than hydrodynamics of hierarchical
networks, that does not require the
existence of such explicit structures and that
can account for the pervasive quarter-power
scaling in biology.
We conjecture that organisms have been
selected to maximize fitness by maximizing
metabolic capacity, namely, the rate at which
energy and material resources are taken up
from the environment and allocated to some
combination of survival and reproduction.
This is equivalent to maximizing the scaling
of whole-organism metabolic rate, B. It follows
that B is limited by the geometry and
scaling behavior of the total effective surface
area, a, across which nutrients and energy are
exchanged with the external or internal environment.
Examples include the total leaf area
of plants, the area of absorptive gut or capillary
surface area of animals, and the total area
of mitochondrial inner membranes within
cells. In general, therefore, B a. It is important
to distinguish a from the relatively
smooth external surface, or "skin," enclosing
many organisms. We further conjecture that
natural selection has acted to maximize a
subject to various constraints while maintaining
a compact shape. This is equivalent to
minimizing the time and resistance for delivery
of resources by minimizing some characteristic
length or internal linear distance of
the hierarchical network.
Broadly speaking, two sets of variables
can be used to describe the size and shape of
an organism: a conventional Euclidean set
describing the external surface, A, enclosing
the total volume, V; and a "biological" set
describing the internal structure, which includes
the effective exchange area, a, and the
1
Theoretical Division, MS B285, Los Alamos National
Laboratory, Los Alamos, NM 87545, USA. 2
The Santa
Fe Institute, 1399 Hyde Park Road, Santa Fe, NM
87501, USA. 3
Department of Biology, University of
New Mexico, Albuquerque, NM 87131, USA.
*To whom correspondence should be addressed. Email:
gbw@lanl.gov
Table 1. Examples of the biological network variables l, a, and v in plant, mammalian, and unicellular
systems.
Variable Plant Mammal Unicellular
l Mean path length from root
to leaf, or between leaves
Mean circulation
distance from heart to
capillary, or between
capillaries
Mean distance from cell
surface to mitochondria
and between
mitochondria
a Total area of leaves; area of
absorptive root surface
Total area of capillaries;
gut surface area
Actual cell surface area;
total surface area of
mitochondrial inner
membranes
v Total wood volume; total
cell volume
Total blood volume;
total tissue, or cell,
volume
Volume of cytoplasm
R E P O R T S
www.sciencemag.org SCIENCE VOL 284 4 JUNE 1999 1677
total volume of biologically active material, v
(Table 1). Although it is clearly a very difficult
technical problem to calculate a, there
are some general scaling properties that it
must obey regardless of the detailed dynamics.
Before examining these, it is instructive
to consider the simpler case of how the area
of skin, or external physical surface, of an
organism or any Euclidean object, scales.
We first show how, and under what conditions,
the classic 2/3-power Euclidean scaling
law for A arises (3). In general, A is some
complicated function of the various length
scales, L1, L2, L3, . . ., which parameterize
size and shape: A A(L1, L2, L3, . . .). On
purely dimensional grounds this can be expressed
as A(L1, L2, L3, . . .) L1
2
(L2/L1,
L3/L1, . . .), where is a dimensionless function
of the dimensionless ratios L2/L1, and so
on. Suppose that we change the overall size
by making a uniform scale transformation on
all the lengths, Li: Li 3 Li Li (i 1, 2,
3, . . .), where is some arbitrary number.
This similarity transformation preserves the
shape of the object as its size varies. In this
case clearly does not change, so A responds
in the following manner:
A 3 A A( L1, L2, L3, . . .)
2
A(L1, L2, L3, . . .) (1)
The Euclidean volume of the object, V
V(L1, L2, L3, . . .), can be treated similarly; on
dimensional grounds, V L1
3
(L2/L1, L3/
L1, . . .), where is a dimensionless function
of the dimensionless ratios L2/L1, and so on.
After the scale transformation, which leaves
unchanged,
V 3 V V( L1, L2, L3, . . .)
3
V(L1, L2, L3, . . .) (2)
From Eqs. 1 and 2, it is clear that A /V 2/3
A/V2/3
, that is, A V2/3
; similarly, Li V1/3
.
Notice that these are consistent with writing
V AL, where L is some length that is a
function of the Li and scales as L 3 L L.
Assuming a size-invariant uniform density,
these then give the conventional Euclidean
geometric scaling results L Li M1/3
and A
M2/3
. These should apply, for example, to
the body length and skin area of vertebrates.
The above argument ignores two basic
facts of biology. First, the metabolic process
relies on the hierarchical fractal-like nature of
resource distribution networks. Examples include
the macroscopic branching vascular
networks of plants and animals and the complicated
ultrastructure within cells. We emphasize
that the network can be "virtual"; it
need not be a physical system of branching
tubes, so long as it exhibits hierarchical pathways
of material flow. Second, although organisms
vary widely in size, these networks
terminate at invariant units of fixed size that
can be characterized by a biological length
scale, l0. At the whole-organism level they
include capillaries of mammals and leaves of
plants. At the cellular and molecular levels,
they include mitochondria and chloroplasts,
and the metabolic rate-limiting cytochrome oxidase
and RuBisCo (ribulose-1,5-bisphosphate
carboxylase-oxygenase) molecules within these
organelles. We now modify the above scaling
argument by incorporating these two important
biological features.
For a given type of organism the effective
surface area is a function of the invariant
length, l0, together with various independent
length scales, li, that parameterize its fractallike
structure. It is important to distinguish
biological length scales, li, which characterize
the interior networks of the organism,
from Euclidean ones, Li, which characterize
its exterior shape. For example, in a mammal
one of the li is the length of the aorta, whereas
one of the Li is the overall body length;
similarly, in unicellular organisms one of the
li is the distance between mitochondria,
whereas one of the Li is the cell radius.
Working as before, the effective exchange
area, a, can be expressed as
a(l0, l1, l2, . . .) l1
2
l0
l1
,
l2
l1
, . . . (3)
where is a dimensionless function of the
dimensionless ratios l2/l1, and so on. Now, as
the size of the organism changes, l0 remains
fixed. Consider, then, an arbitrary scale transformation
on the network: li 3 li li (i
1, 2, 3, . . .) keeping l0 fixed. The analog of
Eq. 1 reads
a 3 a a(l0, l1, l2, l3, . . .)
2
l1
2
l0
l1
,
l2
l1
,
l3
l1
, . . . (4)
Because l0 is fixed, the right-hand side is
no longer simply proportional to 2
as in
Eq. 1. Although we do not know the -dependence
of , we can parameterize it as a
power law reflecting the hierarchical fractal-like
organization:
l0
l1
,
l2
l1
,
l3
l1
, . . . a
l0
l1
,
l2
l1
,
l3
l1
, . . . (5)
where a is an "arbitrary" exponent. In this
case
a 3 a a(l0, l1, l2, l3, . . .)
2 a
a(l0, l1, l2, l3, . . .) (6)
The crucial point here is that, because of
the presence of l0, a does not scale simply
as 2
. The assumption of a power law does
not require the existence of an idealized
mathematical self-similar fractal, which
has no "fundamental" length scale such as
l0. Even though the actual physical network
is not a pure fractal because it has terminal
units of fixed size and can be asymmetric, it
is still natural to use the fractal language.
We can therefore interpret the exponent in
Eq. 6, (2 a) da, as the fractal dimension
of a (4). As such, it satisfies 0 a
1. The lower limit, a 0, is the conventional
Euclidean case discussed above; the
upper limit, a 1, represents the "maximum
fractality" of a volume-filling structure
in which the effective area scales like
a conventional volume.
Similarly, the biological volume, v, associated
with a, can be expressed as v(l0, l1, l2,
l3, . . .) l1
3
(l0/l1, l2/l1, l3/l1, . . .), where
is a dimensionless function of the dimensionless
ratios l2/l1, and so on. This represents the
volume of protoplasm or biologically active
material in the organism. It is not necessarily
identical to V, because most organisms contain
empty spaces enclosed by the skin; however,
v V. By analogy with , we assume
that, under a scale transformation, transforms
as a power with an exponent v: (l0/
l1, l2/l1, l3/l1, . . .) v
(l0/l1, l2/l1, l3/
l1, . . .). Consequently, v scales as
v 3 v v(l0, l1, l2, l3, . . .)
3 v
v(l1, l2, l3, . . .) (7)
with 0 v 1. Combining Eqs. 6 and 7
straightforwardly leads to a v(2 a)/(3 v)
.
Now v can always be expressed as v al,
where l is some length characteristic of the
internal structure of the organism. We can
therefore relate the scaling behaviour of v to
that of a and l, with l expected to be proportional
to one of the li. It is instructive, however,
to consider the more general case and
write l l(l0, l1, l2, . . .) l1 (l0/l1, l2/
l1, . . .), as was done with a and v; is a
dimensionless function, analogous to and
. This scales as l 3 l 1 l
l, where dl
1 l is the fractal dimension of l, with 0
l 1. Consequently, v 3 v 3 a l
v
which, when compared to Eq. 7, gives v
a l (4). Assuming a uniform constant
density, so that v M, then gives
a
2 a
3 a l M
2 a
3 a l (8)
Table 2. The scaling of length, area, and volume associated with biological networks compared to the
conventional Euclidean case. Allometric relations with M assume that tissue density is constant.
Variable Conventional Euclidean Fractal biological
Length L A1/2
V1/3
M1/3
l a1/3
v1/4
M1/4
Area A L2
V2/3
M2/3
a l3
V3/4
M3/4
Volume V L3
M v l4
M
R E P O R T S
4 JUNE 1999 VOL 284 SCIENCE www.sciencemag.org1678
Our conjecture that organisms have evolved
so as to maximize the scaling of a implies
that the exponent, b (2 a )/(3 a
l), must be maximized. It is straightforward
to verify that this occurs when a 1 and l
0, thereby giving b 3/4. Metabolic rate
should therefore scale as B M3/4
, regardless
of the details of the branching architecture (5)
and dynamics governing the metabolic process
and distribution of resources.
This has several important consequences.
First, because a M3/4
, the number of invariant
units in the network also scales as
M3/4
. Second, the result l 0, which gives
dl 1, implies that internal distances associated
with the network are not themselves
fractal. This is consistent with the constraint
that times for supply of resources, and therefore
path lengths, should be minimized.
Third, and perhaps most significant, is that a
1, which implies that the fractal dimension
of a is da 3 rather than the canonical
Euclidean value of 2. Thus, the effective
surface area is "maximally fractal" and the
network structure is volume-filling. It is in
this sense that organisms have exploited a
fourth spatial dimension (6) by evolving hierarchical
fractal-like structures to maximize
resource acquisition and allocation. More
specifically, the area of the effective exchange
surface scales as if it were a volume:
a 3 a 3
a, (rather than 2
a), whereas
characteristic internal lengths associated with
the fractal-like structure scale as l 3 l l.
Consequently, the biological volume scales
as v 3 v 4
v, so that in addition to a
M3/4
, we also have l li M1/4
.
These relationships should apply to all
organisms that have been selected to maximize
metabolic power under the constraint of
minimizing internal transport distances and
thereby having a maximally compact threedimensional
body shape (Table 2). For organisms
such as roundworms and flatworms,
which may be functionally one- or two-dimensional,
these geometric relationships can
be appropriately modified. In D dimensions,
for example, our argument straightforwardly
generalizes to give a B MD/(D 1)
as in (2)
and l M1/(D 1)
for the biological variables,
and A M(D 1)/D
and L M1/D
for the
Euclidean ones. These relationships are not
expected to apply to a few organisms, such as
filamentous algae and fungi, that have been
selected to maximize linear dimensions so as
to sparsely occupy a maximal volume.
The present derivation is more general
than our original model in which it was assumed
that resource distribution networks
were volume-filling and that energy dissipated
was minimized. Incorporating dynamics
led to a complete description of the physics
and geometry of the networks that were
shown to be fractal-like with 1/4-power scaling
(2, 7). Versions of this physically explicit
model show how the universal geometric derivation
given here is realized in a variety of
systems in different kinds of organisms. It is
no accident, therefore, that many biological
networks exhibit area-preserving branching,
even though different anatomical designs exploit
different hydrodynamic principles (2,
7). Unlike the genetic code, which has
evolved only once in the history of life, fractal-like
distribution networks that confer an
additional effective fourth dimension have
originated many times. Examples include extensive
surface areas of leaves, gills, lungs,
guts, kidneys, chloroplasts, and mitochondria,
the whole-organism branching architectures
of trees, sponges, hydrozoans, and crinoids,
and the treelike networks of diverse
respiratory and circulatory systems. It is not
surprising, therefore, that even unicellular organisms
exhibit quarter-power scaling, including
the 3/4-power scaling law for metabolic
rate. Although living things occupy a
three-dimensional space, their internal physiology
and anatomy operate as if they were
four-dimensional.
Quarter-power scaling laws are perhaps as
universal and as uniquely biological as the
biochemical pathways of metabolism, the
structure and function of the genetic code,
and the process of natural selection. The vast
majority of organisms exhibit scaling exponents
very close to 3/4 for metabolic rate and
to 1/4 for internal times and distances. These
are the maximal and minimal values, respectively,
for the effective surface area and linear
dimensions for a volume-filling fractal-like
network. On the one hand, this is testimony to
the power of natural selection, which has
exploited variations on this fractal theme to
produce the incredible variety of biological
form and function. On the other hand, it is
testimony to the severe geometric and physical
constraints on metabolic processes,
which have dictated that all of these organisms
obey a common set of quarter-power
scaling laws. Fractal geometry has literally
given life an added dimension.
References and Notes
1. K. Schmidt-Nielsen, Scaling: Why is Animal Size So
Important? (Cambridge Univ. Press, Cambridge,
1984); W. A. Calder III, Size, Function and Life History
(Harvard Univ. Press, Cambridge, MA, 1984); R. H.
Peters, The Ecological Implications of Body Size (Cambridge
Univ. Press, Cambridge, 1983); K. J. Niklas,
Plant Allometry: The Scaling of Form and Process
(Univ. of Chicago Press, Chicago, IL, 1994); J. H.
Brown and G. B. West, Eds., Scaling in Biology (Oxford
Univ. Press, Oxford, in press).
2. G. B. West, J. H. Brown, B. J. Enquist, Science 276, 122
(1997).
3. Rubner originally suggested that metabolic rate
scales like the external Euclidean surface area, A,
erroneously leading to a 2/3-power law [M. Rubner,
Z. Biol. Munich 19, 535 (1883)].
4. B. B. Mandelbrot, The Fractal Geometry of Nature
(Freeman, New York, 1977); H. Takayasu, Fractals in
the Physical Sciences (Wiley, Chichester, UK, 1992).
5. In particular, this shows that the derivation for mammalian
and plant systems presented in (2) does not
depend on details of the network such as symmetric
branching. This was confirmed numerically by D. L.
Turcotte, J. D. Pelletier, and W. I. Newman [J. Theor.
Biol. 193, 577 (1998)].
6. Blum earlier noted that, in four Euclidean dimensions,
the surface area of a sphere would scale as the
3/4-power of its four-dimensional volume, and that
this might in some way be related to the 3/4 exponent
in Kleiber's law. Hainsworth subsequently proposed
that this extra dimension be identified with
time. Neither of these authors, however, gave any
argument to support their conjectures [J. J. Blum, J.
Theor. Biol. 64, 599 (1977); F. R. Hainsworth, Animal
Physiology; Adaptions in Function (Addison-Wesley,
Reading, MA), p. 170.
7. G. B. West, J. H. Brown, B. J. Enquist, in Scaling in
Biology, J. H. Brown and G. B. West, Eds. (Oxford
Univ. Press, Oxford, in press).
8. Supported by a University of New Mexico Faculty
Research Semester (J.H.B.), by NSF grant GER-
9553623 and an NSF postdoctoral fellowship (B.J.E.),
and by U.S. Department of Energy contract ERWE161
and NSF grant PHY-9873638 (G.B.W.). We also acknowledge
the generous support of the Thaw Charitable
Trust.
26 January 1999; accepted 26 April 1999
NEW! Science Online Content Alert Service:
instantly
Mind the gap.,
s
R E P O R T S
www.sciencemag.org SCIENCE VOL 284 4 JUNE 1999 1679