Optimality and evolutionary tuning of the
expression level of a protein
Erez Dekel1& Uri Alon1
Different proteins have different expression levels. It is unclear to
ment. Evolutionary theories suggest that protein expression levels
maximize fitness1–11, but the fitness as a function of protein level
lactose for growth12. We experimentally measured the growth
(cost), as well as the growth advantage (benefit) conferred by the
the difference between the benefit and the cost, predicts that for
each lactose environment there exists an optimal Lac expression
level that maximizes growth rate. We then performed serial
dilution evolution experiments at different lactose concen-
trations. In a few hundred generations, cells evolved to reach the
predicted optimal expression levels. Thus, protein expression
from the lac operon seems to be a solution of a cost–benefit
optimization problem, and can be rapidly tuned by evolution to
function optimally in new environments.
The expression level of a protein can change over evolutionary
timescales by two main processes15. The first is neutral evolutionary
drift7, and the second is selection of mutations that increase fitness.
The latter is often viewed as an optimization process1–11,16–18. Many
cases of optimization have been demonstrated on the level of the
organism phenotype. For example, bacteria evolve to increase their
E. coli towards optimal metabolic fluxes3,21.
Here we ask whether evolutionary optimization can predict the
expression level of a protein in a given environment. We use the
lactose for use as an energy and carbon source, and LacY, which
transports lactose into the cell12,22–24. Decades of study have provided
a quantitative characterization of this system, making it an excellent
starting point for a theoretical and experimental study.
The present study had three stages: first, we measured the cost and
benefit of Lac protein expression in wild-type E. coli. Second, we
found that the cost and benefit functions predict that there is an
optimal expression level that maximizes growth in a given lactose
environment. Third, we monitored the evolution of E. coli in
different lactose environments and compared its Lac expression to
the predicted optimum in each environment (Fig. 1).
To form a cost–benefit theory, we began with direct measurement
of the cost and benefit of Lac protein expression in wild-type E. coli.
The cost is defined as the relative reduction in growth rate due to the
burden placed on the cells by production and presence of the Lac
Figure 1 | The lac operon of E. coli and the experimental design in the
present study. The lac operon encodes the enzyme LacZ, denoted Z, which
uses the sugar lactose, L, to increase growth rate. In the experiments, we
consider environments with constant concentrations of lactose. The
repressor of the lac system LacI, denoted I, was deactivated by the presence
of lactose. To examine evolution at zero lactose under the same conditions,
we added the non-metabolized inducer IPTG, which also deactivates LacI,
and glycerol as an additional carbon source, in all cases. Measurements on
burden of Lac protein production or maintenance (cost h) and the increase
in growth generated by Lac proteins in the presence of lactose (benefit B).
Evolution experiments were performed by serial dilution of cells growing in
tubes for several hundred generations. Cells undergoing evolution in the
absence of lactose were predicted to lose lac expression. Cells undergoing
evolution at low (high) lactose concentrations were predicted to evolve to
optimal low (high) Lac protein levels relative to wild type.
1Department of Molecular Cell Biology and Department of Physics of Complex Systems, The Weizmann Institute of Science, Rehovot 76100, Israel.
Vol 436|28 July 2005|doi:10.1038/nature03842
© 2005 Nature Publishing Group
proteins13,14. To measure the cost of Lac expression, we measured the
growth rate at various concentrations of the inducer isopropyl-b-D-
thiogalactoside (IPTG; Fig. 2a) in the absence of lactose. IPTG
induces the lac system, but no benefit is gained because IPTG is
not metabolized. We found a nonlinear, concave dependence of the
increase with the amount of proteins produced.
The nonlinear cost function may reflect the fact that the cell has
limited resources, so that production of Lac proteins reduces the
capacity to produce other important proteins. For example, rapidly
synthesizing ribosomal proteins and other translation factors, and
most of their transcription capacity to produce ribosomal related
RNA25. To describe the cost as afunction ofproteinlevelwe used two
nonlinear functions. The first is a quadratic function
h1ðZÞ ¼ h0Z þh
where Z is the expression level of the Lac proteins and h0and h
parameters. The second cost function assumes that there is an
effective maximal capacity M for producing non-essential proteins.
Production of Lac proteins at levels that approach M would inhibit
production of essential systems, so that cell growth would be
significantly slowed down. Perhaps the simplest phenomenological
cost function that increases strongly as an effective limit M is
approached is (see Methods for a derivation of this function):
The parameters of cost function 2 that describe the measured
reduction in growth rate are h0ZWT¼ 0:02^0:003; where ZWTis
the fully induced expression of the wild-type Lac proteins and
M ¼ (1.8 ^ 0.3)ZWT. We note that the parameter M is expected to
depend on the growth rate of the cells: at slower growth rates, more
resources are known to be free for protein expression25and M is
higher (see Supplementary Information for more details). We also
note that a linear cost function does not fit the direct cost measure-
ment of the wild-type strain (Fig. 2a) and is not able to explain the
results presented below. We find that the cost of fully induced wild-
type expression of the lac system is about 4.5%.
In addition to the cost of Lac proteins, we also directly measured
their benefit. For this purpose, we provided saturating IPTG for full
induction and various concentrations of the sugar lactose. The non-
metabolized inducer IPTG kept the Lac protein levels constant, and
hence the cost constant, allowing measurement of the benefit due to
use of lactose. The growth rate of the cells increased with lactose
concentration, reflecting the benefit gained by the action of the Lac
proteins that transport and use this sugar (Fig. 2b).
We find that the form of the benefit function is well described by
the established transport and catabolism kinetics of this system (see
Figure 2 | Cost and benefit functions of lac expression in wild-type E. coli.
a, Cost, defined as relative reduction in growth of E. coli wild-type cells
grownin definedglycerolmediumwithvarying amountsofIPTG relativeto
cells grown with no IPTG. The x axis is LacZ protein level relative to LacZ
protein level at saturating IPTG (ZWT). Also shown are the cost of strains
evolved at 0.2mM lactose for 530 generations (data point at the 0.4 point of
the 1.12 point of the x axis; open triangle). The green solid line represents
cost function 1 (equation (1) with h0ZWT¼ 0.09 ^ 0.01 and
¼ 3 ^ 0.1). The red line represents cost function 2 (equation (2)).
The green dotted line represents a linear cost function. b, Benefit of Lac
proteins as a function of lactose. Cells were grown with saturating levels of
IPTG and varying levels of lactose. The growth rate difference is shown
relative to cells grown with no IPTG or lactose. h(ZWT) is the cost of the lac
system at zero lactose, and dZWTis the benefit of lac induction at saturating
lactose levels. The red line indicates the theoretical growth rate, using g ¼
2hðZWTÞþdZWTL=ðKYþLÞ; with KY¼ 0.4mM, h(ZWT) ¼ 0.044 and
dZWT¼ 0.17 (equation (5)). Error bars are the experimental standard
function of Lac protein expression. The fitness function, given by the
difference of cost and benefit, is shown for different concentrations of
lactose. The x axis is the ratio of protein level to the fully induced wild-type
to the uninduced wild-type strain for environments with lactose levels
L ¼ 0.1mM (blue line), L ¼ 0.6mM (green line) and L ¼ 5mM (red line),
according to equation (5). The dot on each line is the predicted optimal
expression level that provides maximal growth (equation (6)). Cells grown
in lactose levels above 0.6mM are predicted to evolve to increased Lac
protein expression (arrow a), whereas cells grown at lactose levels lower
than 0.6mM are predicted to evolve to decreased Lac protein expression
NATURE|Vol 436|28 July 2005
© 2005 Nature Publishing Group
Methods and Supplementary Information). The increase in growth
rate is proportional to the rate of lactose use, and hence to the
enzymatic rate of LacZ:
BðZÞ ¼ d½ZLin?
where Linis the concentration of lactose in the cell, [ZLin] is the
concentration of LacZ bound to lactose, and d is the relative growth
advantage per LacZ molecule at saturating lactose concentration.
Using the known biochemical kinetic parameters of LacYand LacZ
an experimental estimate of the benefit parameter d under the
present conditions: d ¼ 0:17 Z21
system confers a 17% growth advantage at saturating lactose relative
to cells with no lactose.
WT: Thus, full induction of the lac
protein levels in the wild-type bacterium. The growth rate relative to
the growth rate with no Lac expression is given by the difference
between the cost and benefit functions
g ¼ 2hðZÞþBðZÞð4Þ
We find that the cost and benefit functions are such that at each level
of lactose in the environment there exists a well-defined optimal
protein expression level that maximizes growth. The reason for this
optimum is that benefit increases linearly with expression level,
whereas cost increases nonlinearly and thus high expression levels
become unfavourable. The optimal protein expression level depends
on the concentration of lactose in the environment, L (Fig. 3). To
compute the optimum we calculated the protein level Zopt, which
maximizes growth (see equation (6) in Methods for an analytical
solution); that is, where dg/dZ ¼ 0.
the optimal expression is Zopt¼ 0, because the Lac proteins bear
only a cost and no benefit. At high lactose levels, Zoptis greater
than the wild-type expression level ZWT, because increased lac
expression brings additional growth benefit to the cells. The cost–
benefit analysis predicts that the wild-type unrepressed expression
level is optimal (Zopt¼ ZWT) at a lactose concentration of
L0¼ 0.6 ^ 0.1mM.
Does E. coli actually evolve to these optimal expression levels if
supplied with a constant lactose environment? To test this, we
performed laboratory evolution experiments using serial dilution
protocols19. Cells were grown in 10-ml cultures in a defined medium
supplemented with a given lactose concentration. Each day, the cells
log2100 ¼ 6.6 generations of growth per day. Experiments at seven
lactose concentrations were performed in parallel for over 500
At different times during the experiment we measured LacZ
activity using an enzymatic assay (ONPG assay; see Supplementary
Information) and LacZ protein level using quantitative gel electro-
phoresis (see Supplementary Information). We found that, in all
cases, the activity per LacZ enzyme did not measurably change
over the course of the evolutionary experiments. Hence, changes in
LacZ activity were proportional to changes in LacZ protein level
(Supplementary Fig. 6).
We found that LacZ activity and protein level of the cells changed
heritablyduring thecourseoftheexperiment, andapproachedanew
adapted state after 300–500 generations (Fig. 4). Cells grown at low
lactose concentrations (L ¼ 0, 0.1mM and 0.2mM) showed a
decrease in LacZ activity and expression (Fig. 4). Cells grown in
the absence of lactose lost enzyme expression and activity altogether
due to mutations such as a 764-base-pair deletion that included the
entire lac promoter (see Supplementary Information). Cells growing
at L ¼ 0.5mM showed little change in LacZ activity and expression.
Cells growing at lactose concentrations of 1mM, 2mM and 5mM
showed an increase in LacZ activity and expression, although the
increase saturated at about 12% of the wild-type level.
The evolved cells reached LacZ expression levels close to the
predicted optimal levels (Fig. 4). All cost functions give good
predictions at low to intermediate lactose levels (L , 1mM)
(Fig. 4b), predicting, for example, that the wild-type protein levels
are optimal in an environment with L < 0.5mM, as observed. The
cost function that best explains the data at high lactose levels is cost
function 2 (equation (2)). This cost function shows a large cost per
protein at high expression levels, and thus limits the optimal protein
levels in high lactose environments. Hence, the cost and benefit
fitness function measured in the wild-type strain can be used to
evaluate the optimal protein level reached by evolutionary selection
in different environments.
The rate at which the cells converged to their adapted expression
level is shown in Fig. 4a. The adapted expression is reached within
Figure 4 | Experimental evolutionary adaptation of E. coli cells to different
grown for 530 generations in serial dilution experiments with different
lactose levelsis shown asa function of generationnumber.Cells were grown
in 0, 0.1mM, 0.5mM, 2mM and 5mM lactose in a glycerol minimal
medium supplemented with 0.15mM IPTG. Lines are population genetics
simulations. The only fitting parameter in these simulations is the
probability per generation per cell of a mutation that yields the predicted
optimal LacZ level (see Methods). b, Adapted LacZ activity of cells in serial
dilution experiments as a function of lactose concentration, L, relative to
wild-type cells. Data are for more than 530 generations, except for the data
point at 5mM lactose, which is at generation 400. The red line indicates the
theoretical prediction for optimal expression level using cost function 2
(equation (6)). The green solid line indicates the quadratic cost function.
The green dotted line indicates the linear cost function. Error bars are the
experimental standard errors.
NATURE|Vol 436|28 July 2005
© 2005 Nature Publishing Group
about 350 generations for high lactose levels and within 400–500
generations at low lactose levels. These rates are well described by a
population genetics model in which cells grow exponentially and
produce, with probability p, mutants that have the predicted protein
leveland the measured adapted growth rate. The only free parameter
generation, p, of producing a mutant, which ranged between 1027
and 3 £ 1027. The simulations suggest that the mutation occurred
and waslost during dilution many times during theexperiment until
by a single mutation that changed lac expression rather than a
combination of several mutations11. The curves in every run of the
stochastic simulation are nearly identical, showing that the selection
dynamics can be described by a deterministic approximation (see
Supplementary Information). Because the mutation rate in E. coli is
on the order of 1029per base pair per generation, the observed
mutant probabilities, p < 1027, suggest that there is an effective
‘mutational target’ on the order of 100 different base pairs that can
give rise to each of the optimal phenotypes.
At the highest level of lactose, 5mM, cells first converged to the
predictedoptimal expressionlevel,and then, around generation 400,
mutations in other systems that provide an advantage in the present
Indeed, direct measurements of the cost function of this mutant
suggest a threefold increase in the parameter M (Supplementary
Fig. 8). In contrast, direct measurement of the cost functions of
strains that evolved at lower lactose levels shows a cost function that
is the same as that of the wild-type strain (Fig. 2a).
This study provides evidence that cells can rapidly evolve towards
protein expression levels that are optimal solutions of a cost–benefit
problem. The cost and benefit fitness function measured in the wild-
type strain was found to predict optimal protein levels in different
lactose environments, which were then reached in direct evolution-
ary experiments. It would be intriguing to test whether cost–benefit
to form and test a theory of optimal gene circuit design6,8,26–28. This
could explain the observed convergence of different organisms to
similar network motifs16,28,29. More generally, this study suggests
that small expression differences between microorganisms can be
due to selected biological functionality rather than random drift.
of the interaction between biological circuitry, evolution and the
Strains and media. E. coli MG1655 (E. coli genetic stock centre) was used. All
experiments were in M9 defined medium consisting of M9 salts, 1mM MgSO4,
0.1mM CaCl2, 0.05% casamino acids, 0.1% glycerol, 0.15mM IPTG and
specified concentrations of D-lactose (Sigma). This concentration of IPTG had
no measurable effect on growth of a strain deleted for the lac operon (data not
shown). Use of glycerol in the medium allowed study of evolution also in the
absence of lactose. The bacteria used lactose and glycerol simultaneously
throughout growth (Supplementary Fig. 1). Previous studies have shown that
cells grown in a chemostat with limiting lactose evolve to greatly increase the
the present experiments.
Growth rate measurements. The exponential growth rate difference of two
strains was measured by comparing 48 cultures of each strain grown in a
checkerboard pattern ona 96-well plate,yielding anaccuracyofabout0.8% (see
Supplementary Information for more details). Multiple assays were averaged in
cases where lower standard errors were required. Exponential growth rate in the
Serial dilution experiments. Ten-millilitre cultures were grown in 50-ml tubes
fresh tube. Samples were frozen (2808C) every 3days. Lactose levels in five of
thetubes(5mM,2mM,1mM,0.5mM,0.2mM)werehigh enough tocausefull
maximal in all tubes, including those with 0mM and 0.1mM lactose, each tube
also contained a saturating level of IPTG.
b-Galactosidase activity and protein level measurements. A substrate of
b-galactosidase (LacZ) that gives optical readout (ONPG) was used in a
multi-well plate reader to obtain high accuracy measurements of activity (see
Supplementary Information for details). LacZ protein levels were quantified
using SDS gel electrophoresis, and relative expression was quantified by
comparison to control lanes with appropriate mixtures of induced and
uninduced wild-type cell lysates (Supplementary Fig. 6).
a growth rate that is a saturated function30of an internal resource R, given by
g ¼ bR/(K þ R) where b is the maximal growth rate and K is the resource
level for half maximal growth. Production of a unit of protein Z reduces the
resource R by 1. Hence, gðZÞ ¼ bðR21ZÞ=ðK þR21ZÞ: This yields a cost
hðZÞ ¼ ðgð0Þ2gðZÞÞ=gð0Þ ¼ h0Z=ð12Z=MÞ; as in equation (2), where h0¼
1K=RðK þRÞ and M ¼ (K þ R)/1. Note that the cost function can not diverge
but rather reaches a maximum value of h(Z) ¼ 1 when 1Z ¼ R, resulting in
g ¼ 0.
model of the lac system22–24(see Supplementary Information) suggests that, to
through the LacY permease at all but the highest lactose concentrations:
Vz½ZLin? < VY½YL? < VY
where Vzis the velocity of LacZ, VYis the velocity of LacYand KY¼ 0.4mM is
the Michaelis constant of LacY24. This yields a cost–benefit fitness function,
parameter d includes the ratio Y/Z as well as VYand Vz):
g ¼ 2
The optimum of this, dg/dZ ¼ 0, occurs at an optimal protein level of
Zopt¼ M 12
Note that Zopt¼ 0 for lactose levels lower than Lc¼ KYðd=h021Þ21<
0:057mM because cost exceeds benefit. The parameters are M ¼ 1.8ZWT,
h0/d ¼ 0.02/0.17 ¼ 0.12 and KY¼ 0.4mM.
Similar results for Zoptwere found by numerically solving the detailed
transport model and finding the optimal expression by varying LacY and
LacZ levels (Supplementary Fig. 9). Note that the analytical approximation
(equations (5) and (6)) underestimates the growth advantage of the adapted
strains relative to the wild-type strain in the same environment, except near
L ¼ 0 and L ¼ 0.6mM, where predicted growth advantage matches the
experimental results. The detailed simulation gives better predictions for
the growth advantage of the adapted strains (Supplementary Fig. 9) at all
Simulations ofevolution rate.Populationgeneticssimulationswere performed
as follows7(see Supplementary Information for details): a population of wild-
typecellsgrewexponentiallyatgrowthrateg0,growing eachsimulatedday from
N0¼ 108cells to Nf¼ 1010cells. Mutants were formed with a probability p per
generation per cell. The mutants grew at rate g0þ Dg, with relative LacZ
expression DA. The parameter Dg was set equal to the measured growth rates
of the adapted strains (Supplementary Fig. 9), and DA was used from the
optimum of the cost and benefit fitness function (equations (5) and (6)). At the
end of each simulated day, 1/100 of the population was passed to the next
simulated day, of which the fraction of mutants was determined by a random
binomialprocess. The resulting dynamics show that the mutants eventually take
over thepopulation.Thesimulationshaveonlyonefreeparameter, p, whichwas
fitted to the data of Fig. 4a, resulting in p ¼ 6:5£1026^2£1026; 3£1027^
1£1027; 3£1027^1£1027and 3£1027^1£1027for L ¼ 0mM, 0.1mM,
2mMand 5mM, respectively. Simulations suggested that adaptationwas due to
a single mutation, except at L ¼ 0, in which two mutations occurred: one
that affected LacZ protein level and the other that increased growth rate in
glycerol3, in agreement with measurements on this adapted strain (Supplemen-
tary Fig. 11).
Received 11 April; accepted 19 May 2005.
1.Elena, S. F. & Lenski, R. E. Evolution experiments with microorganisms: the
dynamics and genetic bases of adaptation. Nature Rev. Genet. 4, 457– -469
NATURE|Vol 436|28 July 2005
© 2005 Nature Publishing Group
2. Orr, H. A. The genetic theory of adaptation: a brief history. Nature Rev. Genet. Download full-text
6, 119– -127 (2005).
Ibarra, R. U., Edwards, J. S. & Palsson, B. O. Escherichia coli K-12 undergoes
adaptive evolution to achieve in silico predicted optimal growth. Nature 420,
186– -189 (2002).
Hartwell, L. H., Hopfield, J. J., Leibler, S. & Murray, A. W. From molecular to
modular cell biology. Nature 402, C47– -C52 (1999).
Rosen, R. Optimality Principles in Biology (Butterworths, London, 1967).
Savageau, M. A. Biochemical Systems Analysis: a Study of Function and Design in
Molecular Biology (Addison-Wesley, Reading, Massachusetts, 1976).
Hartl, D. L. & Clark, A. G. Principles of Population Genetics (Sinauer, Sunderland,
Heinrich, R. & Schuster, S. The Regulation of Cellular Systems (Chapman and
Hall, New York, 1996).
Maynard Smith, J. & Szathmary, E. The Major Transitions in Evolution (Oxford
Univ. Press, Oxford, 1997).
10. Hartl, D. L. & Dykhuizen, D. E. The population genetics of Escherichia coli. Annu.
Rev. Genet. 18, 31– -68 (1984).
11.Liebermeister, W., Klipp, E., Schuster, S. & Heinrich, R. A theory of optimal
differential gene expression. Biosystems 76, 261– -278 (2004).
12. Muller-Hill, B. The lac Operon: a Short History of a Genetic Paradigm (Walter de
Gruyter, New York, 1996).
13. Koch, A. L. The protein burden of lac operon products. J. Mol. Evol. 19,
455– -462 (1983).
14. Nguyen, T. N., Phan, Q. G., Duong, L. P., Bertrand, K. P. & Lenski, R. E. Effects of
carriage and expression of the Tn10 tetracycline-resistance operon on the
fitness of Escherichia coli K12. Mol. Biol. Evol. 6, 213– -225 (1989).
15. Fay, J. C., McCullough, H. L., Sniegowski, P. D. & Eisen, M. B. Population
genetic variation in gene expression is associated with phenotypic variation in
Saccharomyces cerevisiae. Genome Biol. 5, R26 (2004).
16. Conant, G. C. & Wagner, A. Convergent evolution of gene circuits. Nature
Genet. 34, 264– -266 (2003).
17. Stephanopoulos, G. & Kelleher, J. Biochemistry. How to make a superior cell.
Science 292, 2024– -2025 (2001).
18. Segre, D., Vitkup, D. & Church, G. M. Analysis of optimality in natural and
perturbed metabolic networks. Proc. Natl Acad. Sci. USA 99, 15112– -15117 (2002).
19. Cooper, T. F., Rosen, D. E. & Lenski, R. E. Parallel changes in gene expression
after 20,000 generations of evolution in Escherichia coli. Proc. Natl Acad. Sci.
USA 100, 1072– -1077 (2003).
20. Dykhuizen, D. E., Dean, A. M. & Hartl, D. L. Metabolic flux and fitness. Genetics
115, 25– -31 (1987).
21. Honisch, C., Raghunathan, A., Cantor, C. R., Palsson, B. O. & van den Boom, D.
High-throughput mutation detection underlying adaptive evolution of
Escherichia coli-K12. Genome Res. 14, 2495– -2502 (2004).
22. Kremling, A. et al. The organization of metabolic reaction networks. III.
Application for diauxic growth on glucose and lactose. Metab. Eng. 3, 362– -379
23. Wong, P., Gladney, S. & Keasling, J. D. Mathematical model of the lac operon:
inducer exclusion, catabolite repression, and diauxic growth on glucose and
lactose. Biotechnol. Prog. 13, 132– -143 (1997).
24. Yildirim, N., Santillan, M., Horike, D. & Mackey, M. C. Dynamics and bistability
in a reduced model of the lac operon. Chaos 14, 279– -292 (2004).
25. Bremer, H. & Dennis, P. P. in Escherichia coli and Salmonella (ed. Neidhardt,
F. C.) 1553 (American Society for Microbiology, Washington DC, 1996).
26. Yokobayashi, Y., Weiss, R. & Arnold, F. H. Directed evolution of a genetic
circuit. Proc. Natl Acad. Sci. USA 99, 16587– -16591 (2002).
27. Endy, D., You, L., Yin, J. & Molineux, I. J. Computation, prediction, and
experimental tests of fitness for bacteriophage T7 mutants with permuted
genomes. Proc. Natl Acad. Sci. USA 97, 5375– -5380 (2000).
28. Dekel, E., Mangan, S. & Alon, U. Environmental selection of the feed-forward
loop circuit in gene-regulation networks. Phys. Biol. 2, 81– -88 (2005).
29. Milo, R. et al. Network motifs: simple building blocks of complex networks.
Science 298, 824– -827 (2002).
30. Monod, J. The growth of bacterial cultures. Annu. Rev. Microbiol. 3, 371– -394
Supplementary Information is linked to the online version of the paper at
Acknowledgements We thank M. Elowitz, R. Kishony, G. Sela, B. Shraiman and
all members of our laboratory for discussions. We thank the NIH, ISF and
Minerva for support. E.D. thanks the Clore postdoctoral fellowship for support.
Author Information Reprints and permissions information is available at
npg.nature.com/reprintsandpermissions. The authors declare no competing
financial interests. Correspondence and requests for materials should be
addressed to U.A. (firstname.lastname@example.org).
NATURE|Vol 436|28 July 2005
© 2005 Nature Publishing Group