Redox-Active Antibiotics Control Gene Expression and Community Behavior in Divergent Bacteria
It is thought that bacteria excrete redox-active pigments as antibiotics to inhibit competitors. In Pseudomonas aeruginosa, the endogenous antibiotic pyocyanin activates SoxR, a transcription factor conserved in Proteo- and Actinobacteria. In Escherichia coli, SoxR regulates the superoxide stress response. Bioinformatic analysis coupled with gene expression studies in P. aeruginosa and Streptomyces coelicolor revealed that the majority of SoxR regulons in bacteria lack the genes required for stress responses, despite the fact that many of these organisms still produce redox-active small molecules, which indicates that redox-active pigments play a role independent of oxidative stress. These compounds had profound effects on the structural organization of colony biofilms in both P. aeruginosa and S. coelicolor, which shows that “secondary metabolites” play important conserved roles in gene expression and development.
Redox-Active Antibiotics Control Gene
Expression and Community Behavior
in Divergent Bacteria
Lars E. P. Dietrich,
Tracy K. Teal,
Dianne K. Newman
It is thought that bacteria excrete redox-active pigments as antibiotics to inhibit competitors. In
Pseudomonas aeruginosa, the endogenous antibiotic pyocyanin activates SoxR, a transcription
factor conserved in Proteo- and Actinobacteria. In Escherichia coli, SoxR regulates the superoxide
stress response. Bioinformatic analysis coupled with gene expression studies in P. aeruginosa and
Streptomyces coelicolor revealed that the majority of SoxR regulons in bacteria lack the genes
required for stress responses, despite the fact that many of these organisms still produce
redox-active small molecules, which indicates that redox-active pigments play a role independent
of oxidative stress. These compounds had profound effects on the structural organization of colony
biofilms in both P. aeruginosa and S. coelicolor, which shows that Wsecondary metabolitesW
play important conserved roles in gene expression and development.
he opportunistic pathogen Pseudomonas
aeruginosa releases phenazines, redox-
active antibiotics (1, 2). Historically, at-
tention has focused on their toxicity in bacteria
and eukaryotes, which arises from the production
of superoxide (3, 4). More recently, however, it
has been recognized that these compounds have
diverse physiological functions, particularly un-
der oxygen-limited conditions (2, 5–7). We found
that the blue phenazine pyocyanin is an inter-
cellular signal that triggers a specific response in
P. aeruginosa, with only 22 genes up-regulated,
including the complete SoxR regulon (8). The
transcription factor SoxR is well-characterized in
the enteric bacteria Escherichia coli and Salmo-
nella enterica serovar Typhimurium as a stress-
response regulator. In these bacteria, SoxR
activates the transcrip tion factor SoxS, which con-
trols genes involved in the removal of superoxide
and nitric oxide and protection from organic sol-
vents and antibiotics. That SoxR-regulated genes
were triggered by pyocyanin was, therefore, ini-
Department of Biology, Massachusetts Institute of Technol-
ogy, 77 Massachusetts Avenue, Cambridge, MA 01239, USA.
Department of Earth and Planetary Science, Massachusetts
Institute of Technology, 77 Massachusetts Avenue, Cam-
bridge, MA 01239, USA.
Howard Hughes Medical Institute,
Massachusetts Institute of Technology, 77 Massachusetts
Avenue, Cambridge, MA 01239, USA.
Division of Biology,
California Institute of Technology, 1200 East California
Boulevard, Pasadena, CA 91125, USA.
*To whom correspondence should be addressed. E-mail:
Fig. 1. (A) Distribution of SoxR and
SoxS among phyla of the domain
Bacteria. A BLAST search for E. coli
SoxR and SoxS was performed, and
SoxS was found only in enterics. SoxR
homologs were identified in 176 a-, b-,
d-, and g-Proteobacteria and Actino-
bacteria. All of these homologs con-
tain the SoxR-specific cysteine motif
CI[G/Q]CGC[L/M][S/L]XXXC required for
binding of the [2Fe-2S] cluster (31).
The number of hits within respective
phyla are indicated, followed by the
total number o f genomes surveyed.
Members of these phyla (in black) are
noted for their ability to produce and
excrete redox-active small molecules,
such as phenazines (18)andactino-
rhodin (20). Representative structures
are shown. The tree was constructed
using the ARB neighbor joining
method from 16S ribosomal RNAs of
604 bacterial species. The bar repre-
sents 0.1 base substitutions per nucle-
otide. (B) Gene categories regulated by
SoxR. Only in enterics are soxRboxes
located upstream of soxS, which con-
firms the uniqueness of this network. In
all other soxR-containing Proteo- and
Actinobacteria, soxRboxes are mainly
found upstream of five gene types as
indicated; 100% correspond to 16 a-
Proetobacteria, 18 b-Proteobacteria,
27 enteric, 38 non-enteric g-Proteo-
bacteria, or 22 Actinobacteria.
“Dehydr.” stands for putative dehy-
drogenases; “oxygen.” for putative
mono- or dioxygenases; “L-PSP” pu-
tative L-PSP endoribonucleases, a family of ribonucleases that share homology with the rat liver
perchloric acid-soluble protein, L-PSP; and “methyl./acetylase” putative methyl- or acetyltrans-
ferases. Additional annotation information can be found at http://soxRbox.mit.edu.
www.sciencemag.org SCIENCE VOL 321 29 AUGUST 2008
tially not surprising, as this would be consistent
with the conventional view of phenazines as
toxic compounds (9–11).
However , recent studies of the SoxR regulons
in pseudomonads indicate an alternative role to
the E. coli SoxR-SoxS paradigm. First, su-
peroxide is not the sole activator of SoxR, as
P. aeruginosa pyocyanin also induces the ex-
pression of its regulon under anoxic conditions
(8). Second, SoxRs from Pseudomonas putida
(12)andP. aeruginosa (8, 13, 14) do not control
any of the genes typically involved in superoxide
resistance and detoxification, rather, SoxR from
P. aeruginosa up-regulates expression of two
transporters and a putative monooxygenase (fig.
S1A). Third, P. aeruginosa soxR mutants show
no decrease in resistance to superoxide, unlike
E. coli soxR mutants (14). These observations led
us to hypothesize that redox-active signaling mol-
ecules, such as phenazines, might control other
aspects of microbial behavior.
In this study, we investigated the distribu-
tion of the E. coli–type oxidative stress re-
sponse by performing a BLAST search for SoxR
and SoxS in the bacterial domain (15). SoxR
was found in sequences from 176 strains in the
phyla Proteobacteria and Actinobacteria (Fig.
1A), 123 of which come from completed ge-
nomes. The occurrence of SoxS was restricted
to the family Enterobacteriaceae. T o identify
alternative SoxR targets in non-enterics, we
searched all available complete bacterial ge-
nomes (616) for the presence of soxRboxes (i.e.,
SoxR-binding sites in the promoter regions of
target genes) using a position weight matrix
(PWM) derived from the soxRbox sequences of
12 diverse SoxR-containing bacteria (fig. S1B).
This PWM permits statistically robust predictions
of SoxR binding to a soxRbox. Of the 123 soxR-
containing genomes, 121 contain soxRboxes.
SoxRboxes were also found in 27 genomes (19
were Firmicutes) that do not contain a soxR homo-
log. The results of our analysis (table S1 and http://
soxRbox.mit.edu) were consistent with gene
expression studies made in the Gram-negative
bacteria E. coli, S. enterica (10), P. aeruginosa
(8, 13, 14), and Agrobacterium tumefaciens
(16), which validates our search algorithm.
The organization found in E. coli (fig. S1),
with one soxRbox upstream of soxS and no other
soxRboxes in the genome, occurred only in en-
terics (27 genomes) (Fig. 1B). Two enterics
contained an additional soxRbox upstream of
putative dioxygenases. The remaining organisms
contained one or more soxRboxes upstream of
genes other than soxS. These SoxR target genes
fell into five main categories, including trans-
porters, oxygenases, dehydrogenases, putative
acetyl- or methyltransferases, and L-PSP endo-
ribonucleases (L-PSP is defined in the legend to
Fig. 1B), all of which are potentially involved in
the transformation or transport of small mole-
cules, such as antibiotics (17). The occurrence of
soxR upstream of soxS in enterics thus appears to
be an evolutionary exception confirmed by the
unique branching of the enteric orthologs on a
SoxR phylogenetic tree (fig. S2).
Given that many of the bacteria that contain
soxRboxes are producers of redox-active anti-
biotics (18) (Fig. 1A), it seems reasonable that
SoxR may have evolved to regulate their trans-
port and/or turnover. We chose to work with
the Gram-positive actinomycete Str eptomyces
coelicolor A3(2) to test whether the SoxR
regulon is up-regulated in response to endoge-
nous small molecules, because members of this
phylum are widely recognized as important
sources of antibiotics (19). S. coelicolor A3(2)
produces the blue pigment actinorhodin and
the red undecylprodigiosin (fig. S3) (20). On
the basis of our analysis, we predicted a SoxR
regulon comprising two genes for S. coelicolor
A3(2), encoding putative redox enzymes (Fig.
2A). Expression of these genes in the wild type
(strain M145) was compared with that in a mu-
Fig. 2. The putative S. coelicolor A3(2) SoxR
regulon is specifically up-regulated by pigments.
(A) Genes predicted to be regulated by SoxR are
shown in gray. (B) RNA extracted from plate-
grown S. coelicolor A3(2) M145 and the pigment-
null mutant M512 was used to generate cDNA for
quantitative RT-PCR (15). Signals were stan-
dardized to SCO4548 (32). The experiment was
done in triplicate, and data reported represent
the mean TSD. SoxR itself (SCO1697) was also
tested for changes in gene expression.
Fig. 3. Phenazine production modulates colony morphology in P. aeruginosa PA14. P. aeruginosa
cultures were spotted onto agar plates containing Congo Red and Coomassie Blue, and incubated
at 20°C for 6 days. The phenazine null strain (Dphz) started to wrinkle on day 2, the wild type (wt)
wrinkled on day 3, and the soxR and mexGHI-opmD deletion strains wrinkled on day 5, whereas a
pyocyanin overproducer (DKN370) remained smooth and white after 6 days.
29 AUGUST 2008 VOL 321 SCIENCE www.sciencemag.org
tant that does not produce the two pigments
(st rain M512) ( 21). Both predicted SoxR-
regulated genes were significantly up-regulated
in the wild type relative to the pigment-null mu-
tant (~250- to 6000-fold), as determined by quan-
titative reverse transcription polymerase chain
reaction (R T-PCR) (Fig. 2B), which confirmed
that pigment production stimulated gene expres-
sion via SoxR. Hence, the primary function of
SoxR in S. coelicolor,asinP. aeruginosa, is not
to activate a response to superoxide but to me-
diate a response to endogenous pigments.
Recently, we showed that phenazines are
terminal signals in P. aeruginosa’s quorum-
sensing cascade (8). The importance of quorum
sensing for the coordination of many bacterial
communities is well established (22). Moreover , a
phenazine-dependent effect on biofilm formation
has been reported in P. aur eofaciens (23). Together
with our bioinformatic SoxR results, these obser-
vations led us to hypothesize that redox-active
pigments might act as signals to modulate the
structural organization of cellular communities.
To investigate the effect of extracellular pig-
ments on community development, we began
by focusing on P. aeruginosa PA14. We spotted
10-ml aliquots of late exponential-phase cultures
onto agar plates and incubated them at room
temperature for 8 days. Under these conditions,
wild-type cells initially formed smooth colonies
(Fig. 3). After 4 days of incubation, the colo-
nies began to wrinkle and reached a maximum
area of ~2.5 cm
(Figs. 3 and 4A). However, the
phenazine-null mutant formed severely wrinkled
colonies within 2 days of incubation (Fig. 3),
which subsequently flattened and spread to ~3.5
(Fig. 4A). In contrast, a mutant that over-
produced pyocyanin (DKN370) remained smooth
and compact (Figs. 3 and 4A). These results
demonstrated a role for phenazines in control-
ling bacterial colony size and structure.
Phenazines are diffusible molecules that
may influence phenotype over distance. Indeed,
we found that adding pyocyanin to the growth
medium (fig. S4A) or spotting the phenazine
overproducer next to the phenazine deletion
mutant (fig. S4B) resulted in the formation of
smooth compact colonies. We tested the role of
SoxR in mediating the effect of phenazines on
colony morphology by making a SoxR-deletion
mutant. However, the mutant behaved similarly
to the pyocyanin overproducer, and colonies re-
mained smooth for 4 days (Fig. 3). As for the
overproducer, DsoxR released more pyocyanin
into the agar than the wild type (Fig. 4B). There
thus appears to be a direct correlation between
pyocyanin release and colony smoothness.
To further analyze the DsoxR phenotype, we
tested P. aeruginosa mutants disrupted in the
SoxR target genes PA14_35160 (encoding a pu-
tative monooxygenase), mexGHI-opmD [encod-
ing a resistance-nodulation–cell division (RND)
efflux pump], and PA14_16310 [encoding a
major facilitator superfamily (MFS) transporter].
Deletions of PA14_35160 and PA14_16310 did
not affect colony morphology; however, the loss
of mexGHI-opmD produced a phenotype that
looked like the DsoxR mutant, i.e., wrinkling was
slow (Fig. 3), and was accompanied by a slightly
elevated pyocyanin release (Fig. 4B). By con-
trast, the release of the yellow phenazine-1-
carboxylate (PCA) and an unidentified red
phenazine (possibly 5-methyl-PCA), decreased
by 10 and 60%, respectively, in the mexGHI-
opmD mutant relative to the wild type, which
indicates that mexGHI-opmD is a general phen-
azine transporter. Antibiotic biosynthetic genes
are often found adjacent to their cognate trans-
porter (24), so it is interesting to note that the
mexGHI-opmD operon is clustered with the phen-
azine biosynthetic genes phzM, phzA1-G1,and
phzS (fig. S5A).
PhzA1-G1 synthesizes the yellow phenazine
PCA, and PhzM methylates PCA to yield the
red phenazine 5-methyl-PCA, which is then hy-
droxylated by PhzS to form pyocyanin. Trans-
poson insertion mutants in mexI and opmD of
P. aeruginosa PAO1 are known to accumulate an
unidentified toxic compound that causes an
elongated lag phase in planktonic cultures (25).
We found a similar phenotype in P. aeruginosa
PA14 (fig. S5B), which is probably caused by an
intracellular accumulation of phenazines. Our
experiments showed that SoxR target genes do
not directly influence colony development; in-
stead, SoxR regulates the efflux of phenazines
via the RND transporter MexGHI-OpmD. Al-
though yellow PCA and red phenazine are
retained in the mexGHI-opmD mutant, the re-
lease of pyocyanin indicates an alternative efflux
mechanism favoring pyocyanin. Compensatory
changes in expression of RND efflux pumps are
well known to occur in P. aeruginosa (26).
To determine whether the phenotypic effects
of pigment production observed for P. aeruginosa
were unique to this organism or more general-
izable, we performed analogous experiments
with S. coelicolor A3(2). As for P. aeruginosa,a
pigment-defective mutant of S. coelicolor
adopted a more wrinkled morphology than the
respective wild type (Fig. 4C). The mechanisms
whereby pigments control colony morphology
are not understood, but are likely to be complex.
For P. aeruginosa PA14, we know that pyocy-
anin affects the expression of at least 35 genes
other than those in the SoxR regulon (8)andhas
profound effects on the cell’s physiology, in-
Fig. 4. (A) Surface coverage of 35 colonies per strain monitored over 8 days
(TSD). (B) Concentration of pyocyanin release from three colonies into 10 ml
agar supplemented with Congo Red and Coomassie Blue. After 5 days of
growth at room temperature, the cells were scraped off, pyocyanin was
extracted from the agar using chloroform, and extracts were analyzed by high-
performance liquid chromatography. The data reported represent the mean
TSD. (C) Spore suspensions of S. coelicolor A3(2) M145 and the pigment
mutant M512 were spotted and incubated for 5 days on R5
medium at room
temperature. The pigment mutant exhibits a wrinkled morphology, whereas
the wild type takes on a smoother phenotype. Scale bar is 0.5 cm.
www.sciencemag.org SCIENCE VOL 321 29 AUGUST 2008
cluding the redox state of the intracellular ni-
cotinamide adenine dinucleotide [NAD(H)] pool
(27). Any number of these effects may contrib-
ute, both directly and indirectly, to the ultimate
architectures observed. One component that is
likely involved is extracellular polysaccharide
(EPS). Congo Red, a constituent of the agar used
in the experiments shown in Fig. 3, is known to
bind the gluose-rich exopolysaccharide PEL
(28). Because the phenazine-null mutant is bright
red, whereas the pyocyanin overproducer is pale,
we infer there is an inverse relationship between
phenazine and PEL production (Fig. 3). How
phenazines affect the pel genes and how such
changes in EPS composition contribute to colony
morphogenesis remain to be determined.
Pigments excreted by bacteria have long been
assumed to be “secondary” metabolites or even
waste products, owing to the sporadic strain- and
condition-dependent nature of their production
(29). Many of these redox-active compounds are
known to have antibiotic activities toward compet-
ing cells (1, 20), but until recently , their potential to
directly participate in the physiology of the produc-
ing organism has been largely neglected (7). We
now know that small molecules initially character-
ized as antibiotics allow intercellular communica-
tion within bacterial populations (30), and this work
implies a conserved function for redox-active pig-
ment antibiotics of the Gram-negative bacterium
P. aeruginosa and the Gram-positive bacterium
S. coelicolor A3(2). These pigments influence
transcriptional regulation and modulate the phys-
ical characteristics of communities of their pro-
ducers at later stages in their development. Rather
than being “secondar y,” diverse redox-active anti-
biotics may share similar functions of primary
importance throughout the bacterial domain.
References and Notes
1. D. V. Mavrodi, W. Blankenfeldt, L. S. Thomashow,
Annu. Rev. Phytopathol. 44, 417 (2006).
2. A. Price-Whelan, L. E. Dietrich, D. K. Newman,
Nat. Chem. Biol. 2, 71 (2006).
3. S. Mahajan-Miklos, M. W. Tan, L. G. Rahme,
F. M. Ausubel, Cell 96 , 47 (1999).
4. M. Mazzola, R. J. Cook, L. S. Thomashow, D. M. Weller,
L. S. Pierson 3rd, Appl. Environ. Microbiol. 58, 2616
5. M. E. Hernandez, A. Kappler, D. K. Newman, Appl.
Environ. Microbiol. 70, 921 (2004).
6. Y. Wang, D. K. Newman, Environ. Sci. Technol. 42, 2380
7. M. E. Hernandez, D. K. Newman, Cell. Mol. Life Sci. 58,
8. L. E. Dietrich, A. Price-Whelan, A. Petersen, M. Whiteley,
D. K. Newman, Mol. Microbiol. 61, 1308 (2006).
9. S. I. Liochev, L. Benov, D. Touati, I. Fridovich, J. Biol. Chem.
274, 9479 (1999).
10. P. J. Pomposiello, B. Demple, J. Bacteriol. 182, 23 (2000).
11. I. R. Tsaneva, B. Weiss, J. Bacteriol. 172, 4197 (1990).
12. W. Park, S. Pena-Llopis, Y. Lee, B. Demple, Biochem. Biophys.
Res. Commun. 341, 51 (2006).
13. K. Kobayashi, S. Tagawa, J. Biochem. 136, 607 (2004).
14. M. Palma et al., Infect. Immun. 73, 2958 (2005).
15. Materials and methods are available as supporting
material on Science Online.
16. W. Eiamphungporn, N. Charoenlap, P. Vattanaviboon,
S. Mongkolsuk, J. Bacteriol. 188, 8669 (2006).
17. E. Cundliffe, Annu. Rev. Microbiol. 43, 207 (1989).
18. J. M. Turner, A. J. Messenger, Adv. Microb. Physiol. 27,
19. D. A. Hopwood, Streptomyces in Nature and Medicine:
The Antibiotic Makers (Oxford Univ. Press, Oxford, 2007) .
20. K. F. Chater, Philos. Trans. R. Soc. Lond. B Biol. Sci. 361,
21. B. Floriano, M. Bibb, Mol. Microbiol. 21, 385 (1996).
22. D. G. Davies et al., Science 280, 295 (1998).
23. V. S. Maddula, Z. Zhang, E. A. Pierson, L. S. Pierson 3rd,
Microb. Ecol. 52, 289 (2006).
24. K. Tahlan et al., Mol. Microbiol. 63, 951 (2007).
25. S. Aendekerk et al., Microbiology 151, 1113 (2005).
26. X. Z. Li, N. Barre, K. Poole, J. Antimicrob. Chemother. 46,
27. A. Price-Whelan, L. E. Dietrich, D. K. Newman,
J. Bacteriol. 189, 6372 (2007).
28. L. Friedman, R. Kolter, Mol. Microbiol. 51, 675 (2004).
29. R. P. Williams, Bacteriol. Rev. 20, 282 (1956).
30. G. Yim, H. H. Wang, J. Davies, Philos. Trans. R. Soc.
London B Biol. Sci. 362, 1195 (2007).
31. Single-letter abbreviations for the amino acid residues are as
I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg;
S, Ser; T, Thr; V, Val; W, Trp; Y, Tyr; and X, any amino acid.
32. S. Mehra et al., J. Ind. Microbiol. Biotechnol. 33, 159 (2006).
33. We thank P. Straight (Harvard University); M. Bibb and
A. Hesketh (John Innes Centre, UK) for providing
S. coelicolor A3(2) strains M145 and M512; and
N. C. Caiazza, C. T. Brown, and B. Wold for helpful
discussions. This work was supported by an EMBO Long
Term Fellowship (L.E.P.D.) and grants from the Packard
Foundation and Howard Hughes Medical Institute (D.K.N.).
Supporting Online Material
Materials and Methods
Figures S1 to S5
19 May 2008; accepted 24 July 2008
Solution Structure of the Integral
Human Membrane Protein VDAC-1
in Detergent Micelles
Robert G. Garces,
* Thomas J. Malia,
*† Vladislav Y. Orekhov,
The voltage-dependent anion channel (VDAC) mediates trafficking of small molecules and ions
across the eukaryotic outer mitochondrial membrane. VDAC also interacts with antiapoptotic
proteins from the Bcl-2 family, and this interaction inhibits release of apoptogenic proteins from
the mitochondrion. We present the nuclear magnetic resonance (NMR) solution structure of
recombinant human VDAC-1 reconstituted in detergent micelles. It forms a 19-stranded b barrel
with the first and last strand parallel. The hydrophobic outside perimeter of the barrel is covered by
detergent molecules in a beltlike fashion. In the presence of cholesterol, recombinant VDAC-1 can
form voltage-gated channels in phospholipid bilayers similar to those of the native protein. NMR
measurements revealed the binding sites of VDAC-1 for the Bcl-2 protein Bcl-x
, for reduced
b–nicotinamide adenine dinucleotide, and for cholesterol. Bcl-x
interacts with the VDAC
barrel laterally at strands 17 and 18.
he integral membrane protein VDAC forms
the primary path for diffusion of metabo-
lites between the mitochondrial intermem-
brane space and the cytosol ( 1, 2). VDAC is
conserved across eukaryotes, with about 30%
sequence identity between yeast and human. The
three isoforms, VDAC-1, VDAC-2, and VDAC-3,
found in humans are 68% to 75% pairwise iden-
tical. All three isoforms allow the exchange of
metabolites through the membrane but have dis-
tinct physiological roles and expression profiles
Numerous reports have suggested that VDAC-1
is involved in mitochondrial apoptosis (5–7).
Apopt otic signals lead to the formation of a mito-
chondrial exit channel that allows the release of
apoptogenic proteins, which in turn cause cell
death by activating executioner caspase or through
other mechanisms (8–10). Functional studies in-
dicate that VDAC-1 closure leads to the opening
of the mitochondrial exit channel (11). The anti-
apoptotic protein Bcl-x
opens the VDAC-1 channel
for trafficking of metabolites and thus inhibits the
release of apoptogenic proteins (12). Direct inter-
action between VDAC-1 and Bcl-x
demonstrated (11, 13).
Insights into VDAC organization have come
from biochemical and biophysical studies (14, 15)
and low-resolution electron microscopy (EM) data
showing that VDAC-1 is a cylindrical channel
withadiameterof20to30Å(16, 17). Electro-
physiological experiments revealed that, at low
membrane potentials of 10 mV, VDAC is in the
open state, but it switches to the closed state at
Department of Biological Chemistry and Molecular Pharma-
cology, Harvard Medical School, Boston, MA 02115, USA.
Department of Biology, University of Maryland, College Park,
MD 20742, USA.
Swedish NMR Centre, University of Goth-
enburg, Box 465, 40530 Gothenburg, Sweden.
*These authors contributed equally to this work.
†Present address: Centocor, Incorporated, Radnor, PA 19087,
‡To whom correspondence should be addressed. E-mail:
29 AUGUST 2008 VOL 321 SCIENCE www.sciencemag.org1206