Genomic footprints of a cryptic plastid endosymbiosis in diatoms.

DOI: 10.1126/science.1172983
, 1724 (2009); 324Science
et al.Ahmed Moustafa,
Endosymbiosis in Diatoms
Genomic Footprints of a Cryptic Plastid
This copy is for your personal, non-commercial use only.
. clicking herecolleagues, clients, or customers by
, you can order high-quality copies for yourIf you wish to distribute this article to others
. herefollowing the guidelines
can be obtained byPermission to republish or repurpose articles or portions of articles
(this information is current as of March 7, 2010 ):
The following resources related to this article are available online at www.sciencemag.org
http://www.sciencemag.org/cgi/content/full/324/5935/1724
version of this article at:
including high-resolution figures, can be found in the onlineUpdated information and services,
http://www.sciencemag.org/cgi/content/full/324/5935/1724/DC1
can be found at: Supporting Online Material
found at:
can berelated to this articleA list of selected additional articles on the Science Web sites
http://www.sciencemag.org/cgi/content/full/324/5935/1724#related-content
http://www.sciencemag.org/cgi/content/full/324/5935/1724#otherarticles
, 8 of which can be accessed for free: cites 20 articlesThis article
7 article(s) on the ISI Web of Science. cited byThis article has been
http://www.sciencemag.org/cgi/content/full/324/5935/1724#otherarticles
6 articles hosted by HighWire Press; see: cited byThis article has been
http://www.sciencemag.org/cgi/collection/microbio
Microbiology
: subject collectionsThis article appears in the following
registered trademark of AAAS.
is aScience2009 by the American Association for the Advancement of Science; all rights reserved. The title
CopyrightAmerican Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005.
(print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by theScience

o
n
M
ar
ch
7
, 2
01
0
w
w
w
.s
ci
en
ce
m
ag
.o
rg
D
ow
nl
oa
de
d
fro
m

Genomic Footprints of a Cryptic
Plastid Endosymbiosis in Diatoms
Ahmed Moustafa,1* Bánk Beszteri,2* Uwe G. Maier,3 Chris Bowler,4,5
Klaus Valentin,2 Debashish Bhattacharya1,6†
Diatoms and other chromalveolates are among the dominant phytoplankters in the world’s
oceans. Endosymbiosis was essential to the success of chromalveolates, and it appears that the
ancestral plastid in this group had a red algal origin via an ancient secondary endosymbiosis.
However, recent analyses have turned up a handful of nuclear genes in chromalveolates that are of
green algal derivation. Using a genome-wide approach to estimate the “green” contribution to
diatoms, we identified >1700 green gene transfers, constituting 16% of the diatom nuclear
coding potential. These genes were probably introduced into diatoms and other chromalveolates
from a cryptic endosymbiont related to prasinophyte-like green algae. Chromalveolates appear
to have recruited genes from the two major existing algal groups to forge a highly successful,
species-rich protist lineage.
Diatoms are well-studied members ofthe putative supergroup Chromalveolata[fig. S1 and supporting online material
(SOM) text] and comprise unicellular, photo-
synthetic, dominant taxa in the marine phyto-
plankton. Diatoms are central to understanding
oceanic primary production and biogeochemistry
(1). Much effort is currently being expended
to develop some taxa as models for genetic
and genomic research as well as sources for
biofuel (2) and nanotechnology (3). We con-
ducted a phylogenomic analysis of the diatom
proteome using complete genome data from
Thalassiosira and Phaeodactylum. This proce-
dure identified 2423 and 2533 (2423/2533)
Phaeodactylum and Thalassiosira genes, re-
spectively (this order of results is used through-
out the paper and SOM), that are derived from
red or green algal sources. Contrary to the ex-
pectation of the chromalveolate hypothesis
(4), however, >70% of these genes are of green
(not red) lineage provenance (Fig. 1, table S1,
and fig. S7). This green gene contribution con-
stitutes ≈16% of the diatom proteome. Two of
the major topological classes that were uncov-
ered are shown in fig. S2. The first class (fig.
S2A) contains 442/442 trees in which both red
and green algae are present, but there is robust
bootstrap support for the green algae plus dia-
tom (and other chromalveolates) clade. Of
these trees, 144/133 show the green algal and
diatom sequences to diverge within the Plantae
kingdom, which is composed of green algae and
plants, glaucophytes, and red algae (5, 6). An
example is phytoene desaturase (fig. S2A), which
is an early enzyme in plastid carotenoid biosyn-
thesis. It was previously reported that 5 of the
16 genes in this photoprotective pathway are
of green algal origin (7), and their occurrence
in chromalveolates probably ensures a high
photosynthetic efficiency under fluctuating light
(8). The remaining 298/309 trees indicate an in-
dependent origin of the gene in the donor green
algae, with respect to other Plantae. A second
major class of trees shows an independent gene
origin in prasinophytes relative to other green
algae and plants, before being transferred to
diatoms and other chromalveolates (Fig. 2). The
absence of red algal homologs in some trees
in this class may be explained by gene loss in
the reduced nuclear genome of the red algal
representative in our database, Cyanidioschyzon
merolae (SOM text). An example tree from this
class is a member of the isoprenylcysteine car-
boxyl methyltransferase superfamily (fig. S2B).
We found four genes that encode the following
gene products: naphthoate synthase (GenBank GI
number 219114006), heme oxygenase (GenBank
GI number 219117865), pyruvate dehydrogenase
(GenBank GI number 219119135), and GUN4-
like protein (GenBank GI number 219127880),
which are retained in red algal plastid genomes
but absent from this red-derived organelle ge-
nome in diatoms. These sequences are present in
the diatom nucleus but are of green algal der-
ivation. This suggests that red plastid–encoded
genes were lost if green homologs were already
present in the host nucleus.
To identify the putative sources of the dia-
tom green genes, we examined their distribution
among the green lineages (Viridiplantae). The
Viridiplantae comprise two well-supported phyla,
the Chlorophyta (most green algae, such as
Chlamydomonas, in the core chlorophytes and
the prasinophytes) and the Streptophyta (char-
ophyte green algae and all land plants; Fig. 2A).
The prasinophytes include the world’s smallest
eukaryotes (the picoeukaryote Ostreococcus; cell
diameter ≈1 mm), which are part of a morpho-
logically diverse group of paraphyletic lineages
diverging at the base of the Chlorophyta (9). We
found that 637/716 diatom green genes (36/41%)
trace their origin to the prasinophytes in our data
set (Micromonas and Ostreococcus; Mamiellales
clade) of which 167/175 are shared with other
Chlorophyta (71/67 genes; Chlamydomonas
and Volvox) or Streptophyta (23/40 genes; Ara-
bidopsis, Oryza, Physcomitrella, and Zea) or by
both phyla (73/68 genes; Fig. 2B). These 167/175
genes have a putative ancient origin in Viridi-
plantae. Streptophyte- and core chlorophyte–
specific donors account for 192/177 and 145/170
genes, respectively (Fig. 2C). Many of these
genes may ancestrally have been present in the
Viridiplantae and lost by prasinophytes and/or
other green lineage members, whereas the re-
mainder represent independent horizontal gene
transfers (HGTs) into streptophytes and core
chlorophytes. In spite of the reduced nuclear ge-
nome of the prasinophytes in our study (≈9000
protein-encoding genes) as compared to the
larger genomes of core chlorophytes and strepto-
phytes (≈15,000 and ≈30,000 protein-encoding
genes, respectively), 470/541 genes are shared ex-
clusively between prasinophytes and diatoms (Fig.
2, B and C), of which 462/502 (98 and 93%) are
present in expressed sequence tag (EST) libraries
from Phaeodactylum and Thalassiosira. Because
of their specific affiliation with picoprasinophytes,
these genes are unlikely to represent missing se-
quences from Cyanidioschyzon. This diatom green
gene set may therefore be gene recruitments via
HGT in picoprasinophytes that were later trans-
1Interdisciplinary Program in Genetics, University of Iowa,
Iowa City, IA 52242, USA. 2Alfred Wegener Institute for
Polar and Marine Research, Am Handelshafen 12, 27570
Bremerhaven, Germany. 3Zellbiologie, Philipps-Universität
Marburg, Marburg, Germany. 4CNRS UMR8186, Depart-
ment of Biology, Ecole Normale Supérieure, 46 rue d’Ulm,
75005 Paris, France. 5Stazione Zoologica Anton Dohrn,
Villa Comunale, I-80121 Naples, Italy. 6Department of Biol-
ogical Sciences and the Roy J. Carver Center for Comparative
Genomics, University of Iowa, Iowa City, IA 52242, USA.
*These authors contributed equally to this work.
†To whom correspondence should be addressed. E-mail:
debashi-bhattacharya@uiowa.edu
0
500
1000
1500
2000
2500
3000 Viridiplantae
Rhodophyta
Unresolved
G
en
e
fa
m
ili
es
Th
al
as
si
os
ira
Ph
ae
od
ac
ty
lu
m
N
um
be
r o
f g
en
es
Fig. 1. Diatom genes of a red or green algal
origin that were identified using phylogenomic
analysis of complete genome data. Each bar rep-
resents the total number of algal genes in the
corresponding diatom species. The “gene fami-
lies” bar indicates the total number of transferred
genes in both diatoms after clustering the data
into gene families through single-linkage hierar-
chical clustering. The “unresolved” category indi-
cates that red and green algae are sisters of each
other in the tree and monophyletic with diatoms
(and other chromalveolates).
26 JUNE 2009 VOL 324 SCIENCE www.sciencemag.org1724
REPORTS

o
n
M
ar
ch
7
, 2
01
0
w
w
w
.s
ci
en
ce
m
ag
.o
rg
D
ow
nl
oa
de
d
fro
m

ferred to the diatom (chromalveolate) nucleus.
These sequences could hold clues to the evolu-
tion of prasinophyte green algae and their great
success in different aquatic environments (10).
The fourfold higher abundance of green versus
red genes in diatoms raises questions about the
timing of the transfer of the green genes and
whether these sequences were introduced via
a single or multiple endosymbioses, or by un-
precedented levels of HGT in chromalveolates.
In order to address this issue, we determined
the distribution of diatom green genes among
chromalveolates using complete genome data.
Here, the distinction between gene origins via
endosymbiotic gene transfer (EGT) (11) versus
HGT reflects whether the genes can be traced
back to a point source (prasinophyte-like algae)
and are found in most if not all chromalveolates,
versus sporadic gene origin in particular lineages
and from multiple different sources, respectively.
Neither of these outcomes is proof but rather
argues for or against one hypothesis. Using this
approach, we find that 85% of the green genes
can be traced back to the ancestor of both dia-
toms and other Stramenopiles (Fig. 3). Diatoms
share 46/55 green genes with the obligate par-
asites apicomplexans and 54/63 genes with
the plastid-lacking ciliates. Analysis of genome
data from the distantly related photosynthetic
coccolithophorid Emiliania huxleyi, which is a
haptophyte sister to cryptophytes (fig. S1),
identified >400 green genes shared with dia-
toms. The inclusion of ESTs from dinoflagel-
lates and cryptophyte algae shows that even
when these partial data are used, 10 and 3% of
the diatom green genes are shared with these
groups, respectively (fig. S3). Given these re-
sults, we suggest that despite extensive gene
losses among nonphotosynthetic lineages such
as ciliates and apicomplexans, the most likely
explanation is that a large proportion of the dia-
tom green genes is of an ancient provenance and
predate the split of cryptophytes and haptophytes
from other chromalveolates.
Taken together, our results provide evidence
of a prasinophyte-like endosymbiont in the com-
mon ancestor of chromalveolates. As discussed
above, prasinophytes are an anciently diverged
paraphyletic group of green algae (12) that was
present early on in chromalveolate evolution. In
the fossil record, prasinophytes are widely dis-
tributed by the Early Cambrian (13). These cells
may well have been an abundant prey source for
the chromalveolate ancestor. The alternative ex-
planation of chromalveolate polyphyly would
imply an unprecedented number of independent
gains (≈400) of the same green genes by diatoms
and haptophytes. Therefore, our results provide
strong support for a shared evolutionary history
for these disparate chromalveolate lineages. In
substantiated cases of serial endosymbiosis, the
most recent endosymbiont provides the plastid,
whereas the nuclear genome bears the footprints
of past events. The dinoflagellates provide sev-
eral independent examples of this phenomenon
with the replacement of the broadly distributed
red algal (peridinin-containing) plastid in differ-
ent taxa with one of green, cryptophyte, or dia-
tom origin (14, 15). Therefore, the presence of a
red algal–derived plastid in most photosynthetic
chromalveolates is most easily explained by the
green algal endosymbiosis having predated the
red algal capture (7, 16, 17).
A different interpretation of our green gene
data is that these sequences did not derive from
EGT and HGT but rather support a bona fide
sister-group relationship between chromalveo-
lates and green algae. Under this scenario, the
chromalveolate ancestor contained a plastid of
primary endosymbiotic origin [cyanobacterial
(18)] that was shared with the green lineage
and subsequently replaced by one of second-
ary (red algal) derivation. Although possible,
this scenario is highly implausible because it
not only argues against Plantae monophyly,
which has been supported by recent phyloge-
nomic studies (6, 19), but more importantly,
demands that the vast majority of chromal-
veolate nuclear genes with nonplastid functions
A B
C
Charophytes
Land plants
Core chlorophytes
se
ty
hp
ot
pe
rt
S
se
ty
hp
or
ol
h
C
Mamiellales
Other prasinophytes se
ty
hp
o
ni
sa
r
P
Prasinophytes
Chlorophytes
Streptophytes
73/68
71/67
23/40
470/541
543/530
199/228
84/99
137/136
457/522
192/177
145/170
Prasinophytes
Chlorophytes
Streptophytes
Fig. 2. Phylogenetic distribution of diatom genes of green algal origin among Viridiplantae. (A)
Schematic tree that illustrates well-accepted phylogenetic relationships within the green lineage.
(B) Venn diagram depicting the distribution of diatom green genes of prasinophyte origin. These
genes support a specific sister-group relationship between prasinophytes and diatoms (and other
chromalveolates). The two broad categories of gene sharing are as follows: (i) the gene is exclusive
to prasinophytes (470/541), and (ii) the gene is shared with other Viridiplantae. (C) Venn diagram
depicting the distribution of all diatom green genes among Viridiplantae. Here, 192/177 genes are
of chlorophyte origin, whereas 145/170 genes are apparently derived from streptophytes. It should
be noted that these are provisional values and will be affected by the strength of the phylogenetic
signal in any given protein or the absence of data from particular groups; that is, some apparently
streptophyte-specific diatom green genes may simply be explained by the loss of the genes in other
Viridiplantae (such as prasinophytes).
Fig. 3. The distribution of
diatom green genes among
different chromalveolates.
The value for each major
chromalveolate lineage rep-
resents the number of proteins
that satisfy two phylogenetic
criteria: (i) monophyly of dia-
toms and the chromalveolate
lineage in question, and (ii)
monophyly of this clade
with Viridiplantae. The cat-
egory “other Stramenopiles”
includes the pelagophyte
Aureococcus anophagefferens
and the oomycetes Phytoph-
thora capsici, P. ramorum,
and P. sojae, which have com-
plete nuclear genome data
available. Data from the re-
maining taxa in this category
are organelle- or EST-derived.
0
500
1000
1500
2000
HaptophytesCiliatesApicomplexansOther
Stramenopiles
Diatoms
Thalassiosira
Phaeodactylum
N
um
be
r o
f g
en
es
Chromalveolate lineages
AlveolatesStramenopiles
www.sciencemag.org SCIENCE VOL 324 26 JUNE 2009 1725
REPORTS

o
n
M
ar
ch
7
, 2
01
0
w
w
w
.s
ci
en
ce
m
ag
.o
rg
D
ow
nl
oa
de
d
fro
m

(actin and tubulins) be directly related to
Viridiplantae. Although most single- and multi-
gene trees clearly demonstrate Viridiplantae
monophyly (20), they do not, however, sup-
port a specific affiliation between greens and
chromalveolates. There is no reason to expect
that this phylogenetic signal would have been
lost from chromalveolate genomes while being
retained by Viridiplantae. Therefore, given the
known proclivity of endosymbiosis to drive in-
tracellular gene transfer (21, 22) and the absence
of evidence for a specific phylogenetic relation-
ship between Viridiplantae and chromalveolates,
outside of the 16% reported here, we suggest that
the green “footprint” in chromalveolates (although
substantial) probably reflects a combination of
EGT and HGT rather than a host affiliation.
The rise to prominence in the oceans by
diatoms and other chromalveolates such as
dinoflagellates and haptophytes after the end-
Permian mass extinction (250 million years
ago) has been interpreted as the victory of red
plastid lineages over the predominant green
plastid taxa such as prasinophytes. Changing
nearshore ocean chemistry is thought to underlie
this globally important phenomenon (13, 23).
In contrast to current thinking, our findings show
that chromalveolates were already green be-
fore they acquired the red plastid. Although
ancient, these two endosymbioses that were
supplemented by subsequent HGTs supplied
chromalveolates such as diatoms [≈100,000 ex-
tant species (24)] with the genetic potential to
become some of the most ecologically suc-
cessful and dominant marine primary producers
on our planet.
References and Notes
1. C. B. Field, M. J. Behrenfeld, J. T. Randerson, P. Falkowski,
Science 281, 237 (1998).
2. G. C. Dismukes, D. Carrieri, N. Bennette, G. M. Ananyev,
M. C. Posewitz, Curr. Opin. Biotechnol. 19, 235 (2008).
3. N. Kroger, N. Poulsen, Annu. Rev. Genet. 42, 83 (2008).
4. T. Cavalier-Smith, J. Eukaryot. Microbiol. 46, 347 (1999).
5. A. Reyes-Prieto, D. Bhattacharya, Mol. Phylogenet. Evol.
45, 384 (2007).
6. N. Rodriguez-Ezpeleta et al., Curr. Biol. 15, 1325 (2005).
7. R. Frommolt et al., Mol. Biol. Evol. 25, 2653 (2008).
8. S. Coesel, M. Obornik, J. Varela, A. Falciatore, C. Bowler,
PLoS One 3, e2896 (2008).
9. M. Turmel, M. C. Gagnon, C. J. O'Kelly, C. Otis,
C. Lemieux, Mol. Biol. Evol. 26, 631 (2008).
10. E. Derelle et al., Proc. Natl. Acad. Sci. U.S.A. 103, 11647
(2006).
11. W. Martin, H. Brinkmann, C. Savonna, R. Cerff,
Proc. Natl. Acad. Sci. U.S.A. 90, 8692 (1993).
12. J. Steinkötter, D. Bhattacharya, I. Semmelroth, C. Bibeau,
M. Melkonian, J. Phycol. 30, 340 (1994).
13. P. G. Falkowski, A. H. Knoll, Evolution of Primary Producers
in the Sea (Elsevier Academic, Amsterdam, 2007).
14. K. Ishida, B. R. Green, Proc. Natl. Acad. Sci. U.S.A. 99,
9294 (2002).
15. T. Tengs et al., Mol. Biol. Evol. 17, 718 (2000).
16. J. Petersen, R. Teich, H. Brinkmann, R. Cerff, J. Mol. Evol.
62, 143 (2006).
17. A. Reyes-Prieto, A. Moustafa, D. Bhattacharya, Curr. Biol.
18, 956 (2008).
18. M. M. Hauber, S. B. Muller, V. Speth, U. G. Maier,
Bot. Acta 107, 383 (1994).
19. F. Burki, K. Shalchian-Tabrizi, J. Pawlowski, Biol. Lett. 4,
366 (2008).
20. H. S. Yoon et al., BMC Evol. Biol. 8, 14 (2008).
21. W. Martin et al., Proc. Natl. Acad. Sci. U.S.A. 99, 12246
(2002).
22. A. Reyes-Prieto, J. D. Hackett, M. B. Soares,
M. F. Bonaldo, D. Bhattacharya, Curr. Biol. 16, 2320
(2006).
23. H. R. Thierstein, J. R. Young, Coccolithophores: From
Molecular Processes to Global Impact (Springer, Berlin,
2004).
24. F. E. Round, R. M. Crawford, D. G. Mann, The Diatoms:
Biology and Morphology of the Genera (Cambridge Univ.
Press, Cambridge, 1990).
25. D.B. was supported by grants from NSF and NIH
(EF 04-31117 and R01ES013679, respectively). A.M. was
supported by an Institutional National Research Service
Award (T 32 GM98629) from NIH. U.G.M. thanks the
Deutsche Forschungsgemeinschaft (grant SFB-TR1) for
support. We thank T. Mock for providing tiling
array–generated diatom transcripts and J. E. DeReus at
the High Performance Computing Facility at the
University of Iowa for technical support.
Supporting Online Material
www.sciencemag.org/cgi/content/full/324/5935/1724/DC1
Materials and Methods
SOM Text
Figs. S1 to S7
Tables S1 and S2
References
3 March 2009; accepted 22 May 2009
10.1126/science.1172983
Solution Nuclear Magnetic Resonance
Structure of Membrane-Integral
Diacylglycerol Kinase
Wade D. Van Horn,1* Hak-Jun Kim,1,2* Charles D. Ellis,1 Arina Hadziselimovic,1 Endah S. Sulistijo,1
Murthy D. Karra,1 Changlin Tian,1,3 Frank D. Sönnichsen,4 Charles R. Sanders1†
Escherichia coli diacylglycerol kinase (DAGK) represents a family of integral membrane enzymes
that is unrelated to all other phosphotransferases. We have determined the three-dimensional
structure of the DAGK homotrimer with the use of solution nuclear magnetic resonance. The third
transmembrane helix from each subunit is domain-swapped with the first and second
transmembrane segments from an adjacent subunit. Each of DAGK’s three active sites resembles a
portico. The cornice of the portico appears to be the determinant of DAGK’s lipid substrate
specificity and overhangs the site of phosphoryl transfer near the water-membrane interface.
Mutations to cysteine that caused severe misfolding were located in or near the active site,
indicating a high degree of overlap between sites responsible for folding and for catalysis.
Escherichia coli diacylglycerol kinase(DAGK) is encoded by the dgkA geneand catalyzes the direct phosphorylation
of diacylglycerol (DAG) by Mg(II)–adenosine
triphosphate (MgATP) to form phosphatidic acid
as part of the membrane-derived oligosaccharide
(MDO) cycle (1–3). In Gram-positive organisms,
the dgkA homolog encodes an undecaprenol ki-
nase, indicating a role in oligosaccharide assem-
bly or in related signaling pathways (4). The
DAGK homolog in Streptococcus mutans is
known to be a virulence factor for smooth-
surface dental caries (5). DAGK was among the
first integral membrane enzymes to be solubi-
lized, purified, and mechanistically characterized
(6). The wild-type protein is very stable (7, 8)
and can spontaneously insert into lipid bilayers to
adopt its functional fold (9, 10). Paradoxically,
DAGK resembles many disease-linked human
membrane proteins because it is highly suscep-
tible to mutation-induced misfolding (10–12).
DAGK functions as a 40-kD homotrimer, with
a total of nine transmembrane (TM) helices and
three active sites (13).
The structure of DAGKwas determined using
solution nuclearmagnetic resonance (NMR)meth-
ods (14) under conditions in which the enzyme is a
functional homotrimer solubilized in ~100-kD
dodecylphosphocholine micelles; this work ex-
tends other solution NMR studies of >20-kDmul-
tispan membrane proteins (15–22). The backbone
structure of the helical TM domain of DAGK
(residues 26 to 121) was precisely determined by
the data (Fig. 1, fig. S1, and table S4); however,
motions associated with the N terminus (residues
1 to 25) have hindered determination of its con-
formation beyond confirming the presence of two
stable amphipathic helices.
DAGK’s structure bears no resemblance to
that of the water-soluble DAGK (23) (Fig. 1).
The three-fold symmetry axis lies at the center of
a parallel left-handed bundle formed by the sec-
ond transmembrane (TM2) helices of the three
subunits. TM2 has previously been proposed to
play a central role in DAGK’s folding and sta-
bility (24, 25) and contains several highly con-
served residues (Fig. 1), particularly near the
1Department of Biochemistry and Center for Structural Biology,
Vanderbilt University, Nashville, TN 37232, USA. 2Korea Polar
Research Institute, Incheon 406-840, Korea. 3School of Life
Science, University of Science and Technology of China, Hefei,
Anhui 230026, P. R. China. 4Otto Diels Institute for Organic
Chemistry, Christian Albrechts University of Kiel, D-24098 Kiel,
Germany.
*These authors contributed equally to this work.
†To whom correspondence should be addressed. E-mail:
chuck.sanders@vanderbilt.edu
26 JUNE 2009 VOL 324 SCIENCE www.sciencemag.org1726
REPORTS

o
n
M
ar
ch
7
, 2
01
0
w
w
w
.s
ci
en
ce
m
ag
.o
rg
D
ow
nl
oa
de
d
fro
m