to the mechanisms of membrane integration
for the chloride channel CLIC1 (25) or diph-
theria toxin (26), both of which exist alternately
in soluble and membrane-integrated forms. Mo-
lecular interplay between lipid composition and
membrane insertion of IM protein structures is
another intriguing possibility (27). Although
complete understanding of the integration
dynamics of Mistic requires further study, all
available data suggest that it must autono-
mously associate with the bacterial membrane
and that this property alone accounts for its
high efficiency in chaperoning the production
and integration of downstream cargo proteins
(fig. S6). Taken together with the NMR tech-
niques and protocols developed and used for
Mistic structure determination, Mistic_s unique
ability to assist in the production of IM proteins
opens new avenues around traditional obstacles
in the study of IM proteins, particularly those
of eukaryotic origin.
References and Notes
1. C. G. Tate, FEBS Lett. 504, 94 (2001).
2. R. Laage, D. Langosch, Traffic 2, 99 (2001).
3. H. Kiefer, R. Vogel, K. Maier, Receptors Channels 7,
4. J. Tucker, R. Grisshammer, Biochem. J. 317, 891
5. W. C. Wimley, S. H. White, Biochemistry 39, 4432
6. K. Wuthrich, Nature Struct. Biol. 5, 492 (1998).
7. G. M. Clore, A. M. Gronenborn, Nature Struct. Biol. 4,
8. V. K. Rastogi, M. E. Girvin, Nature 402, 263 (1999).
9. K. R. MacKenzie, J. H. Prestegard, D. M. Engelman,
Science 276, 131 (1997).
10. C. Fernandez, C. Hilty, G. Wider, P. Guntert, K.
Wuthrich, J. Mol. Biol. 336, 1211 (2004).
11. P. M. Hwang et al., Proc. Natl. Acad. Sci. U.S.A. 99,
12. A. Arora, F. Abildgaard, J. H. Bushweller, L. K. Tamm,
Nature Struct. Biol. 8, 334 (2001).
13. See supporting data on Science Online.
14. V. Ramamurthy, D. Oliver, J. Biol. Chem.272, 23239 (1997).
15. K. Pervushin, R. Riek, G. Wider, K. Wuthrich, Proc.
Natl. Acad. Sci. U.S.A. 94, 12366 (1997).
16. M. Salzmann, G. Wider, K. Pervushin, K. Wuthrich,
J. Biomol. NMR 15, 181 (1999).
17. C. Ritter, T. Luhrs, W. Kwiatkowski, R. Riek, J. Biomol.
NMR 28, 289 (2004).
18. P. A. Kosen, Methods Enzymol. 177, 86 (1989).
19. J. L. Battiste, G. Wagner, Biochemistry 39, 5355
20. P. Guntert, Methods Mol. Biol. 278, 353 (2004).
21. C. Hilty, G. Wider, C. Fernandez, K. Wuthrich,
Chembiochem 5, 467 (2004).
22. T. P. Roosild et al., data not shown.
23. R. B. Bass, P. Strop, M. Barclay, D. C. Rees, Science
298, 1582 (2002).
24. Y. Jiang et al., Nature 423, 33 (2003).
25. D. R. Littler et al., J. Biol. Chem. 279, 9298 (2004).
26. S. Choe et al., Nature 357, 216 (1992).
27. W. Zhang, M. Bogdanov, J. Pi, J. Pittard, W. Dowhan,
J. Biol. Chem. 278, 50128 (2003).
28. We thank E. Wiater for help in TGF-b receptor binding
studies and C.Park for Edman degradation sequencing
of proteins. Supported by NIH grant GM056653. R.R.
is a Pew scholar. The bundle of 10 conformers
representing the NMR structure is deposited in the
PDB database with accession code 1YGM. The coding
sequence of Mistic has been deposited in GenBank
with accession code AY874162.
Supporting Online Material
Materials and Methods
Figs. S1 to S6
Tables S1 and S2
14 October 2004; accepted 6 January 2005
The Genome of the
Basidiomycetous Yeast and Human
Pathogen Cryptococcus neoformans
Brendan J. Loftus,1* Eula Fung,2Paola Roncaglia,3Don Rowley,2
Paolo Amedeo,1Dan Bruno,2Jessica Vamathevan,1
Molly Miranda,2Iain J. Anderson,1James A. Fraser,4
Jonathan E. Allen,1Ian E. Bosdet,5Michael R. Brent,6
Readman Chiu,5Tamara L. Doering,7Maureen J. Donlin,8
Cletus A. D’Souza,9Deborah S. Fox,4,10Viktoriya Grinberg,1
Jianmin Fu,11Marilyn Fukushima,2Brian J. Haas,1James C. Huang,4
Guilhem Janbon,12Steven J. M. Jones,5Hean L. Koo,1
Martin I. Krzywinski,5June K. Kwon-Chung,13Klaus B. Lengeler,4,14
Rama Maiti,1Marco A. Marra,5Robert E. Marra,4,15
Carrie A. Mathewson,5Thomas G. Mitchell,4Mihaela Pertea,1
Florenta R. Riggs,1Steven L. Salzberg,1Jacqueline E. Schein,5
Alla Shvartsbeyn,1Heesun Shin,5Martin Shumway,1
Charles A. Specht,16Bernard B. Suh,17Aaron Tenney,6
Terry R. Utterback,18Brian L. Wickes,11Jennifer R. Wortman,1
Natasja H. Wye,5James W. Kronstad,9Jennifer K. Lodge,8
Joseph Heitman,4Ronald W. Davis,2
Claire M. Fraser,1Richard W. Hyman2
Cryptococcus neoformans is a basidiomycetous yeast ubiquitous in the
environment, a model for fungal pathogenesis, and an opportunistic human
pathogen of global importance. We have sequenced its È20-megabase
genome, which contains È6500 intron-rich gene structures and encodes a
transcriptome abundant in alternatively spliced and antisense messages. The
genome is rich in transposons, many of which cluster at candidate centro-
meric regions. The presence of these transposons may drive karyotype
instability and phenotypic variation. C. neoformans encodes unique genes
that may contribute to its unusual virulence properties, and comparison of
two phenotypically distinct strains reveals variation in gene content in
addition to sequence polymorphisms between the genomes.
With an increased immunocompromised
population as a result of AIDS and wide-
spread immunosuppressive therapy, Crypto-
coccus neoformans has emerged as a major
pathogenic microbe in patients with impaired
immunity (1). C. neoformans elaborates two
Fig. 6. Mutational dis-
ture and function. (A)
Residues forming the
core of Mistic, with
those mutated in struc-
tural disruption studies
highlighted with arrows.
(B) Mistic mutated sin-
gly at three core resi-
dues displays varying
structural stability and
of Trp13to Ala (W13A)
reduces the overall yield
of fused aKv1.1 by a
factor of 2 to 3. More important, mutation of Met75to Ala (M75A) destabilizes the structure of
Mistic sufficiently such that, when expressed by itself, it partitions substantially into the cytoplasm
(fourth lane from left), in stark contrast to wild-type Mistic or any of the other mutants analyzed.
This results in a functionally disabled protein; thus, when M75A is fused to aKv1.1, there is no
detectable yield of this protein (rightmost lane).
R E P O R T S
www.sciencemag.org SCIENCEVOL 30725 FEBRUARY 2005
specialized virulence factors, a polysaccharide
capsule (2) and the antioxidant pigment
melanin (3), which enhance human infection
and central nervous system colonization. Here,
we report the genome sequence of two related
strains of C. neoformans serotype D (JEC21
and B-3501A) as an important step in the
elucidation of the genomic basis for virulence
in this pathogenic yeast.
The 19-Mb genome sequence of C. neo-
formans JEC21 Eexcluding the ribosomal
RNA (rDNA) repeats region constituting
È5% of the genome^ spans 14 chromosomes
from 762 kb to 2.3 Mb (table S1), whereas
the 18.5-Mb sequence of the B-3501A strain
consists of 14 linked assemblies (scaffolds).
Unlike S. cerevisiae, the genome of C.
neoformans shows no evidence for a whole-
genome duplication (4). However, a chro-
mosomal translocation and an exact È60-kb
segmental duplication are present in JEC21
compared with B-3501A (5). Almost 5% of
the genome consists of transposons, the
majority clustered on each chromosome in
single blocks that span 40 to 100 kb that may
represent sequence-independent regional
centromeres, similar to those in S. pombe
and N. crassa (6) (Fig. 1). Each block is
unique but all contain at least one copy of
the Tcn5 or Tcn6 transposons, which may
represent functional elements or target the
centromeres. Transposons are also clustered
adjacent to the rDNA repeats and within the
mating-type (MAT) locus (Fig. 1). In con-
trast to the other transposons, the long inter-
spersed nuclear element–like (LINE-like)
retroelement Cnl1 shows a marked prefer-
ence for telomeric regions.
To ensure accurate gene structure annota-
tion, sequence data were obtained from both
ends of more than 23,000 cDNA clones of a
full-length normalized cDNA library from C.
neoformans JEC21 cells grown under various
conditions (7). A total of 6572 protein-
encoding genes were identified, which contain
an average of 6.3 exons of 255 base pairs (bp)
and 5.3 introns of 67 bp (table S2). The mean
transcript size of 1.9 kb contains an average
of 15% noncoding sequence from both the 5¶
and 3¶ ends. The gene organization in C.
neoformans is thus considerably more com-
plex than that of ascomycetes for which ge-
nome sequence (table S2) is available and is
comparable to that observed in Arabadopsis
thaliana or Caenorhabditis elegans.
A conspicuous feature to emerge from
comparing cDNA and genome sequence data
is evidence for alternative splicing and en-
dogenous antisense transcripts, in some cases
emanating from the same gene locus (Fig. 2).
Alternative splicing and natural antisense
RNA transcribed in cis were identified in
genes encoding diverse functions distributed
genome-wide, which suggests that both are
widespread genetic regulatory mechanisms
in C. neoformans (tables S3 to S5). Alterna-
tive splice forms were predicted for 277
genes, or 4.2% of the transcriptome (table
S4), and a variety of mechanisms could be
identified (e.g., exon skipping, truncation,
and extension at both 5¶ and 3¶ ends).
Antisense transcripts were identified for 53
genes; however, they appear to have no ap-
preciable coding potential and are usually
completely overlapped by their sense coun-
terparts (table S5). The presence and fre-
quency of these antisense transcripts and the
presence of the molecular components nec-
essary for RNA interference extend previous
studies (8) and indicate that regulation by
double-stranded RNA is likely a general reg-
ulatory mechanism in this organism.
JEC21 and B-3501A are highly related
inbred strains of the alpha mating type, the
most prevalent mating type in environmental
1The Institute for Genomic Research, 9712 Medical
Center Drive, Rockville, MD 20850, USA.
Genome Technology Center, Stanford University, 855
California Avenue, Palo Alto, CA 94304, USA.
3Neurobiology Sector, International School for Ad-
vanced Studies (SISSA-ISAS), Via Beirut 2-4, 34014
Trieste, Italy.4Department of Molecular Genetics and
Microbiology, Duke University Medical Center, 322
CARL Building, Research Drive, Box 3546, DUMC,
Durham, NC 27710, USA.5Genome Sciences Centre,
100-570 West 7th Avenue, Vancouver, BC V5Z 4S6,
Washington University, One Brookings Drive, St.
Louis, MO 63130, USA.
Microbiology, Washington University School of Med-
icine, 660 South Euclid Avenue, St. Louis, MO 63110,
Biology, Saint Louis University School of Medicine,
1402 S. Grand Boulevard, St. Louis, MO 63104, USA.
9The Michael Smith Laboratories, The University of
British Columbia, 2185 East Mall, Vancouver, BC V6T
the Department of Pediatrics, Louisiana State Health
Science Center, Children’s Hospital, 200 Henry Clay
Avenue, New Orleans, LA 70118, USA.11University of
Texas Health Science Center, 7703 Floyd Curl Drive,
San Antonio, TX 78229, USA.12Unite ´ de Mycologie
Mole ´culaire, Institut Pasteur, 25 rue du Docteur Roux,
Cedex 15, Paris, France.
Section, Laboratory of Clinical Investigation, National
Institutes of Health (NIAID/NIH), 9000 Rockville Pike,
Bethesda, MD 20892, USA.14Institut fu ¨r Mikrobiologie,
Heinrich-Heine-Universita ¨t, Universita ¨tsstraße 1/
26.12, Du ¨sseldorf, Germany.
Ecology, The Connecticut Agricultural Experiment
Station, 123 Huntington Street, New Haven, CT
University, 650 Albany Street, EBRC-625, Boston, MA
02118, USA.17Department of Biomolecular Engineer-
ing, University of California, Santa Cruz, 1156 High
Street, Santa Cruz, CA 95064 USA.18Joint Technology
Center, J. Craig Venter Foundation, 5 Research Place,
Rockville, MD 20850, USA.
6Laboratory for Computational Genomics,
7Department of Molecular
8Department of Biochemistry and Molecular
10Research Institute for Children and
15Plant Pathology and
16Department of Medicine, Boston
*To whom correspondence should be addressed.
Fig. 1. The C. neoformans JEC21 genome with each chromosome
represented as a colored bar. Specific features are pseudocolored, from
red (high density) to deep blue (low density) and plotted on a log
scale. These include the density of genes, transposons, expressed
sequence tags (ESTs), and predicted SNPs and indels. Candidate
centromeric regions and the MAT locus are represented as red bars and
a blue bar, respectively. The location of the rDNA repeat is represented by
a green bar.
R E P O R T S
25 FEBRUARY 2005 VOL 307SCIENCEwww.sciencemag.org
and clinical isolates (9). As a result of back-
crossing during strain construction, the se-
quence differences that distinguish these
strains are restricted to 50% of their ge-
nomes, which overall are 99.5% identical at
the sequence level. The predicted single-
nucleotide polymorphisms (SNPs) and inser-
tion and deletion polymorphisms (indels) are
distributed in blocks of high and low se-
quence polymorphism, reflecting the recombi-
nation events that occurred during production
of these sibling strains (Fig. 1). The pheno-
types of JEC21 and B-3501 differ markedly,
with B-3501A being more thermotolerant
and more virulent in animal models than
JEC21. To investigate the genetic basis for
these differences, genomic regions en-
compassing JEC21 genes were compared
directly with the B-3501A assembly. The
vast majority (99.7%) of genes share 998%
nucleotide identity (fig. S1). Strain-specific
genes were experimentally verified by poly-
merase chain reaction and included a Ras
guanosine triphosphatase–activating protein
and two proteins of unknown function
specific to B-3501A, whereas four proteins
of unknown function were specific to
JEC21. These genes, in addition to 22
duplicated genes in JEC21 located on the
È60-kb segmental duplication, delineate the
A remarkable feature of C. neoformans is
the link between virulence and mating type,
which is governed by a specialized genomic
region, the MAT locus (10). Genome analysis
revealed several additional genes in MAT.
Numerous other genes involved in mating are
not in MAT or on the MAT chromosome and
are scattered throughout the genome. Con-
sistent with classification as a heterothallic
fungus that does not switch mating type,
there are no silent mating-type cassettes.
The major virulence factor of C. neoformans
is its extensive polysaccharide capsule, an
elaborate and dynamic structure that surrounds
the fungal cell wall that is unique among fungi
that affect humans (2). Genome analysis
identified more than 30 new genes likely
involved in capsule biosynthesis, including a
family containing seven members of the
capsule-associated (CAP64) gene. The CAP64
family appears to be restricted to basidiomy-
cetes, and two members encode alternatively
spliced forms (table S5). A second family of six
capsule-associated (CAP10) genes appears
restricted to a subset of fungi and is absent
from other yeasts.
The cell wall is an essential and unique
component of fungi, and most of the genes
involved in the biosynthesis of cell-wall
polysaccharides are conserved between the
ascomycetes and C. neoformans, making
them attractive targets for broad-spectrum
antifungal drugs. However, S. cerevisiae and
C. neoformans manifest notable differences
in their mechanisms of cell-wall protein as-
sociation. In S. cerevisiae, two major classes
of proteins are covalently bound to the cell
wall: the Pir proteins and a set of proteins
that are covalently attached to the cell wall
by a glycosylphosphatidylinositol (GPI) an-
chor. C. neoformans lacks both Pir-related
genes and several genes that have been
implicated in attachment of the GPI anchors
to the b-1,6-glucan in the cell wall (11).
Genome analysis also predicts more than 50
extracellular mannoproteins that may be
associated with the cell wall, most of which
are unique to C. neoformans.
The phylum Basidiomycota last shared a
common ancestor with the ascomycetes È900
million years ago, and the two phyla have
diverged considerably (12). Overall, 65% of
C. neoformans genes have conserved se-
quence homologs in a sampling of completed
fungal genomes (table S2), and of these 12%
are restricted to the basidiomycete genome
Phanerochaete chrysosporium. Another 10%
appear to be unique to C. neoformans, based
on the absence of identifiable homologs in the
current public databases, whereas the remain-
ing 25% match nonfungal sequences (7).
Lineage-specific gene family expansions do
not represent the most abundant protein
domains within the C. neoformans genome,
which are similar to those of ascomycetous
fungi (tables S6 and S7). Two of the 11 gene
families that appear unique to C. neoformans
are involved in capsule formation, and
another encodes nucleotide sugar epimerases
associated with cell-wall formation. About
60% of the C. neoformans genes could be
assigned gene ontology terms for molecular
function (7), and comparison with S. cerevisiae
reveals a similar distribution of genes across
nearly all functional categories (fig. S2). One
exception is an expansion of the drug-efflux
transporters of the major facilitator superfamily
in C. neoformans, which suggests enhanced
transport capability in this environmental
Recently, the Candida albicans genome
was reported (13), enabling a comparison
between these divergent pathogenic fungi. C.
neoformans is an environmental organism that
infects through inhalation, whereas C. albicans
is part of normal human microbiota and infects
by bloodstream invasion. Myriad cell-surface
proteins implicated in C. albicans adhesion to
epithelial cells are absent in C. neoformans,
which suggests that C. neoformans binds host
cells by distinct mechanisms. C. neoformans
elaborates both capsule and melanin; C.
albicans makes neither and lacks genes for
The C. neoformans genome sequence
provides new insights into this important
fungal human pathogen. The genome encodes
a core complement of genes common to other
fungi and, despite a large divergence time, the
functional distribution of many C. neoformans
genes mirrors that of S. cerevisiae. By
contrast with S. cerevisiae, however, the C.
neoformans genome displays an intron-rich
gene tapestry and a transcriptome rife with
alternative splicing and antisense transcripts.
These genome sequence data, together
with those from another basidiomycete, P.
chrysosporium (14), suggest that more com-
plex gene structures may be a general feature
of basidiomycetes (table S2). The genome
sequence data described herein from two
closely related strains of C. neoformans provide
a foundation to explore the molecular basis of
virulence in this pathogen and reveal differ-
ences in virulence strategies between C.
neoformans and other pathogenic fungi.
References and Notes
1. A. Casadevall, J. R. Perfect, Cryptococcus neoformans
(ASM Press, Washington, DC, 1998).
2. I. Bose, A. J. Reese, J. J. Ory, G. Janbon, T. L. Doering,
Eukaryot. Cell 2, 655 (2003).
3. A. Casadevall, A. L. Rosas, J. D. Nosanchuk, Curr.
Opin. Microbiol. 3, 354 (2000).
4. M. Kellis, B. W. Birren, E. S. Lander, Nature 428, 617
5. J. A. Fraser et al., in preparation.
6. E. B. Cambareri, R. Aisner, J. Carbon, Mol. Cell. Biol.
18, 5465 (1998).
7. Materials and methods are available as supporting
material on Science Online.
Fig. 2. Gene structures that display evidence for both alternative splicing and natural in cis
antisense transcripts based on JEC21 cDNA alignments to the genome sequence. Colored boxes
represent exonic regions. Each gene structure represents an alternative spliced form. The black line
represents the genomic sequence.
R E P O R T S
www.sciencemag.orgSCIENCEVOL 30725 FEBRUARY 2005
8. J. M. Gorlach, H. C. McDade, J. R. Perfect, G. M. Cox, Download full-text
Microbiol. 148, 213 (2002).
9. K. J. Kwon-Chung, J. E. Bennett, Am. J. Epidemiol.
108, 337 (1978).
10. K. B. Lengeler et al., Eukaryot. Cell 1, 704 (2002).
11. S. Shahinian, H. Bussey, Mol. Microbiol. 35, 477
12. S. B. Hedges, J. E. Blair, M. L. Venturi, J. L. Shoe, BMC
Evol. Biol. 4, 2 (2004).
13. T. Jones et al., Proc. Natl. Acad. Sci. U.S.A. 101, 7329
14. D. Martinez et al., Nature Biotechnol. 22, 695 (2004).
15. We thank J. Perfect, F. Dietrich, and J. Murphy for their
invaluable and ongoing support for the C. neoformans
genome project. Funding was provided by National
Institute of Allergy and Infectious Diseases (NIAID)
cooperative agreements AI48594 (C.M.F.) and AI47087
(R.W.D.). Accession numbers for the JEC21 genome
(AE017341-AE017353, AE017356), the B-3501A ge-
nome (AAEY00000000), and the JEC21 cDNA sequences
(CF675703.1-CF722528.1) have been submitted to
Supporting Online Material
Materials and Methods
Figs. S1 to S3
Tables S1 to S9
9 August 2004; accepted 5 January 2005
Published online 13 January 2005;
Include this information when citing this paper.
Control of Excitatory and
Inhibitory Synapse Formation
Ben Chih, Holly Engelman, Peter Scheiffele*
The normal function of neural networks depends on a delicate balance
between excitatory and inhibitory synaptic inputs. Synapse formation is
thought to be regulated by bidirectional signaling between pre- and
postsynaptic cells. We demonstrate that members of the Neuroligin family
promote postsynaptic differentiation in cultured rat hippocampal neurons.
Down-regulation of neuroligin isoform expression by RNA interference results
in a loss of excitatory and inhibitory synapses. Electrophysiological analysis
revealed a predominant reduction of inhibitory synaptic function. Thus,
neuroligins control the formation and functional balance of excitatory and
inhibitory synapses in hippocampal neurons.
Adhesion molecules bridge the pre- and
postsynaptic compartments of synapses in
the central nervous system. Neuroligin-1
(NL-1), a member of the Neuroligin family
of postsynaptic adhesion molecules, can
trigger formation of functional presynaptic
terminals in axons through interaction with
its axonal receptor b-neurexin E(1–3), re-
viewed in (4–6)^. To explore whether the
b-neurexin–neuroligin complex acts bidirec-
tionally and controls postsynaptic differenti-
ation, we overexpressed NL-1 in cultured
hippocampal neurons (7). Analysis of den-
dritic morphology, postsynaptic scaffolding
molecules, and postsynaptic glutamate recep-
tor distribution revealed that NL-1 promotes
assembly of the postsynaptic apparatus (Fig. 1).
NL-1–overexpressing neurons showed a 68 T
7% increase in the density of dendritic spine–
like protrusions. Spines in NL-1–expressing
cells frequently exhibited irregular, hand-
shaped heads with multiple presynaptic ter-
minals labeled for the vesicular glutamate
transporter 1 (vGlut1), a marker of excitatory
synapses (Fig. 1A; fig. S1). The density of
synaptic puncta containing the scaffolding
proteins PSD-95 and Homer was increased
significantly (Fig. 1B). Moreover, staining for
the NR1 subunit of N-methyl-D-aspartate
(NMDA) receptors revealed that NL-1 strong-
ly promotes NMDA receptor recruitment (Fig.
1D). We also observed recruitment of AMPA-
type glutamate receptors, as indicated by
clustering of GluR2/3 subunits in some NL-
1–expressing cells. However, high NL-1 levels
Department of Physiology and Cellular Biophysics,
Center for Neurobiology and Behavior, Columbia
University, New York, NY 10032, USA.
*To whom correspondence should be addressed.
Fig. 1. NL-1 promotes postsynaptic differentiation.
Hippocampal neurons were cotransfected with
expression vectors for hemagglutinin (HA)–tagged
NL-1 and EGFP or with EGFP vectors only. (A)
Immunostaining for vGlut1 and EGFP in control
cells expressing EGFP (left column) and cells
coexpressing EGFP and NL-1 (right column). NL-
1–induced spine structures contacting multiple
presynaptic terminals (right). (B) Immunostaining
for PSD-95 and EGFP (left) or HA epitope to detect
NL-1 (right). (C) Immunostaining for Homer and
EGFP (left) or HA epitope to detect NL-1 (right).
(D) Immunostaining for NMDA-receptor subunit 1
(NR1) and EGFP (left) or HA epitope to detect NL-
1 (right). Scale bar, 5 mm. (E) Quantification of
postsynaptic protein recruitment, dendritic spine
induction, and synapse formation in cells express-
ing NL-1 and EGFP-transfected control cells.
vGlut1/PSD-95 shows density of puncta with
colocalizing pre- and postsynaptic markers (SEM,
n 0 10, ***P G 0.001).
R E P O R T S
25 FEBRUARY 2005VOL 307SCIENCE www.sciencemag.org