Macrophage Apoptosis by
Anthrax Lethal Factor Through
p38 MAP Kinase Inhibition
Jin Mo Park, Florian R. Greten, Zhi-Wei Li, Michael Karin*
The bacterium Bacillus anthracis causes the death of macrophages, which may
allow it to avoid detection by the innate immune system. We found that B.
anthracis lethal factor (LF) selectively induces apoptosis of activated macro-
phages by cleaving the amino-terminal extension of mitogen-activated protein
kinase (MAPK) kinases (MKKs) that activate p38 MAPKs. Because macrophages
that are deficient in transcription factor nuclear factor B (NF-B) are also
sensitive to activation-induced death and p38 is required for expression of
certain NF-B target genes, p38 is probably essential for synergistic induction
of those NF-B target genes that prevent apoptosis of activated macrophages.
This dismantling of the p38 MAPK module represents a strategy used by B.
anthracis to paralyze host innate immunity.
Bacillus anthracis, the causative agent of an-
thrax, has gained notoriety as a potential bio-
warfare and bioterrorism agent. During inhala-
tion anthrax, the most lethal form of the disease,
B. anthracis spores are engulfed by alveolar
macrophages (1). The spores, however, survive
phagocytosis and germinate within phago-
somes, and the bacteria spread to regional
lymph nodes and eventually the bloodstream. In
this late stage of bacteremia, the infected indi-
vidual is subjected to fatal systemic shock (2).
For successful infection, B. anthracis must
evade the host innate immune system by killing
macrophages (3), a strategy used by other high-
ly virulent bacteria (4). It is unclear how B.
anthracis interacts with macrophages in such a
contradictory manner.
Three proteins secreted by B. anthracis are
central to its pathogenicity: protective antigen
(PA), edema factor (EF), and lethal factor (LF)
(5). By binding a specific cell-surface receptor,
PA translocates EF and LF into the cytosol (6).
EF is an adenylate cyclase that causes tissue
edema (7), whereas LF is a metalloprotease that
exhibits unique specificity toward MKKs,
cleaving between their NH
2
-terminal extension
and the catalytic domain (8). Because the NH
2
-
terminal extension is required for interactions
with both MAPKs and MKK kinases
(MKKKs) (9), this cleavage prevents MAPK
activation (10). Lethal toxin (LT), a complex of
PA and LF, is the major factor responsible for
the lethality of anthrax (1). Ex vivo, LT exhibits
cytotoxicity toward macrophages, an activity
likely to be important for evasion of host de-
fenses (1, 4). Together, all three toxin compo-
nents also inhibit neutrophil migration and
phagocytosis (11), further helping immune sys-
tem evasion. A direct causal relation between
dismantling of MAPK signaling and LT-medi-
ated cytotoxicity is currently lacking. Although
in culture, LT was mostly described as inducing
macrophage necrosis (12), necropsy of inhala-
tion anthrax victims revealed extensive macro-
phage apoptosis (13). Furthermore, mouse
strains whose macrophages are resistant to LT-
induced necrosis are more susceptible to B.
anthracis infection (14), thus questioning the
role of the necrotic response.
We sought physiologically relevant
conditions under which LT could induce mac-
rophage apoptosis. Treatment of J774A.1 mac-
rophage-like cells with the protein phosphatase
inhibitor calyculin A was found to sensitize
them to LT-induced apoptosis (12). Treatment
of different cells with such phosphatase inhib-
itors activates numerous protein kinases, in-
cluding MAPKs (15) and inhibitor of NF-B
(IB) kinase (IKK) (16). Hence, calyculin A
could activate J774A.1 macrophages. We there-
fore tested the effect of the general macrophage
activator lipopolysaccharide (LPS) on the re-
sponse of J774A.1 cells and bone marrow–
derived macrophages (BMDMs) to LT. Titra-
Laboratory of Gene Regulation and Signal Transduc-
tion, Department of Pharmacology, School of Medi-
cine, University of California, San Diego, 9500 Gilman
Drive, La Jolla, CA 92093– 0636, USA.
*To whom correspondence should be addressed. E-
mail: karinoffice@ucsd.edu
Fig. 1. Anthrax LT induces apoptosis of activated macrophages. (A) J774A.1 cells were incubated
without or with LPS (100 ng/ml) in the absence or presence of purified recombinant PA
63
(2.5
g/ml) and the indicated amounts of LF. After 8 hours, the cells were analyzed by terminal
deoxynucleotidyl transferase–mediated dUTP nick end labeling (TUNEL) staining for presence of
apoptotic bodies. (B) J774A.1 cells were treated as in (A). After 18 hours, genomic DNA was isolated
and analyzed by agarose gel electrophoresis and ethidium bromide staining. (C) J774A.1 cells were
incubated with LPS or other microbial components, including peptidoglycan (PGN; 10 g/ml),
synthetic bacterial lipopeptide (SBLP; Pam
3
CSK
4
;1g/ml), and lipoteichoic acids from S. aureus
(LTA-S; 10 g/ml) or B. subtilis (LTA-B; 10 g/ml) in the absence or presence of LT, and apoptosis
was analyzed by DNA fragmentation. (D to F) J774A.1 cells and BMDMs were treated with (Œ, ‚)
or without (■, 䊐) LPS and the indicated concentrations of LF, and the extent of total (necrotic and
apoptotic) cell death was determined after 8 hours by staining with Hoechst 33258 dye (Œ, ■) and
compared with the extent of apoptotic cell death determined by TUNEL staining (‚, 䊐).
R EPORTS
20 SEPTEMBER 2002 VOL 297 SCIENCE www.sciencemag.org2048
tion experiments revealed that in a manner
strictly dependent on PA
63
(17), a mature and
active form of PA (18), LF caused rapid apo-
ptosis of LPS-activated macrophages at 200
ng/ml, a suboptimal concentration for inducing
necrosis in J774A.1 and BMDMs from
C57BL/6 mice (Fig. 1, A, B, and F). No apo-
ptosis was detected in resting (nonactivated)
macrophages. In addition to LPS derived from
Gram-negative bacteria, lipoteichoic acids
(LTAs) from the Gram-positive bacteria Staph-
ylococcus aureus and B. subtilis also induced
apoptosis of LT-treated cells (Fig. 1C), suggest-
ing that a similar component of B. anthracis,a
Gram-positive bacterium, can activate macro-
phages and trigger apoptosis in the presence of
LT. Apoptosis induced by LTAs was not inhib-
ited by polymyxin B (17), indicating that it is
not mediated by contaminating LPS. At 200
ng/ml, most of the LF-induced cell death in
activated J774A.1 cells was apoptotic in nature,
whereas in activated BALB/c BMDMs, only
50% of the observed cell death was due to
apoptosis (Fig. 1, D and E). At higher concen-
trations, LT caused the necrotic death of both
resting and activated macrophages, and the ap-
optotic response was attenuated. As previously
reported (19), LT did not cause necrosis of
C57BL/6 BMDMs. Nonetheless, LT effective-
ly induced the apoptosis of these cells (Fig. 1F).
Therefore, unlike necrosis (19), LT-induced ap-
optosis of activated macrophages is not con-
fined to a subset of mouse strains.
Treatment of macrophages with LPS acti-
vates extracellular signal–regulated kinase
(ERK), c-Jun NH
2
-terminal kinase ( JNK),
and p38 MAPKs (20), as well as IKK and
NF-B(21). Titration experiments revealed
that LF (together with PA
63
) inhibited ERK
activation by LPS in BMDMs at a concentra-
tion as low as 40 ng/ml (Fig. 2A). Inhibition
of JNK1 and p38 activation required a higher
LF concentration (200 ng/ml), similar to that
needed for induction of apoptosis in activated
macrophages. However, LF did not inhibit
JNK2 nor LPS-induced IB␣ degradation,
suggesting that it does not inhibit IKK acti-
vation. Using inhibitors that are selective for
each MAPK cascade [PD98059, a MEK1/
MEK2 inhibitor for ERK (22); SP600125 for
JNK (23); SB202190 for p38 (24)], we ex-
amined the contribution of MAPK inhibition
to LF-induced apoptosis of activated macro-
phages. Only treatment with the specific p38
inhibitor SB202190 induced apoptosis of
LPS-treated BMDMs (Fig. 2B). SB202190
was not cytotoxic toward resting macro-
phages (17).
To determine whether cleavage of the
MKKs responsible for p38 activation, MKK3
and MKK6 (25), is required for LF-induced
apoptosis, we generated mutant forms of
MKK3 and MKK6 that lack specific resi-
dues required for recognition by LF (26, 27).
Both MKK3b
R26Q/I27G
(MKK3CR) and
MKK6
K14Q/I15G
(MKK6CR) were resistant to
LF cleavage (Fig. 2C). Stable expression of
either mutant in RAW264.7 macrophages
(which are more amenable to transfection than
J774A.1) revealed that only MKK6CR partially
protected p38 from inhibition by LF (Fig. 2D).
Most importantly, the pooled population of
MKK6CR-expressing cells exhibited consider-
able resistance to LF-induced apoptosis after
activation (Fig. 2, E and F). Neither MKK6CR
nor MKK3CR protected activated RAW264.7
cells from apoptosis induced by SB202190
(Fig. 2E). Thus, the ability of LT to induce
apoptosis of activated macrophages depends on
inhibition of p38 activation.
Even though LF does not inhibit the IKK to
Fig. 2. Inhibition of p38 MAPK induces apopto-
sis of activated macrophages. (A) BMDMs
(C57BL/6) were left unstimulated or stimulat-
ed with LPS in the presence of PA
63
(2.5 g/ml)
and the indicated amounts of LF. After 20 min,
cell lysates were prepared and analyzed by
immunoblotting with antibodies specific to
different MAPKs ( Tot) or their phosphorylated
(activated) forms (Phos). (B) BMDMs (C57BL/
6) were stimulated with LPS in the presence of
dimethylsulfoxide (DM), SB202190 (SB; 10 M),
PD98059 (PD; 20 M), or SP600125 (SP; 20 M)
and analyzed as in Fig. 1B. (C)
35
S-labeled wild-
type (WT) and mutant MKK proteins were syn-
thesized in vitro and then incubated with indi-
cated amounts of LF for 30 min. MKK cleavage
was analyzed by SDS–polyacrylamide gel elec-
trophoresis and autoradiography. FL, full length;
Clv, cleavage product. (D) RAW264.7 cells were
transfected with cytomegalovirus-driven expression vectors for green fluo-
rescent protein (Mock), MKK3bCR (3b CR), and MKK6bCR (6b CR). Stable
transfectants were left unstimulated or stimulated with LPS in the presence
or absence of PA
63
(2.5 g/ml) and LF (500 ng/ml) and analyzed for p38
activation as in (A). (E and F) The RAW264.7-derived transfected cell pools
were treated with LPS, LT, and SB as indicated. After 8 hours, the number of
apoptotic and necrotic cells was determined by staining with Hoechst 33258
(blue) and annexin V-Alexa568 (red), respectively.
R EPORTS
www.sciencemag.org SCIENCE VOL 297 20 SEPTEMBER 2002 2049
NF-B pathway, survival of LPS-activated
macrophages depends on IKK and NF-B acti-
vation. Using a conditional Ikk allele in which
exon 3, which encodes part of the kinase do-
main, was flanked by binding sites (loxp) for the
Cre recombinase, we generated IKK-deficient
myeloid cells by crossing Ikk
loxp/loxp
mice with
mice that express a lysozyme M promoter–
driven Cre recombinase (28). The frequency of
Ikk deletion in BMDMs of Ikk
loxp/loxp
LysM-
Cre mice was ⬃50% (Fig. 3A), resulting in only
a partial decrease in IKK (Fig. 3B) and NF-B
(Fig. 3C) activation in the mixed cell popula-
tion. To circumvent difficulties associated with
this heterogeneity, we examined LPS-induced
apoptosis in individual macrophages and corre-
lated it with the presence or absence of the p65
NF-B subunit in the nucleus. These experi-
ments revealed that LPS only induced apoptosis
of those cells lacking nuclear p65 (Fig. 3D). No
apoptosis was detected upon incubation of
Ikk
loxp/loxp
BMDMs lacking Cre with LPS. In
addition, expression of a degradation-resistant
form of IB in RAW264.7 cells sensitized them
to LPS-induced apoptosis (17 ).
The marked sensitivity of BMDMs that lack
either p38 or NF-B activity to activation-in-
duced death suggests that p38 may be required
to activate a transcription factor or recruit a
coactivator that synergizes with NF-Btoin-
duce transcription of a gene(s) whose product
inhibits apoptosis. Previous analysis of NF-
B–mediated gene expression in dendritic cells
revealed that p38 is required to induce some
NF-B target genes (29). To investigate this
possibility, we used real-time polymerase chain
reaction (PCR) (17) to examine the require-
ment of p38 for the expression of known NF-
B target genes in J774A.1 cells and BMDMs.
Expression of many NF-B–regulated genes,
such as IB␣, iNOS, A20, and GADD45, was
effectively induced by LPS in untreated cells as
well as in cells treated with either SB202190 or
LT (Fig. 4A). However, expression of other
NF-B target genes, including those encoding
interleukin-1␣ (IL-1␣), IL-1, and COX-2, was
induced by LPS in untreated macrophages but
was inhibited by either SB202190 or LT. Ex-
pression of only one gene, that encoding tumor
necrosis factor–␣ (TNF-␣), was partially inhib-
ited by LT but not by SB202190. Inhibition of
LPS-induced TNF-␣, IL-1␣, and IL-1 expres-
sion by LT was observed previously (30). Be-
cause p38 is also involved in mRNA stabiliza-
tion (25), we used nuclear run-off experiments
to confirm that the effect of its inhibitor is
transcriptional (Fig. 4B). On the basis of these
results, we suggest that through phospho-
rylation of an as-yet unidentified target, p38
synergizes with NF-B to induce the expres-
sion of a subset of target genes, which in
macrophages includes inhibitors(s) of activa-
tion-induced death. Inhibition of either p38 or
NF-B is sufficient to sensitize macrophages
to activation-induced death by preventing in-
duction of this antiapoptotic factor.
Our results uncover a strategy by which B.
anthracis paralyzes the innate immune system
to promote its undisturbed spread toward sys-
temic infection. By inhibiting activation of p38
MAPK, this deadly pathogen switches the sig-
Fig. 3. Defective NF-B activation sensitizes macrophages to activation-induced death. (A)
Southern blot analysis of Hind III–digested genomic DNA from BMDMs of Ikk
loxp/loxp
or LysM-
Cre-Ikk
loxp/loxp
mice. The 3.2- and 1.6-kilobase (kb) fragments correspond to the floxed and
deleted alleles, respectively. (B and C) BMDMs of the two genotypes were treated with LPS for the
indicated times, and IKK (B) and NF-B (C) activation was assayed. Gel loading and extract
preparation were controlled by Western blotting (WB) for IKK␣ (B) or a mobility shift assay for
NF-1 (C). Fold increase in IKK and NF-B activities was determined by densitometry. (D) BMDMs
were stimulated with LPS (100 ng/ml) for 8 hours and stained with annexin V to visualize apoptotic
cells. Nuclear p65 was identified by immunofluorescence, using a rabbit polyclonal antibody against
p65 (Santa Cruz Biotechnology, Santa Cruz, CA).
Fig. 4. Induction of certain
NF-B target genes during
macrophage activation is
p38 dependent. (A) C57BL/6
BMDMs (white bars) and
J774A.1 cells (black bars)
were stimulated or not with
LPS in the absence or pres-
ence of SB202190 (10 M)
or LT (500 ng/ml LF and 2.5
g/ml PA
63
). After 4 hours,
cellular RNA was isolated,
and relative expression of
the indicated genes was de-
termined by real-time PCR.
The results shown are aver-
ages of three determina-
tions. (B) The transcription
rates of the indicated genes
in 774A.1 cells that were or
were not treated with LPS
and SB202190, as indicated,
were determined by nuclear run-off assays.
R EPORTS
20 SEPTEMBER 2002 VOL 297 SCIENCE www.sciencemag.org2050
nal for macrophage activation to a trigger of
rapid cell death. Selective killing of activated
macrophages prevents the secretion of chemo-
kines and cytokines that alert the remainder of
the immune system to the presence of the
pathogen. This may explain why anthrax infec-
tions proceed undetected until the terminal
stage, when vast bacteremia occurs. Future re-
search should focus on the balance between
macrophage activation and apoptosis, as it
seems to play a key role in the pathogenesis of
anthrax and other deadly infections.
References and Notes
1. T. C. Dixon, M. Meselson, J. Guillemin, P. C. Hanna,
N. Engl. J. Med. 341, 815 (1999).
2. P. C. Hanna, J. A. Ireland, Trends Microbiol. 7, 180
(1999).
3. Y. Singh, S. H. Leppla, R. Bhatnagar, A. M. Friedlander,
J. Biol. Chem. 264, 11099 (1989).
4. Y. Weinrauch, A. Zychlinsky, Annu. Rev. Microbiol.
53, 155 (1999).
5. M. Mock, A. Fouet, Annu. Rev. Microbiol. 55, 647 (2001).
6. K. A. Bradley et al., Nature 414, 225 (2001).
7. S. H. Leppla, Proc. Natl. Acad. Sci. U.S.A. 79, 3162 (1982).
8. N. S. Duesbery et al., Science 280, 734 (1998).
9. Y. Xia et al., Genes Dev. 12, 3369 (1998).
10. N. S. Duesbery et al., Proc. Natl. Acad. Sci. U.S.A. 98,
4089 (2001).
11. J. Keppie, P. W. Harris-Smith, H. Smith, Br. J. Exp.
Pathol. 44, 446 (1963).
12. C. G. Lin et al., Curr. Microbiol. 33, 224 (1996).
13. L. M. Grinberg et al., Mod. Pathol. 14, 482 (2001).
14. S. L. Welkos, T. J. Keener, P. H. Gibbs, Infect. Immun.
51, 795 (1986).
15. K. D. Chen et al., J. Cell. Biochem. 76, 585 (2000).
16. J. A. DiDonato et al., Nature 388, 548 (1997).
17. Supplementary figures and details of experimental
procedures are available on Science Online.
18. K. R. Klimpel, S. S. Molloy, G. Thomas, S. H. Leppla,
Proc. Natl. Acad. Sci. U.S.A. 89, 10277 (1992).
19. J. E. Roberts, J. W. Watters, J. D. Ballard, W. F.
Dietrich, Mol. Microbiol. 29, 581 (1998).
20. C. Dong, R. J. Davis, R. A. Flavell, Annu. Rev. Immunol.
20, 55 (2002).
21. W. M. Chu et al., Cell 103, 909 (2000).
22. D. T. Dudley et al., Proc. Natl. Acad. Sci. U.S.A. 92,
7686 (1995).
23. B. L. Bennett et al., Proc. Natl. Acad. Sci. U.S.A. 98,
13681 (2001).
24. J. C. Lee et al., Nature 372, 739 (1994).
25. L. Chang, M. Karin, Nature 410, 37 (2001).
26. G. Vitale, L. Bernardi, G. Napolitani, M. Mock, C.
Montecucco, Biochem. J. 352, 739 (2000).
27. A. D. Pannifer et al., Nature 414, 229 (2001).
28. B. E. Clausen et al., Transgenic Res. 8, 265 (1999).
29. S. Saccani, S. Pantano, G. Natoli, Nature Immunol. 3,
69 (2002).
30. J. L. Erwin et al., Infect. Immun. 69, 1157 (2001).
31. We thank B. Liddington for critical review of the
manuscript and gift of LF and PA
63
and C. Adams
for manuscript preparation. J.M.P., F.R.G., and Z.-
W.L. were supported by postdoctoral fellowships
from the Irvington Institute for Immunological Re-
search, the Deutsche Forschungsgemeinschaft, and
the Cancer Research Institute, respectively. Work
was supported by NIH grants AI43477, ES04151,
and ES06376 and the Superfund basic research
program (ES10337). M.K. is an American Cancer
Society Research Professor.
Supporting Online Material
www.sciencemag.org/cgi/content/full/1073163/DC1
Materials and Methods
Figs. S1 to S4
Reference
22 April 2002; accepted 7 August 2002
Published online 29 August 2002;
10.1126/science.1073163
Include this information when citing this paper.
Enhanced Tumor Formation in
Mice Heterozygous for Blm
Mutation
Kathleen Heppner Goss,
1,2
* Mary A. Risinger,
1
Jennifer J. Kordich,
1,2
Maureen M. Sanz,
4
Joel E. Straughen,
1
Lisa E. Slovek,
1
Anthony J. Capobianco,
1
James German,
4
Gregory P. Boivin,
3
Joanna Groden
1,2
†
Persons with the autosomal recessive disorder Bloom syndrome are predisposed
to cancers of many types due to loss-of-function mutations in the BLM gene,
which encodes a recQ-like helicase. Here we show that mice heterozygous for
a targeted null mutation of Blm, the murine homolog of BLM, develop lym-
phoma earlier than wild-type littermates in response to challenge with murine
leukemia virus and develop twice the number of intestinal tumors when crossed
with mice carrying a mutation in the Apc tumor suppressor. These observations
indicate that Blm is a modifier of tumor formation in the mouse and that Blm
haploinsufficiency is associated with tumor predisposition, a finding with im-
portant implications for cancer risk in humans.
Bloom syndrome (BS) is characterized by
small stature, immunodeficiency, male infer-
tility, and predisposition to cancer of many
tissue types (1). Cells from persons with BS
show increased somatic recombination, chro-
mosome breakage, and site-specific muta-
tions (1–3). The BS locus, BLM, encodes
BLM, an adenosine triphosphate–dependent,
3⬘-5⬘ helicase with homology to the recQ
DEXH-box–containing DNA and RNA heli-
cases (4); loss of BLM helicase activity is
responsible for the genomic instability of BS
cells (5, 6 ). BLM resolves Holliday junc-
tions, suppresses recombination in vitro, and
is required for the fidelity of DNA double-
strand break repair (7–9).
We have used gene targeting by homol-
ogous recombination to disrupt the mouse
Blm gene to simulate BLM
Ash
, a BS-causing
mutant allele of BLM carried by approxi-
mately 1% of Ashkenazi Jews (4, 10, 11).
BLM
Ash
contains a frameshift mutation in
exon 10 of BLM that results in premature
translation termination (4). In contrast to
work with two mouse models of BS previ-
ously reported (12, 13), we used a gene-
targeting construct in which exons 10, 11,
and 12 of Blm were replaced with a hypo-
xanthine phosphoribosyltransferase (Hprt)
cassette (fig. S1A). Germ line transmission
of this mutant allele, Blm
Cin
, followed blas-
tocyst injection of targeted embryonic stem
cells to generate heterozygous mice (fig.
S1B). Crosses to generate Blm
Cin/Cin
mice
were unsuccessful, indicating that homozy-
gous disruption of Blm results in embryonic
lethality (14). Western blots of protein ly-
sates from Blm
⫹/⫹
testes, an abundant
source of Blm RNA and BLM protein (12,
13), displayed a specific band of approxi-
mately 190 kD when probed with a COOH-
terminal antiserum to BLM (fig. S1C).
Lysates from Blm
Cin/⫹
testes had an ap-
proximately 50% reduction in BLM in
comparison to Blm
⫹/⫹
testes. Lysates of
heterozygous tissues were similarly evalu-
ated with an NH
2
-terminal antibody to
BLM and revealed no smaller immunore-
active proteins (fig. S1D). This reduction of
full-length BLM and the absence of trun-
cated BLM in Blm
Cin/⫹
mice confirm that
we had generated a null allele. This allele
allowed us to examine the biological con-
sequences of Blm haploinsufficiency, that
is, a reduction in wild-type (WT) Blm gene
dosage and its gene product.
BS somatic cells exhibit increases in
chromosome aberrations, sister chromatid
exchanges (SCEs), homologous chromatid
interchanges, and micronuclei that are a con-
sequence of chromosome breakage (1, 15,
16). Although the cytogenetic analysis of
somatic cells from human BLM heterozy-
gotes remains to be completed (17), sperma-
tozoa from two of three obligate heterozy-
gotes have been shown to display excess
numbers of chromosome breaks and rear-
rangements (18 ). To learn whether Blm hap-
loinsufficiency affects genomic stability, we
cultured primary lung fibroblasts from
1
Department of Molecular Genetics, Biochemistry,
and Microbiology;
2
Howard Hughes Medical Institute;
3
Department of Pathology and Laboratory Medicine;
University of Cincinnati College of Medicine, 231
Albert Sabin Way, Cincinnati, OH 45267, USA.
4
De-
partment of Pediatrics, Weill Medical College of Cor-
nell University, New York, NY 10021, USA.
*Present address: Department of Surgery, Division of
Epithelial Pathobiology, University of Cincinnati Col-
lege of Medicine, Cincinnati, OH 45267, USA.
†To whom correspondence should be addressed. E-
mail: Joanna.Groden@uc.edu
R EPORTS
www.sciencemag.org SCIENCE VOL 297 20 SEPTEMBER 2002 2051
