Bruchpilot Promotes Active Zone
Assembly, Ca2þChannel Clustering,
and Vesicle Release
Robert J. Kittel,1* Carolin Wichmann,1,2* Tobias M. Rasse,1* Wernher Fouquet,1
Manuela Schmidt,1Andreas Schmid,1Dhananjay A. Wagh,3Christian Pawlu,2
Robert R. Kellner,4Katrin I. Willig,4Stefan W. Hell,4Erich Buchner,3
Manfred Heckmann,2† Stephan J. Sigrist1,5†
The molecular organization of presynaptic active zones during calcium influx–triggered
neurotransmitter release is the focus of intense investigation. The Drosophila coiled-coil domain
protein Bruchpilot (BRP) was observed in donut-shaped structures centered at active zones of
neuromuscular synapses by using subdiffraction resolution STED (stimulated emission depletion)
fluorescence microscopy. At brp mutant active zones, electron-dense projections (T-bars) were
entirely lost, Ca2þchannels were reduced in density, evoked vesicle release was depressed, and
short-term plasticity was altered. BRP-like proteins seem to establish proximity between Ca2þ
channels and vesicles to allow efficient transmitter release and patterned synaptic plasticity.
at the active zone, a process triggered by
Ca2þinflux through clusters of voltage-gated
channels (1, 2). The spacing between Ca2þ
ynaptic communication is mediated by
the fusion of neurotransmitter-filled
vesicles with the presynaptic membrane
channels and vesicles at active zones is par-
ticularly thought to influence the dynamic
properties of synaptic transmission (3).
The larval Drosophila neuromuscular junc-
tion (NMJ) is frequently used as a model of
glutamatergic synapses (4, 5). The monoclonal
antibody Nc82 specifically stains individual
active zones (fig. S1A) (6, 7) by recognizing a
coiled-coil domain protein of roughly 200 kD
named Bruchpilot (BRP) (6). BRP shows ho-
mologies to the mammalian active zone compo-
nents CAST Ecytoskeletal matrix associated with
the active zone (CAZ)–associated structural
protein^ (8), also called ERC (ELKS, Rab6-
interacting protein 2, and CAST) (9). Whereas
confocal microscopy recognized diffraction
limited spots, the subdiffraction resolution of
stimulated emission depletion (STED) fluores-
cence microscopy (10, 11) revealed donut-
shaped BRP structures at active zones (Fig.
1A). Viewed perpendicular to the plane of
synapses, both single and multiple Brings[ were
uncovered, of similar size to freeze-fracture-
derived estimates of fly active zones (12)
(average length of isolated rings was 0.191 T
0.002 mm, n 0 204; average length of single
rings of double ring structures was 0.148 T 0.002
mm, n 0 426; average length of double rings was
0.297 T 0.005, n 0 213) (fig. S1B). The donuts
were up to 0.16 mm high, as judged by images
taken parallel to the synaptic plane (Fig. 1A).
BRP seemed to demark individual active
zones associated with clusters of Ca2þchan-
Fig. 1. Junctional and
in mutants of the active
Unlike confocal, STED
recognized by Nc82.
Viewed from above, both
single (white arrows) and
clusters of multiple rings
(arrowheads) were identi-
fied. The red arrow indi-
cates a synapse viewed
parallel to the synaptic
plane. (B) Individual
synapses of control ani-
mals were labeled by
Nc82, whereas brp mu-
tant synapses completely
lacked the Nc82 signal,
which could be partially
restored by re-expressing
the brp cDNA in the brp
mutant background with
useof the neuron-specific
driver line ok6-GAL4. (C)
Staining with a neuronal
membrane marker (anti–
tion of brp NMJs. (D) Receptor fields were surrounded by the typical perisynaptic
expression of the neuronal cell adhesion molecule (NCAM) homolog FasciclinII
micrograph of a control type Ib boutonwith synapses (arrowheads) presynaptically
decorated with T-bars (arrows). (F) A brp mutant bouton showing an overall
normal organization but without T-bars. (G) Serial sections of a control synapse. A
T-bar can be observed in two consecutive sections (arrows). (H) Serial sections of a
representative brp mutant synapse completely lacking a T-bar and revealing
presynaptic membrane rufflings (asterisks). Scale bars in (G) and (H), 250 nm.
www.sciencemag.orgSCIENCEVOL 312 19 MAY 2006
us to isolate a mutant chromosome (brp69) in
which nearly the entire open reading frame of
BRP was deleted (fig. S1C). brp mutants
Ebrp69/df(2R)BSC29^ developed into mature
larvae but did not form pupae. The Nc82
label was completely lost from the active
zones of brp mutant NMJs but could be re-
stored by re-expressing the brp cDNA (6) in
the brp mutant background with use of the
neuron-specific driver lines ok6-GAL4 (Fig.
1B) or elav-GAL4. This also rescued larval
lethality. Mutants had slightly smaller NMJs
(average control size was 780.0 T 35.8mm2, n 0
14; average brp size was 593.3 T 29.1 mm2, n 0
12; P 0 0.0013) (Fig. 1C) and somewhat fewer
individual synapses (average synapse number
for control was 411.1 T 41.5, n 0 9; for brp,
296.3 T 28.9; n 0 8; P 0 0.036). However,
individual receptor fields, identified by the
glutamate receptor subunit GluRIID (13), were
enlarged in brp mutants (average field size in
control was 0.43 T 0.02 mm2, n 0 9; in brp,
0.64 T 0.03 mm2; n 0 8; P G 0.001) (Fig. 1D).
Thus, principal synapse formation occurred in
brp mutants, with individual postsynaptic
receptor fields increased in size but moderate-
ly decreased in number.
In electron micrographs of brp mutant
NMJs, synapses with pre- and postsynaptic
membranes in close apposition were present at
regular density (Fig. 1, F and H), and consistent
with the enlarged glutamate receptor fields (Fig.
1D) postsynaptic densities appeared larger
while otherwise normal (Fig. 1F). However, in-
termittent rufflings of the presynaptic mem-
brane were noted (Fig. 1H), and brp mutants
completely lacked presynaptic dense projections
(T-bars). Occasionally, very little residual
electron-dense material attached to the pre-
synaptic active zone membrane was identified
partially restored (fig. S2C), although these
structures were occasionally somewhat aberrant
in shape. Thus, BRP assists in the ultrastructural
assembly of the active zone and is essential for
In brp mutant larvae we noted a drastic
decrease in evoked excitatory junctional cur-
rent (eEJC) amplitudes (Fig. 2A) by using
two-electrode voltage clamp recordings of
postsynaptic currents at low stimulation frequen-
cies (elav-GAL4 background control, –89.3 T 3.4
nA; brp, –32.1 T 5.9 nA; n 0 10 each; P G 0.001;
ok6-GAL4 background control, –89.6 T 4.4 nA;
n 0 9; brp, –32.8 T 3.7 nA; n 0 10; P G 0.001).
This drop in current amplitude could be partially
rescued through brp re-expression within the
presynaptic motoneurons by using either elav-
GAL4 or ok6-GAL4 (elav-GAL4, –55.5 T 4.3 nA;
n 0 11; P 0 0.01; ok6-GAL4, –62.2 T 5.3 nA; n 0
10; P 0 0.002) (Fig. 2A). In contrast, the
amplitude of miniature excitatory junctional
currents (mEJCs) in response to single, spon-
taneous vesicle fusion events was increased
over control levels (control, –0.84 T 0.06 nA;
brp, –1.17 T 0.05 nA; n 0 10 each; P 0 0.004)
(Fig. 2B). This is consistent with the enlarged
individual glutamate receptor fields of brp
mutants (Fig. 1D) and excludes a lack of post-
synaptic sensitivity as the cause of the reduced
1European Neuroscience Institute Go ¨ttingen, Grisebachstrasse
5, 37077 Go ¨ttingen, Germany.2Institut fu ¨r Klinische Neuro-
biologie, Josef Schneider Strasse 11, 97080 Wu ¨rzburg,
Hubland, 97074 Wu ¨rzburg, Germany.4Department of Nano-
Biophotonics, Max Planck Institute for Biophysical Chemistry,
Am Fassberg 11, 37077 Go ¨ttingen, Germany.5Institut fu ¨r
Klinische Neurobiologie, Rudolf Virchow Zentrum, 97080
Wu ¨rzburg, Germany.
3Lehrstuhl fu ¨r Genetik und Neurobiologie, Am
*These authors contributed equally to this work.
†To whom correspondence should be addressed. E-mail:
Heckmann_M@klinik.uni-wuerzburg.de (M.H.); firstname.lastname@example.org
Fig. 2. Electrophysiological characterization of brp mutant NMJs. (A)
(Top) Average traces of eEJCs at 0.2 Hz nerve stimulation and (bottom)
mean eEJC amplitudes of control (dark gray), brp mutant (white), and
rescued animals (light gray) carrying either a copy of elav-GAL4 or ok6-
GAL4. (B) Sample traces of mEJCs and a cumulative histogram of the
amplitude distribution (0.05 nA bins). The average mEJC amplitude was
increased in brp mutants, whereas the frequency was not significantly
altered. Quantal content of brp NMJs was significantly reduced with
respect to controls. (C) Average scaled eEJCs (control, black; brp, gray)
illustrate the delayed release in brp mutants compared to controls. Al-
thoughtherisetimeofeEJCswassignificantlyincreasedatbrp NMJs, the rise
time of mEJCs was indistinguishable from the control. The decay time
constant (t) of eEJCs was not significantly altered at brp synapses (t control:
7.7 T 0.6 ms; t brp: 8.9 T 0.7 ms; n 0 10 each, P 0 0.104), whereas mEJCs
decayed with a slightly but significantly longer t in the mutant than in the
control (t control: 7.0 T 0.3 ms; t brp: 8.0 T 0.4 ms; n 0 10 each, P 0
0.045). One asterisk indicates P e 0.05, two asterisks, P e 0.01; and three
asterisks, P e 0.001. Error bars indicate SEM.
19 MAY 2006VOL 312SCIENCE www.sciencemag.org
It follows that the number of vesicles released
per presynaptic action potential (AP) (quantal
content) was severely compromised at brp mu-
tant NMJs (control, 109 T 5.7; brp, 28 T 5.2; n 0
10 each; P G 0.001) (Fig. 2B) and could not be
attributed solely to the moderate decrease in
synapse number. The ultrastructural defects of
brp mutant synapses may interfere with the
proper targeting of vesicles to the active zone
membrane and thereby impair exocytosis. The
number of vesicles directly docked to active
zone membranes was slightly decreased in brp
mutants (control average of 1.10 T 0.13 from 51
active zones, n 0 3; brp average of 0.87 T 0.09
from 89 active zones, n 0 4; P 0 0.53). How-
ever, the amplitude distribution and sustained
frequency of mEJCs (control, 1.55 T 0.33 Hz;
brp, 1.87 T 0.15 Hz; n 0 10 each; P 0 0.186)
(Fig. 2B) illustrated that brp mutant synapses
did not appear to suffer from extrasynaptic
release, as would be caused by a misalignment
of vesicle fusion sites with postsynaptic recep-
tors. Consistent with the appropriate deposition
of exo- and endocytotic proteins, an apparently
normal distribution of Syntaxin, Dap160, and
Dynamin (fig. S3) was observed at brp mutant
The exact amplitude and time course of AP-
triggered Ca2þinflux in the nerve terminal
governs the amplitude and time course of ves-
icle release (14). Nerve-evoked responses of
brp mutants were delayed (rise time of 2.53 T
0.37 ms, n 0 10) when compared with controls
(rise time of 1.11 T 0.05 ms, n 0 10, P G 0.001),
whereas in contrast mEJC rise times were un-
changed (control, 1.06 T 0.04 ms; brp, 1.06 T
0.03 ms; n 0 10 each) (Fig. 2C). Thus, evoked
vesicle fusion events were less synchronized
with the invasion of the presynaptic terminal by
an AP. Spatiotemporal changes in Ca2þinflux
have a profound effect on short-term plasticity
(15–17). Whereas at 10 Hz controls (n 0 18)
exhibited substantial short-term depression of
eEJC amplitudes, brp mutants (n 0 15) showed
strong initial facilitation before stabilizing at a
slightly lower but frequency-dependent steady-
state current (control at 10 Hz, –54.7 T 3.3 nA;
brp, –35.6 T 3.0 nA; P G 0.001) (Fig. 3A). As
judged by the initial facilitation at 10 Hz, neither
a reduction in the number of releasable vesicles
nor available release sites could fully account
for the low quantal content of brp mutants at
moderate stimulation frequencies. Further-
more, the altered short-term plasticity of brp
mutant synapses suggested a change in the
highly Ca2þ-dependent vesicle release probability
(18). Paired-pulse protocols were applied to
the NMJ (Fig. 3B). Closely spaced stimuli lead
to a buildup of residual Ca2þin the vicinity of
presynaptic Ca2þchannels, enhancing the prob-
ability of a vesicle within this local Ca2þ
domain to undergo fusion after the next pulse
(19). The absence of marked facilitation at
control synapses (ratio at 30-ms interval of 1.1 T
ready vesicles (20). At brp mutant NMJs,
however, the prominent facilitation at short
interpulse intervals (ratio at 30-ms interval of
2.0 T 0.13, P G 0.001) illustrated that the
enhancement of release probability strongly out-
weighed the depletion of releasable vesicles.
Thus, initial vesicle release probability was low,
andreleaseatbrp synapses particularly benefited
from the accumulation of intracellular Ca2þ.
between Ca2þchannels and vesicles at release
sites (3). It has been calculated that doubling this
distance from 25 to 50 nm decreases the release
probability threefold (21), and the larger this dis-
tance, the more effective the slow synthetic Ca2þ
buffer EGTA Eethyleneglycol-bis(b-aminoethyl)-
N,N,N¶,N¶-tetraacetic acid^ should become in
suppressing release (22). Indeed, after bath
application of membrane permeable EGTA-AM
(tetraacetoxymethyl ester of EGTA), the reduc-
tion of evoked vesicle release was more pro-
nounced at brp mutant than at control NMJs
(control, 64.2 T 13.8%; brp, 16.7 T 8.8%; n 0 6
each; P 0 0.026) (Fig. 3C).
The Ca2þ-channel subunit Cacophony gov-
erns release at Drosophila NMJs (23, 24). By
using a fully functional, GFP (green fluorescent
protein)–labeled variant (CacGFP) (25), we visu-
alized Ca2þchannels in vivo (26). Consistently,
Ca2þchannel expression was severely reduced
over the entire NMJ and at synapses lacking
BRP (Fig. 3D).
Thus, we conclude that brp mutants suffered
from a diminished vesicle release probability
Fig. 3. Impaired vesicle release
in brp mutants is caused by a
mislocalization of presynaptic
Ca2þchannels. (A) A 10-Hz
stimulation revealed transient
short-term facilitation of brp mu-
tant currents (white circles) and
the absence of a frequency-
dependent depression of steady-
state current amplitudes when
compared with controls (black
circles) (n Q 10 per genotype
at each frequency). (B) Average
currents after paired-pulse stimu-
lation at an interval of 30 ms
normalized to the amplitude of
the first pulse (control, black; brp,
gray) and paired-pulse ratios at
varying intervals demonstrate
pronounced potentiation at brp
NMJs (n 0 9 per genotype at
each interval). (C) Examples of
nerve-evoked local postsynaptic
currents recorded with a focal
electrode (36, 37) at indicated
time points (in seconds) after
bath application of EGTA-AM.
The bar chart illustrates the severe reduction of current amplitudes in brp mutants
5000 s after EGTA-AM wash-in. The values are normalized to the initial eEJC
amplitude. (D) Projections of confocal stacks displaying the NMJ (top images; scale
bar, 10 mm) and several boutons (lower images; scale bar, 2 mm) reveal weak
CacGFPsignal at brp mutant synapses. Quantification of CacGFPintensity averaged
over the entire NMJs [control, 31.1 T 2.4 arbitrary units (a.u.); n 0 13; brp, 18.0 T
2.0 a.u.; n 0 10; P 0 0.0017] or only synaptic areas (control, 52.6 T 1.2 a.u.; n 0
421 synapses; brp, 25.3 T 0.8 a.u.; n 0 320 synapses; P G 0.001, student t test)
included as bar charts. One asterisk indicates P e 0.05; two asterisks, P e 0.01; and
three asterisks, P e 0.001. Error bars indicate SEM.
www.sciencemag.orgSCIENCEVOL 31219 MAY 2006
due to a decrease in the density of presynaptic Download full-text
Ca2þchannel clusters. It is conceivable that
BRP tightly surrounds but is not part of the T-
bar structure, contained within the unlabeled
center of donuts. BRP may establish a matrix,
required for both T-bar assembly as well as the
appropriate localization of active zone compo-
nents including Ca2þchannels, possibly by
mediating their integration into a restricted
number of active zone slots (27). Related mech-
anisms might underlie functional impairments of
mammalian central synapses lacking active zone
components (28) and natural physiological dif-
ferences between synapse types (17). Electron
microscopy has identified regular arrangements
at active zones of mammalian CNS (central
nervous system) synapses (Bparticle web[) (29)
and frog NMJs (Bribs[) (30), where these struc-
tures have also been proposed to organize Ca2þ
channel clustering. At calyx of Held synapses,
both a fast and a slow component of exocytosis
have been described (31). The fast component
recovers slowly and is believed to owe its pro-
perties to vesicles attached to a matrix tightly
associated with Ca2þchannels (32), whereas the
slow component recovers faster (31) and is
thought to be important for sustaining vesicle
release during tetanic stimulation. In agreement
with this concept, the absence or impairment of
such a matrix at brp synapses has a profound
effect on vesicle release at low stimulation fre-
quencies, but this effect subsides as the fre-
quency increases (Fig. 3A). The sustained
frequency of mEJCs at brp synapses could be
explained if spontaneous fusion events arise
from the slow release component (33) or a
pathway independent of evoked vesicle fu-
Synapses lacking BRP and T-bars exhibited
a defective coupling of Ca2þinflux with vesicle
fusion, whereas the vesicle availability did not
appear rate-limiting under low frequency stim-
ulation. The activity-induced addition of pre-
synaptic dense bodies has been proposed to
elevate vesicle release probability (35). Our
work supports this hypothesis and suggests an
involvement of BRP or related factors in syn-
aptic plasticity by promoting Ca2þchannel
clustering at the active zone membrane.
References and Notes
1. R. G. Zhai, H. J. Bellen, Physiology (Bethesda) 19, 262
2. B. Katz, R. Miledi, Proc. R. Soc. London Ser. B 161, 496
3. E. Neher, Neuron 20, 389 (1998).
4. H. L. Atwood, S. Karunanithi, Nat. Rev. Neurosci. 3, 497
5. Y. H. Koh, L. S. Gramates, V. Budnik, Microsc. Res. Tech.
49, 14 (2000).
6. D. A. Wagh et al., Neuron 49, 833 (2006).
7. T. Wucherpfennig, M. Wilsch-Brauninger, M. Gonzalez-
Gaitan, J. Cell Biol. 161, 609 (2003).
8. T. Ohtsuka et al., J. Cell Biol. 158, 577 (2002).
9. Y. Wang, X. Liu,T. Biederer,T. C. Sudhof, Proc. Natl. Acad.
Sci. U.S.A. 99, 14464 (2002).
10. T. A. Klar, S. Jakobs, M. Dyba, A. Egner, S. W. Hell, Proc.
Natl. Acad. Sci. U.S.A. 97, 8206 (2000).
11. S. W. Hell, Nat. Biotechnol. 21, 1347 (2003).
12. C. J. Feeney, S. Karunanithi, J. Pearce, C. K. Govind, H. L.
Atwood, J. Comp. Neurol. 402, 197 (1998).
13. G. Qin et al., J. Neurosci. 25, 3209 (2005).
14. E. F. Barrett, C. F. Stevens, J. Physiol. 227, 691
15. R. S. Zucker, W. G. Regehr, Annu. Rev. Physiol. 64, 355
16. H. L. Atwood, Nature 215, 57 (1967).
17. A. Rozov, N. Burnashev, B. Sakmann, E. Neher, J. Physiol.
531, 807 (2001).
18. J. S. Dittman, A. C. Kreitzer, W. G. Regehr, J. Neurosci.
20, 1374 (2000).
19. B. Katz, R. Miledi, J. Physiol. 195, 481 (1968).
20. H. von Gersdorff, R. Schneggenburger, S. Weis, E. Neher,
J. Neurosci. 17, 8137 (1997).
21. M. R. Bennett, L. Farnell, W. G. Gibson, Biophys. J. 78,
22. E. M. Adler, G. J. Augustine, S. N. Duffy, M. P. Charlton,
J. Neurosci. 11, 1496 (1991).
23. F. Kawasaki, R. Felling, R. W. Ordway, J. Neurosci. 20,
24. H. Kuromi, A. Honda, Y. Kidokoro, Neuron 41, 101
25. F. Kawasaki, B. Zou, X. Xu, R. W. Ordway, J. Neurosci. 24,
26. T. M. Rasse et al., Nat. Neurosci. 8, 898 (2005).
27. Y. Q. Cao et al., Neuron 43, 387 (2004).
28. W. D. Altrock et al., Neuron 37, 787 (2003).
29. G. R. Phillips et al., Neuron 32, 63 (2001).
30. M. L. Harlow, D. Ress, A. Stoschek, R. M. Marshall, U. J.
McMahan, Nature 409, 479 (2001).
31. T. Sakaba, E. Neher, Neuron 32, 1119 (2001).
32. T. Sakaba, A. Stein, R. Jahn, E. Neher, Science 309, 491
33. J. Trommershauser, R. Schneggenburger, A. Zippelius,
E. Neher, Biophys. J. 84, 1563 (2003).
34. Y. Sara, T. Virmani, F. Deak, X. Liu, E. T. Kavalali, Neuron
45, 563 (2005).
35. J. M. Wojtowicz, L. Marin, H. L. Atwood, J. Neurosci. 14,
36. J. Dudel, Pflugers Arch. 391, 35 (1981).
37. C. Pawlu, A. DiAntonio, M. Heckmann, Neuron 42, 607
38. We thank E. Neher for comments on the manuscript and
C. Quentin for technical support. S.J.S. (SI 849/2-1,
SFB406/A16, and Research Center for Molecular Physi-
ology of the Brain Go ¨ttingen), M.H. (HE 2621/4-1), and
E.B. and D.A.W. (SFB581/B6 and GRK200/3) were
supported by the Deutsche Forschungsgemeinschaft;
S.W.H., by the German Ministry of Research (BMBF); and
T.M.R., by a Max Planck Fellowship. The European
Neuroscience Institute Go ¨ttingen (ENI-G) is jointly funded
by the Go ¨ttingen University Medical School, the Max
Planck Society, and Schering AG.
Supporting Online Material
Materials and Methods
Figs. S1 to S3
16 February 2006; accepted 4 April 2006
Published online 13 April 2006;
Include this information when citing this paper.
A Systems Approach to Mapping DNA
Damage Response Pathways
Christopher T. Workman,1* H. Craig Mak,1* Scott McCuine,1Jean-Bosco Tagne,2Maya Agarwal,1
Owen Ozier,2Thomas J. Begley,3Leona D. Samson,4Trey Ideker1†
Failure of cells to respond to DNA damage is a primary event associated with mutagenesis and
environmental toxicity. To map the transcriptional network controlling the damage response, we
measured genomewide binding locations for 30 damage-related transcription factors (TFs) after
exposure of yeast to methyl-methanesulfonate (MMS). The resulting 5272 TF-target interactions
revealed extensive changes in the pattern of promoter binding and identified damage-specific
binding motifs. As systematic functional validation, we identified interactions for which the target
changed expression in wild-type cells in response to MMS but was nonresponsive in cells lacking
the TF. Validated interactions were assembled into causal pathway models that provide global
hypotheses of how signaling, transcription, and phenotype are integrated after damage.
cer, aging, immune deficiencies, and other de-
xposure of cells to chemical and physical
damaging agents can result in DNA
generative diseases (1). DNA damage is sensed
by a highly conserved mechanism involving the
ATM/ATR protein kinases in humans (ataxia-
telangiectasia mutated/ataxia-telangiectasia and
Rad3-related; homologous to Tel1 and Mec1 in
yeast). These aggregate at DNA lesions (2) and
activate signaling cascades that include the
yeast). Chk kinases, in turn, trigger both tran-
scriptional and transcription-independent re-
sponses, including activation of DNA repair
machinery and cell-cycle arrest (1).
Beyond the known DNA repair genes, ge-
nomewide expression profiling in yeast has
identified several hundred genes (3–5) whose
expression is increased or decreased in response
to alkylation damage by methyl-methanesulfonate
(MMS). At the level of growth phenotype, sys-
1University of California San Diego, La Jolla, CA 92093,
bridge, MA 02139, USA.
University at New York, Rensselaer, NY 12144, USA.
4Massachusetts Institute of Technology, Cambridge, MA
2Whitehead Institute for Biomedical Research, Cam-
3University of Albany–State
*These authors contributed equally to this work.
†To whom correspondence should be addressed. E-mail:
19 MAY 2006VOL 312 SCIENCE www.sciencemag.org