, 1194 (2004); 306
Zachary E. Perlman,
Multidimensional Drug Profiling By Automated
www.sciencemag.org (this information is current as of March 26, 2009 ):
The following resources related to this article are available online at
version of this article at:
including high-resolution figures, can be found in the onlineUpdated information and services,
can be found at: Supporting Online Material
can berelated to this articleA list of selected additional articles on the Science Web sites
, 10 of which can be accessed for free: cites 28 articlesThis article
126 article(s) on the ISI Web of Science. cited byThis article has been
23 articles hosted by HighWire Press; see: cited byThis article has been
: subject collections This article appears in the following
in whole or in part can be found at: this article
permission to reproduce of this article or about obtaining reprintsInformation about obtaining
registered trademark of AAAS.
2004 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 the
on March 26, 2009
(NMR) spectroscopy of skeletal muscle in
this patient demonstrated normal tricarboxylic
acid cycle flux but reduced adenosine tri-
phosphate (ATP) production, suggesting im-
paired coupling of these processes (fig. S5, D
and E). Additional studies of other kindred
members will be required to establish the fre-
quency and severity of these manifestations.
These findings establish a causal relation-
ship between a mitochondrial mutation and
hypertension, hypercholesterolemia, and hy-
pomagnesemia. The mitochondrial origin of
this disorder is of particular interest given
recent evidence implicating mitochondrial
dysfunction in type 2 diabetes mellitus and
insulin resistance, other components of the
metabolic syndrome. Rare mitochondrial mu-
tations cause diabetes with deafness (25). In
vivo NMR of skeletal muscle has linked
loss of mitochondrial function to insulin re-
sistance (26). Finally, expression of genes
involved in oxidative phosphorylation is re-
duced among patients with type 2 diabetes
mellitus and insulin resistance (27). Thus, al-
though insulin resistance, obesity, and hy-
pertriglyceridemia are absent in K129, these
traits have been previously linked to loss
of mitochondrial function. These observa-
tions raise the possibility that all the fea-
tures of the metabolic syndrome can result
from pleiotropic effects of impaired mito-
chondrial function; we speculate that the loss
of mitochondrial function with aging (26, 28)
might commonly contribute to all compo-
nents of the metabolic syndrome.
The variation in the phenotypic conse-
quences of this homoplasmic mitochondrial
mutation is notable. Hypomagnesemia, hyper-
tension, and hypercholesterolemia each show
È50% penetrance among adults on the ma-
ternal lineage. Incomplete penetrance arising
from homoplasmic mutations is well de-
scribed and has been attributed to nuclear
genome and/or environmental modifiers (29).
The nearly stochastic distributions of these
traits on the maternal lineage (fig. S2) and
the nonsignificant correlations among their
quantitative values on the maternal lineage
suggests that these are independent, pleio-
tropic effects of the mitochondrial mutation.
Prior studies suggest potential mecha-
nisms linking each trait to impaired mito-
chondrial function. Cells of the DCT have the
highest energy consumption of the nephron
(30), and Mg2þreabsorption in the DCT re-
quires ATP-dependent Naþreabsorption (31).
Inhibitors of mitochondrial ATP production
increase cholesterol biosynthesis while inhib-
iting clearance in vitro (32). Finally, reduced
ATP production has been reported in animal
models of hypertension (33). Further work
will be required to elucidate the molecular
mechanisms linking genotype and phenotype.
The results of this study suggest that
the loss of mitochondrial function with age
(26, 28) could contribute to the characteristic
age-related increase in blood pressure (34)
and to its clustering with hypocholesterol-
emia in the general population. The mutation
in K129 results in a complex pattern of phe-
notypic clustering that is reminiscent of the
frequent but not obligatory clustering seen
in the general population. This highlights
the complexity that can arise from a single
mutation because of the combined effects of
reduced penetrance and pleiotropy and un-
derscores the value of studying very large
kindreds. The present findings motivate fur-
ther investigation of a potential role for mito-
chondrial dysfunction in common forms of
hypertension and hypercholesterolemia.
References and Notes
1. J. Stamler, D. Wentworth, J. D. Neaton, JAMA 256,
2. A. Mosterd et al., N. Engl. J. Med. 340, 1221 (1999).
3. D. L. Wingard, E. Barrett-Connor, M. H. Criqui, L. Suarez,
Am. J. Epidemiol. 117, 19 (1983).
4. G. M. Reaven, Diabetes 37, 1595 (1988).
5. M. H. Criqui et al., Circulation 73, I40 (1986).
6. R. R. Williams et al., JAMA 259, 3579 (1988).
7. S. Mizushima, F. P. Cappuccio, R. Nichols, P. Elliott,
J. Hum. Hypertens. 12, 447 (1998).
8. J. M. Peacock, A. R. Folsom, D. K. Arnett, J. H. Eckfeldt,
M. Szklo, Ann. Epidemiol. 9, 159 (1999).
9. F. Guerrero-Romero, M. Rodriguez-Moran, Acta Dia-
betol. 39, 209 (2002).
10. H. Masuzaki et al., Science 294, 2166 (2001).
11. R. P. Lifton, A. G. Gharavi, D. S. Geller, Cell 104, 545
12. J. L. Goldstein, M. S. Brown, Science 292, 1310 (2001).
13. G. I. Bell, K. S. Polonsky, Nature 414, 788 (2001).
14. S. George et al., Science 304, 1325 (2004).
15. S. O’Rahilly, I. S. Farooqi, G. S. Yeo, B. G. Challis,
Endocrinology 144, 3757 (2003).
16. M. Konrad, K. P. Schlingmann, T. Gudermann, Am. J.
Physiol. Renal Physiol. 286, F599 (2004).
17. Mutations in PPAR, and Akt2 may be an exception,
as the few patients reported have both insulin re-
sistance and hypertension.
18. Materials and methods are available as supporting
material on Science Online.
19. MITOMAP: A Human Mitochondrial Genome Data-
base, available at www.mitomap.org.
20. M. Sprinzl, C. Horn, M. Brown, A. Ioudovitch, S. Steinberg,
Nucleic Acids Res. 26, 148 (1998).
21. S. H. Kim et al., Science 179, 285 (1973).
22. G. J. Quigley, A. Rich, Science 194, 796 (1976).
23. S. S. Ashraf et al., RNA 5, 188 (1999).
24. D. C. Wallace, Science 283, 1482 (1999).
25. P. Maechler, C. B. Wollheim, Nature 414, 807 (2001).
26. K. F. Petersen et al., Science 300, 1140 (2003).
27. V. K. Mootha et al., Nature Genet. 34, 267 (2003).
28. A. Trifunovic et al., Nature 429, 417 (2004).
29. V. Carelli, C. Giordano, G. d’Amati, Trends Genet. 19,
30. R. F. Reilly, D. H. Ellison, Physiol. Rev. 80, 277 (2000).
31. D. B. Simon et al., Nature Genet. 12, 24 (1996).
32. R. A. Zager, A. C. Johnson, S. Y. Hanson, Am. J. Physiol.
Renal Physiol. 285, F1092 (2003).
33. A. Atlante et al., Int. J. Mol. Med. 1, 709 (1998).
34. R. S. Vasan et al., JAMA 287, 1003 (2002).
35. We thank the members of K129 for their generous
participation in this project; I. Beerman, C. Mendenhall,
and F. Niazi for assistance with patient evaluation;
D. Befroy and S. Dufour for assistance with spec-
troscopy; C. Ariyan and J. Kim for help with muscle
biopsy; C. Garganta for measurement of urinary
amino acids and organic acids; the staff of the Yale
General Clinical Research Center; and A. Gharavi for
helpful discussions. Supported by NIH grant nos.
MO1 RR-00125, P50 HL-55007, and R01 DK-49230.
A.H. is the recipient of an American Heart Associa-
tion Fellowship (no. 0475003N).
Supporting Online Material
Materials and Methods
Figs. S1 to S5
Tables S1 to S3
References and Notes
8 July 2004; accepted 6 September 2004
Published online 21 October 2004;
Include this information when citing this paper.
Multidimensional Drug Profiling
By Automated Microscopy
Zachary E. Perlman,1,2* Michael D. Slack,3*. Yan Feng,1*-
Timothy J. Mitchison,1,2Lani F. Wu,3`
Steven J. Altschuler3`
We present a method for high-throughput cytological profiling by microscopy.
Our system provides quantitative multidimensional measures of individual cell
states over wide ranges of perturbations. We profile dose-dependent phe-
notypic effects of drugs in human cell culture with a titration-invariant
similarity score (TISS). This method successfully categorized blinded drugs and
suggested targets for drugs of uncertain mechanism. Multivariate single-cell
analysis is a starting point for identifying relationships among drug effects at a
systems level and a step toward phenotypic profiling at the single-cell level. Our
methods will be useful for discovering the mechanism and predicting the
toxicity of new drugs.
High-throughput methods for describing cell
phenotype such as transcriptional and pro-
teomic profiling allow broad, quantitative,
and machine-readable measures of the re-
sponses of cell populations to perturbation
(1–4). Automated microscopy has the poten-
tial to complement these profiling approaches
by allowing fast and cheap collection of data
describing protein behaviors and biological
pathways within individual cells (5–9). Ac-
cessing these data to produce useful profiles
of cell phenotype will require new image
R E P O R T S
12 NOVEMBER 2004 VOL 306SCIENCEwww.sciencemag.org
on March 26, 2009
analysis methods, the development of which
has so far lagged behind the adoption of high-
throughput imaging technologies.
In the context of drug discovery, profiling
technologies are useful in measuring both
drug action on a desired target in the cellular
milieu and drug action on other targets.
Ideally, such profiling should be performed
as a function of drug concentration, because
several factors make the effects of drugs
highly dose dependent. For example, the
degree to which a primary target is perturbed
may affect different downstream pathways
differently, and drugs can bind to multiple
targets with different affinities. In some
cases, the therapeutic mechanism may in-
volve binding to more than one target with
differing affinity (10, 11). To date, drug
effects have been broadly profiled with
transcript analysis, proteomics, and measure-
ment of cell line dependence of toxicity (11–
21). In these studies, multidimensional profil-
ing methods were only applied at a single-
drug concentration. The only studies in which
drug dose has been explicitly considered as a
variable used the degree of cell proliferation,
an essentially one-dimensional (1D) readout of
phenotype (12, 13). Two recent reviews have
highlighted the possibility of using combina-
tions of targeted phenotypic imaging screens
togenerateprofilesofdrugactivity(6, 22). Here,
we suggest that large sets of unbiased measure-
ments might serve as high-dimensional cytol-
ogical profiles analogous to transcriptional
profiles. We present a method based on
hypothesis-free molecular cytology that pro-
vides multidimensional single-cell phenotypic
information yet is simple and inexpensive
enough to allow extensive dose-response
profiles for many drugs.
We assembled a test set of 100 com-
pounds (table S1). Of these, 90 were drugs of
known mechanism of action, six were
blinded alternate titrations from this set of
known drugs, one (didemnin B) was a toxin
reported to have multiple biological targets
(23), and three were drugs of unknown
mechanism. The known drug set was chosen
to cover common mechanisms of toxicity or
therapeutic action in cancer and other dis-
eases and to include several groups with a
common target (macromolecule or pathway)
but unrelated structures. We analyzed 13
threefold dilutions of each drug, covering a
final concentration range on cells from
micromolar to picomolar Etable S2 and
supporting online material (SOM) text A^.
HeLa (human cancer) cells were cultured in
384-well plates to near confluence, treated
fluorescent probes for variouscellcomponents
and processes. We chose 11 distinct probes
ing a DNA stain and two antibodies per well
Ethe probe sets are SC35, anillin; "-tubulin,
actin; phospho-p38, phospho–extracellular
signal–regulated kinase (ERK); p53, c-Fos;
phospho–adenosine 3¶,5¶-monophosphate re-
sponse element–binding protein (CREB),
calmodulin^. Using automated fluorescence
microscopy, we collected images of up to
È8000cellsfromeach well.Oneach plate,26
wells were treated only with dimethyl sulfox-
ide (DMSO) to generate a control population
(SOM text A). The experiment was per-
formed twice in parallel to provide a replicate
used to automatically identify nuclei and
nuclear organelles, and cytoplasmic regions
were approximated as an annulus surrounding
each identified nucleus (Fig. 1A and SOM
text B). For each cell, region, and probe, a
set of descriptors was measured. These
included measures of size, shape, and inten-
sity, as well as ratios of intensities between
regions (93 descriptors total, table S3). In all,
È7 ? 107individual cells were identified
from 9600,000 images, yielding È109data
We can examine the population response
of each descriptor to increasing concentra-
tions of a given drug, which we show with
the genotoxic compound camptothecin (24)
(Fig. 1B). At low concentrations, the histo-
gram for the total DNA content has the
characteristic bimodal shape reflecting a
mixture of G1, S, and G2/M cell populations.
G2and M populations may be distinguished
by 2D display of total DNA signal against
nuclear area (25). As drug concentration
increases, the cells arrest with S/G2DNA
content (24). The measured DNA content
distribution shifts leftward as dose increases,
and at the highest concentrations apoptosis is
widely induced. Anillin, a cytokinesis protein
whose levels reflect cell cycle progression
Fig. 1. Key steps in algorithm for reducing image data to compound profile. (A) Image
segmentation. For each image [examples show DNA (blue), SC35 (red), and anillin (green)], we
generated a nuclear region (blue) and a set of associated regions [shown here are cytoplasmic
annulus (yellow) and SC35 speckles (red)]. For each defined nuclear region, we measure multiple
descriptors. (B) Quantification of population response. For a given compound, titration, and
descriptor, we generated a population histogram and related cumulative distribution function (cdf)
(black) to be compared with the control population (blue). Shown is a threefold dilution series
ranging from 65 pM to 35 6M camptothecin. We reduced each experimental cdf to a single
dependent variable through comparison with a control population with the nonparametric KS
statistic against a control population (SOM text C). Each vertical red or green line indicates the
position and sign of the maximal height difference between the curves; this height is the KS
statistic. (C) Heat map of compound profile. A z score is calculated for each KS statistic (SOM text
C), and the vector of z scores for all descriptors and all titrations is displayed for rapid visual
assessment. Increased scores are represented in red and decreased in green, with intensity
encoding magnitude. Arrowheads to the right indicate descriptors shown in (B), and the arrowhead
at the bottom indicates the dose shown in (A).
1Institute of Chemistry and Cell Biology, Harvard
Medical School, Boston, MA 02115, USA.
ment of Systems Biology, Harvard Medical School,
Boston, MA 02115, USA.3Bauer Center for Genomics
Research, Harvard University, Cambridge, MA 02138,
*These authors contributed equally to this work.
.Present address: Alphatech, Inc., San Diego, CA
-Present address: Novartis Institutes for BioMedical
Research, Cambridge, MA 02139, USA.
`To whom correspondence should be addressed.
E-mail: firstname.lastname@example.org (S.J.A.); lwu@
R E P O R T S
www.sciencemag.orgSCIENCEVOL 30612 NOVEMBER 2004
on March 26, 2009
(26), shows marked nuclear accumulation in
the G2arrested state (Fig. 1A). p53, a trans-
cription factor that is part of the genotoxic
response pathway, is strongly induced at high
camptothecin concentrations, but much less so
at concentrations sufficient to promote G2
arrest (Fig. 1B).
For profiling studies, it is useful to reduce
each population of descriptor values to a
single number. Our study made several
demands of this reduction: It must be able
to compare distributions of arbitrary shape
(Fig. 1B); it must be robust to variation in
dynamic range and noise levels among
different descriptors; it must convert differ-
ent types of measurement into a common
unit for comparison; it must be descriptor
parameterization independent (e.g., an inten-
sity ratio should behave the same as its
reciprocal); and it must be insensitive to the
precise quantitative relationship between
antibody-staining intensity and antigen den-
sity. We devised a measure based on the
Kolmogorov-Smirnov (KS) statistic, allow-
ing nonparametric comparison of experimen-
tal and control distributions from the same
plate (Fig. 1B, fig. S1, and SOM text C).
Dividing by a measure of the variability
within the control population yielded a z
score, which can be displayed as a function
of descriptor and drug concentration in a
heat plot to allow rapid visual comparison of
compound response profiles (Fig. 1C). These
plots represent a family of dose-response
curves for a single drug but differ from tra-
ditional curves reflecting changes in a bio-
chemical measurement. In particular, the
relationship between z score and the original
physical measure may be nonlinear. For
example, the statistically significant res-
ponses of p53 to low doses of camptothecin
(Fig. 1C) reflect subtle effects not easily
discerned by eye in the source images.
The heat plots typically have a sharp
transition, reflecting a concentration at
which many descriptors become different
from control values. We will refer to this as
the primary effective concentration (PEC)
for the drug. The isolated responses observed
at some low concentrations represent noise
that could be reduced by increasing repli-
cates, improving experimental procedures,
and normalizing for local variation in cell
density. For 39 drugs, we saw no strong
effect, leaving a heat plot dominated by
noise. Those drugs either lack a target in
HeLa cells, were used at inactive dosages, or
effected changes not detectable with our
antibody set. For almost all of the 61 drugs
that showed a strong response, some descrip-
tors responded at concentrations other than
the PEC (Fig. 2). This may reflect varying
biological consequences of low and high
saturation of a single target, or it may reflect
interactions with multiple targets with dif-
ferent affinities. For example, camptothecin
binds primarily to DNA complexes with
topoisomerase I, promoting DNA strand
breaks and S-phase arrest at low concen-
trations, but it also blocks transcription and a
number of other cellular processes at higher
concentrations (24). Other drugs in our test
set are known to have multiple targets, such
as histone deacetylase inhibitors (27) and the
general kinase inhibitor staurosporine (28)
and were thus expected to show complex
dose-response behavior. Such phenotypic
Fig. 2. Comparison of compound profiles. As in Fig. 1C, the x axis shows increasing dose and the y
axis encodes descriptors. Dose ranges are shown from 65 pM to 35 6M for all drugs except
epothilone B, which is shown from .65 pM to .35 6M. Color scale is as in Fig. 1C. For ease of
visualization, descriptors in all profiles are sorted in decreasing order of camptothecin response.
(A) Compounds of similar mechanism show similar profiles. Shown are representative compound
profiles. HDAC, histone deacetylase; ALLN, N-acetyl-Leu-Leu-norleucinal. (B) Compound profiles
can distinguish differences between drugs with similar mechanisms. Wells with too few cells for
analysis are represented in white.
Table 1. Assessment of TISS by literature categories. For each category that has more than two
compounds, we computed two sets of TISS scores: pairwise TISS comparisons between members of the
category (intrapair) and comparisons in which only one element of the pair is in the category (interpair).
As a crude in silico comparison to other cell-based assays such as fluorescence-activated cell sorting
(single-cell based) and cytoblots (whole-population based), we repeated this procedure with a descriptor
set consisting of only total intensity measures and compared it with either our KS-based TISS values or a
mean-based TISS values (SOM text C). P values (columns 2 to 4) describe the probability that the rank
ordering of the two sets of TISS values would have been seen by random draws from the same
distribution (SOM text C). KS, KS-based TISS (P value); mean, mean-based TISS (P value).
No. pairwise TISS
KS KSMeanIntrapair Interpair
3.86 ? 10j20
6.02 ? 10j5
9.81 ? 10j6
R E P O R T S
12 NOVEMBER 2004VOL 306SCIENCEwww.sciencemag.org
on March 26, 2009
complexity may help explain why toxicity at
high doses is common even for therapeutic
drugs that are apparently highly selective at
the level of target binding.
Drugs with common reported targets but
diverse chemical structures often showed
similar profiles readily distinguished from
those of drugs of different mechanism (Fig.
2A). In other cases, markedly different
profiles were evident within a family, most
notably the protein synthesis inhibitors (Fig.
2B). This may reflect different cell responses
to alternative biochemical mechanisms of
poisoning ribosomes (29) or perhaps the
existence of alternate targets (23).
in specificity (and thus phenotype) are rele-
vant, but changes in affinity (and thus PEC)
are not. Two different dosage series of the
same drug should result in similar heat plots
shifted along the concentration axis. We
developed a titration-invariant similarity
score (TISS) to allow comparison between
dose-response profiles independent of start-
ing dose (SOM text C). TISS values were
generated for the 61 compounds that showed
significant signal, and these were used for
unsupervised clustering (Fig. 3). TISS was
successful at grouping compounds with sim-
ilar reported targets (Table 1). As expected,
clustering reflected biological mechanism
rather than chemical similarity. For example,
kinase inhibitors, most of which are adeno-
sine 5¶-triphosphate–mimetic compounds, did
not cluster as a group. Clustering was poor
even within a set of kinase inhibitors with
overlapping targets Ecyclin-dependent kinase
(CDK) inhibitors^, perhaps reflecting variable
inhibition of other kinases. The CDK inhib-
itors related by structure and reported target
(purvalanol, roscovitine, and olomoucine) did
Of the blinded alternate titrations of known
drugs, scriptaid, hydroxyurea, emetine, and
two alternate series of nocodazole showed sig-
nificant responses. These clustered closely with
their unblinded counterparts and compounds of
similar reported mechanism. Didemnin B, for
which the reported range of activities includes
inhibition of protein synthesis (23), clustered
with ribosome inhibitors (Fig. 2B). Two of the
three poorly characterized compounds showed
strong responses. One, concentramide, is
difficult to interpret. The other, austocystin,
clusters with transcription and translation
inhibitors. Preliminary experiments suggest
that this compound inhibits transcription in
vitro (25). Thus, our methods can group com-
pounds of like mechanism and thereby
suggest mechanism for new drugs.
Extensions of cytological profiling to
reflect dependencies among descriptors will
allow more sophisticated analysis of drug
Fig. 3. Hierarchical clustering of the 61 most responsive compound profiles
by TISS values. Compound stock concentrations (6M) are in parentheses
(fig. S3). Left panel shows mechanism of compound as described in litera-
ture. In blue are compounds that were blinded or are of unknown mech-
anism. Middle panel shows matrix of P values derived from pairwise TISS
values (SOM text C). Dendrogram at top shows degree of association.
R E P O R T S
www.sciencemag.orgSCIENCE VOL 30612 NOVEMBER 2004
on March 26, 2009
responses at a systems level. For example, Download full-text
both p53 and c-Fos, a transcription factor
involved in mitogen-activated protein kinase
(MAPK) signaling, are involved in cell stress
responses, but the interrelationship of the
p53 and MAPK pathways is poorly under-
stood (30). Single-cell profiling reveals that
different drug mechanisms induce different
relative patterns of response by these two
pathways (Fig. 4). The proteasome inhibitor
MG132 causes increased correlated induc-
tion in these pathways, whereas responses to
camptothecin are anticorrelated. Anticorre-
lated responses observed in fixed-time
images may reflect switching of mutually
exclusive cell states in response to different
degrees of stress or might reflect a dynamic
temporal response, such as oscillation, that is
not synchronized among cells (31). These
data help establish a concentration and time
window, but live imaging will be required to
distinguish between these hypotheses.
Cytometric dose-response profiling is a
fast and cheap method for quantitatively
surveying broad ranges of individual cell
responses. We have used our methods to
assign mechanism to blinded and uncharac-
terized drugs and to suggest systems-level
relationships between signaling pathways.
The complex dose-response curves and large
cell-to-cell variability we frequently observed
reinforce the utility of unbiased multidimen-
sional characterization of drug effects over
wide ranges of doses.
Many improvements and extensions of
this work are possible. These include better
lab automation, broader drug reference sets,
different types of perturbation (such as RNA
interference), improved strategies for cell
segmentation, more sophisticated feature ex-
traction (5, 9), different sets of antibody
probes and cells, the inclusion of more time
points and live cell imaging, and the inte-
gration of complementary profiling strategies.
Additionally, our methods may be extended to
allow the characterization of responses by
subpopulations defined by such variables as
cell cycle state, cell density, or neighboring
environment. This analysis, extended to work
in tissues or clinical samples, offers the po-
tential to speed the identification of toxic
compounds during therapeutic drug develop-
ment and the targeting of drug effects to
specific subtypes of cells.
References and Notes
1. M. B. Eisen, P. T. Spellman, P. O. Brown, D. Botstein,
Proc. Natl. Acad. Sci. U.S.A. 95, 14863 (1998).
2. A. C. Gavin et al., Nature 415, 141 (2002).
3. Y. Ho et al., Nature 415, 180 (2002).
4. P. Uetz et al., Nature 403, 623 (2000).
5. R. F. Murphy, M. Velliste, G. Porreca, J. VLSI Signal
Process. 35, 311 (2003).
6. J. H. Price et al., J. Cell. Biochem. Suppl. 39, 194
7. W. K. Huh et al., Nature 425, 686 (2003).
8. T. L. Saito et al., Nucleic Acids Res. 32, D319 (2004).
9. C. Conrad et al., Genome Res. 14, 1130 (2004).
10. J. G. Hardman, L. E. Limbird, A. G. Gilman, Eds., The
Pharmacological Basis of Therapeutics (McGraw-Hill,
New York, ed. 10, 2001), pp. 39–42.
11. M. J. Marton et al., Nature Med. 4, 1293 (1998).
12. J. N. Weinstein et al., Science 275, 343 (1997).
13. K. D. Paull, C. M. Lin, L. Malspeis, E. Hamel, Cancer Res.
52, 3892 (1992).
14. U. Scherf et al., Nature Genet. 24, 236 (2000).
15. E. C. Gunther, D. J. Stone, R. W. Gerwien, P. Bento,
M. P. Heyes, Proc. Natl. Acad. Sci. U.S.A. 100, 9608
16. D. Leung, C. Hardouin, D. L. Boger, B. F. Cravatt,
Nature Biotechnol. 21, 687 (2003).
17. M. A. Lindsay, Nature Rev. Drug Discov. 2, 831 (2003).
18. P. Y. Lum et al., Cell 116, 121 (2004).
19. G. Giaever et al., Proc. Natl. Acad. Sci. U.S.A. 101,
20. S. J. Haggarty, P. A. Clemons, S. L. Schreiber, J. Am.
Chem. Soc. 125, 10543 (2003).
21. D. E. Root, S. P. Flaherty, B. P. Kelley, B. R. Stockwell,
Chem. Biol. 10, 881 (2003).
22. V. C. Abraham, D. L. Taylor, J. R. Haskins, Trends
Biotechnol. 22, 15 (2004).
23. M. D. Vera, M. M. Joullie, Med. Res. Rev. 22, 102
24. C. J. Thomas, N. J. Rahier, S. M. Hecht, Bioorg. Med.
Chem. 12, 1585 (2004).
25. Z. E. Perlman et al., data not shown.
26. C. M. Field, B. M. Alberts, J. Cell Biol. 131, 165 (1995).
27. M. Yoshida et al., Cancer Chemother. Pharmacol. 48
(suppl. 1), S20 (2001).
28. M. E. Noble, J. A. Endicott, L. N. Johnson, Science 303,
29. J. D. Laskin, D. E. Heck, D. L. Laskin, Toxicol. Sci. 69,
30. B. Kaina, Biochem. Pharmacol. 66, 1547 (2003).
31. G. Lahav et al., Nature Genet. 36, 147 (2004).
32. We thank A. Daneau, M. Ethier, and B. Mantenuto at
the Bauer Center for Genomics Research for their
assistance with the use of the Bauer Center compu-
tational cluster and K. Maciag, A. Murray, O. Rando,
A. Salic, and A. Yonetani for helpful discussions.
Supported in part by National Cancer Institute PO1
CA078048. Z.E.P. is a Howard Hughes Medical Institute
Supporting Online Material
Figs. S1 to S5
Tables S1 to S4
25 May 2004; accepted 8 September 2004
Fig. 4. Single-cellanalysisshowsdif-
fering patterns of dose-dependent
p53 and c-Fos responses to differ-
ent drugs. (A) Scatter plot of
average nuclear p53 intensity
versus average c-Fos intensity in
a typical control well and repre-
sentative image. The bright cells
at the top of the image are in
mitosis. (B) Dose-dependent in-
creases in response to MG132
shown in heat maps are correlat-
ed in scatter plots and images
(orange nuclei), shown for the four
highest concentrations. (C) Dose-
dependent increases in response to campto-
thecin shown in heat maps are anticorre-
lated in scatter plots and images. The black
(c-Fos) and green (p53) heat map values
for the highest dose reflect the contribu-
tion of apoptotic cells with negligible p53
and c-Fos nuclear staining.
R E P O R T S
12 NOVEMBER 2004VOL 306SCIENCEwww.sciencemag.org
on March 26, 2009