-
Therapeutic Discovery
A Robust High-Content Imaging Approach for Probingthe Mechanism
of Action and Phenotypic Outcomes ofCell-Cycle Modulators
Jeffrey J. Sutherland1, Jonathan Low2, Wayne Blosser2, Michele
Dowless2, Thomas A. Engler3, andLouis F. Stancato2
AbstractHigh-content screening is increasingly used to elucidate
changes in cellular biology arising from treat-
ment with small molecules and biological probes. We describe a
cell classifier for automated analysis of
multiparametric data from immunofluorescence microscopy and
characterize the phenotypes of 41 cell-cycle
modulators, including several protein kinase inhibitors in
preclinical and clinical development. This method
produces a consistent assessment of treatment-induced phenotypes
across experiments done by different
biologists and highlights the prevalence of nonuniform and
concentration-dependent cellular response to
treatment. Contrasting cell phenotypes from high-content
screening to kinase selectivity profiles from cell-free
assays highlights the limited utility of enzyme potency ratios
in understanding the mechanism of action
for cell-cycle kinase inhibitors. Our cell-level approach for
assessing phenotypic outcomes is reliable,
reproducible and capable of supporting medium throughput
analyses of a wide range of cellular perturba-
tions. Mol Cancer Ther; 10(2); 242–54. �2011 AACR.
Introduction
The characterization of cell populations with
immuno-fluorescence microscopy, or high-content screening,allows a
detailed understanding of the effect of smallmolecule modulators on
the mitotic cell cycle. By quan-tifying the intensity,
localization, and morphology of 4 to5 markers in individual cells,
such experiments typicallyproduce on the order of 105 to 106
observations pertreatment. In recognition of the heterogeneity of
cellpopulations (1), several methods have been proposedfor
analyzing treatment-induced perturbations by cellu-lar imaging.
Such approaches assign a phenotype toindividual cells, using rules
to emulate their classificationby biologists (2–4), or by applying
nonsupervised (5–9) orsupervised (10, 11) multivariate analysis of
cytologicalfeatures. Methods for analysis of high-content
imagingexperiments have been reviewed elsewhere (12, 13).
Chemical modulation of the mitotic cell cycle hasproven to be
effective in treating cancer. A number ofapproved chemotherapeutic
agents disrupt DNA repli-cation [e.g., the topoisomerase inhibitors
(14) topotecanand camptothecin] or microtubule dynamics (15;
e.g.,the tubulin modulators paclitaxel and vinca alkaloids).The
mitotic cell cycle is regulated by many kinases, andthe search for
small molecule inhibitors has produced anumber of agents in
preclinical and clinical development.
The roles of kinases such as CDK1, AURKA, AURKB,and PLK1 in the
G2-M checkpoint are well-established(16). The CDK1-cyclin B complex
regulates entry intomitosis; loss of CDK1 function results in
arrest at theG2-M boundary and enrichment of cell populationshaving
large nuclei with 4N DNA present as diffusechromatin. Because of
their role at the G2-M checkpoint,inhibition of other G2-M kinases
in addition to CDK1 isexpected to result in a phenotype
consistentwith selectiveCDK1 inhibition (17). The kinase AURKA is
involved incentrosome regulation, and its inhibition manifests
itselfvia enrichment of cells in prometaphase. AURKA pro-motes
bipolar spindle assembly (18), andmutations in thiskinase prevent
centrosome separation leading to the for-mation of monopolar
spindles (19). In addition to its rolein cytokinesis, AURKB
activates the spindle-assemblycheckpoint, and manifestation of
AURKA inhibitionrequires a functional spindle-assembly checkpoint.
Forthis reason, dual aurora A/B inhibitors are expected toyield a
phenotype consistent with selective AURKB inhi-bition (20).
Finally, PLK1 functions in spindle formation,chromosome segregation
and cytokinesis, the inhibition
Authors' Affiliation: 1Lilly Research Labs Information
Technology, 2Can-cer Biology & Patient Tailoring, and
3Discovery Chemistry Research andTechnology, Eli Lilly and Company,
Indianapolis, Indiana
Note: Supplementary material for this article is available at
MolecularCancer Therapeutics Online
(http://mct.aacrjournals.org/).
Corresponding Author: Jeffrey J. Sutherland, Eli Lilly and
Company, LillyCorporate Center, Indianapolis, IN, 46285. Phone:
317-655-0833; Fax:317-276-6545. E-mail: [email protected]; or
Louis F. Stancato, Can-cer Biology & Patient Tailoring,
Indianapolis, IN, 46285. E-mail:[email protected]
doi: 10.1158/1535-7163.MCT-10-0720
�2011 American Association for Cancer Research.
MolecularCancer
Therapeutics
Mol Cancer Ther; 10(2) February 2011242
on June 1, 2021. © 2011 American Association for Cancer
Research. mct.aacrjournals.org Downloaded from
Published OnlineFirst January 7, 2011; DOI:
10.1158/1535-7163.MCT-10-0720
http://mct.aacrjournals.org/
-
of which results in failure to establish a bipolar spindle
inprometaphase (21).Protein kinases have a high degree of
structural homol-
ogy, and activity of small molecule inhibitors againstproteins
other than the intended target (i.e., off-targetactivity) is
frequently observed in cell-free assays. Tounderstand the
relationship between selectivity andeffects on cell populations, we
developed a classifier ofcellular phenotype in HCT-116 cells (a
cell line for color-ectal carcinoma), and applied it to kinase and
nonkinasemodulators of the cell cycle. The approach yields
ahighly-reproducible assessment of changes in cell popu-lations
induced by different treatments or different con-centrations of the
same treatment, and highlights thelimited utility of kinase
selectivity panels in selectinginhibitors that exhibit the desired
phenotype.
Materials and Methods
Cell-cycle modulators and high-content imagingVarious inhibitors
of proteins involved in cell-cycle
regulation were selected from the Lilly corporate collec-tion
(Table 1). Where available, compounds were pur-chased from
commercial vendors. Several reportedkinase inhibitors not available
for purchase were synthe-sized internally, using synthetic schemes
described in thepublic domain. All compounds have � 95%
purity.Experiments were done by 3 biologists over a 6-monthperiod,
to study phenotypes induced by cell-cycle mod-ulators. Experiments
2 and 3 were designed to specifi-cally probe reproducibility of
phenotypes, whereasexperiment 1 was designed to characterize a
large collec-tion of kinase inhibitors. HCT-116 cells were plated
onto96-well dishes, treated with compounds in 10-point
con-centration curves, and imaged on the Arrayscan VTIplatform (see
Supplementary Methods). Cytological fea-tures of cells were
captured with the Target Activationbio-application bundled with the
imaging instrument.The selected cytological features quantify a
number ofimportant changes in cells undergoing mitosis
(Supple-mentary Table 1). Objects are defined from nuclei
identi-fication, and mostly correspond to individual cells;clusters
of nuclei from treatments causing polyploidyare captured as 1
object with �8N DNA content anddaughter nuclei in anaphase are
captured as 2 objects.
Normalization of cellular featuresNumerical values of features
from the instrument soft-
ware are log2 transformed to increase the normality
ofdistribution across cell populations (base 2 is convenientfor
counting doublings of intensity, e.g., DNA content of2N, 4N, 8N,
etc.). To account for plate-to-plate variationin cytological
features (i.e., changes arising from variationin antibody staining
intensity, incubator conditions, etc.),individual values for each
feature are converted to Z-scores, using the mean and standard
deviation obtainedby pooling dimethyl sulfoxide (DMSO)-treated
cells from8 negative control wells (i.e., normalization of features
on
a per-plate basis). The intensity distributions for cells in
8positive control wells containing 0.2 mmol/L of nocoda-zole are
visually assessed for each plate and channel toverify for
consistency across plates after feature scaling(Supplementary Fig.
1). This method of normalization isadequate to control for
plate-to-plate variability within anexperiment. All further
analysis uses scaled values ofcytological features.
Quantifying antiproliferative effects of compoundtreatments
For each well, the cell density is calculated by count-ing the
number of objects (cells) per field of view, andaveraging across
all fields for a given well. For a treat-ment compound, cell
density is converted to a percent-age relative to the
plate-averaged cell density fromDMSO treatment (i.e., 100%
corresponds to the averagecell density for DMSO treatment).
Logistic regressioncurve fits were done using TIBCO Spotfire
(Version 2.1;TIBCO Software, Inc.), and the concentration at
whichthe curve crosses 50% is reported as the EC50 of
thecompound.
Quantifying cell phenotypes induced by compoundtreatments
A set of 8 reference compounds of reportedmechanismwere selected
for the purpose of classifying cells byphenotype (designated in
bold type in Table 1). Thesewere the CDK inhibitors AG-024322 and
R-547, the aur-ora kinase inhibitors AZD-1152 and tozasertib, the
PLK1inhibitor BI-2536, and the microtubule modulators ON-01910,
nocodazole and paclitaxel. ON-01910 was origin-ally reported as a
non-ATP competitive PLK1 inhibitor; insubsequent reports it has
been found to not inhibit PLK1biochemically and generates a cell
phenotype consistentwith microtubule modulation (22, 23). The
compoundswere selected to represent a diversity of
phenotypesobserved for G2-Mmodulators. Visual analysis of
imagesreveals cell populations consistent with themechanism
ofaction of the compounds at all concentrations above
theantiproliferation EC50, allowing those wells to be pooledfor the
purpose of training a classifier. Cells in these wellsare described
by the cytological features in Supplemen-tary Table 1, and assigned
the class label of the treatmentcompound. To our knowledge, there
are no cell-cyclemodulators that arrest cells in metaphase or
anaphase. Totrain the classifier in the identification of these
states,images for DMSO-treated cells in experiment 3 werereviewed
and �30 cells of each type identified andlabeled accordingly. The
total number of cells acrosspooled wells used for training the
classifier was 63,575,58,298, and 78,552 for experiments 1 to
3.
The reference compounds were used to develop aclassifier of
cells for each experiment (i.e., 3 experiments,3 classifiers). The
classifier was developed with therecursive partitioning algorithm
in Jmp (Version 7.0.2,SAS Institute, Inc), using the compound class
label asresponse variable and cytological features as factors.
The
High-Content Imaging of Cell-Cycle Modulators
www.aacrjournals.org Mol Cancer Ther; 10(2) February 2011
243
on June 1, 2021. © 2011 American Association for Cancer
Research. mct.aacrjournals.org Downloaded from
Published OnlineFirst January 7, 2011; DOI:
10.1158/1535-7163.MCT-10-0720
http://mct.aacrjournals.org/
-
Table 1. Compounds characterized by high-content imaging in
HCT-116
Namea EC50, mmol/L Dominant phenotypec
Experimentb 1 Experiment 2 Experiment 3
Aurora inhibitorsAZD1152 0.026 0.023 0.019 EndoAZD1152
metabolite 0.034 0.025 0.030 EndoCYC116 0.649 EndoENMD-2076 0.519
proM#, endo"MLN-8054 0.247 >2 0.380 proM#, endo"PHA-739358 0.036
0.075 0.056 endo#, proM þ M-apopt"Tozasertib (VX-680; MK0457) 0.045
0.050 0.066 Endo
CDK inhibitorsAG-024322 0.173 0.116 0.102 G2
AG-12286 (WO 9921845 A2 223784-75-6) 0.249 G2 þ proMAlvocidib
(flavopiridol) 0.153 0.256 0.266 G2 þ proMBMI-1026 (39) 0.030
G2BMS-265246 (40) 0.293 0.492 0.463 G2JNJ-7706621 (41, 42) 0.524
G2PD-171851 (43) 0.570 G2 þ proMPD-0332991 >2
G1-SAminopurvalanol 2.500 G2R-547 0.077 0.144 0.104 G2SCH-727965
0.018 0.008 0.010 G2 þ proMSNS-032 0.074 0.119 0.128 G2 þ proMWO
2001064655 A1 358788-29-1 0.292 G2 þ proMWO 2001064656 A1
358789-50-1 0.250 G2 þ endo
PLK1 inhibitorsBI-2536 0.009 0.008 0.007 proM þ
M-apoptGSK-461364 0.012 0.014 0.012 proM þ M-apoptHMN-176 (28)
0.164 0.260 0.216 proM þ M-apoptWO 2006049339 A1 886856-66-2 0.101
0.072 0.070 G2 þ proM þ M-apoptWO 2006066172 A1 893440-87-4 1.861
proM þ M-apopt
Wee1 inhibitorUS 2007254892 A1 955365–24-9 2.340 proM þ
M-apopt
DNA intercalators; Topoisomerase inhibitorsAclarubicin 0.238
0.255 G1-SCamptothecin 0.011 0.008 G2#, G2þproM"Doxorubicin 0.254
0.247 G2Topotecan 0.020 0.033 G2#, G2 þ proM"
DNA synthesis inhibitors5-Fluoro-20-deoxyuridine (floxuridine)
>2 >2 G1-SHerboxidiene 0.010 0.011 G2 þ proMIlludin S 0.015
0.019 G1-SMitomycin 0.187 0.253 G2
Microtubule modulatorsON-01910 0.100 0.052 0.025 proM þ
M-apoptAlbendazole 0.448 0.305 proM þ M-apoptCiclobendazole 0.547
0.691 proM þ M-apoptMebendazole 0.318 0.483 proM þ
M-apoptNocodazole 0.055 0.072 proM þ M-apoptPaclitaxel 0.003 0.007
proM þ M-apopt
Abbreviations: G1-S, G1 or S-phase; proM, prometaphase; M-apopt,
M-phase apoptotic; endo, endoreduplication.aStructures are given in
Supplementary scheme 1; compounds in bold are used as references in
calibrating the cell classifier; kinaseinhibitors with no reference
given are described at http://www.clinicaltrials.gov; inhibitors
described in patents, but not the journalliterature, are identified
by the CAS number retrieved in SciFinder (Version 2007.3, American
Chemical Society).bExperiments in which the compound was
characterized, each of which was done by a different biologist on a
different date.cPhenotype from cell-classifier for dominant cell
population(s), that is, polyploidy; " and # indicate phenotypes at
lower and higherconcentrations, respectively.
Sutherland et al.
Mol Cancer Ther; 10(2) February 2011 Molecular Cancer
Therapeutics244
on June 1, 2021. © 2011 American Association for Cancer
Research. mct.aacrjournals.org Downloaded from
Published OnlineFirst January 7, 2011; DOI:
10.1158/1535-7163.MCT-10-0720
http://mct.aacrjournals.org/
-
maximum significance rule was used for selecting
splits.Recursive partitioning is a greedy approach, selecting
thebest feature for each split in an incremental mannerregardless
of its appeal from a biological perspective.To emulate the manner
in which biologists analyze cells,the first splits are obtained by
visual binning of DNAintensity into 2N, 4N, 8N, and >8N
categories. This isdone manually via histogram plots in TIBCO
Spotfire,allowing the boundaries between 2N, 4N, etc. to changefrom
experiment to experiment (Supplementary Fig. 2).For experiment 3,
the subsequent splits used features andrules selected by the
recursive partitioning algorithm.The terminal nodes are assigned
names consistent withthe values of cytological features and review
of images.For the other experiments, the features selected in
experi-ment 3were retained for each split, but the split
valuewasallowed to change. The rules for classifying metaphaseand
anaphase cells are not updated in experiments 1 to 2because of
their rarity in the study of cell-cycle modula-tors, and the need
to identify representative cells byreview of images. The complete
decision tree as depictedin Jmp is shown in Supplementary Figure
3.
Comparing treatment wellsFor the purpose of contrasting the cell
classifier to other
approaches, it is necessary to quantitatively assess
thesimilarity of phenotypes observed under different treat-ment
conditions (i.e., comparing cell phenotypes betweenwells). The
phenotype profile for awell is represented as avector of length 9,
indicating the proportion of cellsbelonging to each phenotype. For
comparison, a vectorof length5 represents thewell-averagedvalues
for 5DNA-related features (excludingObjectVarIntenDNAdue to itshigh
correlation with ObjectAvgIntenDNA; Pearson r >0.9 for the 3
experiments).We employed theEuclidian andcosine distance metrics
for comparing pairs of vectors:
p ¼ p1; p2; . . . ; pn� �
; q ¼ q1; q2; . . . ; qn� �
dis tan ce; Euclidian
¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPni¼1
pi � qið Þ2s
dis tan ce; cos ine ¼Pni¼1
piqiffiffiffiffiffiffiffiffiffiPni¼1
p2i
r ffiffiffiffiffiffiffiffiffiPni¼1
q2i
r
Results
Antiproliferation effects of cell-cycle inhibitorsA total of 41
cell-cycle modulators were characterized
via high-content imaging of HCT-116 cells. Cells wereimaged on
the Arrayscan VTI platform, quantifyingintensity, localization
and/or morphology of Hoechststain (DNA), terminal deoxynucleotidyl
transferasedUTP nick end labeling (TUNEL; amarker for
apoptosis),and antibodies for cyclin B1, phospho-histone H3(pHH3),
and a-tubulin (see Supplementary Methods).
To assess the reproducibility of imaging experiments,29
compounds were tested in 2 or 3 separate experiments,done by
different biologists over a 6-month period. Mostcompounds inhibit
cell proliferation at a concentration of1 mmol/L or less (Table 1).
Using the 58 pairs of definedEC50s for the same compound (i.e.,
EC50s not prefixedwith the symbol ">"), the fold difference
between experi-ments ranges from 1.0 to 3.9, with an average of
1.5. Theassessment of antiproliferative properties of
treatments,measured by quantifying changes in cell density (i.e.,
cellcount per field of view), is highly reproducible
acrossexperiments.
Classification of cell phenotypes fromimmunofluorescence
microscopy
A classifier of cell phenotypes was developed using 8reference
compounds, selected to represent a diversity ofmechanisms among
G2-M modulators (see the Methodssection). A visual assessment of
images from treatmentconcentrations above the antiproliferation
EC50 revealscell populations consistent with published reports
onstandard tubulin modulators (15) and RNAi approachesfor kinase
targets (17, 20, 23). Wells containing a referencecompound at
concentrations above the antiproliferationEC50 were pooled, and
cells described using cytologicalfeatures (e.g., nuclear area,
cyclin B1 intensity, etc.) listedin Supplementary Table 1. The
classifier begins withassessment of DNA content (2N, 4N, 8N,
>8N) usingboundaries defined by visual analysis of Hoechst
DNAstaining intensity (Supplementary Fig. 2), and creates adecision
tree by applying recursive partitioning usingcytological features
as factors and the compoundmechanism of action as response
variable. The rules thatconstitute the decision tree are selected
incrementally in amanner that distinguishes cells treated with
compoundshaving different mechanisms. A cell’s phenotype isdefined
by the terminal node into which it falls(Fig. 1). As such, each
cell is assigned 1 of 9 possiblephenotypes, including the phenotype
"other" that corre-sponds mostly to cell debris. The model is
recalibrated ineach experiment by analysis of cells treated with
refer-ence compounds, without the need to examine and iden-tify
cells for training as required with other approaches.
For the CDK1 inhibitors AG-024322 and R-547, thedominant cell
populations are G2-arrested cells having4N DNA content, large round
nuclei, and low DNAintensity (i.e., diffuse chromatin). By
contrast, treatmentwith the PLK1 inhibitor BI-2536 results mostly
in cellscharacteristic of prometaphase arrest with 4N DNA con-tent
and high DNA intensity (i.e., condensed chromatin).The aurora
kinase inhibitors AZD-1152 and tozasertibinduce polyploidy (via
endoreduplication), consistentwith the dominance of the AURKB
phenotype overAURKA. With a 48-hour incubation (allowing 2
cellnumber doublings), the dominant population shouldconsist of
cells with 8N DNA content; smaller popula-tions with 4N and >8N
DNA content arise from misseg-mentation of nuclei clusters and
mostly represent
High-Content Imaging of Cell-Cycle Modulators
www.aacrjournals.org Mol Cancer Ther; 10(2) February 2011
245
on June 1, 2021. © 2011 American Association for Cancer
Research. mct.aacrjournals.org Downloaded from
Published OnlineFirst January 7, 2011; DOI:
10.1158/1535-7163.MCT-10-0720
http://mct.aacrjournals.org/
-
artifacts of image analysis (some >8N cells are
present).Representative images from each mechanistic class areshown
in Figure 2 and Supplementary Figure 4.
The classifier makes significant use of features derivedfrom DNA
staining (Fig. 1). In characterizing G2-M mod-ulators, other
markers such as cyclin B1 and pHH3 arefrequently examined in
conjunction with DNA content.While the proportion of cells for a
given phenotype varieswidely across mechanistic classes, cyclin B1
and pHH3staining intensities of cells for a given phenotype
aresimilar across mechanistic classes (SupplementaryFig. 5), except
for DMSO-treated cells having lower cyclin
B1 intensity for all phenotypes, and higher pHH3 inten-sity for
prometaphase andmetaphase cells. TUNEL stain-ing, a measure of
apoptosis through DNA end-labelingfollowing DNA fragmentation,
shows greater differentia-tion across mechanistic classes for cells
of a given phe-notype: cells treated with CDK1 and PLK1
inhibitorshave high induction of apoptosis compared to DMSOor
aurora inhibitor treatments. Most aurora inhibitorsinduce
cytokinesis defects, but cells continue cyclingbeyond 48 hours. The
intensity of a-tubulin immunos-taining from treatment with
paclitaxel, a microtubulestabilizer, is increased compared to
destabilizers such
Figure 1. Decision tree used for classifying cells according to
cytological features (described in Supplementary Table 1). DNA
content is defined bymanual binning of ObjectTotalIntenDNA (left
rectangles; see Supplementary Fig. 2). Other splits are obtained
from recursive partitioning in Jmp, with those ingray defined by
partitioning cells treated with 8 calibration compounds, using the
compound mechanism of action as class label. The splits in black
aredefined by partitioning examples of cells in G1, metaphase, and
anaphase; approximately 30 metaphase and 30 anaphase cells were
identified bymanual review of images from DMSO treatment in
experiment 3. A cell is assigned the phenotype for the terminal
node (rectangles) into which it falls.Cells in the nodes endo 4N,
endo 8N, and endo >8N denote multinucleated cells; all are
combined into 1 phenotype. Likewise, cells assigned to
M-apoptoticand M-apoptotic brightest are combined, since they arise
from staining variation with Hoechst 33258 in apoptotic cells (see
text). With the exception of thenodes in black, numerical values
used in the decision tree for each cytological feature are adjusted
by recalibrating the model in each experiment; thenotation used to
identify nodes [e.g., S1: 1.2, 1.0, 1.1 (0.95)] indicates the split
number (S1), the numerical values of a cytological feature used
forseparating cells in 3 experiments (1.2, 1.0, and 1.1 for
experiments 1–3, respectively), and the range observed in 17
subsequent experiments done withcompounds from lead optimization
programs (0.95).
Sutherland et al.
Mol Cancer Ther; 10(2) February 2011 Molecular Cancer
Therapeutics246
on June 1, 2021. © 2011 American Association for Cancer
Research. mct.aacrjournals.org Downloaded from
Published OnlineFirst January 7, 2011; DOI:
10.1158/1535-7163.MCT-10-0720
http://mct.aacrjournals.org/
-
as nocodazole, allowing some degree of differentiation;all
mechanistic classes result in higher tubulin intensitythan DMSO
treatment.In addition to phenotypes commonly observed in non-
treated cells (e.g., G2, prometaphase, etc.), the
classifieridentifies "apoptotic" phenotypes for G2 and M
states(Fig. 1). These cells have apparent 8N or >8N DNAcontent,
and occur with higher prevalence in CDK1(G2-apoptotic cells) and
PLK1 or microtubule modulators(M-phase apoptotic cells), suggesting
the presence ofpolyploidy for these classes. However, this
observationis inconsistent with G2-M arrest. Staining with
TUNELindicates a high induction of apoptosis, and suggests
thatHoechst dye has higher affinity for unwinding chromatinin
apoptotic cells. DNA stainingwith propidium iodide, aDNA
intercalator that binds with stoichiometry of 1 dyeper 4–5 base
pairs, is not affected by DNA coiling andreveals a single 4N peak
for treatments such as nocoda-zole, vs. the 2 peak population
distribution for Hoechst(Supplementary Fig. 1). As such, apparent
DNA intensityof 8N or >8N, in the absence of other cytological
featuressuch as a high DNAperimeter-to-area ratio (due to
multi-lobed nuclei) or low pHH3 intensity (consistent withAURKB
inhibition), cannot be used to infer polyploidy.
In spite of this, Hoechst is preferred due to signal quench-ing
from other fluorescent channels that occurs withpropidium
iodide.
Summarizing concentration-dependent phenotypesThe classifier
assigns a phenotype to every imaged cell.
A population of cells within a well can be summarized asthe
percentage of cells exhibiting each of the 9 phenotypesreported by
the classifier. Changes in populations along
aconcentration-response curve can be summarized viastacked bar
graphs: for the aurora inhibitor PHA-739358, the classifier reveals
a mixed aurora A/B cellpopulation at the antiproliferation EC50,
which becomesconsistent with AURKB inhibition at intermediate
con-centrations, and exhibits a dominant AURKA profile athigher
concentrations (Fig. 3). This representation isuseful for
elucidating the structure-phenotype relation-ship in lead
optimization programs.
Comparing phenotypes from the cell classifier toaveraged
cytological features
A simple approach for quantifying treatment-inducedphenotypes
consists of averaging cytological featureswithin a well (e.g., the
average DNA content of cells).
Figure 2. Scatter plots showingselected cytological features,
withcells colored according tophenotype from the cell
classifier(left) and fields of view for DNAalone (top right) or all
4 channelscombined (bottom right; DNA,blue; CyclinB1, green; pHH3,
red;a-tubulin, yellow). Representativecells treated with the CDK
inhibitorR-547 at 0.25 mmol/L are identifiedusing circles colored
as indicatedin the scatter plot (seeSupplementary Fig. 4 for
otherinhibitors).
High-Content Imaging of Cell-Cycle Modulators
www.aacrjournals.org Mol Cancer Ther; 10(2) February 2011
247
on June 1, 2021. © 2011 American Association for Cancer
Research. mct.aacrjournals.org Downloaded from
Published OnlineFirst January 7, 2011; DOI:
10.1158/1535-7163.MCT-10-0720
http://mct.aacrjournals.org/
-
To understand how this approach differs from our phe-notype
classifier, we describe every treatment well using3 approaches: (a)
the average value for each of 12 cyto-logical features used alone,
(b) a vector of length 5 con-sisting of the average values of
DNA-related cytologicalfeatures (DNA well-average features), and
(c) a vector oflength 9 indicating the percentage of cells for each
phe-notype reported by the classifier (cell phenotype
profiles).Wells for treatment concentrations below the
antiproli-feration EC50 are discarded to focus on those wells
inwhich cells are responding to treatment. For approach 1, 2wells
are compared by taking the absolute value of thedifference between
wells. For approaches 2 and 3 thatdescribe each well as a vector,
the difference betweenwells is calculated using the Euclidian and
cosine dis-tances (see Materials and Methods). Small differences
ordistances indicate consistent assessment of phenotype.
For the purpose of assessing reproducibility betweenexperiments,
we identified 16 compounds tested in all 3experiments, and set
aside the 6 compounds used formodel calibration. This yielded 159
pairs of wells fromdifferent experiments that contain the same
compound atthe same (or similar) concentration. Because
themeasuresabove have different natural scales, the 3 approaches
forcomparing wells are normalized by converting differ-ences to
Z-scores. A method that yields reproducibleassessments of phenotype
should produce scores withlarge negative values in the left-tail of
the distribution (i.e., much more similar than the average pair of
wells).Some cytological features are poorly reproduced
acrossexperiments, especially total and variation for cyclin B1and
pHH3 intensities (Fig. 4). Cell phenotype profiles aresignificantly
more reproducible (P < 0.0001) whether
Euclidian or cosine distance metrics are used, and areless
sensitive to the distance metric than the DNA well-average profiles
(1-sided t tests assuming unequal var-iance, rejecting the null
hypothesis that the distances aredrawn from the same distribution;
N ¼ 159). This arisesby virtue of recalibration using reference
compounds,and compensates for variation in experimental
conditionssuch as light source intensity, Hoechst and
antibodystaining, signal quenching arising from use of
additionalfluorescent markers, biologist technique, etc. Over
thecourse of 17 experiments in support of internal leadoptimization
efforts, the variation in values used for rulesin the classifier
approaches 1 Z-score unit (after normal-ization of raw data;
seeMaterials andMethods) for DNA-related features, and 2 units for
cyclin B1 and pHH3-related features (Fig. 1). Sources of
experimental variationacross experiments cannot be fully controlled
using nor-malization to DMSO control alone, or normalizationusing a
signal window defined by the negative andpositive control (results
not shown).
In addition, we compared the utility of the cell classifierin
assessing the similarity of phenotypes from 2 treat-ments. The 16
compounds repeated across all experi-ments yielded ca. 4,000 pairs
of wells per experiment;each paired well contains a different
compound at aconcentration above the EC50. Although distances
fromcell phenotype profiles are correlated with those fromDNA
well-average profiles, some treatments appearmore similar using 1
method over the other (Fig. 4). Asan example, the aurora kinase
inhibitor tozasertibappears somewhat similar to the CDK1 inhibitor
R-547using the well-average measure, but is much less similarusing
the cell phenotype profiles (23rd percentile for well
Figure 3. Proportion of cellsclassified into each of 9phenotypes
versus concentrationfor the aurora kinase inhibitorPHA-739359. The
curve indicatesinhibition of proliferationmeasured by counting
cells perfield of view; representativeimages at 3 concentrations
revealphenotypes consistent withlargest cell populations from
theclassifier: at 0.063 mmol/L, there isevidence of multinucleated
cells(aurora B) and prometaphaserosettes (aurora A); at0.125
mmol/L, the dominant cellpopulation is consistent withaurora B,
whereas at higherconcentrations the phenotype isconsistent with
aurora A inhibition.
Sutherland et al.
Mol Cancer Ther; 10(2) February 2011 Molecular Cancer
Therapeutics248
on June 1, 2021. © 2011 American Association for Cancer
Research. mct.aacrjournals.org Downloaded from
Published OnlineFirst January 7, 2011; DOI:
10.1158/1535-7163.MCT-10-0720
http://mct.aacrjournals.org/
-
average vs. 84th percentile for cell phenotype profiles).This
recalls the aphorism "the average cell does not exist"(1), where
the tozasertib average arises from a bimodaldistribution of cells
in prometaphase (small bright nuclei)and polyploidy (very large
diffuse nuclei), and appears
similar to the CDK1 inhibitor with predominantly G2-arrested
cells having nuclei of intermediate size andintensity
(Supplementary Fig. 6). Other examples includethe CDK1 inhibitors
R-547 vs. BMS-265246, where theformer has a higher proportion of
G2-apoptotic cells, and
A
B
Figure 4. Reproducibility and comparison to standard approaches
of population profiles from the cell classifier. A, comparing
cell-level analysis to averagevalues of cytological features for
cells in a given well: 159 pairs of wells are compared, each well
contains the same compound at the same or similarconcentration, but
from different experiments. Because the measures have different
natural scales, a numerical value obtained by comparing 2 wells
isconverted to a Z-score using the mean and standard deviation
obtained from all pairs of wells; only wells with concentrations
above the antiproliferation EC50are analyzed to avoid the dominance
of cells in G1-S. Large negative values denote high reproducibility
of phenotype across experiments (see text). The first 12comparisons
simply take the absolute difference of means for each cytological
feature, where the last 4 describe each well using a vector of
well-average DNAparameters or proportion of cells belonging to each
phenotype and quantify similarity using the cosine and Euclidian
metrics; experiment 1 used 2-folddilutions starting at 5 mmol/L and
experiments 2 to 3 used 2-fold dilutions starting at 2 mmol/L; for
experiments 1 versus 2 to 3, we compared concentrationswithin 20%
of each other; that is, 2.5 mmol/L from experiment 1 versus 2
mmol/L from experiments 2 to 3, etc. B, comparison of Euclidian
distances fromaveraged DNA features versus cell phenotype profiles
for 3,730 pairs of wells above the antiproliferation EC50 in
experiment 3; R
2 ¼ 0.79; selected treatmentcomparisons which appear less
similar by cell population profiles are highlighted.
High-Content Imaging of Cell-Cycle Modulators
www.aacrjournals.org Mol Cancer Ther; 10(2) February 2011
249
on June 1, 2021. © 2011 American Association for Cancer
Research. mct.aacrjournals.org Downloaded from
Published OnlineFirst January 7, 2011; DOI:
10.1158/1535-7163.MCT-10-0720
http://mct.aacrjournals.org/
-
the PLK1 inhibitors HMN-176 vs. BI-2536, where theformer has a
higher fraction of M-phase apoptotic cells.
Cell population responses to chemotherapeuticagents
A large number of chemotherapeutic agents are knownto affect the
G2-M transition of the mitotic cell cycle. Weinterrogated the
connection between the mechanism ofaction of these modulators and
the effect on HCT-116 cellpopulations. The phenotypic profiles were
assessed in 2experiments, and yielded highly consistent results
(Sup-plementary Fig. 7)
The quinoline alkaloids camptothecin and topotecanstabilize the
topoisomerase I-DNA complex, resulting insingle-strand DNA breaks.
Both agents produce cellpopulations arrested in G2 at
concentrations near theantiproliferation EC50, consistent with
their reportedcell-cycle effects (24, 25). However, an increasing
popula-tion of cells in prometaphase is apparent at higher
con-centrations, suggesting cellular effects unrelated
totopoisomerase I inhibition. The alkylating agent mitomy-cin also
results in a large enrichment of cells in G2. Whilewe observed
consistent phenotypes for structurallyrelated camptothecin and
topotecan, the anthracyclineantibiotics doxorubicin and aclarubicin
result in distinctphenotypes, even though both stabilize the
topoisome-rase II-DNA complex. Doxorubicin, at concentrationshigher
than what is necessary for topoisomerase II inhibi-tion, is capable
of inhibiting topoisomerase I and will alsocompete for binding
sites for the various DNA stains andwill therefore elicit multiple
concentration-dependentphenotypic outcomes (unpublished results).
Most cellsfrom aclarubicin treatment have G1-S properties,
possiblythrough more potent inhibition of topoisomerase I.
Asobserved for aclarubicin, the antimetabolite
5-fluoro-20-deoxyuridine (5-FUDR or floxuridine) significantly
inhi-bits proliferation without resulting in a large enrichmentof
cells in G2-M. The classifier is not optimized forcharacterizing
treatments that modulate the G1/S mitoticcell cycle; cells in G1
are difficult to distinguish from thosein S phase using Hoechst
staining and light microscopy,necessitating additional markers such
as 5-ethyl-2-deox-yuridine (EDU) or bromodeoxyuridine (BrdU) or
G1-specific markers such as phosphor retinoblastoma pro-tein 1
(pRB1). The tubulin depolymerizers mebendazole,albendazole,
ciclobendazole, and nocodazole, and thetubulin stabilizer
paclitaxel all produce populationsenriched in prometaphase and
apoptotic cells (15). Theproportion of prometaphase cells increases
with concen-tration, presumably due to a diminishing proportion
ofnonviable cells in the population.
Cell population responses to inhibitors of cell-cyclekinases
Inhibitors of kinases tend to exhibit varying degrees
ofoff-target activity (26), making it difficult to anticipate
thelevel of selectivity from cell-free assays expected to yielda
phenotype consistent with modulation of the intended
target. The cell classifier was used to characterize changesin
cell populations induced by treatment with several G2-M kinase
inhibitors (Fig. 5 for selected inhibitors; Sup-plementary Fig. 8).
Most CDK1 and pan-CDK inhibitorscause enrichment of cells in G2,
consistent with the role ofCDK1 in cell-cycle regulation. Many of
these moleculesinhibit interphase CDKs (CDK2/4/5/6). In
particular,the inhibitors AG-024322, JNJ-7706621, and
aminopurva-lanol have potencies vs. interphase CDKs that are
similaror greater than that for CDK1, yet show a
phenotypeconsistent with CDK1 inhibition. The residual G1-S
cellsmay be nonresponders or G1/S-arrested cells arisingfrom
inhibition of interphase CDKs. For example, theexquisitely
selective CDK4 inhibitor PD-0332991 inhibitsproliferation, and
induces mostly G1-S cells according tothe classifier (Supplementary
Fig. 8). Additional markerssuch as pRB1 staining and/or EDU
labeling are requiredfor effective characterization of compounds
having domi-nant G1/S mechanisms.
In contract to the dominance of the CDK1 phenotypenoted above,
AG-12286, alvocidib, PD-171851, SCH-727965, and SNS-032 potently
inhibit CDK9 in additionto other CDKs, and producemixed populations
of cells inG2, prometaphase, and advanced states of apoptosis.
Therole of CDK9 in transcriptional regulation (27), coupledwith
this observation, suggests that manifestation of theCDK1 phenotype
is distorted byCDK9 activity.However,other inhibitors (e.g.,
AG-024322, BMI-1026, BMS-265246,and R-547) significantly inhibit
CDK9 in vitro yet induce aCDK1 phenotype. The inhibitors
JNJ-7706621, 358788-29-1, and 358789–50-1 (reported by Astrazeneca)
exhibitAURKB activity in enzyme assays, which is evident inthe cell
population profile for the latter despite the role ofthat kinase
beyond the G2-M checkpoint. The relationshipbetween enzyme
inhibition in cells and phenotype iscomplex and not fully
understood from cell-free assays.
We investigated the relationship between aurorakinase activity
and cell phenotype using a number ofinhibitors at our disposal. The
prodrug AZD-1152 andits metabolite differ by a phosphate group used
toimprove solubility of the prodrug; both are selectiveAURKB
inhibitors and induce polyploidy in treatedcells, consistent with
their enzyme activity. As theyare the only selective AURKB
inhibitors we have char-acterized, it is interesting to note that
both retain largerpopulations of G1-S cells than the other aurora
kinaseinhibitors. As cells responding to AURKB inhibitioncontinue
cycling beyond 48 hours, the absence of otherarrest mechanisms may
explain this observation. Theinhibitors tozasertib and CYC116
induce polyploidy asexpected for dual A/B inhibitors, unlike
PHA-739358that exhibits a concentration-dependent change
fromdominant polyploidy (consistent with AURKB) to do-minant
prometaphase arrest (consistent with AURKA).In our hands, MLN-8054
and ENMD-2076 both inhibitAURKBmore potently in biochemical assays,
yet inducecell populations consistent with AURKA at lower
con-centrations and AURKB at higher concentrations. For
Sutherland et al.
Mol Cancer Ther; 10(2) February 2011 Molecular Cancer
Therapeutics250
on June 1, 2021. © 2011 American Association for Cancer
Research. mct.aacrjournals.org Downloaded from
Published OnlineFirst January 7, 2011; DOI:
10.1158/1535-7163.MCT-10-0720
http://mct.aacrjournals.org/
-
Figure 5. Kinase profiling data from biochemical enzyme assays
and cell phenotypes in HCT-116 cells. Right, proportion of cells
classified into eachof 9 phenotypes versus treatment concentration;
the curve indicates inhibition of proliferation measured by
counting cells per field of view. Left and middle,results from
kinase enzyme assay profiling, with insets showing detail for CDK,
aurora, and PLK kinases. Changes in markers from green to red
(andsmall to large) indicate increasing binding affinity, on a
log10 scale. Labeled markers indicate the IC50 in mmol/L (no units)
or % inhibition at 20 mmol/L (valuesfollowed by %); the absence of
labels denote inactive results (i.e., IC50 > 10 mmol/L or %
inhibition < 80 for single point results). See Supplementary
Methodsfor details on enzyme assays. Additional inhibitors are
shown in Supplementary Fig. 8. Human kinome provided courtesy of
Cell Signaling Technology, Inc.www.cellsignal.com.
High-Content Imaging of Cell-Cycle Modulators
www.aacrjournals.org Mol Cancer Ther; 10(2) February 2011
251
on June 1, 2021. © 2011 American Association for Cancer
Research. mct.aacrjournals.org Downloaded from
Published OnlineFirst January 7, 2011; DOI:
10.1158/1535-7163.MCT-10-0720
http://mct.aacrjournals.org/
-
tozasertib, its significant binding affinity for CDK1 isnot
apparent in the induced cell population.
In contrast to the variation in phenotypes observed forCDK and
aurora kinase inhibitors, small molecules tar-geting PLK1 generally
exhibit similar phenotypes con-sisting ofmixed populations of cells
in prometaphase andadvanced apoptosis. A notable exception to the
class isthe Banyu inhibitor 886856-66-2 that exhibits a small
butincreasing population of G2 cells with increasing
concen-tration. HMN-176 is not an ATP-competitive inhibitor,but
interferes with the cellular localization of PLK1 (28).ON-01910 was
originally described as a PLK1 inhibitor,but is now thought to be a
tubulin modulator (22, 23;Supplementary Fig. 7).
Relationship between selectivity in cell-free assaysand
phenotype in HCT-116 cells
The variation in phenotypic response among the kinaseinhibitors
in Table 1 prompted us to examine the con-cordance between enzyme
inhibition profiles and phe-notypic response for a larger number of
internalcompounds exhibiting activity against G2-M kinases.For this
purpose, we selected for analysis compoundshaving antiproliferation
EC50 < 1 mmol/L in HCT-116cells and with fully determined G2-M
kinase enzymeprofiles (IC50s vs. CDK1, AURKA, AURKB and PLK1,or �80
percent inhibition in single concentration testingat 20 mmol/L).
For wells having concentrations above theantiproliferation EC50, we
summarized the proportion ofcells belonging to non-G1-S phenotypes
from the cellclassifier (Supplementary Fig. 9). While selective
AURKAinhibitors induce populations dominated by prometa-phase
arrest and advanced apoptosis, the dual A/Binhibitors have a larger
proportion of multinucleatedcells than AURKB-selective inhibitors.
For CDK inhibi-tors, a decreasing proportion of cells in G2 is
apparent asselectivity vs. the aurora kinases and PLK1
increases.However, the corresponding increase in prometaphaseand
M-phase apoptotic cells may be attributed to CDK9inhibition as
noted above, andmost compounds were nottested vs. other CDKs. For
PLK1 inhibitors, there is nodiscernable change in the dominant
prometaphase andM-phase apoptotic phenotype with increasing
selectivityvs. the CDKs and aurora kinases. This suggests a
domi-nant role for PLK1 enzyme inhibition over other
cell-cycletargets, despite a role for PLK1 in mitosis. A
conventionalview holds that simultaneous inhibition of
targetsinvolved earlier in the cell cycle would manifest
overM-phase arrest. The lack of clear relationships betweenenzyme
inhibition profiles and cell phenotype supportsthe importance of
phenotype determination via high-content imaging to verify that
cell death arises frommodulation of the targeted kinase (or
kinases).
Discussion
The cell classifier described in this work enables
theinvestigation of mechanism of action for cell-cycle mod-
ulators by summarizing cell-level results from high-con-tent
imaging experiments. While an increasing body ofliterature
describes approaches that distill millions ofcellular measurements
for interpretation of phenotype,only a few approaches have explored
the reproducibilityof phenotype across experiments (7, 10). The
approachdescribed in this work has been applied to experimentsdone
over 6 months by 3 different biologists, and sig-nificantly reduces
variability in results that often afflictindustrial application of
immunofluorescence-basedmicroscopy. The methodology is applied
within leadoptimization programs at Lilly to understand changesin
phenotype that arise from structural modification oflead series,
and the extent to which activity at otherkinases translates to
deviation from the desired effect(i.e., obtaining a phenotype
consistent with inhibition of agiven kinase target).
The application of our classifier highlights the preva-lence of
concentration-dependent phenotypes amonginhibitors purported to
inhibit the same kinase. Whilethis work does not evaluate the
effects of RNAi treat-ments, the cell morphologies that we identify
as consis-tent with the intended target are informed frompublished
reports using RNAi (17, 20, 22, 23, 29, 30).We postulate that
departure from the expected pheno-type arises from off-target
effects, rather than variableinhibition of the intended target.
Although the effect cansometimes be rationalized from enzyme
activity profiles(e.g., PHA-739358), there is generally no
discernablequantitative relationship between enzyme selectivityand
phenotype. It is noteworthy that exquisitely selectivekinase
inhibitors studied in this work (PD-0332991 andthe AZD-1152
metabolite) induce the expected pheno-types.
Conceptually, our approach is similar to methods thatextract
classification rules from cells representing distinctphenotypes
identified by review of images (2–4). Suchmethods can overcome
variation in staining intensity, etc.by repeating the
identification of representative cells andrule training in every
experiment. By recalibrating thecell classifier from reference
molecules selected to repre-sent the relevant cell phenotypes, the
need for manualreview of images is substantially reduced. On the
otherhand, the use of reference inhibitors renders the approachless
effective in the identification of novel or unexpectedcell
morphologies (6). The simplicity of a decision tree forcell
classification is appealing, but uses artificial rectan-gular
boundaries in cytological feature space. It can bedifficult to
ascertain whether rarer cell populations areartifacts of
classification; most PLK1 inhibitors appear toinduce small
populations of multinucleated cells, butthese are M-phase apoptotic
cells with irregular shapesand lower DNA staining intensity. In our
hands, the useof mixture models (9) does not resolve the phenotype
ofcells falling between major phenotypic classes, asexpected given
the probabilistic assignment of cells tophenotype classes
(unpublished results). Higher resolu-tion imaging, perhaps with
additional markers, may be
Sutherland et al.
Mol Cancer Ther; 10(2) February 2011 Molecular Cancer
Therapeutics252
on June 1, 2021. © 2011 American Association for Cancer
Research. mct.aacrjournals.org Downloaded from
Published OnlineFirst January 7, 2011; DOI:
10.1158/1535-7163.MCT-10-0720
http://mct.aacrjournals.org/
-
required to fully resolve distinct cell populations (10).
Ascellular imaging technology continues to evolve, theability to
monitor additional fluorescence channels willallow for more
detailed analyses of complex cellularprocesses.The imaging assay
and phenotype classifier presented
in this work are configured for
medium-throughputcharacterization of G2-M modulators, and can be
usedto identify the probable mechanism of action in concertwith
biochemical profiling results. The classifier does notdifferentiate
G1 cells from those in early S phase. Forexample, the CDK inhibitor
R-547 was shown to inhibitRb1 phosphorylation in phase I studies
and inducespopulations of G1- and G2-arrested HCT-116 cells
whenstudied via flow cytometry (31). The cell classifier indi-cates
a dominant population of G2 cells, with someresidual G1-S cells
that could in fact be G1/S-arrested(i.e., blocked cell cycle).
Other markers, such as antibo-dies for pRb1 and/or EDU labeling are
probably neces-sary to allow full characterization of cells in
G1-S.Likewise, cells responding to CDK1 and topoisomeraseinhibitors
(e.g., camptothecin and doxorubicin) are notdifferentiated with the
markers employed in this study,but are readily distinguished via
immunostaining forgH2AX, due to chain breaks induced by
topoisomeraseinhibitors (32).In a similar vein, the classifier does
not differentiate
cells responding to AURKA inhibitors, PLK1 inhibitors,tubulin
stabilizers (i.e., paclitaxel) or tubulin destabili-zers, all of
which directly modulate cytoskeleton functionresulting in
prometaphase arrest. High-resolution micro-scopy of paclitaxel
reveals a metaphase-like arrangementof most chromosomes, with some
lagging at the spindlepole (33). Such features are not discernable
at the resolu-tion used in this assay. At best, simple
intensity-derivedcytological features from a-tubulin staining
provideslight differentiation of mechanistic classes, with
pacli-taxel and PLK1 inhibitors having higher staining inten-sity
compared with tubulin destabilizers (whetherconsidering all cells
or only those in prometaphase/M-apototic states; Supplementary Fig.
5). This observationis consistent with the absence of changes in
microtubulemass at lower compound concentration, despite
theirimpact on microtubule dynamics as determined
viahigh-resolution microscopy (34, 35). In cases where
cross-reactivity is suspected (e.g., potent PLK andAURKA enzyme
activity), high-resolution microscopyof centrosomes and microtubule
topology to detectmonopole spindle assemblies having reduced
amountsof centrosomal g-tubulin (PLK1; refs. 22, 23, 36),
circularmonopole spindles (AURKA; refs 20, 29), or the presenceof
multipolar spindles (tubulin destabilizers; refs 15, 33,37) may be
required to further clarify a compound’smechanism of action.
Higher-resolution microscopycan provide greater cellular detail for
understandingmechanism, but reduces throughput due to longer
scantimes while also posing challenges with image
storage.High-resolution microscopy is also not without its
ambi-guities: unlike the monopolar spindles induced by treat-ment
with PLK1 RNAi or BI-2536, the inhibitor GSK-461364 potently
inhibits the PLK1 enzyme in biochemicaland substrate
phosphorylation assays, yet induces multi-polar spindles
characteristic of tubulin modulators inH460 human lung cancer cells
(38).
Although the cell classifier system described herefocuses on
characterization of cell-cycle modulators,the broad concept of
summarizing cell-level results canbe applied across high-content
screenings and phenoty-pic drug discovery. The automated
classification of phe-notypic screening and structure-activity
relationship datarepresents a solution to what is now the principal
diffi-culty in taking raw data from a high content assay andusing
it to make a statistically robust and reproducibledecision of
phenotypic outcome.
Disclosure of Potential Conflicts of Interest
All the authors have Eli Lilly shares received via 401(k) and
bonusplans.
Acknowledgments
The authors thank the Phenotypic DrugDiscoveryWorkingGroup
andPD2Management for guidance on screen implementation, follow-up,
andgeneral scientific discussion; Mark Marshall and Jake Starling
for helpfuldiscussions and project guidance.
The costs of publication of this article were defrayed in part
by thepayment of page charges. This article must therefore be
hereby markedadvertisement in accordance with 18 U.S.C. Section
1734 solely to indicatethis fact.
Received August 2, 2010; revised November 12, 2010;
acceptedNovember 24, 2010; published OnlineFirst January 7,
2011.
References1. Levsky JM, Singer RH. Gene expression and the myth
of the average
cell. Trends Cell Biol 2003;13:4–6.2. Jones TR, Carpenter AE,
Lamprecht MR, Moffat J, Silver SJ, Grenier
JK, et al. Scoring diverse cellular morphologies in
image-basedscreens with iterative feedback and machine learning.
Proc NatlAcad Sci U S A 2009;106:1826–31.
3. Tao CY, Hoyt J, Feng Y. A support vector machine classifier
forrecognizing mitotic subphases using high-content screening
data.J Biomol Screen 2007;12:490–6.
4. Conrad C, Erfle H, Warnat P, Daigle N, Lorch T, Ellenberg J,
et al.Automatic identification of subcellular phenotypes on human
cellarrays. Genome Res 2004;14:1130–6.
5. Low J, Huang S, Blosser W, Dowless M, Burch J, Neubauer B,et
al. High-content imaging characterization of cell cycle
ther-apeutics through in vitro and in vivo subpopulation analysis.
MolCancer Ther 2008;7:2455–63.
6. Yin Z, Zhou X, Bakal C, Li F, Sun Y, Perrimon N, et al. Using
iterativecluster merging with improved gap statistics to perform
online phe-notype discovery in the context of high-throughput RNAi
screens.BMC Bioinformatics 2008;9:264.
7. Young DW, Bender A, Hoyt J, McWhinnie E, Chirn GW, Tao CY,et
al. Integrating high-content screening and ligand-targetprediction
to identify mechanism of action. Nat Chem Biol 2008;4:59–68.
High-Content Imaging of Cell-Cycle Modulators
www.aacrjournals.org Mol Cancer Ther; 10(2) February 2011
253
on June 1, 2021. © 2011 American Association for Cancer
Research. mct.aacrjournals.org Downloaded from
Published OnlineFirst January 7, 2011; DOI:
10.1158/1535-7163.MCT-10-0720
http://mct.aacrjournals.org/
-
8. Perlman ZE, Slack MD, Feng Y, Mitchison TJ, Wu LF, Altschuler
SJ.Multidimensional drug profiling by automated microscopy.
Science2004;306:1194–8.
9. Slack MD, Martinez ED, Wu LF, Altschuler SJ. Characterizing
hetero-geneous cellular responses to perturbations. Proc Natl Acad
Sci U S A2008;105:19306–11.
10. Loo LH, Lin HJ, Steininger RJ 3rd, Wang Y, Wu LF, Altschuler
SJ. Anapproach for extensibly profiling the molecular states of
cellularsubpopulations. Nat Methods 2009;6:759–65.
11. Loo LH, Wu LF, Altschuler SJ. Image-based multivariate
profiling ofdrug responses from single cells. Nat Methods
2007;4:445–53.
12. Feng Y, Mitchison TJ, Bender A, Young DW, Tallarico JA.
Multi-parameter phenotypic profiling: using cellular effects to
characterizesmall-molecule compounds. Nat Rev Drug Discov
2009;8:567–78.
13. Low J, Stancato L, Lee J, Sutherland JJ. Prioritizing hits
from phe-notypic high-content screens. Curr Opin Drug Discov Devel
2008;11:338–45.
14. Pommier Y. DNA topoisomerase I inhibitors: chemistry,
biology, andinterfacial inhibition. Chem Rev 2009;109:2894–902.
15. Jordan MA, Wilson L. Microtubules as a target for anticancer
drugs.Nat Rev Cancer 2004;4:253–65.
16. Lapenna S, Giordano A. Cell cycle kinases as therapeutic
targets forcancer. Nat Rev Drug Discov 2009;8:547–66.
17. Van Horn RD, Chu S, Fan L, Yin T, Du J, Beckmann R, et al.
CDK1activity is required for mitotic activation of aurora A during
G2-Mtransition of human cells. J Biol Chem 2010;285:21849–57.
18. Barr AR, Gergely F. Aurora-A: the maker and breaker of
spindle poles.J Cell Sci 2007;120:2987–96.
19. Glover DM, Leibowitz MH, McLean DA, Parry H. Mutations in
auroraprevent centrosome separation leading to the formation of
monopolarspindles. Cell 1995;81:95–105.
20. Girdler F, Gascoigne KE, Eyers PA, Hartmuth S, Crafter C,
Foote KM,et al. Validating Aurora B as an anti-cancer drug target.
J Cell Sci2006;119:3664–75.
21. Lane HA, Nigg EA. Antibody microinjection reveals an
essential rolefor human polo-like kinase 1 (Plk1) in the functional
maturation ofmitotic centrosomes. J Cell Biol 1996;135:1701–13.
22. Peters U, Cherian J, Kim JH, Kwok BH, Kapoor TM. Probing
cell-division phenotype space and Polo-like kinase function using
smallmolecules. Nat Chem Biol 2006;2:618–26.
23. Steegmaier M, Hoffmann M, Baum A, Lenart P, Petronczki M,
KrssakM, et al. BI 2536, a potent and selective inhibitor of
polo-like kinase 1,inhibits tumor growth in vivo. Curr Biol
2007;17:316–22.
24. TaronM, Plasencia C, Abad A,Martin C, Guillot M. Cytotoxic
effects oftopotecan combined with various active G2-M-phase
anticancerdrugs in human tumor-derived cell lines. Investigational
new drugs2000;18:139–47.
25. Tsao YP, D’Arpa P, Liu LF. The involvement of active DNA
synthesis incamptothecin-induced G2 arrest: altered regulation of
p34cdc2/cyclinB. Cancer research 1992;52:1823–9.
26. Fabian MA, Biggs WH 3rd, Treiber DK, Atteridge CE, Azimioara
MD,Benedetti MG, et al. A small molecule-kinase interaction map
forclinical kinase inhibitors. Nature Biotechnology
2005;23:329–36.
27. Garriga J, Xie H, Obradovic Z, Grana X. Selective control
ofgene expression by CDK9 in human cells. J Cell Physiol
2010;222:200–8.
28. Tanaka H, Ohshima N, Ikenoya M, Komori K, Katoh F, Hidaka
H.HMN-176, an active metabolite of the synthetic antitumor agent
HMN-214, restores chemosensitivity to multidrug-resistant cells by
target-ing the transcription factor NF-Y. Cancer research
2003;63:6942–7.
29. Manfredi MG, Ecsedy JA, Meetze KA, Balani SK, Burenkova O,
ChenW, et al. Antitumor activity of MLN8054, an orally active
small-molecule inhibitor of Aurora A kinase. Proc Natl Acad Sci U S
A2007;104:4106–11.
30. Taylor S, Peters JM. Polo and Aurora kinases: lessons
derived fromchemical biology. Curr Opin Cell Biol
2008;20:77–84.
31. DePinto W, Chu XJ, Yin X, Smith M, Packman K, Goelzer P, et
al.In vitro and in vivo activity of R547: a potent and selective
cyclin-dependent kinase inhibitor currently in phase I clinical
trials. MolCancer Ther 2006;5:2644–58.
32. Furuta T, Hayward RL, Meng LH, Takemura H, Aune GJ, Bonner
WM,et al. p21CDKN1A allows the repair of replication-mediated
DNAdouble-strand breaks induced by topoisomerase I and is
inactivatedby the checkpoint kinase inhibitor
7-hydroxystaurosporine. Oncogene2006;25:2839–49.
33. Kelling J, Sullivan K, Wilson L, Jordan MA. Suppression of
centromeredynamics by Taxol in living osteosarcoma cells. Cancer
Res 2003;63:2794–801.
34. Jordan MA, Thrower D, Wilson L. Effects of vinblastine,
podophyllo-toxin and nocodazole on mitotic spindles. Implications
for the role ofmicrotubule dynamics in mitosis. J Cell Sci
1992;102:401–16.
35. Jordan MA, Toso RJ, Thrower D, Wilson L. Mechanism of
mitoticblock and inhibition of cell proliferation by taxol at low
concentrations.Proc Natl Acad Sci U S A 1993;90:9552–6.
36. Spankuch-Schmitt B, Bereiter-Hahn J, Kaufmann M, Strebhardt
K.Effect of RNA silencing of polo-like kinase-1 (PLK1) on apoptosis
andspindle formation in human cancer cells. J Natl Cancer Inst
2002;94:1863–77.
37. Okouneva T, Hill BT, Wilson L, Jordan MA. The effects of
vinflunine,vinorelbine, and vinblastine on centromere dynamics. Mol
CancerTher 2003;2:427–36.
38. Lansing TJ, McConnell RT, Duckett DR, Spehar GM, Knick
VB,Hassler DF, et al. In vitro biological activity of a novel
small-moleculeinhibitor of polo-like kinase 1. Mol Cancer Ther
2007;6:450–9.
39. Seong YS, Min C, Li L, Yang JY, Kim SY, Cao X, et al.
Characterizationof a novel cyclin-dependent kinase 1 inhibitor,
BMI-1026. Cancerresearch 2003;63: 7384–91.
40. Misra RN, Xiao H, Rawlins DB, Shan W, Kellar KA, Mulheron
JG, et al.1H-Pyrazolo[3,4-b]pyridine inhibitors of cyclin-dependent
kinases:highly potent 2,6-Difluorophenacyl analogues. Bioorg Med
Chem Lett2003;13:2405–8.
41. Emanuel S, Rugg CA, Gruninger RH, Lin R, Fuentes-Pesquera
A,Connolly PJ, et al. The in vitro and in vivo effects of
JNJ-7706621: adual inhibitor of cyclin-dependent kinases and aurora
kinases. CancerRes 2005;65:9038–46.
42. Huang S, Connolly PJ, Lin R, Emanuel S, Middleton SA.
Synthesis andevaluation of N-acyl sulfonamides as potential
prodrugs of cyclin-dependent kinase inhibitor JNJ-7706621. Bioorg
Med Chem Lett2006;16:3639–41.
43. Barvian M, Boschelli DH, Cossrow J, Dobrusin E, Fattaey A,
Fritsch A,et al. Pyrido[2,3-d]pyrimidin-7-one inhibitors of
cyclin-dependentkinases. J Med Chem 2000;43: 4606–16.
Sutherland et al.
Mol Cancer Ther; 10(2) February 2011 Molecular Cancer
Therapeutics254
on June 1, 2021. © 2011 American Association for Cancer
Research. mct.aacrjournals.org Downloaded from
Published OnlineFirst January 7, 2011; DOI:
10.1158/1535-7163.MCT-10-0720
http://mct.aacrjournals.org/
-
2011;10:242-254. Published OnlineFirst January 7, 2011.Mol
Cancer Ther Jeffrey J. Sutherland, Jonathan Low, Wayne Blosser, et
al. Cell-Cycle ModulatorsMechanism of Action and Phenotypic
Outcomes of A Robust High-Content Imaging Approach for Probing
the
Updated version
10.1158/1535-7163.MCT-10-0720doi:
Access the most recent version of this article at:
Material
Supplementary
http://mct.aacrjournals.org/content/suppl/2011/01/07/1535-7163.MCT-10-0720.DC1
Access the most recent supplemental material at:
Cited articles
http://mct.aacrjournals.org/content/10/2/242.full#ref-list-1
This article cites 43 articles, 20 of which you can access for
free at:
E-mail alerts related to this article or journal.Sign up to
receive free email-alerts
SubscriptionsReprints and
[email protected] at
To order reprints of this article or to subscribe to the
journal, contact the AACR Publications
Permissions
Rightslink site. (CCC)Click on "Request Permissions" which will
take you to the Copyright Clearance Center's
.http://mct.aacrjournals.org/content/10/2/242To request
permission to re-use all or part of this article, use this link
on June 1, 2021. © 2011 American Association for Cancer
Research. mct.aacrjournals.org Downloaded from
Published OnlineFirst January 7, 2011; DOI:
10.1158/1535-7163.MCT-10-0720
http://mct.aacrjournals.org/lookup/doi/10.1158/1535-7163.MCT-10-0720http://mct.aacrjournals.org/content/suppl/2011/01/07/1535-7163.MCT-10-0720.DC1http://mct.aacrjournals.org/content/10/2/242.full#ref-list-1http://mct.aacrjournals.org/cgi/alertsmailto:[email protected]://mct.aacrjournals.org/content/10/2/242http://mct.aacrjournals.org/
/ColorImageDict > /JPEG2000ColorACSImageDict >
/JPEG2000ColorImageDict > /AntiAliasGrayImages false
/CropGrayImages false /GrayImageMinResolution 200
/GrayImageMinResolutionPolicy /Warning /DownsampleGrayImages true
/GrayImageDownsampleType /Bicubic /GrayImageResolution 150
/GrayImageDepth -1 /GrayImageMinDownsampleDepth 2
/GrayImageDownsampleThreshold 1.50000 /EncodeGrayImages true
/GrayImageFilter /DCTEncode /AutoFilterGrayImages true
/GrayImageAutoFilterStrategy /JPEG /GrayACSImageDict >
/GrayImageDict > /JPEG2000GrayACSImageDict >
/JPEG2000GrayImageDict > /AntiAliasMonoImages false
/CropMonoImages false /MonoImageMinResolution 600
/MonoImageMinResolutionPolicy /Warning /DownsampleMonoImages true
/MonoImageDownsampleType /Average /MonoImageResolution 600
/MonoImageDepth -1 /MonoImageDownsampleThreshold 1.50000
/EncodeMonoImages true /MonoImageFilter /CCITTFaxEncode
/MonoImageDict > /AllowPSXObjects false /CheckCompliance [ /None
] /PDFX1aCheck false /PDFX3Check false /PDFXCompliantPDFOnly false
/PDFXNoTrimBoxError true /PDFXTrimBoxToMediaBoxOffset [ 0.00000
0.00000 0.00000 0.00000 ] /PDFXSetBleedBoxToMediaBox true
/PDFXBleedBoxToTrimBoxOffset [ 0.00000 0.00000 0.00000 0.00000 ]
/PDFXOutputIntentProfile (None) /PDFXOutputConditionIdentifier ()
/PDFXOutputCondition () /PDFXRegistryName () /PDFXTrapped
/False
/CreateJDFFile false /Description > /Namespace [ (Adobe)
(Common) (1.0) ] /OtherNamespaces [ > /FormElements false
/GenerateStructure false /IncludeBookmarks false /IncludeHyperlinks
false /IncludeInteractive false /IncludeLayers false
/IncludeProfiles false /MarksOffset 18 /MarksWeight 0.250000
/MultimediaHandling /UseObjectSettings /Namespace [ (Adobe)
(CreativeSuite) (2.0) ] /PDFXOutputIntentProfileSelector /NA
/PageMarksFile /RomanDefault /PreserveEditing true
/UntaggedCMYKHandling /LeaveUntagged /UntaggedRGBHandling
/LeaveUntagged /UseDocumentBleed false >> > ]>>
setdistillerparams> setpagedevice