A Robust High-Content Imaging Approach for Probing the Mechanism of Action and Phenotypic Outcomes of Cell-Cycle Modulators

Published OnlineFirst January 7, 2011; DOI: 10.1158/1535-7163.MCT-10-0720

Molecular Cancer Therapeutic Discovery Therapeutics

Jeffrey J. Sutherland1, Jonathan Low2, Wayne Blosser2, Michele Dowless2, Thomas A. Engler3, and Louis F. Stancato2

Abstract High-content screening is increasingly used to elucidate changes in cellular biology arising from treatment 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 perturbations. Mol Cancer Ther; 10(2); 242–54. 2011 AACR.

Introduction Chemical modulation of the mitotic cell cycle has proven to be effective in treating cancer. A number of The characterization of cell populations with immuno- approved chemotherapeutic agents disrupt DNA repli- fluorescence microscopy, or high-content screening, cation [e.g., the topoisomerase inhibitors (14) topotecan allows a detailed understanding of the effect of small and camptothecin] or microtubule dynamics (15; e.g., molecule modulators on the mitotic cell cycle. By quan- the tubulin modulators paclitaxel and vinca alkaloids). tifying the intensity, localization, and morphology of 4 to The mitotic cell cycle is regulated by many kinases, and 5 markers in individual cells, such experiments typically the search for small molecule inhibitors has produced a produce on the order of 105 to 106 observations per number of agents in preclinical and clinical development. treatment. In recognition of the heterogeneity of cell The roles of kinases such as CDK1, AURKA, AURKB, populations (1), several methods have been proposed and PLK1 in the G2-M checkpoint are well-established for analyzing treatment-induced perturbations by cellu- (16). The CDK1-cyclin B complex regulates entry into lar imaging. Such approaches assign a phenotype to mitosis; loss of CDK1 function results in arrest at the individual cells, using rules to emulate their classification G2-M boundary and enrichment of cell populations by biologists (2–4), or by applying nonsupervised (5–9) or having large nuclei with 4N DNA present as diffuse supervised (10, 11) multivariate analysis of cytological chromatin. Because of their role at the G2-M checkpoint, features. Methods for analysis of high-content imaging inhibition of other G2-M kinases in addition to CDK1 is experiments have been reviewed elsewhere (12, 13). expected to result in a phenotype consistent with selective CDK1 inhibition (17). The kinase AURKA is involved in centrosome regulation, and its inhibition manifests itself via enrichment of cells in prometaphase. AURKA pro- Authors' Affiliation: 1Lilly Research Labs Information Technology, 2Can- cer Biology & Patient Tailoring, and 3Discovery Chemistry Research and motes bipolar spindle assembly (18), and mutations in this Technology, Eli Lilly and Company, Indianapolis, Indiana kinase prevent centrosome separation leading to the for- Note: Supplementary material for this article is available at Molecular mation of monopolar spindles (19). In addition to its role Cancer Therapeutics Online (http://mct.aacrjournals.org/). in cytokinesis, AURKB activates the spindle-assembly Corresponding Author: Jeffrey J. Sutherland, Eli Lilly and Company, Lilly checkpoint, and manifestation of AURKA inhibition Corporate Center, Indianapolis, IN, 46285. Phone: 317-655-0833; Fax: 317-276-6545. E-mail: [email protected]; or Louis F. Stancato, Can- requires a functional spindle-assembly checkpoint. For cer Biology & Patient Tailoring, Indianapolis, IN, 46285. E-mail: this reason, dual aurora A/B inhibitors are expected to [email protected] yield a phenotype consistent with selective AURKB inhi- doi: 10.1158/1535-7163.MCT-10-0720 bition (20). Finally, PLK1 functions in spindle formation, 2011 American Association for Cancer Research. chromosome segregation and cytokinesis, the inhibition

242 Mol Cancer Ther; 10(2) February 2011

High-Content Imaging of Cell-Cycle Modulators

of which results in failure to establish a bipolar spindle in a per-plate basis). The intensity distributions for cells in 8 prometaphase (21). positive control wells containing 0.2 mmol/L of nocoda- Protein kinases have a high degree of structural homol- zole are visually assessed for each plate and channel to ogy, and activity of small molecule inhibitors against verify for consistency across plates after feature scaling proteins other than the intended target (i.e., off-target (Supplementary Fig. 1). This method of normalization is activity) is frequently observed in cell-free assays. To adequate to control for plate-to-plate variability within an understand the relationship between selectivity and experiment. All further analysis uses scaled values of effects on cell populations, we developed a classifier of cytological features. cellular phenotype in HCT-116 cells (a cell line for color- ectal carcinoma), and applied it to kinase and nonkinase Quantifying antiproliferative effects of compound modulators of the cell cycle. The approach yields a treatments highly-reproducible assessment of changes in cell popu- For each well, the cell density is calculated by count- lations induced by different treatments or different con- ing the number of objects (cells) per field of view, and centrations of the same treatment, and highlights the averaging across all fields for a given well. For a treat- limited utility of kinase selectivity panels in selecting ment compound, cell density is converted to a percent- inhibitors that exhibit the desired phenotype. age relative to the plate-averaged cell density from DMSO treatment (i.e., 100% corresponds to the average Materials and Methods cell density for DMSO treatment). Logistic regression curvefitsweredoneusingTIBCOSpotfire(Version2.1; Cell-cycle modulators and high-content imaging TIBCO Software, Inc.), and the concentration at which Various inhibitors of proteins involved in cell-cycle the curve crosses 50% is reported as the EC50 of the regulation were selected from the Lilly corporate collec- compound. tion (Table 1). Where available, compounds were pur- chased from commercial vendors. Several reported Quantifying cell phenotypes induced by compound kinase inhibitors not available for purchase were synthe- treatments sized internally, using synthetic schemes described in the A set of 8 reference compounds of reported mechanism public domain. All compounds have 95% purity. were selected for the purpose of classifying cells by Experiments were done by 3 biologists over a 6-month phenotype (designated in bold type in Table 1). These period, to study phenotypes induced by cell-cycle mod- were the CDK inhibitors AG-024322 and R-547, the aur- ulators. Experiments 2 and 3 were designed to specifi- ora kinase inhibitors AZD-1152 and tozasertib, the PLK1 cally probe reproducibility of phenotypes, whereas inhibitor BI-2536, and the microtubule modulators ON- experiment 1 was designed to characterize a large collec- 01910, nocodazole and paclitaxel. ON-01910 was origin- tion of kinase inhibitors. HCT-116 cells were plated onto ally reported as a non-ATP competitive PLK1 inhibitor; in 96-well dishes, treated with compounds in 10-point con- subsequent reports it has been found to not inhibit PLK1 centration curves, and imaged on the Arrayscan VTI biochemically and generates a cell phenotype consistent platform (see Supplementary Methods). Cytological fea- with microtubule modulation (22, 23). The compounds tures of cells were captured with the Target Activation were selected to represent a diversity of phenotypes bio-application bundled with the imaging instrument. observed for G2-M modulators. Visual analysis of images The selected cytological features quantify a number of reveals cell populations consistent with the mechanism of important changes in cells undergoing mitosis (Supple- action of the compounds at all concentrations above the mentary Table 1). Objects are defined from nuclei identi- antiproliferation EC50, allowing those wells to be pooled fication, and mostly correspond to individual cells; for the purpose of training a classifier. Cells in these wells clusters of nuclei from treatments causing polyploidy are described by the cytological features in Supplemen- are captured as 1 object with 8N DNA content and tary Table 1, and assigned the class label of the treatment daughter nuclei in anaphase are captured as 2 objects. compound. To our knowledge, there are no cell-cycle modulators that arrest cells in metaphase or anaphase. To Normalization of cellular features train the classifier in the identification of these states, Numerical values of features from the instrument soft- images for DMSO-treated cells in experiment 3 were ware are log2 transformed to increase the normality of reviewed and 30 cells of each type identified and distribution across cell populations (base 2 is convenient labeled accordingly. The total number of cells across for counting doublings of intensity, e.g., DNA content of pooled wells used for training the classifier was 63,575, 2N, 4N, 8N, etc.). To account for plate-to-plate variation 58,298, and 78,552 for experiments 1 to 3. in cytological features (i.e., changes arising from variation The reference compounds were used to develop a in antibody staining intensity, incubator conditions, etc.), classifier of cells for each experiment (i.e., 3 experiments, individual values for each feature are converted to Z- 3 classifiers). The classifier was developed with the scores, using the mean and standard deviation obtained recursive partitioning algorithm in Jmp (Version 7.0.2, by pooling dimethyl sulfoxide (DMSO)-treated cells from SAS Institute, Inc), using the compound class label as 8 negative control wells (i.e., normalization of features on response variable and cytological features as factors. The

www.aacrjournals.org Mol Cancer Ther; 10(2) February 2011 243

Sutherland et al.

Table 1. Compounds characterized by high-content imaging in HCT-116

a c Name EC50, mmol/L Dominant phenotype

Experimentb 1 Experiment 2 Experiment 3

Aurora inhibitors AZD1152 0.026 0.023 0.019 Endo AZD1152 metabolite 0.034 0.025 0.030 Endo CYC116 0.649 Endo ENMD-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 inhibitors

AG-024322 0.173 0.116 0.102 G2 AG-12286 (WO 9921845 A2 223784-75-6) 0.249 G2 þ proM Alvocidib (flavopiridol) 0.153 0.256 0.266 G2 þ proM BMI-1026 (39) 0.030 G2 BMS-265246 (40) 0.293 0.492 0.463 G2 JNJ-7706621 (41, 42) 0.524 G2 PD-171851 (43) 0.570 G2 þ proM PD-0332991 >2G1-S Aminopurvalanol 2.500 G2 R-547 0.077 0.144 0.104 G2 SCH-727965 0.018 0.008 0.010 G2 þ proM SNS-032 0.074 0.119 0.128 G2 þ proM WO 2001064655 A1 358788-29-1 0.292 G2 þ proM WO 2001064656 A1 358789-50-1 0.250 G2 þ endo PLK1 inhibitors BI-2536 0.009 0.008 0.007 proM þ M-apopt GSK-461364 0.012 0.014 0.012 proM þ M-apopt HMN-176 (28) 0.164 0.260 0.216 proM þ M-apopt

WO 2006049339 A1 886856-66-2 0.101 0.072 0.070 G2 þ proM þ M-apopt WO 2006066172 A1 893440-87-4 1.861 proM þ M-apopt Wee1 inhibitor US 2007254892 A1 955365–24-9 2.340 proM þ M-apopt DNA intercalators; Topoisomerase inhibitors

Aclarubicin 0.238 0.255 G1-S Camptothecin 0.011 0.008 G2#,G2þproM" Doxorubicin 0.254 0.247 G2 Topotecan 0.020 0.033 G2#,G2 þ proM" DNA synthesis inhibitors 0 5-Fluoro-2 -deoxyuridine (floxuridine) >2 >2G1-S Herboxidiene 0.010 0.011 G2 þ proM Illudin S 0.015 0.019 G1-S Mitomycin 0.187 0.253 G2 Microtubule modulators ON-01910 0.100 0.052 0.025 proM þ M-apopt Albendazole 0.448 0.305 proM þ M-apopt Ciclobendazole 0.547 0.691 proM þ M-apopt Mebendazole 0.318 0.483 proM þ M-apopt Nocodazole 0.055 0.072 proM þ M-apopt Paclitaxel 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; kinase inhibitors with no reference given are described at http://www.clinicaltrials.gov; inhibitors described in patents, but not the journal literature, 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 higher concentrations, respectively.

244 Mol Cancer Ther; 10(2) February 2011 Molecular Cancer Therapeutics

High-Content Imaging of Cell-Cycle Modulators

maximum significance rule was used for selecting splits. To assess the reproducibility of imaging experiments, Recursive partitioning is a greedy approach, selecting the 29 compounds were tested in 2 or 3 separate experiments, best feature for each split in an incremental manner done by different biologists over a 6-month period. Most regardless of its appeal from a biological perspective. compounds inhibit cell proliferation at a concentration of To emulate the manner in which biologists analyze cells, 1 mmol/L or less (Table 1). Using the 58 pairs of defined the first splits are obtained by visual binning of DNA EC50s for the same compound (i.e., EC50s not prefixed intensity into 2N, 4N, 8N, and >8N categories. This is with the symbol ">"), the fold difference between experi- done manually via histogram plots in TIBCO Spotfire, ments ranges from 1.0 to 3.9, with an average of 1.5. The allowing the boundaries between 2N, 4N, etc. to change assessment of antiproliferative properties of treatments, from experiment to experiment (Supplementary Fig. 2). measured by quantifying changes in cell density (i.e., cell For experiment 3, the subsequent splits used features and count per field of view), is highly reproducible across rules selected by the recursive partitioning algorithm. experiments. The terminal nodes are assigned names consistent with the values of cytological features and review of images. Classification of cell phenotypes from For the other experiments, the features selected in experi- immunofluorescence microscopy ment 3 were retained for each split, but the split value was A classifier of cell phenotypes was developed using 8 allowed to change. The rules for classifying metaphase reference compounds, selected to represent a diversity of and anaphase cells are not updated in experiments 1 to 2 mechanisms among G2-M modulators (see the Methods because of their rarity in the study of cell-cycle modula- section). A visual assessment of images from treatment tors, and the need to identify representative cells by concentrations above the antiproliferation EC50 reveals review of images. The complete decision tree as depicted cell populations consistent with published reports on in Jmp is shown in Supplementary Figure 3. standard tubulin modulators (15) and RNAi approaches for kinase targets (17, 20, 23). Wells containing a reference Comparing treatment wells compound at concentrations above the antiproliferation For the purpose of contrasting the cell classifier to other EC50 were pooled, and cells described using cytological approaches, it is necessary to quantitatively assess the features (e.g., nuclear area, cyclin B1 intensity, etc.) listed similarity of phenotypes observed under different treat- in Supplementary Table 1. The classifier begins with ment conditions (i.e., comparing cell phenotypes between assessment of DNA content (2N, 4N, 8N, >8N) using wells). The phenotype profile for a well is represented as a boundaries defined by visual analysis of Hoechst DNA vector of length 9, indicating the proportion of cells staining intensity (Supplementary Fig. 2), and creates a belonging to each phenotype. For comparison, a vector decision tree by applying recursive partitioning using of length 5 represents the well-averaged values for 5 DNA- cytological features as factors and the compound related features (excluding ObjectVarIntenDNA due to its mechanism of action as response variable. The rules that high correlation with ObjectAvgIntenDNA; Pearson r > constitute the decision tree are selected incrementally in a 0.9 for the 3 experiments). We employed the Euclidian and manner that distinguishes cells treated with compounds cosine distance metrics for comparing pairs of vectors: having different mechanisms. A cell’s phenotype is defined by the terminal node into which it falls p ¼ p ; p ; ...; p ; q ¼ q ; q ; ...; q 1 2 n 1 2 n (Fig. 1). As such, each cell is assigned 1 of 9 possible sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi n phenotypes, including the phenotype "other" that corre- P 2 dis tan ce; Euclidian ¼ ðÞpi qi sponds mostly to cell debris. The model is recalibrated in i¼1 each experiment by analysis of cells treated with refer- Pn ence compounds, without the need to examine and iden- piqi dis ce; ine ¼ rffiffiffiffiffiffiffiffiffii¼1 r ffiffiffiffiffiffiffiffiffi tify cells for training as required with other approaches. tan cos Pn Pn 2 2 For the CDK1 inhibitors AG-024322 and R-547, the pi qi i¼1 i¼1 dominant cell populations are G2-arrested cells having 4N DNA content, large round nuclei, and low DNA intensity (i.e., diffuse chromatin). By contrast, treatment Results with the PLK1 inhibitor BI-2536 results mostly in cells characteristic of prometaphase arrest with 4N DNA con- Antiproliferation effects of cell-cycle inhibitors tent and high DNA intensity (i.e., condensed chromatin). A total of 41 cell-cycle modulators were characterized The aurora kinase inhibitors AZD-1152 and tozasertib via high-content imaging of HCT-116 cells. Cells were induce polyploidy (via endoreduplication), consistent imaged on the Arrayscan VTI platform, quantifying with the dominance of the AURKB phenotype over intensity, localization and/or morphology of Hoechst AURKA. With a 48-hour incubation (allowing 2 cell stain (DNA), terminal deoxynucleotidyl transferase number doublings), the dominant population should dUTP nick end labeling (TUNEL; a marker for apoptosis), consist of cells with 8N DNA content; smaller popula- and antibodies for cyclin B1, phospho-histone H3 tions with 4N and >8N DNA content arise from misseg- (pHH3), and a-tubulin (see Supplementary Methods). mentation of nuclei clusters and mostly represent

www.aacrjournals.org Mol Cancer Ther; 10(2) February 2011 245

Sutherland et al.

Figure 1. Decision tree used for classifying cells according to cytological features (described in Supplementary Table 1). DNA content is defined by manual binning of ObjectTotalIntenDNA (left rectangles; see Supplementary Fig. 2). Other splits are obtained from recursive partitioning in Jmp, with those in gray defined by partitioning cells treated with 8 calibration compounds, using the compound mechanism of action as class label. The splits in black are

defined by partitioning examples of cells in G1, metaphase, and anaphase; approximately 30 metaphase and 30 anaphase cells were identified by manual 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-apoptotic and M-apoptotic brightest are combined, since they arise from staining variation with Hoechst 33258 in apoptotic cells (see text). With the exceptionofthe nodes in black, numerical values used in the decision tree for each cytological feature are adjusted by recalibrating the model in each experiment; the notation 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 for separating 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 with compounds from lead optimization programs (0.95).

artifacts of image analysis (some >8N cells are present). B1 intensity for all phenotypes, and higher pHH3 inten- Representative images from each mechanistic class are sity for prometaphase and metaphase cells. TUNEL stain- shown in Figure 2 and Supplementary Figure 4. ing, a measure of apoptosis through DNA end-labeling The classifier makes significant use of features derived following DNA fragmentation, shows greater differentia- from DNA staining (Fig. 1). In characterizing G2-M mod- tion across mechanistic classes for cells of a given phe- ulators, other markers such as cyclin B1 and pHH3 are notype: cells treated with CDK1 and PLK1 inhibitors frequently examined in conjunction with DNA content. have high induction of apoptosis compared to DMSO While the proportion of cells for a given phenotype varies or aurora inhibitor treatments. Most aurora inhibitors widely across mechanistic classes, cyclin B1 and pHH3 induce cytokinesis defects, but cells continue cycling staining intensities of cells for a given phenotype are beyond 48 hours. The intensity of a-tubulin immunos- similar across mechanistic classes (Supplementary taining from treatment with paclitaxel, a microtubule Fig. 5), except for DMSO-treated cells having lower cyclin stabilizer, is increased compared to destabilizers such

246 Mol Cancer Ther; 10(2) February 2011 Molecular Cancer Therapeutics

High-Content Imaging of Cell-Cycle Modulators

Figure 2. Scatter plots showing selected cytological features, with cells colored according to phenotype from the cell classifier (left) and fields of view for DNA alone (top right) or all 4 channels combined (bottom right; DNA, blue; CyclinB1, green; pHH3, red; a-tubulin, yellow). Representative cells treated with the CDK inhibitor R-547 at 0.25 mmol/L are identified using circles colored as indicated in the scatter plot (see Supplementary Fig. 4 for other inhibitors).

as nocodazole, allowing some degree of differentiation; In spite of this, Hoechst is preferred due to signal quench- all mechanistic classes result in higher tubulin intensity ing from other fluorescent channels that occurs with than DMSO treatment. propidium iodide. In addition to phenotypes commonly observed in non- treated cells (e.g., G2, prometaphase, etc.), the classifier Summarizing concentration-dependent phenotypes identifies "apoptotic" phenotypes for G2 and M states The classifier assigns a phenotype to every imaged cell. (Fig. 1). These cells have apparent 8N or >8N DNA A population of cells within a well can be summarized as content, and occur with higher prevalence in CDK1 the percentage of cells exhibiting each of the 9 phenotypes (G2-apoptotic cells) and PLK1 or microtubule modulators reported by the classifier. Changes in populations along a (M-phase apoptotic cells), suggesting the presence of concentration-response curve can be summarized via polyploidy for these classes. However, this observation stacked bar graphs: for the aurora inhibitor PHA- is inconsistent with G2-M arrest. Staining with TUNEL 739358, the classifier reveals a mixed aurora A/B cell indicates a high induction of apoptosis, and suggests that population at the antiproliferation EC50, which becomes Hoechst dye has higher affinity for unwinding chromatin consistent with AURKB inhibition at intermediate con- in apoptotic cells. DNA staining with propidium iodide, a centrations, and exhibits a dominant AURKA profile at DNA intercalator that binds with stoichiometry of 1 dye higher concentrations (Fig. 3). This representation is per 4–5 base pairs, is not affected by DNA coiling and useful for elucidating the structure-phenotype relation- reveals a single 4N peak for treatments such as nocoda- ship in lead optimization programs. zole, vs. the 2 peak population distribution for Hoechst (Supplementary Fig. 1). As such, apparent DNA intensity Comparing phenotypes from the cell classifier to of 8N or >8N, in the absence of other cytological features averaged cytological features such as a high DNA perimeter-to-area ratio (due to multi- A simple approach for quantifying treatment-induced lobed nuclei) or low pHH3 intensity (consistent with phenotypes consists of averaging cytological features AURKB inhibition), cannot be used to infer polyploidy. within a well (e.g., the average DNA content of cells).

www.aacrjournals.org Mol Cancer Ther; 10(2) February 2011 247

Sutherland et al.

Figure 3. Proportion of cells classified into each of 9 phenotypes versus concentration for the aurora kinase inhibitor PHA-739359. The curve indicates inhibition of proliferation measured by counting cells per field of view; representative images at 3 concentrations reveal phenotypes consistent with largest cell populations from the classifier: at 0.063 mmol/L, there is evidence of multinucleated cells (aurora B) and prometaphase rosettes (aurora A); at 0.125 mmol/L, the dominant cell population is consistent with aurora B, whereas at higher concentrations the phenotype is consistent with aurora A inhibition.

To understand how this approach differs from our phe- Euclidian or cosine distance metrics are used, and are notype classifier, we describe every treatment well using less sensitive to the distance metric than the DNA well- 3 approaches: (a) the average value for each of 12 cyto- average profiles (1-sided t tests assuming unequal var- logical features used alone, (b) a vector of length 5 con- iance, rejecting the null hypothesis that the distances are sisting of the average values of DNA-related cytological drawn from the same distribution; N ¼ 159). This arises features (DNA well-average features), and (c) a vector of by virtue of recalibration using reference compounds, length 9 indicating the percentage of cells for each phe- and compensates for variation in experimental conditions notype reported by the classifier (cell phenotype profiles). such as light source intensity, Hoechst and antibody Wells for treatment concentrations below the antiproli- staining, signal quenching arising from use of additional feration EC50 are discarded to focus on those wells in fluorescent markers, biologist technique, etc. Over the which cells are responding to treatment. For approach 1, 2 course of 17 experiments in support of internal lead wells are compared by taking the absolute value of the optimization efforts, the variation in values used for rules difference between wells. For approaches 2 and 3 that in the classifier approaches 1 Z-score unit (after normal- describe each well as a vector, the difference between ization of raw data; see Materials and Methods) for DNA- wells is calculated using the Euclidian and cosine dis- related features, and 2 units for cyclin B1 and pHH3- tances (see Materials and Methods). Small differences or related features (Fig. 1). Sources of experimental variation distances indicate consistent assessment of phenotype. across experiments cannot be fully controlled using nor- For the purpose of assessing reproducibility between malization to DMSO control alone, or normalization experiments, we identified 16 compounds tested in all 3 using a signal window defined by the negative and experiments, and set aside the 6 compounds used for positive control (results not shown). model calibration. This yielded 159 pairs of wells from In addition, we compared the utility of the cell classifier different experiments that contain the same compound at in assessing the similarity of phenotypes from 2 treat- the same (or similar) concentration. Because the measures ments. The 16 compounds repeated across all experi- above have different natural scales, the 3 approaches for ments yielded ca. 4,000 pairs of wells per experiment; comparing wells are normalized by converting differ- each paired well contains a different compound at a ences to Z-scores. A method that yields reproducible concentration above the EC50. Although distances from assessments of phenotype should produce scores with cell phenotype profiles are correlated with those from large negative values in the left-tail of the distribution (i. DNA well-average profiles, some treatments appear e., much more similar than the average pair of wells). more similar using 1 method over the other (Fig. 4). As Some cytological features are poorly reproduced across an example, the aurora kinase inhibitor tozasertib experiments, especially total and variation for cyclin B1 appears somewhat similar to the CDK1 inhibitor R-547 and pHH3 intensities (Fig. 4). Cell phenotype profiles are using the well-average measure, but is much less similar significantly more reproducible (P < 0.0001) whether using the cell phenotype profiles (23rd percentile for well

248 Mol Cancer Ther; 10(2) February 2011 Molecular Cancer Therapeutics

High-Content Imaging of Cell-Cycle Modulators

Figure 4. Reproducibility and comparison to standard approaches of population profiles from the cell classifier. A, comparing cell-level analysis to average values 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 similar concentration, but from different experiments. Because the measures have different natural scales, a numerical value obtained by comparing 2 wells is converted to a Z-score using the mean and standard deviation obtained from all pairs of wells; only wells with concentrations above the antiproliferation EC50 are analyzed to avoid the dominance of cells in G1-S. Large negative values denote high reproducibility of phenotype across experiments (see text). The first 12 comparisons simply take the absolute difference of means for each cytological feature, where the last 4 describe each well using a vector of well-average DNA parameters or proportion of cells belonging to each phenotype and quantify similarity using the cosine and Euclidian metrics; experiment 1 used 2-fold dilutions 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 concentrations within 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 from 2 averaged DNA features versus cell phenotype profiles for 3,730 pairs of wells above the antiproliferation EC50 in experiment 3; R ¼ 0.79; selected treatment comparisons which appear less similar by cell population profiles are highlighted.

average vs. 84th percentile for cell phenotype profiles). similar to the CDK1 inhibitor with predominantly G2- This recalls the aphorism "the average cell does not exist" arrested cells having nuclei of intermediate size and (1), where the tozasertib average arises from a bimodal intensity (Supplementary Fig. 6). Other examples include distribution of cells in prometaphase (small bright nuclei) the CDK1 inhibitors R-547 vs. BMS-265246, where the and polyploidy (very large diffuse nuclei), and appears former has a higher proportion of G2-apoptotic cells, and

www.aacrjournals.org Mol Cancer Ther; 10(2) February 2011 249

Sutherland et al.

the PLK1 inhibitors HMN-176 vs. BI-2536, where the target. The cell classifier was used to characterize changes former has a higher fraction of M-phase apoptotic cells. in cell populations induced by treatment with several G2- M kinase inhibitors (Fig. 5 for selected inhibitors; Sup- Cell population responses to chemotherapeutic plementary Fig. 8). Most CDK1 and pan-CDK inhibitors agents cause enrichment of cells in G2, consistent with the role of A large number of chemotherapeutic agents are known CDK1 in cell-cycle regulation. Many of these molecules to affect the G2-M transition of the mitotic cell cycle. We inhibit interphase CDKs (CDK2/4/5/6). In particular, interrogated the connection between the mechanism of the inhibitors AG-024322, JNJ-7706621, and aminopurva- action of these modulators and the effect on HCT-116 cell lanol have potencies vs. interphase CDKs that are similar populations. The phenotypic profiles were assessed in 2 or greater than that for CDK1, yet show a phenotype experiments, and yielded highly consistent results (Sup- consistent with CDK1 inhibition. The residual G1-S cells plementary Fig. 7) may be nonresponders or G1/S-arrested cells arising The quinoline alkaloids camptothecin and topotecan from inhibition of interphase CDKs. For example, the stabilize the topoisomerase I-DNA complex, resulting in exquisitely selective CDK4 inhibitor PD-0332991 inhibits single-strand DNA breaks. Both agents produce cell proliferation, and induces mostly G1-S cells according to populations arrested in G2 at concentrations near the the classifier (Supplementary Fig. 8). Additional markers antiproliferation EC50, consistent with their reported such as pRB1 staining and/or EDU labeling are required cell-cycle effects (24, 25). However, an increasing popula- for effective characterization of compounds having domi- tion of cells in prometaphase is apparent at higher con- nant G1/S mechanisms. centrations, suggesting cellular effects unrelated to In contract to the dominance of the CDK1 phenotype topoisomerase I inhibition. The alkylating agent mitomy- noted above, AG-12286, alvocidib, PD-171851, SCH- cin also results in a large enrichment of cells in G2. While 727965, and SNS-032 potently inhibit CDK9 in addition we observed consistent phenotypes for structurally to other CDKs, and produce mixed populations of cells in related camptothecin and topotecan, the anthracycline G2, prometaphase, and advanced states of apoptosis. The antibiotics doxorubicin and aclarubicin result in distinct role of CDK9 in transcriptional regulation (27), coupled phenotypes, even though both stabilize the topoisome- with this observation, suggests that manifestation of the rase II-DNA complex. Doxorubicin, at concentrations CDK1 phenotype is distorted by CDK9 activity. However, higher than what is necessary for topoisomerase II inhibi- other inhibitors (e.g., AG-024322, BMI-1026, BMS-265246, tion, is capable of inhibiting topoisomerase I and will also and R-547) significantly inhibit CDK9 in vitro yet induce a compete for binding sites for the various DNA stains and CDK1 phenotype. The inhibitors JNJ-7706621, 358788-29- will therefore elicit multiple concentration-dependent 1, and 358789–50-1 (reported by Astrazeneca) exhibit phenotypic outcomes (unpublished results). Most cells AURKB activity in enzyme assays, which is evident in from aclarubicin treatment have G1-S properties, possibly the cell population profile for the latter despite the role of through more potent inhibition of topoisomerase I. As that kinase beyond the G2-M checkpoint. The relationship observed for aclarubicin, the antimetabolite 5-fluoro-20- between enzyme inhibition in cells and phenotype is deoxyuridine (5-FUDR or floxuridine) significantly inhi- complex and not fully understood from cell-free assays. bits proliferation without resulting in a large enrichment We investigated the relationship between aurora of cells in G2-M. The classifier is not optimized for kinase activity and cell phenotype using a number of characterizing treatments that modulate the G1/S mitotic inhibitors at our disposal. The prodrug AZD-1152 and cell cycle; cells in G1 are difficult to distinguish from those its metabolite differ by a phosphate group used to in S phase using Hoechst staining and light microscopy, improve solubility of the prodrug; both are selective necessitating additional markers such as 5-ethyl-2-deox- AURKB inhibitors and induce polyploidy in treated yuridine (EDU) or bromodeoxyuridine (BrdU) or G1- cells, consistent with their enzyme activity. As they specific markers such as phosphor retinoblastoma pro- are the only selective AURKB inhibitors we have char- tein 1 (pRB1). The tubulin depolymerizers mebendazole, acterized, it is interesting to note that both retain larger albendazole, ciclobendazole, and nocodazole, and the populations of G1-S cells than the other aurora kinase tubulin stabilizer paclitaxel all produce populations inhibitors. As cells responding to AURKB inhibition enriched in prometaphase and apoptotic cells (15). The continue cycling beyond 48 hours, the absence of other proportion of prometaphase cells increases with concen- arrest mechanisms may explain this observation. The tration, presumably due to a diminishing proportion of inhibitors tozasertib and CYC116 induce polyploidy as nonviable cells in the population. expected for dual A/B inhibitors, unlike PHA-739358 that exhibits a concentration-dependent change from Cell population responses to inhibitors of cell-cycle dominant polyploidy (consistent with AURKB) to do- kinases minant prometaphase arrest (consistent with AURKA). Inhibitors of kinases tend to exhibit varying degrees of In our hands, MLN-8054 and ENMD-2076 both inhibit off-target activity (26), making it difficult to anticipate the AURKB more potently in biochemical assays, yet induce level of selectivity from cell-free assays expected to yield cell populations consistent with AURKA at lower con- a phenotype consistent with modulation of the intended centrations and AURKB at higher concentrations. For

250 Mol Cancer Ther; 10(2) February 2011 Molecular Cancer Therapeutics

High-Content Imaging of Cell-Cycle Modulators

Figure 5. Kinase profiling data from biochemical enzyme assays and cell phenotypes in HCT-116 cells. Right, proportion of cells classified into each of 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 (and small 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 (values followed by %); the absence of labels denote inactive results (i.e., IC50 > 10 mmol/L or % inhibition < 80 for single point results). See Supplementary Methods for details on enzyme assays. Additional inhibitors are shown in Supplementary Fig. 8. Human kinome provided courtesy of Cell Signaling Technology, Inc. www.cellsignal.com.

www.aacrjournals.org Mol Cancer Ther; 10(2) February 2011 251

Sutherland et al.

tozasertib, its significant binding affinity for CDK1 is ulators by summarizing cell-level results from high-con- not apparent in the induced cell population. tent imaging experiments. While an increasing body of In contrast to the variation in phenotypes observed for literature describes approaches that distill millions of CDK and aurora kinase inhibitors, small molecules tar- cellular measurements for interpretation of phenotype, geting PLK1 generally exhibit similar phenotypes con- only a few approaches have explored the reproducibility sisting of mixed populations of cells in prometaphase and of phenotype across experiments (7, 10). The approach advanced apoptosis. A notable exception to the class is described in this work has been applied to experiments the Banyu inhibitor 886856-66-2 that exhibits a small but done over 6 months by 3 different biologists, and sig- increasing population of G2 cells with increasing concen- nificantly reduces variability in results that often afflict tration. HMN-176 is not an ATP-competitive inhibitor, industrial application of immunofluorescence-based but interferes with the cellular localization of PLK1 (28). microscopy. The methodology is applied within lead ON-01910 was originally described as a PLK1 inhibitor, optimization programs at Lilly to understand changes but is now thought to be a tubulin modulator (22, 23; in phenotype that arise from structural modification of Supplementary Fig. 7). lead series, and the extent to which activity at other kinases translates to deviation from the desired effect Relationship between selectivity in cell-free assays (i.e., obtaining a phenotype consistent with inhibition of a and phenotype in HCT-116 cells given kinase target). The variation in phenotypic response among the kinase The application of our classifier highlights the preva- inhibitors in Table 1 prompted us to examine the con- lence of concentration-dependent phenotypes among cordance between enzyme inhibition profiles and phe- inhibitors purported to inhibit the same kinase. While notypic response for a larger number of internal this work does not evaluate the effects of RNAi treat- compounds exhibiting activity against G2-M kinases. ments, the cell morphologies that we identify as consis- For this purpose, we selected for analysis compounds tent with the intended target are informed from having antiproliferation EC50 < 1 mmol/L in HCT-116 published reports using RNAi (17, 20, 22, 23, 29, 30). cells and with fully determined G2-M kinase enzyme We postulate that departure from the expected pheno- profiles (IC50s vs. CDK1, AURKA, AURKB and PLK1, type arises from off-target effects, rather than variable or 80 percent inhibition in single concentration testing inhibition of the intended target. Although the effect can at 20 mmol/L). For wells having concentrations above the sometimes be rationalized from enzyme activity profiles antiproliferation EC50, we summarized the proportion of (e.g., PHA-739358), there is generally no discernable cells belonging to non-G1-S phenotypes from the cell quantitative relationship between enzyme selectivity classifier (Supplementary Fig. 9). While selective AURKA and phenotype. It is noteworthy that exquisitely selective inhibitors induce populations dominated by prometa- kinase inhibitors studied in this work (PD-0332991 and phase arrest and advanced apoptosis, the dual A/B the AZD-1152 metabolite) induce the expected pheno- inhibitors have a larger proportion of multinucleated types. cells than AURKB-selective inhibitors. For CDK inhibi- Conceptually, our approach is similar to methods that tors, a decreasing proportion of cells in G2 is apparent as extract classification rules from cells representing distinct selectivity vs. the aurora kinases and PLK1 increases. phenotypes identified by review of images (2–4). Such However, the corresponding increase in prometaphase methods can overcome variation in staining intensity, etc. and M-phase apoptotic cells may be attributed to CDK9 by repeating the identification of representative cells and inhibition as noted above, and most compounds were not rule training in every experiment. By recalibrating the tested vs. other CDKs. For PLK1 inhibitors, there is no cell classifier from reference molecules selected to repre- discernable change in the dominant prometaphase and sent the relevant cell phenotypes, the need for manual M-phase apoptotic phenotype with increasing selectivity review of images is substantially reduced. On the other vs. the CDKs and aurora kinases. This suggests a domi- hand, the use of reference inhibitors renders the approach nant role for PLK1 enzyme inhibition over other cell-cycle less effective in the identification of novel or unexpected targets, despite a role for PLK1 in mitosis. A conventional cell morphologies (6). The simplicity of a decision tree for view holds that simultaneous inhibition of targets cell classification is appealing, but uses artificial rectan- involved earlier in the cell cycle would manifest over gular boundaries in cytological feature space. It can be M-phase arrest. The lack of clear relationships between difficult to ascertain whether rarer cell populations are enzyme inhibition profiles and cell phenotype supports artifacts of classification; most PLK1 inhibitors appear to the importance of phenotype determination via high- induce small populations of multinucleated cells, but content imaging to verify that cell death arises from these are M-phase apoptotic cells with irregular shapes modulation of the targeted kinase (or kinases). and lower DNA staining intensity. In our hands, the use of mixture models (9) does not resolve the phenotype of Discussion cells falling between major phenotypic classes, as expected given the probabilistic assignment of cells to The cell classifier described in this work enables the phenotype classes (unpublished results). Higher resolu- investigation of mechanism of action for cell-cycle mod- tion imaging, perhaps with additional markers, may be

252 Mol Cancer Ther; 10(2) February 2011 Molecular Cancer Therapeutics

High-Content Imaging of Cell-Cycle Modulators

required to fully resolve distinct cell populations (10). As cross-reactivity is suspected (e.g., potent PLK and cellular imaging technology continues to evolve, the AURKA enzyme activity), high-resolution microscopy ability to monitor additional fluorescence channels will of centrosomes and microtubule topology to detect allow for more detailed analyses of complex cellular monopole spindle assemblies having reduced amounts processes. of centrosomal g-tubulin (PLK1; refs. 22, 23, 36), circular The imaging assay and phenotype classifier presented monopole spindles (AURKA; refs 20, 29), or the presence in this work are configured for medium-throughput of multipolar spindles (tubulin destabilizers; refs 15, 33, characterization of G2-M modulators, and can be used 37) may be required to further clarify a compound’s to identify the probable mechanism of action in concert mechanism of action. Higher-resolution microscopy with biochemical profiling results. The classifier does not can provide greater cellular detail for understanding differentiate G1 cells from those in early S phase. For mechanism, but reduces throughput due to longer scan example, the CDK inhibitor R-547 was shown to inhibit times while also posing challenges with image storage. Rb1 phosphorylation in phase I studies and induces High-resolution microscopy is also not without its ambi- populations of G1- and G2-arrested HCT-116 cells when guities: unlike the monopolar spindles induced by treat- studied via flow cytometry (31). The cell classifier indi- ment with PLK1 RNAi or BI-2536, the inhibitor GSK- cates a dominant population of G2 cells, with some 461364 potently inhibits the PLK1 enzyme in biochemical residual G1-S cells that could in fact be G1/S-arrested and substrate phosphorylation assays, yet induces multi- (i.e., blocked cell cycle). Other markers, such as antibo- polar spindles characteristic of tubulin modulators in dies for pRb1 and/or EDU labeling are probably neces- H460 human lung cancer cells (38). sary to allow full characterization of cells in G1-S. Although the cell classifier system described here Likewise, cells responding to CDK1 and topoisomerase focuses on characterization of cell-cycle modulators, inhibitors (e.g., camptothecin and doxorubicin) are not the broad concept of summarizing cell-level results can differentiated with the markers employed in this study, be applied across high-content screenings and phenoty- but are readily distinguished via immunostaining for pic drug discovery. The automated classification of phe- gH2AX, due to chain breaks induced by topoisomerase notypic screening and structure-activity relationship data inhibitors (32). represents a solution to what is now the principal diffi- In a similar vein, the classifier does not differentiate culty in taking raw data from a high content assay and cells responding to AURKA inhibitors, PLK1 inhibitors, using it to make a statistically robust and reproducible tubulin stabilizers (i.e., paclitaxel) or tubulin destabili- decision of phenotypic outcome. zers, all of which directly modulate cytoskeleton function resulting in prometaphase arrest. High-resolution micro- Disclosure of Potential Conflicts of Interest scopy of paclitaxel reveals a metaphase-like arrangement of most chromosomes, with some lagging at the spindle All the authors have Eli Lilly shares received via 401(k) and bonus plans. pole (33). Such features are not discernable at the resolution used in this assay. At best, simple intensity-derived cytological features from a-tubulin staining provide Acknowledgments slight differentiation of mechanistic classes, with pacli- The authors thank the Phenotypic Drug Discovery Working Group and taxel and PLK1 inhibitors having higher staining inten- PD2 Management for guidance on screen implementation, follow-up, and sity compared with tubulin destabilizers (whether general scientific discussion; Mark Marshall and Jake Starling for helpful discussions and project guidance. considering all cells or only those in prometaphase/M- The costs of publication of this article were defrayed in part by the apototic states; Supplementary Fig. 5). This observation payment of page charges. This article must therefore be hereby marked is consistent with the absence of changes in microtubule advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. mass at lower compound concentration, despite their impact on microtubule dynamics as determined via Received August 2, 2010; revised November 12, 2010; accepted high-resolution microscopy (34, 35). In cases where November 24, 2010; published OnlineFirst January 7, 2011.

References 1. Levsky JM, Singer RH. Gene expression and the myth of the average 5. Low J, Huang S, Blosser W, Dowless M, Burch J, Neubauer B, cell. Trends Cell Biol 2003;13:4–6. et al. High-content imaging characterization of cell cycle ther- 2. Jones TR, Carpenter AE, Lamprecht MR, Moffat J, Silver SJ, Grenier apeutics through in vitro and in vivo subpopulation analysis. Mol JK, et al. Scoring diverse cellular morphologies in image-based Cancer Ther 2008;7:2455–63. screens with iterative feedback and machine learning. Proc Natl 6. Yin Z, Zhou X, Bakal C, Li F, Sun Y, Perrimon N, et al. Using iterative Acad Sci U S A 2009;106:1826–31. cluster merging with improved gap statistics to perform online phe- 3. Tao CY, Hoyt J, Feng Y. A support vector machine classifier for notype discovery in the context of high-throughput RNAi screens. recognizing mitotic subphases using high-content screening data. BMC Bioinformatics 2008;9:264. J Biomol Screen 2007;12:490–6. 7. Young DW, Bender A, Hoyt J, McWhinnie E, Chirn GW, Tao CY, 4. Conrad C, Erfle H, Warnat P, Daigle N, Lorch T, Ellenberg J, et al. et al. Integrating high-content screening and ligand-target Automatic identification of subcellular phenotypes on human cell prediction to identify mechanism of action. Nat Chem Biol 2008; arrays. Genome Res 2004;14:1130–6. 4:59–68.

www.aacrjournals.org Mol Cancer Ther; 10(2) February 2011 253

Sutherland et al.

8. Perlman ZE, Slack MD, Feng Y, Mitchison TJ, Wu LF, Altschuler SJ. 28. Tanaka H, Ohshima N, Ikenoya M, Komori K, Katoh F, Hidaka H. Multidimensional drug profiling by automated microscopy. Science HMN-176, an active metabolite of the synthetic antitumor agent HMN- 2004;306:1194–8. 214, restores chemosensitivity to multidrug-resistant cells by target- 9. Slack MD, Martinez ED, Wu LF, Altschuler SJ. Characterizing hetero- ing the transcription factor NF-Y. Cancer research 2003;63:6942–7. geneous cellular responses to perturbations. Proc Natl Acad Sci U S A 29. Manfredi MG, Ecsedy JA, Meetze KA, Balani SK, Burenkova O, Chen 2008;105:19306–11. W, et al. Antitumor activity of MLN8054, an orally active small- 10. Loo LH, Lin HJ, Steininger RJ 3rd, Wang Y, Wu LF, Altschuler SJ. An molecule inhibitor of Aurora A kinase. Proc Natl Acad Sci U S A approach for extensibly profiling the molecular states of cellular 2007;104:4106–11. subpopulations. Nat Methods 2009;6:759–65. 30. Taylor S, Peters JM. Polo and Aurora kinases: lessons derived from 11. Loo LH, Wu LF, Altschuler SJ. Image-based multivariate profiling of chemical biology. Curr Opin Cell Biol 2008;20:77–84. drug responses from single cells. Nat Methods 2007;4:445–53. 31. DePinto W, Chu XJ, Yin X, Smith M, Packman K, Goelzer P, et al. 12. Feng Y, Mitchison TJ, Bender A, Young DW, Tallarico JA. Multi- In vitro and in vivo activity of R547: a potent and selective cyclin- parameter phenotypic profiling: using cellular effects to characterize dependent kinase inhibitor currently in phase I clinical trials. Mol small-molecule compounds. Nat Rev Drug Discov 2009;8:567–78. Cancer Ther 2006;5:2644–58. 13. Low J, Stancato L, Lee J, Sutherland JJ. Prioritizing hits from phe- 32. Furuta T, Hayward RL, Meng LH, Takemura H, Aune GJ, Bonner WM, notypic high-content screens. Curr Opin Drug Discov Devel 2008;11: et al. p21CDKN1A allows the repair of replication-mediated DNA 338–45. double-strand breaks induced by topoisomerase I and is inactivated 14. Pommier Y. DNA topoisomerase I inhibitors: chemistry, biology, and by the checkpoint kinase inhibitor 7-hydroxystaurosporine. Oncogene interfacial inhibition. Chem Rev 2009;109:2894–902. 2006;25:2839–49. 15. Jordan MA, Wilson L. Microtubules as a target for anticancer drugs. 33. Kelling J, Sullivan K, Wilson L, Jordan MA. Suppression of centromere Nat Rev Cancer 2004;4:253–65. dynamics by Taxol in living osteosarcoma cells. Cancer Res 2003;63: 16. Lapenna S, Giordano A. Cell cycle kinases as therapeutic targets for 2794–801. cancer. Nat Rev Drug Discov 2009;8:547–66. 34. Jordan MA, Thrower D, Wilson L. Effects of vinblastine, podophyllo- 17. Van Horn RD, Chu S, Fan L, Yin T, Du J, Beckmann R, et al. CDK1 toxin and nocodazole on mitotic spindles. Implications for the role of

activity is required for mitotic activation of aurora A during G2-M microtubule dynamics in mitosis. J Cell Sci 1992;102:401–16. transition of human cells. J Biol Chem 2010;285:21849–57. 35. Jordan MA, Toso RJ, Thrower D, Wilson L. Mechanism of mitotic 18. Barr AR, Gergely F. Aurora-A: the maker and breaker of spindle poles. block and inhibition of cell proliferation by taxol at low concentrations. J Cell Sci 2007;120:2987–96. Proc Natl Acad Sci U S A 1993;90:9552–6. 19. Glover DM, Leibowitz MH, McLean DA, Parry H. Mutations in aurora 36. Spankuch-Schmitt B, Bereiter-Hahn J, Kaufmann M, Strebhardt K. prevent centrosome separation leading to the formation of monopolar Effect of RNA silencing of polo-like kinase-1 (PLK1) on apoptosis and spindles. Cell 1995;81:95–105. spindle formation in human cancer cells. J Natl Cancer Inst 2002;94: 20. Girdler F, Gascoigne KE, Eyers PA, Hartmuth S, Crafter C, Foote KM, 1863–77. et al. Validating Aurora B as an anti-cancer drug target. J Cell Sci 37. Okouneva T, Hill BT, Wilson L, Jordan MA. The effects of vinflunine, 2006;119:3664–75. vinorelbine, and vinblastine on centromere dynamics. Mol Cancer 21. Lane HA, Nigg EA. Antibody microinjection reveals an essential role Ther 2003;2:427–36. for human polo-like kinase 1 (Plk1) in the functional maturation of 38. Lansing TJ, McConnell RT, Duckett DR, Spehar GM, Knick VB, mitotic centrosomes. J Cell Biol 1996;135:1701–13. Hassler DF, et al. In vitro biological activity of a novel small-molecule 22. Peters U, Cherian J, Kim JH, Kwok BH, Kapoor TM. Probing cell- inhibitor of polo-like kinase 1. Mol Cancer Ther 2007;6:450–9. division phenotype space and Polo-like kinase function using small 39. Seong YS, Min C, Li L, Yang JY, Kim SY, Cao X, et al. Characterization molecules. Nat Chem Biol 2006;2:618–26. of a novel cyclin-dependent kinase 1 inhibitor, BMI-1026. Cancer 23. Steegmaier M, Hoffmann M, Baum A, Lenart P, Petronczki M, Krssak research 2003;63: 7384–91. M, et al. BI 2536, a potent and selective inhibitor of polo-like kinase 1, 40. Misra RN, Xiao H, Rawlins DB, Shan W, Kellar KA, Mulheron JG, et al. inhibits tumor growth in vivo. Curr Biol 2007;17:316–22. 1H-Pyrazolo[3,4-b]pyridine inhibitors of cyclin-dependent kinases: 24. Taron M, Plasencia C, Abad A, Martin C, Guillot M. Cytotoxic effects of highly potent 2,6-Difluorophenacyl analogues. Bioorg Med Chem Lett

topotecan combined with various active G2-M-phase anticancer 2003;13:2405–8. drugs in human tumor-derived cell lines. Investigational new drugs 41. Emanuel S, Rugg CA, Gruninger RH, Lin R, Fuentes-Pesquera A, 2000;18:139–47. Connolly PJ, et al. The in vitro and in vivo effects of JNJ-7706621: a 25. Tsao YP, D’Arpa P, Liu LF. The involvement of active DNA synthesis in dual inhibitor of cyclin-dependent kinases and aurora kinases. Cancer

camptothecin-induced G2 arrest: altered regulation of p34cdc2/cyclin Res 2005;65:9038–46. B. Cancer research 1992;52:1823–9. 42. Huang S, Connolly PJ, Lin R, Emanuel S, Middleton SA. Synthesis and 26. Fabian MA, Biggs WH 3rd, Treiber DK, Atteridge CE, Azimioara MD, evaluation of N-acyl sulfonamides as potential prodrugs of cyclin- Benedetti MG, et al. A small molecule-kinase interaction map for dependent kinase inhibitor JNJ-7706621. Bioorg Med Chem Lett clinical kinase inhibitors. Nature Biotechnology 2005;23:329–36. 2006;16:3639–41. 27. Garriga J, Xie H, Obradovic Z, Grana X. Selective control of 43. Barvian M, Boschelli DH, Cossrow J, Dobrusin E, Fattaey A, Fritsch A, gene expression by CDK9 in human cells. J Cell Physiol 2010; et al. Pyrido[2,3-d]pyrimidin-7-one inhibitors of cyclin-dependent 222:200–8. kinases. J Med Chem 2000;43: 4606–16.

254 Mol Cancer Ther; 10(2) February 2011 Molecular Cancer Therapeutics

A Robust High-Content Imaging Approach for Probing the Mechanism of Action and Phenotypic Outcomes of Cell-Cycle Modulators

Jeffrey J. Sutherland, Jonathan Low, Wayne Blosser, et al.

Mol Cancer Ther 2011;10:242-254. Published OnlineFirst January 7, 2011.

Updated version Access the most recent version of this article at: doi:10.1158/1535-7163.MCT-10-0720

Supplementary Access the most recent supplemental material at: Material http://mct.aacrjournals.org/content/suppl/2011/01/07/1535-7163.MCT-10-0720.DC1

Cited articles This article cites 43 articles, 20 of which you can access for free at: http://mct.aacrjournals.org/content/10/2/242.full#ref-list-1

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://mct.aacrjournals.org/content/10/2/242. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.