Cell Panel Profiling Reveals Conserved Therapeutic Clusters and Differentiates the Mechanism of Action of Different PI3K/Mtor, Aurora Kinase and EZH2 Inhibitors

Home , Aurora inhibitor, Dactolisib, Duvelisib, Mubritinib, Vatalanib

Published OnlineFirst September 1, 2016; DOI: 10.1158/1535-7163.MCT-16-0403

Models and Technologies Molecular Cancer Therapeutics Cell Panel Proﬁling Reveals Conserved Therapeutic Clusters and Differentiates the Mechanism of Action of Different PI3K/mTOR, Aurora Kinase and EZH2 Inhibitors Joost C.M. Uitdehaag1, Jeroen A.D.M. de Roos1, Martine B.W. Prinsen1, Nicole Willemsen-Seegers1, Judith R.F. de Vetter1, Jelle Dylus1, Antoon M. van Doornmalen1, Jeffrey Kooijman1, Masaaki Sawa2, Suzanne J.C. van Gerwen1, Jos de Man1, Rogier C. Buijsman1, and Guido J.R. Zaman1

Abstract

Cancer cell line panels are important tools to characterize which 9 were not reported before, including TTK, BET and two the in vitro activity of new investigational drugs. Here, we clusters of EZH2 inhibitors. To investigate unexpected cluster- present the inhibition profiles of 122 anticancer agents in ings, sets of BTK, Aurora and PI3K inhibitors were profiled proliferation assays with 44 or 66 genetically characterized in biochemical enzyme activity assays and surface plasmon cancer cell lines from diverse tumor tissues (Oncolines). The resonance binding assays. The BTK inhibitor ibrutinib clusters library includes 29 cytotoxics, 68 kinase inhibitors, and 11 with EGFR inhibitors, because it cross-reacts with EGFR. Aurora epigenetic modulators. For 38 compounds this is the first kinase inhibitors separate into two clusters, related to Aurora comparative profiling in a cell line panel. By strictly maintaining A or pan-Aurora selectivity. Similarly, 12 inhibitors in the optimized assay protocols, biological variation was kept to a PI3K/AKT/mTOR pathway separated into different clusters, minimum. Replicate profiles of 16 agents over three years show reflecting biochemical selectivity (pan-PI3K, PI3Kbgd-isoform a high average Pearson correlation of 0.8 using IC50 values and selective or mTOR-selective). Of these, only allosteric mTOR 0.9 using GI50 values. Good correlations were observed with inhibitors preferentially targeted PTEN-mutated cell lines. other panels. Curve fitting appears a large source of variation. This shows that cell line profiling is an excellent tool for Hierarchical clustering revealed 44 basic clusters, of which 26 the unbiased classification of antiproliferative compounds. Mol contain compounds with common mechanisms of action, of Cancer Ther; 15(12); 3097–109. 2016 AACR.

Introduction panel of the National Cancer Institute (NCI-60), which resulted in the discovery of bortezomib and eribulin (3, 4). In parallel, a In cell panel profiling, dose–response curves are determined 39-cell line panel was set up by the Japanese Foundation for of a compound in a panel of cell line proliferation assays, and Cancer Research (JFCR-39; refs. 5, 6). this is an important tool to study the mechanism of action With the advent of genomics, attention increasingly focused and selectivity of novel cancer therapies, in addition to find to coupling cell line biology to compound response. High- drug response biomarkers before initiating expensive and throughput cell line profiling showed that cell panels could time-consuming experiments inanimalmodelsortrialsin identify patient subpopulations, for instance BRAF-mutant patients (1, 2). cell lines are in particular sensitive to BRAF inhibitors and Cell panel profiling has a long history as a tool in cancer drug EGFR-mutant cell lines are relatively sensitive to EGFR inhi- discovery. Over the period between 1989 and 2014, tens of bitors (7, 8). As a result, even larger and more fully genetically thousands of compounds have been profiled in the 60 cell line characterized cell panels were developed, encompassing up to a thousand cell lines (9–11). The Genomics of Drug Sensi- tivity in Cancer (GDSC; ref. 9) and the Cancer Cell Line 1 Netherlands Translational Research Center B.V., Kloosterstraat, the Encyclopedia (CCLE; ref. 10) analyzed 24 drugs in 1,036 cell Netherlands. 2Carna Biosciences, Inc., Kobe, Japan. lines and 138 drugs in 727 cell lines, respectively. Other Note: Supplementary data for this article are available at Molecular Cancer studies include the Cancer Therapeutics Response Portal study Therapeutics Online (http://mct.aacrjournals.org/). of 481 drugs in 860 cell lines (CTRP; refs. 11, 12) and that of Corresponding Author: Guido J.R. Zaman, Netherlands Translational Research GlaxoSmithKline, of 19 compounds in 311 cell lines (GSK; Center B.V., Kloosterstraat 9, 5349 AB, the Netherlands. Phone: 31412700500; refs. 8,13). These studies discovered, for instance, that PARP Fax: 31412700501; E-mail: [email protected] inhibitors were active in cells with the EWS–FLI1 transloca- doi: 10.1158/1535-7163.MCT-16-0403 tion, which has led to clinical evaluation of PARP inhibitors 2016 American Association for Cancer Research. in Ewing sarcoma (9, 14).

www.aacrjournals.org 3097

Uitdehaag et al.

Despite the successes, the use of drug screening in large cell from ATCC who authenticated all cell lines by short tandem line panels has come under debate recently. As the shutdown repeat analysis. Cell proliferation assays were carried out as of the NCI-60 panel was announced, many important drug described (22) using ATPlite 1step (Perkin Elmer). Exposure time candidates and selective tool inhibitors remain unscreened, was 72 hours for all compounds, except for epigenetic com- making it harder to find the best tool inhibitors to probe new pounds, for which exposure time was 120 hours (Supplementary biological pathways (1, 15, 16). Second, a comparison of the Table S2). Percentage growth was calculated, relative to the growth CCLE and GDSC panels prompted a discussion about data of unexposed cells, and relative IC50s were fitted using a four- reproducibility (17–21). It was pointed out that the rank parameter logistics curve. From these and seeding cell density correlation between IC50s on the same drug in the same cell data, absolute GI50s were calculated (3, 22). For AUC calculations, lines could vary between 0.1 and 0.6, with a median of 0.28, all percent-effect data within the test range were summed. If ranges which is poor (17). One cause could be that the GDSC were adapted, AUC data were not calculated. maintains universal dose ranges and extrapolates IC50sifthey In the course of three years, 122 inhibitors were tested on fall outside the standard range. For example, for the potent 44 cell lines, and after a panel extension, on 66 cell lines (Fig. 1; compounds bortezomib and docetaxel, about 40% of the Supplementary Table S1). Compounds were dissolved in profiling IC50s consist of extrapolated values. However, also DMSO generally 48 hours before testing. Cisplatin and carbo- when the area-under-the-curve (AUC) activity measures are platin were dissolved on the day of testing. An overview of all used (as e.g. in refs. 10, 11) which are not extrapolated, rank IC50 data is presented in Supplementary Table S3A. correlations remained poor (17). A tell-tale sign is that, after When compounds were inactive, their IC50 was set to 31,600 10 data clustering by the GDSC and CTRP, the highly related nmol/L. Finally, log IC50s(innmol/L)wereusedforall taxanes docetaxel and paclitaxel fell into two different thera- subsequent analyses. When compounds had been profiled peutic subsets (9, 11). This is improbable and inconsistent twice, the data of the largest panel was chosen, and, if equal, with earlier profilings (4, 6). the earliest data set. In defense of the large screens, it has been argued that using rank correlation as measure for data consistency is not appro- Bioinformatics analysis priate, and a Pearson correlation on log IC50s should be used To compare our profiles with each other and with literature, 10 (20). In addition, the data serve their purpose of identifying Pearson correlations between log IC50swereused(r; Figs. 1 genomic markers (20, 21). Care should be taken therefore not to and 2). All further calculations were performed in R (23). Drug- dismiss these valuable data sets. Unfortunately, and in contrast sensitivity scores were calculated in the package DSS (24). to the NCI-60 and JFCR-39 screens, none of the large screens Hierarchical clustering (Fig. 3) used the Ward method and provide internal replicate data, which is essential to get insight 1 r as distance measure. The resulting tree was validated with into the limits of the reproducibility of cell line screening. multiscale bootstrap resampling (R package pvclust; ref. 25). Here, we present a data set of inhibition profiles of 122 A cutoff for the minimally significant Pearson correlation compounds on a panel of 44 or 66 cell line proliferation assays between profiles (rmin ¼ 0.29; P < 0.05) was obtained by (Oncolines). Of these, 61 compounds are registered drugs, 42 converting the Pearson correlation to a t statistic (26). Then t are investigational drugs, while 19 are in discovery phase or are was translated into P values with the Student t distribution tool compounds. Compared with other profiles, this data set function. Affinity propagation clustering (APC) was performed contains 38 inhibitors for which no data have been reported in using the package apcluster, setting a fixed number of 43 clusters the literature, including many inhibitors with novel mechan- (Table 2; Supplementary Table S4A; ref. 27). Network trees were isms of actions, such as EZH2 inhibitors, BET inhibitors, iso- generated using the Fruchterman–Reingold algorithm, in the form-selective PI3K inhibitors and Aurora kinase inhibitors package igraph (Fig. 4; ref. 28). Principal component analysis (Table 1; Supplementary Tables S1 and S2). Compared with was performed using the function princomp (Fig. 4; ref. 23). the larger panels, the panel shows higher data quality, as evidenced by the correlations of duplicate profiles of 16 chem- Kinase activity assays ically and mechanistically different inhibitors, over a period of The inhibitory activity of compounds on Aurora A and C three years. Clustering analyses of all library profiles revealed (Carna Biosciences, Inc.) was determined with LANCE Ultra separate clusters of different therapeutic modalities, i.e., taxanes, TR-FRET assays (PerkinElmer) at KM,ATP and using ULight-labeled platins, topoisomerase and EGFR inhibitors, which is highly PLK substrate peptide (PerkinElmer, cat. no. TRF-0110; ref. 29). consistent with earlier results (4, 7, 11) and new clusters IC50s from 9-point dose–response curves were converted to KD comprising compounds sharing novel biochemical targets. This using the Cheng–Prusoff equation (Fig. 4D). compound response database therefore serves as a valuable reference set that can be used to benchmark other cancer cell Kinase binding assays line panels and provides an up to date and unbiased view of For surface plasmon resonance experiments, biotin-labelled therapeutic clusters available in antiproliferative therapies. kinases (Carna) were immobilized on streptavidin-coated chips (GE Healthcare; Cat. no. BR100531) on a Biacore T200 (GE Healthcare) to level of 4000 response units (RU) Materials and Methods using Biacore buffer (50 mmol/L Tris pH 7.5, 0.05% (v/v) Experimental data Tween-20, 150 mmol/L NaCl and 5 mmol/L MgCl2), except Cell lines were obtained from the American Type Culture PI3Kd,whichwasimmobilizedtoalevelof8,000RU.Remain- Collection (ATCC) from 2011 to 2014 (Supplemental Table ing streptavidin was blocked with biocytin. The kinetic con- S1B) and cultured in ATCC-recommended media. All experi- stants of the compounds were determined by single-cycle ments were carried out within nine passages of the original vials kinetic experiments as described previously (29). Data

3098 Mol Cancer Ther; 15(12) December 2016 Molecular Cancer Therapeutics

Therapeutic Clusters Discovered by Cell Panel Proﬁling

reported are the geometric averages of two independent experi- according to well-defined protocols (22). Biological variation ments (Fig. 4D and E). was monitored by the average proliferation rates of each cell line, showing overall stability of growth rates over the course Cell line genetics of three years (Fig. 1A). Mutation and gene copy number data of our cell lines were VariationinIC50s was probed with the reference doxorubi- based on COSMIC v.75 of the GDSC (9) and were additionally cin, which was chosen because it gives a potent, full sigmoidal validated by DNA sequence analysis of a number of important dose–response curve in nearly all cell lines (Fig. 1B). The cancer genes (Supplementary Table S1C). For the analysis variation in 10logIC s of six replicate profilings of doxorubicin – fi 50 in Fig. 4F H, cell lines were classi ed as having a "wild-type" or was comparable with that of the NCI-60 panel, and similar a "mutated" genotype, and the logIC50s in both sets were sub- to reported values in standardized high-throughput assays t – mitted to a test, followed by a Benjamini Hochberg correction (Supplementary Fig. S1A; ref. 31). for multiple testing (22, 30). Next, Pearson correlations (r) between replicate compound signatures were investigated (20). Any Pearson corre- Processing of literature data lation above 0.6 is considered "fair" and above 0.8 is con- NCI-60 data were downloaded from https://dtp.cancer.gov/ sidered "near-perfect" (17). The doxorubicin replicates show databases_tools/. All other data came from supplementary data r ¼ 0.99 after 6 days and 0.81 after 643 days (Fig. 1C; from indicated references. IC50s were transformed to units of 10 Supplementary Table S3B). The replicate profiling of a total log (nmol/L), in parallel with our data. In case of NCI-60 and of 15 diverse (candidate) drugs, separated by 392 to 854 JFCR-39, data of multiple replicates were averaged (Fig. 2). days, shows correlations between 0.58 and 0.93 (average 0.8; Fig. 1C). Results Because it has been claimed that IC50 is not an optimal Precision of cell line profiling response metric (24, 32), we investigated if using other metrics The high-throughput Oncolines cell panel has been running would improve correlations. We calculated, from the same raw over the past three years with 44, and later 66 cell lines data, three other response measures: AUC (11, 22), DSS (24),

Figure 1. Reproducibility of Oncolines cell panel data. A, Monitoring of cell replication rate for all cell lines in the panel. Each data point represents a profiling experiment. In 0.1% of cases, growth speed deviated more than a factor two from the average, in which cases profilings were redone. Twenty-two cell lines were added later to the panel (from experiment no. 122 onward). B, Overlay of doxorubicin data in cell line A375 measured at three different time points. C, Pearson correlations between replicate profiles. Indicated are compound names and days between replicates. Different bars represent correlations based on different cell response measures. AUC were not calculated when dose ranges differed between replicates. Average correlationsare

0.80 (IC50s), 0.88 (reinterpreted IC50s), 0.88 (GI50), 0.75 (DSS), and 0.76 (AUC).

www.aacrjournals.org Mol Cancer Ther; 15(12) December 2016 3099

Uitdehaag et al.

and GI50 (3). AUC is based on percent-growth signals and does measure a significantly more potent activity (factor of 15) of not use data-fitting. DSS is an area-based method that uses methotrexate than that presented in the literature (Fig. 2C). fitting. GI50 is the concentration of 50% cell growth inhibition One reason might be that our solutions were always freshly and accounts for the starting cell density. For AUC and DSS, made and that methotrexate can easily degrade when light- lower correlations are found (averages 0.76 and 0.75, Fig. 1C; exposed (33, 34). Supplementary Table S3C). For AUC, this is consistent with literature (20), probably because an AUC value is more sensi- Cluster analysis of all 122 inhibitor profiles provides 26 tive to variations in maximum and minimum signals in the validated therapeutic clusters assay. The fitting applied in DSS does not ameliorate this, Next, we analyzed the data of 122 anticancer agents targeting maybe because DSS was implemented without manual curve all important oncogenic signaling pathways. A total of 38 curation. However, GI50-based correlations were higher than of these compounds were not part of any profiling study 10 their IC50 equivalents (average 0.88 compared with 0.80), before (Table 1 and Supplementary Table S1A). The log demonstrating that correcting for cell line growth rates im- IC50 valuesweresubmittedtounsupervisedhierarchicalclus- proves reproducibility (32). tering (Fig. 3). When duplicate profiles are included, nearly all Next, we investigated if other factors also contribute to IC50 replicates are neighbors, showing that the data reproducibility reproducibility. Plotting the standard deviations of replicate is of such quality that we can pick up "identical" compounds log IC50s shows that they depend on cell line growth rate, and (Supplementary Fig. S2A). From the tree without replicates on time between experimental replicates, but not on the (Fig. 3A), we wanted to isolate the minimum amount of standard deviation of cell growth rate in each cell line (Sup- relevant clusters. Therefore, we applied a cutoff at the correla- plementary Fig. S1B–S1D). As this suggested a non-biological tion level that is minimally significant (r ¼ 0.29; P < 0.05; see source of error, we studied the influence of data interpretation, Methods and Supplementary Table S3D). by repeating curve fitting of all replicate profiles, as data The cutoff leads to a total of 44 clusters, of which 35 contain interpretation rules tended to shift over the years of the at least two compounds, and maximally eight (Table 2). In experiments (Fig. 1C; see Supplementary Table S3B for data total, 26 clusters contain two or more compounds with known and Supplementary Fig. S1E–S1G for final rule set and some and similar biochemical targeting, such as EGFR, topoisomer- illustrative case studies). This resulted in significant improve- ase, or MEK inhibitors (Fig. 3; Table 2). That compounds are ments. After refitting, only 2 out of 15 replicates have a grouped according to mechanism, not potency, is illustrated by correlation below 0.8, and none have a correlation below the MEK inhibitors selumetinib, PD-0325901, and trametinib 0.7. The average is 0.88, similar to the GI50-based correlations. that have an average IC50 on the full panel of, respectively, This shows that it is feasible to have "near-perfect" (17, 20) 5.3 mmol/L, 0.9 mmol/L, and 0.49 mmol/L, but which still correlations between profiles in a cell line panel, and that to belongtothesamecluster(Fig.3;Table2).Welabelthe26 increase reproducibility, more effort should be put into uni- mechanistically defined clusters as `highly validated clusters'. form and unambiguous computational determination of This approach to validation conceptually resembles the use of response parameters (refs. 24, 32; Fig. 1C). biochemical target profiles to pinpoint highly validated clusters (11). Our highly validated clusters include for instance a joint Accuracy of cell line profiling group of paclitaxel and docetaxel (Table 2), in contrast with As a next step, we compared the IC50s of our cell panel GDSC and CTRP data, and in accordance with NCI-60 and profiling to those of other studies. The overlap of our 122 JFCR-39 data (4, 6, 911). compounds with other cell line profiling studies is given in Because clusters might depend on the clustering method 10 Supplementary Table S1A. Analysis of logIC50smeasuredon used, we next evaluated affinity propagation clustering (APC; identical compounds in identical cell lines shows the best refs. 9, 27). This results in 46 clusters, of which 42 are highly Pearson correlations with data from NCI-60 (r ¼ 0.82) and similar to the hierarchical clustering (Table 2; Supplementary JFCR-39 (r ¼ 0.85), followed by data from the GDSC (r ¼ Table S4A). Also, with this method docetaxel and paclitaxel 0.73), CCLE (r ¼ 0.75), and GSK studies (r ¼ 0.78;Fig.2Aand cluster together. However, cisplatin and carboplatin do not, B; Supplementary Fig. S1H). Data from CTRP, which consist which is in contrast to the hierarchical clustering and the only of AUC data, were not analyzed because comparison literature (Fig. 3; refs. 4, 6). of AUC metrics requires use of identical dose ranges, which To further compare the robustness of clustering across plat- was not the case. Given the fact that all platforms differ in read- forms, we calculated hierarchical clustering trees and correla- out technology, incubation time, cell line passage, cell densities tion matrices for the 51 compounds that our data set has in and plate formats, this shows that cell line profiling can be commonwiththeNCI-60panel.Despitethedifferencesinthe reproducible across platforms (20, 21). platforms, clear parallels can be seen between groupings of As five cell lines (A549, ACHN, BT-549, NCI-H460, and compounds (Fig. 2D). In the 51-compound NCI-60 data, 11 OVCAR-3) were profiled in all these studies, we next compared "highly validated" clusters appear, i.e., that contain at least two the potency across platforms for eight highly active compounds compounds with similar biochemical mechanism. These are all that were profiled in multiple panels (Fig. 2C). For the Onco- represented in the clustering tree of Fig. 3 (see also Supple- lines data, mitomycin-C and methotrexate deviate from values mentary Table S4B). These clusters comprise EGFR, BRAF/MEK, measured in at least three other panels (Fig. 2C). Data refitting mTOR, and multikinase inhibitors, as well as topoisomerase improved concordance for mitomycin-C, but not for metho- inhibitors, platins, taxanes, purine analogues and histone dea- trexate (Fig. 2C). Therefore, methotrexate was reprofiled (Sup- cetylase (HDAC) inhibitors (Table 2). The fact that these plementary Table S3B). The correlation of the new data with the clusters can be found in two different platforms with different earlier replicate is high (r ¼ 0.87). Thus, it appears that we read-out technology and comprising different cell lines

3100 Mol Cancer Ther; 15(12) December 2016 Molecular Cancer Therapeutics

Therapeutic Clusters Discovered by Cell Panel Proﬁling

Figure 2.

Comparison of Oncolines IC50swithIC50s measured in other large scale panels. A, Correlation with overlapping data from 51 compounds and 18 cell lines from the NCI-60 panel (3, 4). B, Correlation with overlapping data from 39 compounds and 62 cell lines from the GDSC data set (9). For 10 other correlation see Supplementary Fig. S1H. In A and B,Axesrepresent logIC50 (nmol/L). Numbers in frames indicate Pearson values. C, Variation of measured IC50s between eight compounds in five cell lines (A549, ACHN, BT-549, NCI-H460, and OVCAR-3). Data from five different panels are compared. The y-axis shows the difference between the pertinent logIC50 and the average logIC50 of that cell line–compound combination across all data sets. MTX: methotrexate. D, Pearson correlation matrix of 51 compounds analyzed in both the Oncolines and NCI-60 panels (blue: high correlation, red: negative correlation). The left triangle shows clusters and correlations using data from the Oncolines panel. The right triangle is identical to the left one, only based on NCI-60 data. Both data sets reveal similar clusters (some classes are indicated). provides good validation that they are truly distinct mechanis- kinase assays (ref. 35; Supplementary Table S3E), which tic approaches of cancer therapy. showed that aside from BTK, ibrutinib is also a potent EGFR and HER2 inhibitor, as described before (36). Because of these activities, ibrutinib behaves as an EGFR inhibitor in the cell Unexpected clusterings give insight into compound panel. mechanism The registered ALK/MET kinase inhibitor crizotinib clusters The hierarchical clustering groups compounds with similar with danusertib, a pan-Aurora inhibitor (Fig. 3). This may be biochemical targets, but also reveals some surprises. For related to the high Aurora A activity of crizotinib (37) and the instance ibrutinib, an FDA-registered irreversible BTK inhibitor, tyrosine kinase activity of danusertib (38), making a cluster with clusters with the irreversibe EGFR inhibitor afatinib, and other dual Aurora/Tyrosine kinase inhibitors. reversible EGFR and HER2 inhibitors (Fig. 3). To further Another paradoxical cluster is that of the ALK inhibitor investigate this, we profiled ibrutinib on a panel of biochemical ceritinib and the IGF1R inhibitors GSK-1838705A and

www.aacrjournals.org Mol Cancer Ther; 15(12) December 2016 3101

Uitdehaag et al.

Table 1. Compounds tested in this study Compounda Main target Clinical phaseb Compounda Main target Clinical phaseb ABT-737 BCL2 Phase I Lapatinib EGFR Marketed Actinomycin-D Transcription Marketed LGK-974 PORCN Phase I Afatinib EGFR Marketed Masitinib KIT Phase III All-trans retinoic acid RAR Marketed Melphalan DNA alkylating Marketed Alpelisib PI3K Phase II Mercaptopurine Nucleoside analogue Marketed AMG-900 Aurora kinases Phase I Methotrexate Folate synthesis Marketed Apitolisib PI3K Phase II Mitomycin-C DNA crosslinking Marketed AT-7519 CDK Phase I Mitoxantrone Topoisomerase II Marketed Axitinib VEGFR/PDGFR Marketed MK-1775 WEE1 Phase II AZD-8055 mTOR Phase I MK-2206 AKT Phase II BGJ-398 FGFR Phase II MK-5108 Aurora kinases Phase I BI-2536 PLK1 Phase II MLN-8054 Aurora kinases Phase I BIIB021 HSP90 Phase I Momelotinib TBK-1 Phase II BLU-9931 FGFR4 Preclinical MPI-0479605 TTK Preclinical Bortezomib Proteasome Marketed Mps1-IN-1 TTK Preclinical Bosutinib ABL Marketed Mubritinib ERBB2 Phase I Buparlisib PI3K Phase III Navitoclax BCL2 Phase II Busulfan DNA alkylating Marketed Neratinib EGFR Phase II Cabozantinib MET/VEGFR Marketed Nilotinib ABL Marketed Carboplatin DNA damage Marketed Nintedanib VEGFR/FGFR Phase III Carfilzomib Proteasome Marketed NMS-P715 TTK Preclinical Ceritinib ALK Marketed Nutlin 3a MDM2 Preclinical CHIR-124 CHK1 Preclinical NVP-ADW742 IGF1R Preclinical Cisplatin DNA damage Marketed Olaparib PARP Marketed Crizotinib ALK/MET Marketed Paclitaxel Tubulin Marketed Cytarabine Nucleoside analog Marketed Palbociclib CDK4/6 Marketed Dabrafenib RAF Marketed Panobinostat HDAC Marketed Dacarbazine DNA alkylating Marketed Pazopanib VEGFR/PDGFR Marketed Dactolisib PI3K/mTOR Phase II PD-0325901 MEK Phase II Danusertib Aurora kinases Phase II Pelitinib EGFR Phase II Dasatinib ABL/VEGFR Marketed PHA-793887 CDK Phase I Daunorubicin Topoisomerase II Marketed Pictilisib PI3K Phase I Dinaciclib CDK Phase III Ponatinib ABL Marketed Docetaxel Tubulin Marketed Prednisolone GR Marketed Doxorubicin Topoisomerase II Marketed Quizartinib FLT3 Phase II Duvelisib PI3K Phase II Regorafenib VEGFR/PDGFR Marketed Entinostat HDAC Phase III Roscovitine CDK Phase II Epirubicin Topoisomerase II Marketed Ruxolitinib JAK2/JAK3 Marketed Epothilone B Tubulin Phase II SCH-900776 CHK1 Phase II EPZ-005687 EZH2 Preclinical Selumetinib MEK Phase III EPZ-5676 DOT1L Phase I SN-38 Topoisomerase I Marketed EPZ-6438 EZH2 Phase II Sorafenib VEGFR/PDGFR Marketed Erlotinib EGFR Marketed Sunitinib VEGFR/PDGFR Marketed Etoposide Topoisomerase II Marketed Temozolomide DNA alkylating Marketed Everolimus mTOR Marketed Temsirolimus mTOR Marketed Fluorouracil Nucleoside analogue Marketed TGX-221 PI3K Preclinical Gefitinib EGFR Marketed TH-588 MTH1 Preclinical Gemcitabine Nucleoside analogue Marketed Thioguanine Nucleoside analog Marketed GSK-1070916 Aurora kinases Phase I Topotecan Topoisomerase I Marketed GSK-126 EZH2 Preclinical Tozasertib Aurora kinases Phase II GSK-1838705A IGF1R Preclinical Trametinib MEK Marketed GSK-343 EZH2 Preclinical UNC1999 EZH1/EZH2 Preclinical GSK-461364 PLK1 Phase I Vandetanib VEGFR/PDGFR Marketed I-BET-762 BET Preclinical Vatalanib VEGFR Marketed Ibrutinib BTK Marketed Vemurafenib RAF Marketed ICG-001 Wnt pathway Preclinical Venetoclax BCL2 Marketed Idelalisib PI3K Marketed Vincristine Tubulin Marketed Imatinib ABL Marketed Vinflunine Tubulin Marketed Irinotecan Topoisomerase I Marketed Volasertib PLK1 Phase II JQ1 BET Preclinical Vorinostat HDAC Marketed KU-60019 ATM Preclinical XAV-939 TNKS (tankyrase) Preclinical aCompounds also tested, but found inactive, were XL147, SGX-523, fasudil, NVP-LDE225 (erismodegib), tofacitinib, streptozocin, and macitentan. bHighest clinical phase reached (status May 2016, source: clinicaltrials.gov).

NVP-ADW-742. Ceritinib could be expected to cluster with Therefore, this cluster seems to be based on common IGF1R other ALK inhibitors such as crizotinib. However, recently inhibition. In the CTRP panel, the ALK inhibitor NVP-TAE684 it was stressed that ceritinib also has IGF1R activity (39). wasalsoobservedtoclusterwithIGF1Rinhibitors(11).

3102 Mol Cancer Ther; 15(12) December 2016 Molecular Cancer Therapeutics

Therapeutic Clusters Discovered by Cell Panel Proﬁling

Figure 3. Hierarchical clustering of 122 compounds proﬁled on 44 or 66 cell lines. The dotted line represents the Pearson correlation cutoff used for identifying compound clusters (P < 0.05). The red brackets on the right indicate signiﬁcant clusters according to multiscale bootstrap validation (P < 0.05).

As combined ALK/IGF1R inhibition showed synergistic effects, Some compounds with common biochemical mechanism are it was suggested that cluster proximity can predict synergy dispersed over various clusters. For instance, all alkylating agents (11). tested, such as, e.g., dacarbazine, melphalan, and temozolomide

www.aacrjournals.org Mol Cancer Ther; 15(12) December 2016 3103

Table 2. Mechanism-based compound clusters observed in this and other cell proﬁling studies o acrTe;1(2 eebr2016 December 15(12) Ther; Cancer Mol No. Cluster name Compounds APCa GDSCb GSKc NCId JFCRe CTRPf al. et Uitdehaag Downloaded from 1 BET I-BET-762 JQ1 yy 2 no. 2 ﬂuorouracil nutlin 3a `m m 3 purine analogs thioguanine dacarbazine mercaptopurine y y y 4 selective ABL masitinib imatinib nilotinib ponatinib y y y y y 5 BCL2 navitoclax ABT-737 yy 6 no. 6 actinomycin-D PHA-793887 mmm 7 covalent EGFR afatinib ibrutinib y Published OnlineFirstSeptember1,2016;DOI:10.1158/1535-7163.MCT-16-0403

mct.aacrjournals.org 8 ATRA ATRA mmy 9 Aurora B/C AMG-900 tozasertib GSK-1070916 ym y 10 taxane-like AT-7519 docetaxel ICG-001 paclitaxel vincristine y y y y y 11 VEGFR/PDGFR axitinib cabozantinib nintedanib pazopanib vatalanib regorafenib sorafenib y y y y 12 MEK selumetinib PD-0325901 trametinib yy y y 13 competitive mTOR AZD-8055 mm 14 FGFR BGJ-398 dactolisib m 15 PLK BI-2536 carfilzomib dinaciclib GSK-461364 vinflunine volasertib y y y 16 no. 16 BIIB-021 epothilone B KU-60019 TH-588 on September 29, 2021. © 2016American Association for Cancer Research. 17 pan-PI3-kinase buparlisib alpelisib apitolisib pictilisib y 18 FGFR4 BLU9931 m 19 proteasome bortezomib my 20 multikinase ABL bosutinib dasatinib vandetanib y 21 Aurora A busulfan MK-5108 MLN-8054 y 22 PI3-kinase b/g/d idelalisib duvelisib TGX-221 yy 23 platin-like carboplatin cisplatin melphalan MK-2206 yy 24 IGF-1R ceritinib GSK-1838705A NVP-ADW742 y y y y 25 Wee1/CHK1 CHIR-124 MK-1775 SCH-900776 y 26 no. 27 crizotinib cytarabine danusertib m 27 RAF dabrafenib vemurafenib yy y 28 topoisomerase daunorubicin doxorubicin epirubicin etoposide irinotecan mitoxantrone topotecan SN38 y y y y y 29 HDAC/PARP entinostat panobinostat olaparib vorinostat y y 30 EZH2 no. 1 EPZ-005687 GSK-343 UNC-1999 y 31 no. 31 EPZ-5676 temozolomide m 32 EZH2 no. 2 EPZ-6438 GSK-126 m 33 EGFR erlotinib gefitinib lapatinib neratinib pelitinib y y y y y 34 rapalogs everolimus temsirolimus yy yy y 35 no. 35 venetoclax roscovitine m 36 TTK gemcitabine MPI-0479605 NMS-P715 y 37 no. 37 LGK-974 momelotinib m 38 no. 38 methothrexate mmm oeua acrTherapeutics Cancer Molecular 39 multikinase 3 Mps1-IN-1 ruxolitinib sunitinib y 40 no. 40 mitomycin-C prednisolone m 41 CDK 4/6 palbociclib m 42 FLT3 quizartinib m 43 no. 43 mubritinib m 44g no. 44 XAV-939 m Light gray (y): Cluster containing at least two compounds with the described biochemical target. Dark gray (m): Cluster reported that contains at least one compound listed under "compounds." aAlternative (APC) clustering of our data, see Supplementary Table S4A. bBased on APC clustering from GDSC (9). cClustering from GSK (13). dClustering from NCI-60 (4). eClusterings from JFCR (5, 6). fCTRP clusters from their website (11). gFor additional validated clusters described in these works, see Supplementary Table S4C. Published OnlineFirst September 1, 2016; DOI: 10.1158/1535-7163.MCT-16-0403

Therapeutic Clusters Discovered by Cell Panel Proﬁling

cluster in different groups. Another group that is spread across Aurora A selective inhibitors, whereas tozasertib, AMG-900, GSK- clusters are the CDK inhibitors (PHA-793887, dinaciclib, and 1070916, and danusertib are pan-Aurora inhibitors (Fig. 4D). palbociclib). Thus the two biochemical groups overlap with the groups in the network tree (Fig. 4B) and correlate with Aurora A and pan-Aurora Novel clusters distinguish epigenetic approaches such as selectivity, respectively. EZH2 inhibitors Studying cell line profiles, it appears that the four inhibitors In the hierarchical clustering tree, many clusterings are seen in the pan-Aurora cluster (Fig. 4D) are more active than the of compounds that were profiled for the first time. Examples Aurora A inhibitors in diverse lines such as SK-N-AS, LS174T, are the clusters of the two TTK inhibitors NMS-P715 and MPI- BxPC-3 and AN3 CA, but also acute lymphoblastic leukemia 0479605, and the cluster of the two CHK1 inhibitors CHIR-124 (ALL) cell lines such as MOLT-4 and CCRF-CEM. This matches and SCH-900776 (Fig. 3). This confirms that inhibition of these the observation that inhibitors of Aurora B (such as pan-Aurora cell-cycle targets generates distinct cellular profiles (29). inhibitors) are more effective than selective Aurora A inhibitors A substantial set of epigenetic modulators was tested, includ- in ALL patients (43). ing HDAC, BET, DOT1L, and EZH2 inhibitors. The HDAC inhibitors (entinostat, vorinostat, and panobinostat) form one PI3K inhibitors cluster with the PARP inhibitor olaparib. This is based on their The PI3K/AKT/mTOR pathway contains many important inhibition of among others hematopoietic cell lines, consistent drug targets. Particularly, mTOR inhibitors and the PI3Kd with HDAC inhibitors being most frequently approved for use inhibitor idelalisib have been approved for use as anticancer in hematological cancers (40). Furthermore, as HDAC and drugs, while many other compounds are in clinical trials PARP inhibitors work synergistically in many settings (41), (44, 45). We profiled a total of 12 inhibitors, of which 3 were their proximity supports that clustering can be used to identify not characterized in a larger panel before (Supplementary synergistic pairs (11). Table S1A). The hierarchical clustering assigns all inhibitors to The BET inhibitors JQ1 and I-BET-762 (also known as BRD4 the upper part of the tree, as expected for compounds active in inhibitors, which is one of the BET-family members) form a thesamepathway(Fig.3).Theonlyexceptionistheallosteric separate cluster close to the BCL-2 inhibitors (ABT-737 and AKT inhibitor MK-2206, which profile seems most related to the navitoclax). All are very potent inhibitors of leukemia and lym- platins (Fig. 3). phoma lines such as MOLT-4, Jurkat E6-1, CCRF-CEM, and SR, With regard to mTOR inhibitors, the hierarchical clustering which is in line with the clinical trials for BET and BCL-2 inhi- clearly distinguishes the allosteric agents everolimus and temsir- bitors, that focus on leukemias and lymphomas (42). olimus (also known as rapamycin analogues or rapalogs) Of the class of EZH2 inhibitors, we profiled five representa- from two ATP-competitive inhibitors of mTOR (dactolisib and tives, which form two clusters, the first consisting of GSK-343, AZD-8055). EPZ-005687, and UNC-1999. The second consisting of PI3K inhibitors are separated into two clusters, the first contain- GSK-126 and EPZ-6438 (Fig. 3). Both groups are active in ing alpelisib, buparlisib, pictilisib, and apitolisib, the second many leukemic cell lines; the second EZH2 cluster also strongly containing duvelisib, idelalisib, and TGX-221. These groups are inhibits the osteosarcoma line MG-63. It must be noted that the confirmed in a network analysis (Fig. 4C). Interestingly, buparli- first group shares a common indazole scaffold, which is not sib, the most advanced pan-PI3K inhibitor in the clinic (Table 1; seen in compounds from the second group (Supplementary ref. 44) shows correlation with a substantial number of profiles of Table S2), so the clustering might be related to a scaffold- non-PI3K inhibitors (Fig. 4C). specific off-target activity. Because the catalytic subunit of PI3K exists in several iso- To further verify the EZH2 inhibitor subgroups, we constructed forms, we compared the affinity for PI3Ka to d by surface a network tree (Fig. 4A). This tree is an independent analysis plasmon resonance. This shows that alpelisib, buparlisib, picti- from hierarchical clustering because every inhibitor can have lisib, and apitolisib all have pan-PI3K activity and potently more than two neighbours. For clarity, the network contains only inhibit PI3Ka (Fig. 4E). In contrast, TGX-221, duvelisib, and inhibitors of which the profiles highly correlate with the EZH2 idelalisib all have relatively high inhibitory activity on PI3Kb, g, inhibitors, and correlates thereof (r 0.5, Fig. 4A). This tree or d, and low inhibitory activity on PI3Ka (Fig. 4E). Thus, the confirms that there are two kinds of EZH2 inhibitors. two PI3K clusters in Fig. 4C correlate with PI3K isoform selectivity. Aurora inhibitors come in two classes To study the biological relevance of the PI3K clusters, we A total of six Aurora-kinase inhibitors were profiled (tozasertib, characterized our cell lines for mutations in PTEN, PIK3CA, AMG-900, GSK-1070916, danusertib, MK-5108, and MLN- and PIK3R1, which are three of the most frequently mutated 8054), which all have been tested in patients (Table 1). The first genes in cancer, and components of the PI3K pathway (46). three form a separate cluster that borders CDK and TTK inhibitors. Then we studied which of the inhibitors in our library most Danusertib clusters with crizotinib (see above). The other two, selectively inhibited mutant cell lines. For PTEN mutants, MK-5108 and MLN-8054, border BET and BCL2 inhibitors. these are the rapalogs, followed by duvelisib (Fig. 4F). For Follow-up analysis with a network tree shows two groups, one PIK3CA-mutant cell lines, these are MK-2206, temsirolimus, consisting of MK-5108 and MLN-8054, and one of the remaining and the PI3Ka-selective inhibitor alpelisib (Fig. 4G). For four inhibitors (Fig. 4B). PIK3R1-mutant cell lines, this is the PI3Kg/d-selective inhibitor To study if subgroups are related to isoform selectivity, we duvelisib (Fig. 4H). Thus different inhibitors of the PI3K measured Aurora kinase A, B, and C selectivity in enzyme activity pathway selectively inhibit subsets of cell lines dependent assays and binding assays based on surface plasmon resonance on PI3K signalling. Notably, only the association between (Fig. 4D). Consistent with literature, MK-5108 and MLN-8054 are rapalog sensitivity and PTEN-mutation achieves high

www.aacrjournals.org Mol Cancer Ther; 15(12) December 2016 3105

Uitdehaag et al.

3106 Mol Cancer Ther; 15(12) December 2016 Molecular Cancer Therapeutics

Therapeutic Clusters Discovered by Cell Panel Proﬁling

significance (Fig. 4F), and indeed this is the only biomarker the highly validated clusters have been observed in earlier that has achieved success in the clinic so far (44). profilings(refs.4,6,9,11,13;Table2),demonstratingthat cluster identities are independent of cell panel composition Principal components analysis reveals four basic therapeutic and represent a relevant and systematic classification of ther- vectors apeutic modalities. The other 9 highly validated clusters were Next, we investigated the mechanistic commonalities that never described before, including TTK, multikinase ABL, WEE1/ connect the hierarchical clusters. Because the correlations at the CHK1, and the subclasses of Aurora, PI3K, and EZH2 inhibitors root of the hierarchical clustering tree (Fig. 3) are not signif- (Table 2). This suggests that many of the novel antiproliferative icant, we studied the variation in the cellular panel profiles by approaches not only address new biochemical targets but also principal component analysis, which summarizes the variation lead to distinctly new cell line response profiles. It is expected fi in IC50 response profiles in a few essential vectors. Principal that if additional therapies are pro led, new clusters will component 1 (PC1) captures mostly differences in potency appear. To facilitate their identification, we generated an over- (Supplementary Fig. S2B). In contrast, PC2 distinguishes view of the 72 validated clusters identified so far in the liter- between inhibitors of PI3K/AKT/mTOR signalling and RAS/ ature (Supplementary Table S4C). It is clear that cancer cell line RAF/MEK signalling (Fig. 4I). PC3 distinguishes between EGFR profiling can be used to test how distinct a new therapy is, and inhibitors and Aurora inhibitors. When the compound con- to get an overview of the diversity within cancer therapy. tributions to PC2 and PC3 are plotted, it is clear that the four The finding that biochemical targeting is predictive of cluster quadrants are each occupied by PI3K/mTOR, EGFR, RAF/MEK, identity is supported by our analysis of the cellular profiles and Aurora kinase, and Bcl-2 inhibitors(Fig.4I).Thisisconsistent biochemical selectivities of panels of Aurora and PI3K inhibi- withthefourmajorbranchesintheclusteringtree(Fig.3)and tors, which shows that subclusters can relate to isoform selec- indicates that these four are the major therapeutic vectors in tivity of these inhibitor classes. Thus, there are cell line popula- our data. tions that are more sensitive to Aurora A or pan-Aurora inhibitors. Others are more sensitive to pan-PI3K, or PI3Kb-, g-, or d- selective inhibitors. It is possible that more classes of PI3K Discussion inhibitors will appear if the same inhibitor set is tested on a Here, we present the profiling of 122 compounds, of which larger panel, incorporating cell lines representing chronic lym- many are drugs or drug candidates that were never profiled phocytic leukemia, which is one of the target diseases for before (Table 1). Because of the controversies surrounding PI3Kg/d inhibitors. It must be noted that isoform selectivity reproducibility of cell panel profiling, we carefully character- is not relevant for all compounds, as for instance all non- ized variation in our assays. The metrics generated can serve as a covalent EGFR inhibitors clustertogether(Fig.3),whereas useful benchmark for future studies (Fig. 2; Supplementary some of them also have additional ERBB2 activity (35–37) Fig. S1). We showed that, using optimized experimental and and our panel comprises a.o. the ERBB2-overexpressing line data interpretation protocols, an average correlation of 0.8 AU-565 (10). between internal control profilesisattainable(Fig.1C),leading Principal component analysis shows that our data set differ- to data correlations across platforms of >0.7 (Fig. 2A and B), entiates four major different therapy areas: PI3K/mTOR (related which is a clear improvement over the literature (17, 20) and to metabolism), EGFR and MEK (related to growth factor signal- which is sufficient to produce useful clusterings. ling), Aurora (related to cell-cycle signalling), and Bcl2 (related to Correction for cell growth rate significantly improves correla- apoptosis). Intriguingly, this coincides with the antiproliferative tions. In addition to that, data interpretation appears a great subsets of hallmarks of cancer, as defined by Hanahan and source of error, something which is supported by the finding that Weinberg (Fig. 4I; ref. 47). Thus, these theoretical signalling our data correlate best with NCI-60 and JFCR-39 panels, which classes empirically emerge from our unbiased screen, which also comprise carefully manually fitted data despite having tech- supports the classification of cancer therapy with the hallmarks nologically the greatest differences with our panel. concept. Within 122 Oncolines profiles, we identified 44 clusters with In conclusion, we have addressed controversies concerning correlations >0.3. Although 44 might seem a high number of the reproducibility of cancer cell line profiling and shown that clusters, another profiling study, also using a correlation cutoff, data can contain meaningful information even if panels run over found 49 profiles from 130 agents tested (7). several years. The most direct proof is that compound profiles Of our 44 clusters, 26 are "highly validated" i.e., contain at cluster according to biochemical mechanism. Clustering not least two compounds with similar biochemical mechanisms. only reveals side activities for known compounds, such as the This again shows that the molecular target class of inhibitors EGFR activity of ibrutinib, but also shows that isoform-selective can predict the in vitro response profile (11, 13). In total, 17 of compounds are therapeutically distinguishable subclasses, as

Figure 4. Network trees of three important inhibitor classes, showing subclassiﬁcations and biochemical origins. A, EZH2 inhibitors. B, Aurora inhibitors. C, PI3K/Akt/ mTOR inhibitors. In all trees, connections are drawn between compounds with a Pearson correlation 0.5. Only compounds within two connections of the investigated compounds (red) are shown. Other circle colors represent clusters as in Fig. 3. D, Biochemical selectivity of Aurora kinase inhibitors. Literature: refs. 15, 44, 49. E, Biochemical selectivity of PI3K inhibitors. Literature: refs. 45, 46, 50. F, Drug sensitivity for PTEN-andG, PIK3CA-andH,

PIK3R1-mutant cells. The x-axis shows the geometrically averaged IC50 of mutant cells, divided by the geometrically averaged IC50 of the wild-type cells. The y-axis shows the significance (22). Orange: compounds that are significantly more potent in mutant cells. Blue: other compounds of interest. I, Second and third principal components (PC2 and PC3) of the variation in logIC50sintheprofiling data set are related to cancer hallmarks (green).

www.aacrjournals.org Mol Cancer Ther; 15(12) December 2016 3107

Uitdehaag et al.

for PI3K and Aurora inhibitors. Moreover, cell panel profiling Analysis and interpretation of data (e.g., statistical analysis, biostatistics, can identify compounds with synergistic effects, by combining computational analysis): J.C.M. Uitdehaag, J.A.D.M. de Roos, M.B.W. Prinsen, compounds with similar cellular selectivities (48), or by com- N. Willemsen-Seegers, J.R.F. de Vetter, J. Dylus, A.M. van Doornmalen, J. fi Kooijman, G.J.R. Zaman bining cluster neighbours (11). Thorough cell line pro ling is Writing, review, and/or revision of the manuscript: J.C.M. Uitdehaag, therefore a powerful in vitro tool for categorizing and under- J.A.D.M.deRoos,J.Dylus,S.J.C.vanGerwen,J.deMan,R.C.Buijsman, standing novel cancer therapeutics. G.J.R. Zaman Administrative, technical, or material support (i.e., reporting or organizing fl data, constructing databases): J. Dylus, A.M. van Doornmalen, S.J.C. van Disclosure of Potential Con icts of Interest Gerwen, J. de Man fl No potential con icts of interest were disclosed. Study supervision: G.J.R. Zaman

Authors' Contributions The costs of publication of this article were defrayed in part by the payment of advertisement Conception and design: J.C.M. Uitdehaag, G.J.R. Zaman page charges. This article must therefore be hereby marked in Development of methodology: J.C.M. Uitdehaag accordance with 18 U.S.C. Section 1734 solely to indicate this fact. Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.A.D.M. de Roos, M.B.W. Prinsen, N. Willemsen- Received June 22, 2016; revised August 8, 2016; accepted August 21, 2016; Seegers, J.R.F. de Vetter, A.M. van Doornmalen, M. Sawa published OnlineFirst September 1, 2016.

References 1. Sharma SV, Haber DA, Settleman J. Cell line-based platforms to evaluate 17. Haibe-Kains B, El-Hachem N, Birkbak NJ, Jin AC, Beck AH, Quackenbush J, the therapeutic efficacy of candidate anticancer agents. Nat Rev Cancer et al. Inconsistency in large pharmacogenomic studies. Nature 2013;504: 2010;10:241–53. 389–93. 2. Weinstein JN. Cell lines battle cancer. Nature 2012;483:544–5. 18. Fallahi-Sichani M, Honarnejad S, Heiser LM, Gray JW, Sorger PK. Metrics 3. Shoemaker RH. The NCI human tumour cell line anticancer screen. Nat other than potency reveal systematic variation in responses to cancer drugs. Rev Cancer 2006;6:813–23. Nat Chem Biol 2013;9:708–14. 4. Holbeck SL, Collins JM, Doroshow JH. Analysis of Food and Drug Admin- 19. Fourches D, Muratov E, Tropsha A. Curation of chemogenomics data. istration - approved anticancer agents in the NCI-60 panel of human tumor Nat Chem Biol 2015;11:535. cell lines. Mol Cancer Ther 2010;9:1451–60. 20. Cancer Cell Line Encyclopedia Consortium & Genomics of Drug Sensitivity 5. Dan S, Tsunoda T, Kitahara O, Yanagawa R, Zembutsu H, Katagiri T, et al. in Cancer Consortium. Pharmacogenomic agreement between two cancer An integrated database of chemosensitivity to 55 anticancer drugs and gene cell line data sets. Nature 2015;528:84–7. expression profiles of 39 human cancer cell lines. Cancer Res 2002;62: 21. Haverty PM, Lin E, Tan J, Yu Y, Lam B, Lianoglou S, et al. Reproducible 1139–47. pharmacogenomic profiling of cancer cell line panels. Nature 2016;533: 6. Nakatsu N, Nakamura T, Yamazaki K, Sadahiro S, Makuuchi H, Kanno J, 333–7. et al. Evaluation of action mechanisms of toxic chemicals using JFCR-39, a 22. Uitdehaag JCM, de Roos JADM, van Doornmalen AM, Prinsen MBW, de panel of human cancer cell lines. Mol Pharm 2007;72:1171–80. Man J, Tanizawa Y, et al. Comparison of the cancer gene targeting and 7. Neve RM, Chin K, Fridlyand J, Yeh J, Baehner FL, Fevr T, et al. A collection of biochemical selectivities of all targeted kinase inhibitors approved for breast cancer cell lines for the study of functionally distinct cancer subtypes. clinical use. PLoS One 2014;9:e92146. Cancer Cell 2006;10:515–27. 23. R Development Core Team. R: A language and environment for statis- 8. McDermott U, Sharma SV, Dowell L, Greninger P, Montagut C, Lamb J, tical computing. R Foundation for Statistical Computing. Vienna, et al. Identification of genotype-correlated sensitivity to selective kinase Austria. 2008. inhibitors by using high-throughput tumor cell line profiling. Proc Natl 24. Yadav B, Pemovska T, Szwajda A, Kulessiy E, Kontro M, Karjalainen R, et al. Acad Sci U S A 2007;104:19936–41. Quantitative scoring of differential drug sensitivity for individually opti- 9. Garnett MJ, Edelman EJ, Heidorn SJ, Greenman CD, Dastur A, Lau KW, mized anticancer therapies. Sci Rep 2014;4:5193. et al. Systematic identification of genomic markers of drug sensitivity in 25. Suzuki R, Shimodaira H. Pvclust: an R package for assessing the uncertainty cancer cells. Nature 2012;483:570–5. in hierarchical clustering. Bioinformatics 2006;12:1540–2. 10. Barretina J, Caponigro G, Stransky N, Venkatesan K, Margolin AA, Kim S, 26. Fisher RA. On the probable error of a coefficient of correlation deduced et al. The Cancer Cell Line Encyclopedia enables predictive modelling of from a small sample. Metron 1921;4:3–32. anticancer drug sensitivity. Nature 2013;483:603–7. 27. Frey BJ, Dueck D. Clustering by passing messages between data points. 11. Seashore-Ludlow B, Rees MG, Cheah JH, Cokol M, Price EV, Coletti ME, Science 2007;315:972–6. et al. Harnessing connectivity in a large scale small-molecule sensitivity 28. Csardi G, Nepusz T. The igraph software package for complex network dataset. Cancer Discov 2015;5:1210–23. research. Interjournal 2006;1695. 12. Rees MG, Seashore-Ludlow B, Cheah JH, Adams DJ, Price EV, Gill S, et al. 29. Maia ARR, de Man J, Boon U, Janssen A, Song JY, Omerzu M, et al. Correlating chemical sensitivity and basal gene expression reveals mech- Inhibition of the spindle assembly checkpoint kinase TTK enhances the anism of action. Nat Chem Biol 2016;12:109–16. efficacy of docetaxel in a triple negative breast cancer model. Ann Oncol 13. Greshock J, Bachman KE, Degenhardt YY, Jing J, Wen YH, Eastman S, et al. 2015;26:2180–92. Molecular target class is predictive of in vitro response profile. Cancer Res 30. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical 2010;70:3677–86. and powerful approach to multiple testing. J Royal Statistic Soc B 14. Choy E, Butrynski JE, Harmon DC, Morgan JA, George S, Wagner AJ, et al. 1995;57:289–300.

Phase II study of olaparib in patients with refractory Ewing sarcoma 31. Kalliokoski T, Kramer C, Gedeck P. Comparability of mixed IC50 data - a following failure of standard chemotherapy. BMC Cancer 2014;14:813. statistical analysis. PLoS One 2013;8:e61007. 15. Uitdehaag JCM, Verkaar F, Alwan H, de Man J, Buijsman RC, Zaman GJ, 32. Hafner M, Niepel M, Chung M, Sorger PK. Growth rate inhibition metrics et al. A guide to picking the most selective kinase inhibitor tool compounds correct for confounders in measuring sensitivity to cancer drugs. Nat for pharmacological validation of drug targets. Br J Pharmacol 2011; Methods 2016;13:521–7. 166:858–76. 33. Chatterji DC, Gallelli JF. Thermal and photolytic decomposition of meth- 16. Elkins JM, Fedele V, Szklarz M, Abdul-Azeez KR, Salah E, Mikolajczyk J, otrexate in aqueous solutions. J Pharm Sci 1978;67:526–31. et al. Comprehensive characterization of the published kinase inhibitor set. 34. D'Hondt M, Vangheluwe E, van Dorpe S, Boonen J, Bauters T, vanden- Nat Biotech 2016;34:95–103. Broucke J, et al. Stability of extemporaneously prepared cytarabine,

3108 Mol Cancer Ther; 15(12) December 2016 Molecular Cancer Therapeutics

Therapeutic Clusters Discovered by Cell Panel Proﬁling

methotrexate sodium, and methylprednisolone sodium succinate. Am J hematological malignancies: emerging preclinical and clinical evidence. Health Syst Pharm 2012;69:232–40. Ther Adv Hematol 2015;6:128–41. 35. Kitagawa D, Yokota K, Gouda M, Narumi Y, Ohmoto H, Nishiwaki E, et al. 43. Hartsink-Segers SA, Zwaan CM, Exalto C, Luijendijk MW, Calvert VS, Activity-based kinase profiling of approved tyrosine kinase inhibitors. Petricoin EF, et al. Aurora kinases in childhood acute leukemia: the promise Genes Cells 2013;18:110–22. of aurora B as therapeutic target. Leukemia 2013;27:560–8. 36. Liu Q, Sabnis Y, Zhao Z, Zhang T, Buhrlage SJ, Jones LH, et al. Developing 44. Dienstmann R, Rodon J, Serra V, Tabernero J. Picking the point of irreversible inhibitors of the protein kinase cysteinome. Chem Biol 2013; inhibition: a comparative review of PI3K/AKT/mTOR pathway inhibitors. 20:146–59. Mol Cancer Ther 2014;13:1021–31. 37. Davis MI, Hunt JP, Herrgard S, Ciceri P, Wodicak LM, Pallares G, et al. 45. Yap TA, Bjerke L, Clarke PA, Workman P. Drugging PI3K in cancer: refining Comprehensive analysis of kinase inhibitor selectivity. Nat Biotech targets and therapeutic strategies. Curr Opin Pharmacol 2015;23:98–107. 2011;11:1046–51 46. Lawrence MS, Stojanov P, Mermel CH, Robinson JT, Garraway LA, Golub 38. Carpinelli P, Ceruti R, Giorgini ML, Cappella P, Gianellini L, Croci V, et al. TR, et al. Discovery and saturation analysis of cancer genes across 21 PHA-739358, a potent inhibitor of Aurora kinases with a selective target tumour types. Nature 2014;505:495–501. inhibition profile relevant to cancer. Mol Cancer Ther 2007;6:3158–68. 47. Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell 39. Lovly CM, McDonald NT, Chen H, Ortiz-Cuaran S, Heukamp LC, Yan Y, 2011;144:646–74. et al. Rationale for co-targeting IGF-1R and ALK in ALK fusion positive lung 48. Uitdehaag JCM, de Roos JA, van Doornmalen AM, Prinsen MBW, Spijkers- cancer. Nat Med 2014;20:1027–34. Hagelstein JA, de Vetter JR. Selective targeting of CTNNB1-, KRAS- or MYC- 40. West AC, Johnstone RW. New and emerging HDAC inhibitors for cancer driven cell growth by combinations of existing drugs. PLoS One 2015;10: treatment. J Clin Invest 2014;124:30–9. e0125021. 41. Adimoolan S, Sirisawad M, Chen J, Thiemann P, Ford JM, Buggy JJ. HDAC 49. Carpinelli P, Moll J. Aurora kinases and their inhibitors: more than one inhibitor PCI-24781 decreases RAD51 expression and inhibits homolo- target and one drug. Adv Exp Med Biol 2008;610:54–73. gous recombination. Proc Natl Acad Sci U S A 2007;104:19482–7. 50. Marone R, Cmiljanovic V, Giese B, Wymann MP. Targeting phosphoinosi- 42. Chaidos A, Caputo V, Karadimitris A. Inhibition of bromodomain and tide 3-kinase – Moving towards therapy. Biochim Biophys Acta 2008;1784: extra-terminal proteins (BET) as a potential therapeutic approach in 159–85.

www.aacrjournals.org Mol Cancer Ther; 15(12) December 2016 3109

Cell Panel Profiling Reveals Conserved Therapeutic Clusters and Differentiates the Mechanism of Action of Different PI3K/mTOR, Aurora Kinase and EZH2 Inhibitors

Joost C.M. Uitdehaag, Jeroen A.D.M. de Roos, Martine B.W. Prinsen, et al.

Mol Cancer Ther 2016;15:3097-3109. Published OnlineFirst September 1, 2016.

Updated version Access the most recent version of this article at: doi:10.1158/1535-7163.MCT-16-0403

Supplementary Access the most recent supplemental material at: Material http://mct.aacrjournals.org/content/suppl/2016/08/31/1535-7163.MCT-16-0403.DC1

Cited articles This article cites 48 articles, 11 of which you can access for free at: http://mct.aacrjournals.org/content/15/12/3097.full#ref-list-1

Citing articles This article has been cited by 2 HighWire-hosted articles. Access the articles at: http://mct.aacrjournals.org/content/15/12/3097.full#related-urls

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 Department at Subscriptions [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://mct.aacrjournals.org/content/15/12/3097. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.