Author Manuscript Published OnlineFirst on April 26, 2017; DOI: 10.1158/0008-5472.CAN-17-0489 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited. Title Page The National Cancer Institute ALMANAC†: A Comprehensive Screening Resource for the Detection of Anticancer Drug Pairs with Enhanced Therapeutic Activity Susan L. Holbeck1, Richard Camalier1, James A. Crowell1, Jeevan Prasaad Govindharajulu2, Melinda Hollingshead1, Lawrence W. Anderson1, Eric Polley1, Larry Rubinstein1, Apurva Srivastava2, Deborah Wilsker2, Jerry M. Collins1, and James H. Doroshow1,3* †A Large Matrix of Anti-Neoplastic Agent Combinations 1Division of Cancer Treatment and Diagnosis, National Cancer Institute, NIH, Bethesda, Maryland 20892 2Clinical Pharmacodynamics Program, Applied/Developmental Research Directorate, Leidos Biomedical Research Inc., Frederick National Laboratory for Cancer Research, Frederick, Maryland 21702 3Center for Cancer Research, National Cancer Institute, NIH, Bethesda, Maryland 20892 Running title: NCI ALMANAC of approved cancer drug combinations Keywords: cancer drug screening; combination therapy; chemotherapy; molecularly targeted therapy; apoptosis Financial Support: This project has been funded in whole or in part with Federal funds from the National Cancer Institute, National Institutes of Health, under Contract No. HHSN261200800001E. The content of this publication does not necessarily reflect the views or policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the U.S. Government. This research was supported [in part] by the Intramural Research Program of the NIH, National Cancer Institute, Center for Cancer Research. This research was supported [in part] by American Recovery and Reinvestment Act funds. *Corresponding author: James H. Doroshow, Division of Cancer Treatment and Diagnosis, National Cancer Institute, NIH, 31 Center Drive, Bldg. 31 Room 3A-44, Bethesda, MD 20892; Phone: 301-496- 4291; Fax: 301-496-0826; E-mail: [email protected] Disclosure of Potential Conflicts of Interest: The authors declare no potential conflicts of interest. 1 Downloaded from cancerres.aacrjournals.org on April 7, 2019. © 2017 American Association for Cancer Research. Author Manuscript Published OnlineFirst on April 26, 2017; DOI: 10.1158/0008-5472.CAN-17-0489 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited. Abstract To date, over 100 small molecule oncology drugs have been approved by the US Food and Drug Administration. Due to the inherent heterogeneity of tumors, these small molecules are often administered in combination to prevent emergence of resistant cell subpopulations. Therefore, new combination strategies to overcome drug resistance in patients with advanced cancer are needed. In this study, we performed a systematic evaluation of the therapeutic activity of over 5,000 pairs of FDA- approved cancer drugs against a panel of 60 well-characterized human tumor cell lines (NCI-60) to uncover combinations with greater than additive growth-inhibitory activity. Screening results were compiled into a database, termed the NCI-ALMANAC (A Large Matrix of Anti Neoplastic Agent Combinations), publically available at https://dtp.cancer.gov/ncialmanac. Subsequent in vivo experiments in mouse xenograft models of human cancer confirmed combinations with greater than single-agent efficacy. Concomitant detection of mechanistic biomarkers for these combinations in vivo supported the initiation of two phase I clinical trials at the NCI to evaluate clofarabine with bortezomib and nilotinib with paclitaxel in patients with advanced cancer. Consequently, the hypothesis-generating NCI- ALMANAC web-based resource has demonstrated value in identifying promising combinations of approved drugs with potent anticancer activity for further mechanistic study and translation to clinical trials. 2 Downloaded from cancerres.aacrjournals.org on April 7, 2019. © 2017 American Association for Cancer Research. Author Manuscript Published OnlineFirst on April 26, 2017; DOI: 10.1158/0008-5472.CAN-17-0489 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited. Introduction Despite recent advances in translational oncology (1), new therapies for cancer are urgently needed, as are new approaches to facilitate the rapid translation of promising preclinical data to clinical evaluation. Recognizing that combining anticancer drugs—whether cytotoxic or molecularly targeted— with distinct mechanisms of action is the approach most likely to overcome single-agent resistance and produce sustained clinical remissions (2,3), we determined that it would be of value to systematically screen pairwise combinations of 104 US Food and Drug Administration (FDA)-approved oncology drugs in the NCI-60 panel of human tumor cell lines. Our objective was to create a database, the NCI- ALMANAC (A Large Matrix of Anti Neoplastic Agent Combinations), that could be searched to identify combinations with greater antitumor activity than either agent alone. Promising combinations could then be selected for further evaluation in vitro or in human tumor xenografts models. By screening only approved drugs with proven activity and established safety profiles, combinations identified from the NCI-ALMANAC database offer potential for rapid translation into Investigational New Drug (IND)- exempt clinical trials. The NCI-60 panel is one of many drug development resources that are, and will continue to be, supported by the NCI; each of the cell lines has been extensively characterized at the molecular level; exome sequence, mRNA expression, microRNA expression, protein quantification, protein modification, DNA methylation, enzyme activity, and metabolomics data are all publicly available (4,5). These datasets enable molecular data mining in the context of combination drug studies to assess the engagement of multiple potential targets and/or downstream markers of drug action, both in animal model systems and in patients during early phase clinical trials (6). They also support research into the multiple mechanisms driving resistance to cancer therapeutic agents, including mutation or amplification of the gene encoding the drug target, enhanced drug efflux or metabolism, activation of compensatory signaling networks that bypass the effects of target engagement, changes in DNA damage response or epigenetic pathways, or alterations in the tumor microenvironment (7). 3 Downloaded from cancerres.aacrjournals.org on April 7, 2019. © 2017 American Association for Cancer Research. Author Manuscript Published OnlineFirst on April 26, 2017; DOI: 10.1158/0008-5472.CAN-17-0489 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited. Although a variety of active clinical trials are testing drug combinations based on accepted precepts of tumor cell biology, the preclinical hypotheses supporting many of these studies may be difficult to validate in patients without clearly defined mechanisms of action, resistance, and toxicity for each component of the combination, as well as for the combination itself (8). Indeed, the heterogeneity observed across individual tumors hints at the possibility of distinct co-occurring resistance mechanisms within a patient, such that different regions of the same malignancy may vary in their sensitivity (or resistance) to an individual agent at the time that treatment is initiated (9-11). Hence, developing new combination strategies to overcome drug resistance in patients with advanced cancer is a therapeutic imperative (8,12). Promising drug combinations may be discovered with synthetic lethal screens (13) utilizing siRNA or shRNA libraries, or CRISPR-Cas9 genome editing (14), to identify targets for commonly used anticancer drugs (15,16); however, in this study we utilized an unbiased, “hypothesis- free” phenotypic screening approach (8). The results of our comprehensive combination screening effort are available in the NCI-ALMANAC database, a web-based resource intended for hypothesis-generating assessments of oncology drugs combinations. Based on the outcome of this screening program, several promising drug combinations were examined in human tumor xenografts developed from NCI-60 cell lines. From our in vitro-to-in vivo translation effort, two promising drug combinations never before tested in human studies—clofarabine with bortezomib and paclitaxel with nilotinib—were chosen for clinical evaluation. 4 Downloaded from cancerres.aacrjournals.org on April 7, 2019. © 2017 American Association for Cancer Research. Author Manuscript Published OnlineFirst on April 26, 2017; DOI: 10.1158/0008-5472.CAN-17-0489 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited. Materials and Methods Cell Lines and Drug Screening NCI-60 cell lines were obtained from the NCI Developmental Therapeutics Program Tumor Repository in the mid-1990s−early 2000s and distributed for drug screening and xenograft experiments in 2010-2012. For each lot of cells, the Repository performs Applied Biosystems AmpFLSTR® Identifiler® testing with PCR amplification to confirm consistency with the published Identifiler® STR profile (17) for the given cell line. Each cell line was tested for mycoplasma when it was accepted into the repository; routine mycoplasma testing of lots is not performed. Cells are kept in continuous culture at NCI for no more than 20 passages. Oncology drugs were obtained through commercial sources. Experiments were performed at 3 locations (NCI’s Frederick National Laboratory for Cancer Research, SRI International, and University of Pittsburgh). At NCI, drug treatment was performed according to the standard NCI-60 testing protocol (https://dtp.cancer.gov/discovery_development/nci-60/default.htm), using 96-well plates and 2-day drug exposures; the endpoint is determination of Sulphorhodamine B levels in duplicate wells. For each drug pair, both single agents were tested at 5 concentrations. For drug combinations, one agent was tested at 5 concentrations together with one of 3 different concentrations of the second agent, generating 5×3 combination matrices. At SRI International and University of Pittsburgh, this protocol was modified: cells were plated in 384-well plates, fetal bovine serum was increased to 10%, CellTiter-Glo® was used as an endpoint, and drugs were tested at 3 concentrations as single agents and in 3×3 concentration matrices for combinations, in individual wells. Drug concentrations were chosen based on clinical relevance; where possible, concentrations were selected to be below the human peak plasma concentration (C ) at max the FDA-approved clinical dose and also to yield measurable activity in our assay systems. In a few cases, one or more concentrations above the C were selected to achieve some growth inhibition max in vitro. Growth percent was calculated as described in the standard NCI-60 testing protocol. Inclusion of a time zero measurement in all experiments allowed for determination of cell killing and growth 5 Downloaded from cancerres.aacrjournals.org on April 7, 2019. © 2017 American Association for Cancer Research. Author Manuscript Published OnlineFirst on April 26, 2017; DOI: 10.1158/0008-5472.CAN-17-0489 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited. inhibition. Fifty-nine of the NCI-60 cell lines passed pre-defined quality control parameters for > 90% of the drug pairs. The SK-MEL-2 melanoma cell line was an outlier; data for only 65% of the drug pairs met quality control standards as the cells grew poorly in 384-well plates at both testing locations. Development and Use of the NCI ComboScore Determination of combination benefit (“ComboScore”) utilized a modification of Bliss independence (18), as follows: Let YApBq be the growth fraction for the ith cell line exposed to the pth i concentration of drug A and the qth concentration of drug B, defined as: (cid:1846)(cid:3002)(cid:3291)(cid:3003)(cid:3292) −(cid:1846) (cid:1851)(cid:3002)(cid:3291)(cid:3003)(cid:3292) = 100∗ (cid:2869) (cid:2868) (cid:2919) (cid:1846)(cid:2868)−(cid:1846) (cid:2869) (cid:2868) where (cid:1846) is the time zero endpoint measurement, (cid:1846)(cid:3002)(cid:3291)(cid:3003)(cid:3292) is the endpoint measurement after 2 days, and (cid:1846)(cid:2868) (cid:2868) (cid:2869) (cid:2869) is the endpoint measurement after 2 days for the control well. Define YAp; YBq as the growth fractions i i when only exposed to drug A or drug B, respectively. The expected growth fraction for the combination is: min(cid:4672)(cid:1851)(cid:3002)(cid:3291),(cid:1851)(cid:3003)(cid:3292)(cid:4673) (cid:1861)(cid:1858) (cid:1851)(cid:3002)(cid:3291) ≤ 0 (cid:1867)(cid:1870) (cid:1851)(cid:3003)(cid:3292) ≤ 0 (cid:1852)(cid:3002)(cid:3291)(cid:3003)(cid:3292) = (cid:4688) (cid:3036) (cid:3036) (cid:3036) (cid:3036) (cid:3036) 1 (cid:4672)(cid:1851)(cid:3560)(cid:3002)(cid:3291) ∗(cid:1851)(cid:3560)(cid:3003)(cid:3292)(cid:4673) (cid:1867)(cid:1872)ℎ(cid:1857)(cid:1870)(cid:1875)(cid:1861)(cid:1871)(cid:1857) 100 (cid:3036) (cid:3036) where (cid:1851)(cid:3560) = (cid:1865)(cid:1861)(cid:1866)(cid:4666)(cid:1851),100(cid:4667) truncates the growth fraction at 100. The final combination score for the cell (cid:3036) (cid:3036) line and the drug combination is the sum of the differences in expected versus observed growth fractions: (cid:1851)(cid:3002)(cid:3003) = (cid:3533)(cid:1851)(cid:3002)(cid:3291)(cid:3003)(cid:3292) − (cid:1852)(cid:3002)(cid:3291)(cid:3003)(cid:3292) (cid:3036) (cid:3036) (cid:3036) (cid:3043),(cid:3044) These calculations set a ceiling of 100% growth, i.e., the expected combination growth percentage cannot be greater than the control. One screening site had a high incidence of “reversals,” where the growth percent increased as the concentration of one or both drugs was increased, i.e., apparent enhancement of growth by drug treatment. While the phenomenon of hormesis can be real, the frequency with which this was observed, at only one of the screening locations, argued against this being actual hormesis. All data for a drug pair/cell line were removed from the dataset if the growth percent increased by 50% or greater 6 Downloaded from cancerres.aacrjournals.org on April 7, 2019. © 2017 American Association for Cancer Research. Author Manuscript Published OnlineFirst on April 26, 2017; DOI: 10.1158/0008-5472.CAN-17-0489 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited. between any 2 adjacent doses of either drug. ComboScores are available for download at https://dtp.cancer.gov/ncialmanac. Evaluation of Combination Therapies Registered with Clinicaltrials.gov For initial analysis of the clinical status of all NCI-ALMANAC combinations, a drug pair was designated “clinically tested” if the two names were provided as an intervention in the same trial listed in the clinicaltrials.gov database (see Supplementary Methods for further detail). Heat Map JMP 11 (SAS institute) software was utilized for preparation of heat maps and statistical analyses. Two-way hierarchical clustering utilized the Ward method with data standardization and without imputation of missing values. Human Tumor Xenografts Animal experiments were performed at both SRI International and Frederick National Laboratory for Cancer Research. Both SRI International and Frederick National Laboratory are accredited by AAALAC International and follow the Public Health Service Policy for the Care and Use of Laboratory Animals. Animal care was provided in accordance with the procedures outlined in the “Guide for Care and Use of Laboratory Animals” (National Research Council; 2011; National Academies Press; Washington, D.C.). For mouse inoculation, tumor cells from the NCI-60 panel were used at the 4th to 6th in vitro passage of cryopreserved cell stocks. Cells (1×107 cells/0.1 mL/injection) were subcutaneously inoculated bilaterally into female athymic nu/nu NCr mice, and therapeutic studies were initiated upon reaching a target volume of 100 mm3. Sample sizes were n = 10-20 mice for vehicle-treated groups and n = 5-10 mice for single-agent- or combination-treated groups. Mice were treated with drug doses that had been demonstrated to be the MTD for the single agents and the combination in prior experiments by the NCI Developmental Therapeutics Program (see Supplementary Data for dosage regimens); in some 7 Downloaded from cancerres.aacrjournals.org on April 7, 2019. © 2017 American Association for Cancer Research. Author Manuscript Published OnlineFirst on April 26, 2017; DOI: 10.1158/0008-5472.CAN-17-0489 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited. studies, treatments at lower doses were also performed. For each mouse, the time from treatment initiation to tumor volume doubling was estimated using exponential growth interpolation when the event was between two time points. Mice for which tumor volume had not doubled by the end of study were considered censored at the last observation. The distribution of time-to-tumor-volume-doubling was estimated for each treatment cohort using the Kaplan-Meier method, and the log-rank test was used to determine significant differences between combination and single-agent treatment groups. Biomarker Analyses of Human Tumor Xenografts Analyses of apoptosis (19) and DNA damage response biomarkers (20,21) were performed as described previously (see Supplementary Methods for details). 8 Downloaded from cancerres.aacrjournals.org on April 7, 2019. © 2017 American Association for Cancer Research. Author Manuscript Published OnlineFirst on April 26, 2017; DOI: 10.1158/0008-5472.CAN-17-0489 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited. Results The NCI-ALMANAC Dataset and Web Tools Pairwise combinations of 104 FDA-approved anticancer drugs (Supplementary Table S1) were screened in each of the cell lines of the NCI-60 panel to discover those with enhanced growth inhibition or cytotoxicity profiles compared to either single agent (Fig. 1A). Combination activity was reported as a “ComboScore” (22); positive ComboScores indicated greater than additive in vitro activity; negative scores indicated less-than-additive activity (see the Methods). A total of 5,232 drug pairs were evaluated in each of the cell lines; 304,549 experiments were performed to test each drug at either 9 or 15 combination dose points, for a total of 2,809,671 dose combinations. The NCI-ALMANAC is publicly available at https://dtp.cancer.gov/ncialmanac; the data can be visualized as a heat map summarizing the entire dataset, as a bar graph of the ComboScores for a particular drug pair in all cell lines, or as dose- response curves for one drug combination in a given cell line (Supplementary Fig. S1). Finally, users can obtain an overview of which drugs or mechanistic categories of drugs perform best in combination with a given drug of interest. We first researched what fraction of the NCI-ALMANAC drug pairs appeared on clinicaltrials.gov for an estimation of clinical experience (see Methods for details). At the start of this study in June 2011, nearly 75% of the > 5,000 pairs of FDA-approved drugs examined had not been reported to clinicaltrials.gov as being part of an ongoing or completed study (Fig. 1B). Next, to identify patterns in cellular sensitivity and combination activity, we conducted a two-dimensional hierarchical clustering analysis. The resulting heat map of ComboScores across cell lines and drug combinations revealed that most drug combinations have selective activity for a subset of cell lines, as indicated by the blue-red heterogeneity throughout most of the heat map, though some combinations did exhibit patterns of predominantly greater than additive (highly red) or predominantly less-than-additive (highly blue) activity across most of the NCI-60 cell lines (Fig. 2A and B). Additionally, except for leukemia cell lines, 9 Downloaded from cancerres.aacrjournals.org on April 7, 2019. © 2017 American Association for Cancer Research. Author Manuscript Published OnlineFirst on April 26, 2017; DOI: 10.1158/0008-5472.CAN-17-0489 Author manuscripts have been peer reviewed and accepted for publication but have not yet been edited. sensitivity to drug combinations appeared to be independent of cellular histology, as cell lines from the same source tissue did not cluster together. Examination of Promising Anticancer Drug Combinations from the NCI-ALMANAC in Human Tumor Xenografts Because it was not feasible to examine all drug combinations in follow-up animal model experiments, a subset of combinations with greater than additive activity was selected for in vivo analysis based on the ComboScore, the ability of relevant cell lines to grow as xenograft implants, and clinical investigator input regarding clinical utility. An initial study was performed using doses at or near the maximum tolerated dose (MTD) of each single agent to eliminate those combinations with excessive toxicity (requiring dose reduction >50%). For the remaining combinations, the more clinically pertinent criterion of “greater-than-single-agent” activity was applied to the follow-up xenograft analyses to identify combinations that could potentially confer additional efficacy in patients. We also differentiated between “novel” combinations (those that had not been previously tested together exclusively) and “non- novel” combinations (those tested together exclusively in a clinical trial arm), according to clinicaltrials.gov and the literature. Various methods have been used to assess drug efficacy in xenograft studies, although to our knowledge, no method has been validated for comparing results of a high-throughput xenograft screen. For our purposes, we chose Kaplan-Meier and log-rank analyses of time-to-tumor-volume-doubling. Out of 13 “non-novel” combinations that were selected for xenograft analysis based on high in vitro ComboScores (Supplementary Table S2), 5 yielded greater-than-single-agent efficacy in at least one of the xenograft models tested. Furthermore, each of these 5 combinations has demonstrated appreciable efficacy (partial and/or complete responses) in clinical trials (Supplementary Table S3). Despite the small number of combinations examined, this analysis suggested that the NCI-ALMANAC results may be used as a starting point to select novel combinations for further study and potential clinical use. 10 Downloaded from cancerres.aacrjournals.org on April 7, 2019. © 2017 American Association for Cancer Research.