Implementation of the NCI-60 Human Tumor Cell Line Panel to Screen 2260 Cancer Drug Combinations to Generate >3 Million Data Points Used to Populate a Large Matrix of Anti-Neoplastic Agent Combinations (ALMANAC) Database

Abstract

Animal and clinical studies demonstrate that cancer drug combinations (DCs) are more effective than single agents. However, it is difficult to predict which DCs will be more efficacious than individual drugs. Systematic DC high-throughput screening (HTS) of 100 approved drugs in the National Cancer Institute’s panel of 60 cancer cell lines (NCI-60) produced data to help select DCs for further consideration. We miniaturized growth inhibition assays into 384-well format, increased the fetal bovine serum amount to 10%, lengthened compound exposure to 72 h, and used a homogeneous detection reagent. We determined the growth inhibition 50% values of individual drugs across 60 cell lines, selected drug concentrations for 4 × 4 DC matrices (DCMs), created DCM master and replica daughter plate sets, implemented the HTS, quality control reviewed the data, and analyzed the results. A total of 2620 DCMs were screened in 60 cancer cell lines to generate 3.04 million data points for the NCI ALMANAC (A Large Matrix of Anti-Neoplastic Agent Combinations) database. We confirmed in vitro a synergistic drug interaction flagged in the DC HTS between the vinca-alkaloid microtubule assembly inhibitor vinorelbine (Navelbine) tartrate and the epidermal growth factor-receptor tyrosine kinase inhibitor gefitinib (Iressa) in the SK-MEL-5 melanoma cell line. Seventy-five percent of the DCs examined in the screen are not currently in the clinical trials database. Selected synergistic drug interactions flagged in the DC HTS described herein were subsequently confirmed by the NCI in vitro, evaluated mechanistically, and were shown to have greater than single-agent efficacy in mouse xenograft human cancer models. Enrollment is open for two clinical trials for DCs that were identified in the DC HTS. The NCI ALMANAC database therefore constitutes a valuable resource for selecting promising DCs for confirmation, mechanistic studies, and clinical translation.

Keywords

cancer and cancer drugs drug interactions drug combinations synergistic NCI-60

Introduction

Numerous omics studies (genomics, proteomics, transcriptomics, and metabolomics) have unveiled the complexity of mutations, gene copy number alterations, tumor suppressor inactivation, gene expression subtypes, signaling pathway modifications, and altered metabolic pathways that contribute to the development and maintenance of the malignant phenotype.^1–10 Solid tumors comprise genetically diverse and heterogeneous clonal cell populations with distinct genetic profiles and/or differentiation states.^1,2,6,7 Not only do intrinsic resistance factors preexist in subpopulations of tumor cells to confer drug resistance, but also tumor cells respond adaptively to therapy by altering drug target expression, accumulating secondary mutations, or activating feedback loops and compensatory signaling pathways to develop drug resistance.¹¹ Given the inherent genetic heterogeneity and complexity of cancers and their propensity for drug resistance, it is perhaps not surprising that most single-agent cancer treatments fail to provide prolonged patient benefits.^1–4,6,7,11 Cumulative evidence from both animal and clinical studies indicates that drug combinations (DCs) are more effective than single-agent therapies.^1–4,6,7,11 The most effective combinations of cytotoxic chemotherapies increase tumor cell killing, in either an additive or synergistic fashion, by combining agents with different molecular mechanisms and nonoverlapping toxicities to reduce the probability of resistance.^1–4,6,7,11 However, it is difficult to select and/or predict which DCs will effectively provide long-term benefits to patients beyond those afforded by the individual agents.

Traditionally, DCs have been selected empirically and evaluated in clinical trials that often lacked a biological rationale.^1–4,6,7 However, even if we were only to consider 100 Food and Drug Administration (FDA)-approved cancer drugs, there are 4950 possible pairwise DCs, which would be prohibitively expensive and time-consuming to evaluate in animal models let alone in clinical trials. Presumably not every possible DC could be justified mechanistically, thereby reducing the number of DCs under active consideration. Specific DCs might be selected by rational, hypothesis-driven approaches based on the altered oncogenic tumor targets and signaling pathways identified by omics and gene expression studies.^1–4,6,7 Unfortunately, approved cancer drugs have not yet been discovered for every mutated target or altered signaling pathway identified by these approaches. Knowledge of the intrinsic and adaptive drug resistance mechanisms of tumors might also provide a rational basis for the selection of specific DCs.¹¹ However, tumors invariably harbor multiple genetic and signaling pathway alterations, and our understanding of drug resistance and how anticancer agents work remains woefully incomplete.^1–4,6,11 While hypothesis-driven and empiric approaches will continue to be utilized, the systematic high-throughput screening (HTS) of DCs against panels of well-characterized tumor cell lines provides a complementary data-driven strategy to select DCs with the potential to become effective therapies.^1–4,6 Moreover, an unbiased HTS strategy might identify novel unanticipated effective cancer DCs through serendipity.

However, there are also some significant challenges to implementing DC HTS campaigns.^3,7,12–16 The assays must be appropriate for the tumor type of interest and should address the genetic diversity and heterogeneity inherent in those cancers.^1,6,17 This likely requires the development of growth inhibition assays in multiple tumor cell lines that meet the requirements for HTS assays: scalability with a robust and reproducible signal window and acceptable performance statistics.^18–21 Additionally, there are complex data analysis and compound handling issues associated with defining the nature of the pharmacological interactions between two drugs.^15,22–26 Analyses of DC interactions are more complex than the typical HTS data processing methods, such as percent activation, percent inhibition, or z scores. Several models have been developed to analyze and classify whether two drugs interact in an antagonistic, additive, or synergistic manner: Lowe additivity, Bliss independence, and Chou–Talalay median effect models.^15,22–26 Such analyses require testing a matrix of the two drugs over a broad range of concentrations and ratios to provide a pairwise interaction surface that is compared with the individual agent responses.^15,22–26 Additionally, the generation of pairwise DC matrices (DCMs) presents logistical challenges with respect to compound handling and tracking.¹³ There is also the issue of scale, as the number of DCs under consideration increases the unbiased screening of all possible pairwise DCs against multiple tumor cell lines, which has some significant consequences for both the tissue culture burden and amount of data that will be generated.

Since its introduction in the early 1990s, the National Cancer Institute’s panel of 60 cancer cell lines (NCI-60) has been used to screen >100,000 compounds and >50,000 natural product extracts for inhibition of tumor cell line growth and to profile the activity of FDA-approved cancer drugs.^27,28 The NCI-60 cell lines have been molecularly characterized extensively: exome sequence, DNA methylation, mRNA expression, microRNA expression, protein levels and modifications, drug transporter expression, enzyme activity, and metabolomic profiling.^27–30 Compound profiling data and the molecular characterization of the cell lines are publicly available for download from the Developmental Therapeutics Program (DTP) website (https://dtp.cancer.gov/databases_tools/bulk_data.htm).^28,31,32 Patterns of growth inhibition across the NCI-60 cell lines are similar for drugs with closely related mechanisms of action (MOAs), and using the COMPARE algorithm, the drug profiles of known anticancer agents can be used to predict the MOAs of unknown drugs with similar sensitivity and resistance activity profiles, and to provide information on drug resistance mechanisms.^27,28 For example, the ABC transporter family includes several of the best characterized drug efflux pumps that mediate drug resistance, and correlations between the mRNA expression profiles of 48 members of the ABC transporter family in the NCI-60 panel and the growth inhibitory profiles of 1429 drugs were used to identify which transporters likely conferred resistance to selected agents.³³ Whole-exome sequencing of the NCI-60 cell lines showed that the mutation pattern between cell lines was divergent and that specific variants of well-established cancer genes had profound effects on the responses to clinically relevant targeted agents, revealing pharmacogenomic correlations between specific gene variants and drug responses.²⁹ To add further value to the NCI-60 drug screen, the NCI wished to conduct an unbiased DC HTS campaign of 100 FDA-approved anticancer drugs to generate data that would be entered into the ALMANAC (A Large Matrix of Anti-Neoplastic Agent Combinations) database (https://dtp.cancer.gov/ncialmanac), which could be used for hypothesis-generating evaluations of oncology DCs.³⁰ Historically, the NCI-60 anticancer drug screen has been performed in a 96-well format where cells are exposed to compounds for 48 h in media containing 5% fetal bovine serum (FBS) and then fixed and stained with sulforhodamine B.^27,28 To provide the throughput and capacity required for a DC HTS campaign of 2620 DCMs in all 60 cell lines, we needed to miniaturize and automate the growth inhibition assays into 384-well plate format, and to incorporate a homogeneous viability detection reagent to measure growth inhibition. The NCI also stipulated that the compound exposure be lengthened to 72 h and the amount of FBS in the assay be increased to 10%. The current paper describes the reformatting of the NCI-60 cancer drug screen into 384-well plates using the homogeneous CellTiter-Glo (CTG; Promega Corporation, Madison, WI) cellular ATP detection reagent, and how a DC HTS campaign of 2620 DCMs was implemented to generate 3.04 million data points for the NCI ALMANAC database.³⁰ For the first time, we provide detailed descriptions and data for the development and optimization of the growth inhibition assays for the 60 tumor cell lines, how the individual drug concentrations were selected for the DCMs, how the DCM master and replica daughter plates were constructed and utilized in the screen, the HTS assay performance quality control (QC) data for the campaign, and how the drug interaction data were analyzed and scored. We describe a process and present data to confirm a synergistic drug interaction that was flagged in the DC HTS between the vinca-alkaloid microtubule assembly inhibitor vinorelbine (Navelbine) tartrate and the epidermal growth factor-receptor (EGF-R) tyrosine kinase inhibitor gefitinib (Iressa) in the SK-MEL-5 melanoma cell line. Seventy-five percent of the DCs examined in the screen are not currently in the clinical trials database. Selected synergistic drug interactions that were flagged in the DC HTS described herein have subsequently been confirmed by the NCI in vitro and evaluated mechanistically, and were shown to have greater than single-agent efficacy in mouse xenograft human cancer models.³⁰ Enrollment has been opened for two clinical trials for DCs that were identified in the DC HTS.³⁰ The NCI ALMANAC database therefore constitutes a valuable resource for selecting promising DCs for confirmation, mechanistic studies, and clinical translation.

Materials and Methods

Reagents

DMSO (99.9% high-performance liquid chromatography grade, under argon) was from Alfa Aesar (Ward Hill, MA). Dulbecco’s Mg²⁺ and Ca²⁺ free phosphate-buffered saline (PBS) was purchased from Corning (Tewksbury, MA).

FDA-Approved Cancer Drugs

The 100 FDA-approved cancer drugs were obtained through commercial sources and shipped to the University of Pittsburgh by the NCI DTP. The MOAs of the 100 drugs include 61 traditional cytotoxic chemotherapeutics, including DNA-damaging agents (nucleoside analog inhibitors, topoisomerase inhibitors) and microtubule-destabilizing or -stabilizing agents, as well as 37 molecularly targeted agents, inhibitors of angiogenesis, receptor tyrosine kinases, tyrosine kinases, serine/threonine kinases, nuclear receptors, histone deacetylases (HDACs), aromatases, and the proteasome. Ninety-six of the compounds were soluble in 100% DMSO, which was used to prepare stock solutions that were used to generate master and replica daughter plate sets for the individual GI₅₀ (growth inhibition by 50%) determinations and DCMs. Two stock compound solutions were prepared in sterile deionized water and two more were prepared in 0.1 N NaOH. The final concentration of DMSO or other solvents in the NCI-60 growth inhibition assays was ⩽0.2%.

Preparation of Individual Drug GI₅₀ Master and Replica Daughter Plates

Individual drug GI₅₀ values were determined in 5-point 10-fold (log) serial dilution series conducted in single wells per concentration. The maximum starting concentration selected for each drug was approximately 2 logs (100-fold) above the average GI₅₀ of the drugs across all 60 cell lines downloaded from the NCI DTP website. We used liquid handling protocols developed on our Evolution-P3 (EP3; PerkinElmer, Waltham, MA) or Bravo (Agilent Technologies, Santa Clara, CA) automated liquid handing platforms outfitted with 384-well dispensing heads to generate 5-point 10-fold serial dilution master plates in 100% DMSO. Two microliters from each well of the master plates was transferred into barcoded replica daughter plates using the EP3 or Bravo liquid handling robots outfitted with 384-well dispensing heads, and the compound plates were then sealed with aluminum foil and stored at −20 °C until use. In the pilot-phase DC HTS, we screened 20 individual drugs arrayed on one GI₅₀ master and replica daughter plate sets. The 72 drugs added in the first production phase of the DC HTS were arrayed on 2× GI₅₀ master and replica daughter plate sets, and the seven compounds added in the second production phase were arrayed on one GI₅₀ master and replica daughter plate set.

Preparation of DCM Master and Replica Daughter Plates

In the pilot phase, we screened 20 drugs in all possible pairwise DCs for a total of 190 DCMs arrayed on 9.5 DC master plates. We used the Janus MDT Multiprobe (PerkinElmer) automated liquid handling platform to array 20 individual 4 × 4 DCMs onto 384-well master plates. We generated the 10× master plates using DC work lists to direct the Multiprobe to transfer and reformat 20 μL of 1000× drug concentrations from the wells of manually prepared source plates placed on the deck into the wells of the 384-well destination master plates, also on the deck, to make DC wells with 40 µL. Twenty microliters of DMSO was added to single drug wells, and 40 µL to DMSO control wells. The 384-well transfer head of the Bravo was used to transfer 2 μL from the DC master plates into barcoded replica daughter plates, which were then centrifuged at 50× g for 5 min, sealed with aluminum foil, and stored at −20 °C until use.

In the first production phase, we screened the 20 pilot-phase drugs in combination with 72 new drugs, which involved 1440 individual 4 × 4 DCMs arrayed on 72× 384-well DC master plates. To generate DC master plates, we prepared drug A master plates that contained three 1000× drug screening concentrations for each of the 20 pilot phase A drugs by adding 220 µL of each of the three drug concentrations to the appropriate wells in eight deep-well plates. We then used the 384-well transfer head of the Bravo to stamp out 80× drug A 384-well master plates (20 µL) from the deep-well plates. DMSO and doxorubicin (Dox) controls (40 µL) were then added to these drug A master plates, and they were centrifuged (50× g), sealed with foil, and frozen at −20 °C. To generate the drug B master plates, we prepared three 1000× screening concentrations for one of the 72 B drugs and thawed one of the drug A master plates. Using a Matrix multichannel pipettor (Thermo Fisher, Waltham, MA), we transferred 20 µL of drug B to the appropriate wells on the drug A master plate to create a DC master plate. We then used the 384-well transfer head of the Bravo to stamp out 15× 2 µL replica daughter plates from the DC master plate, which were then centrifuged, sealed, and frozen at −20 °C. This process was repeated for each of the 72 B drugs.

In the second production phase, 10 new B drugs were screened in combination with 99 A drugs, which required 990 individual 4 × 4 DCMs arrayed on 49.5× 384-well DC master plates. To generate these DC master plates, we prepared 5× 384-well A drug master plates that contained a matrix of 20 different A drugs at 1000× the screening concentrations by adding 220 µL of each concentration to the appropriate wells in five deep-well plates. We then used the 384-well transfer head of the Bravo to stamp out 10× drug A master plates (20 µL) from the 5 deep-well plates and then centrifuged the plates, sealed them with foil, and froze them at −20 °C. To generate the drug B master plates, we prepared three 1000× screening concentrations for one of the 10 B drugs and transferred 120 µL of each concentration to the appropriate wells of a deep-well plate. We thawed one set of the drug A master plates (five per set) and used a Matrix multichannel pipettor to transfer 20 µL of drug B to the appropriate wells on each of the drug A master plates, thereby creating the DC master plates. The DMSO and Dox controls (40 µL) were then added to the master plates. We then used the 384-well transfer head of the Bravo to stamp out 15× 2 µL replica daughter plates from the DC master plates, which were then centrifuged, sealed with foil, and frozen at −20 °C. This process was repeated for each of the 10 B drugs.

NCI-60 Cell Lines

The NCI-60 cell lines were obtained from the NCI DTP Tumor Repository in 2010. The NCI DTP Tumor Repository performed Applied Biosystems AmpFLSTR Identifiler testing with PCR amplification to confirm consistency with the published Identifiler STR profile for each of the NCI-60 cell lines and tested them for mycoplasma contamination. After receipt of frozen cell pellets from the NCI DTP repository, the 60 cell lines were thawed and placed into tissue culture at 37 °C, 5% CO₂, and 95% humidity and expanded through several passages before centrifugation and resuspension in 10% FBS plus 10% DMSO and freezing in liquid nitrogen. The NCI-60 cell lines were maintained in RPMI 1640 medium with 2 mM l-glutamine (Invitrogen, Carlsbad, CA) supplemented with 10% FBS (Gemini Bio-Products, West Sacramento, CA) and 100 U/mL penicillin and streptomycin (Invitrogen). For QC purposes, we tracked cell doubling times during the cell expansion and HTS process and collected spent media samples for testing using the Universal Mycoplasma PCR testing kit purchased from the American Type Culture Collection (ATCC; Manassas, VA). None of the NCI-60 cell lines tested positive for mycoplasma contamination, and cells were kept in culture for no more than 20 passages.

Determination of Individual Drug GI₅₀ Values in the NCI-60 Panel

The 384-well NCI-60 cell line growth inhibition assays using the CTG homogeneous cellular ATP detection reagent were adapted from similarly formatted assays we had previously developed in melanoma, prostate cancer, and head and neck cell lines.^{19,20,34–36} On day 1 of the assay, the NCI-60 cell lines were harvested by trypsinization, centrifugation, and viable trypan blue excluding cells that were counted using a hemocytometer. Forty-five microliters of cells at the appropriate cell density were seeded into the wells of white, opaque, clear-bottomed, 384-well barcoded time 0 (T0) and time 72 h (T72) assay plates (Greiner BioOne, Monroe, NC, cat. 781098) using a Matrix multichannel pipettor or a Microflo (BioTek, Winooski, VT) bulk reagent dispenser. Assay plates were then incubated at 37 °C in 5% CO₂ and 95% humidity for 24 h. On day 2 (T0), GI₅₀ replica daughter plates containing 2 μL of each compound concentration in 100% DMSO were thawed at 37 °C and the compounds were diluted in 98 μL of serum-free RPMI 1640 media to an intermediate drug concentration (2% DMSO), and then 5 μL was transferred into the test wells of the assay plate using the 384-well transfer head on a Bravo robotic liquid handling platform; plates were centrifuged at 100× g for 1 min and returned to an incubator at 37°C in 5% CO₂ and 95% humidity for 72 h. Also, on day 2, the T0 control cell seeding plates were removed from the incubator and 25 μL of the CTG detection reagent was added to the wells using a Matrix multichannel pipettor, and after a 15 min incubation at room temperature, the relative luminescence units (RLUs) of the T0 control plates was captured on the EnVision (PerkinElmer) microtiter plate reader. On day 5, the compound-treated T72 assay plates were removed from the incubator, 25 μL of CTG was added to the wells using a Microflo bulk reagent dispenser, and after 15 min the RLUs were read on the EnVision plate reader. To analyze the growth inhibition data, we used an ActivityBase (IDBS, Guildford, UK) concentration–response template to process the raw RLU data to percent growth (PG); to generate HTS assay performance statistics, signal-to-background (S/B) ratios, and Z-factor coefficients; and to fit the data to curves and derive the GI₅₀ values. The PG was calculated using the standard NCI-60 protocol, and as described previously.^27,28,35

\begin{array}{l} PG = [(Ti - Tz) / (C - Tz)] \times 100 for concentrations \\ for which Ti > / = Tz \end{array}

\begin{array}{l} PG = [(Ti - Tz) / Tz] \times 100 for concentrations \\ for which Ti < Tz \end{array}

where Ti is the compound well test value at 72 h, Tz is the average of the test values from the T0 control plate at time zero (n = 384), and C is the average of the test values (n = 64) in the T72 DMSO (0.2%) control wells. GraphPad Prism 5 software (GraphPad Software, La Jolla, CA) was used to plot and fit data to curves using the sigmoidal dose–response variable slope equation:

\begin{array}{l} Y = Bottom + [Top - Bottom] / \\ [1 + 10^(LogEC 50 - X) * HillSlope] \end{array}

where Bottom is the Y value at the bottom plateau, Top is the Y value at the top plateau, LogEC50 is the X value when the response is halfway between the bottom and top, and HillSlope describes the steepness of the curve. The GI₅₀ value represents the concentration at which cell growth was inhibited by 50%, the total growth inhibition (TGI) value represents the concentration at which cell growth was fully inhibited, and the LC₅₀ value represents the concentration at which 50% of the cells were killed.

DCM Screening in the NCI-60 Panel

The 384-well NCI-60 cell line DC growth inhibition assays using the CTG detection reagent were adapted from assays we had previously developed in melanoma cell lines.³⁵ On day 1 of the assay, the NCI-60 cell lines were harvested, counted, and seeded at the appropriate cell density into T0 and T72 384-well assay plates using either a Map-C2 (Titertek, Huntsville, AL) or a Multidrop (Thermo Fisher, Waltham, MA) bulk dispenser. On day 2, the T0 control cell seeding plates were removed from the incubator, 25 μL of the CTG was added to the wells, and the RLU signals were acquired as described above. Also on day 2, DC replica daughter plates prepared as described above were thawed at 37 °C and diluted in 98 μL of serum-free RPMI 1640 media using a MAP-C2 dispenser to an intermediate drug concentration (2% DMSO), and then 5 μL was transferred into the test wells of the T72 assay plates using the 384-well transfer head on a Bravo robotic liquid handling platform, and the plates were then returned to an incubator at 37 °C in 5% CO₂ and 95% humidity for 72 h. On day 5, the compound-treated T72 assay plates were removed from the incubator, 25 μL of CTG was added to the wells using a Microflo bulk reagent dispenser, the RLU signals were captured on the EnVision, and the PG was calculated using the standard NCI-60 protocol^27,28 as described above.

Confirmation of DCs Scored Synergistically via HTS

We arrayed two 10 × 10 DCMs onto 384-well master plates. Each DCM included 9 × 9 DC wells (81 total) together with 9 wells (18 total) containing each of the corresponding individual drug concentrations, and one DMSO control well. Two 10 × 10 DCMs were arrayed in columns 3–22 of the 384-well plates, together with DMSO (0.2%) controls in columns 1, 2, 23, and 24, and a 5-point Dox GI₅₀ curve arrayed in columns 23 and 24 as described above. The NCI-60 cell line growth inhibition assays were conducted as described above and previously.³⁵

Drug Interaction Score to Analyze HTS Drug Interactions

The Bliss independence and Loewe additive models for analyzing combinations of two drugs are described by eq 1, which represents a Bliss additive effect on cell growth.^22–25

FAB = FA + FB - FA \times FB

(1)

where the fraction of cells affected (FA) is the cell loss fraction (CLF) or percent growth inhibition (%GI) at X concentration of drug A alone, FB is the CLF at Y concentration of drug B alone, and FAB is the expected CLF of an independent (assumes no interaction between the drugs) theoretical combination of X concentration of drug A with Y concentration of drug B. FAB′ is the experimentally observed CLF for each DC. When FAB′ > FAB, more growth inhibition/cell death was observed than if the drugs were acting independently, and synergism is indicated. If FAB′ < FAB, less growth inhibition/cell death was observed than if the drugs were acting independently, and antagonism is indicated. In cases where FAB′ = FAB, the conclusion is that the two drugs are additive. However due to the inherent variability in experiments, exact FAB, FA, and FB can never be measured.

To take advantage of our HTS strategy where individual drugs are tested in multiple replicates as well as in combination, we developed a drug interaction (DI) score to classify the interaction status of two drugs:

\begin{matrix} DI = (F'_{AB} - (μ_{A} + μ_{B} - μ_{A} \times μ_{B})) /^{2} \\ \sqrt ((N_{1} - 1) {δ_{A}}^{2} + (N_{2} - 1) {δ_{B}}^{2}) / (N_{1} {+ N}_{2} - 2) \end{matrix}

(2)

where F′_AB is the experimentally observed %GI for each DC, µ_A and µ_B are the sample means of measured replicates of %GI or CLF at X and Y concentrations of drug A or drug B alone, δA and δB are the sample standard deviations of measured replicates of drug A or drug B alone, and N1 and N2 are the number of sample replicates of drug A or drug B alone. The DI score analysis was implemented in MATLAB (MathWorks, Natick, MA). The DI score calculation (eq 2) is a modification of the Bliss independence model (eq 1).^22–25 The numerator represents the experimentally observed %GI for each DC in the DCM minus the %GI for a Bliss additive effect calculated from the sample means of all measured replicates of the individual concentrations of drug A and drug B alone. It is the difference between the experimentally observed %GI of a DC and the calculated Bliss additive effect of replicate means of the two drugs individually. The denominator of eq 2 takes advantage of the individual drug concentration replicates and uses the sample standard deviations of measured replicates of drug A and drug B alone to calculate a value for the variability associated with the individual drug measurements, which when divided into the difference between the DC %GI and the calculated Bliss independence additivity for the individual drugs can be used to classify the interactions between the two drugs. A DI >3 indicates synergy, a DI < –3 indicates antagonism, and for −3 < DI < 3 the interaction is additive.

Chou–Talalay Median Effect Model to Analyze Drug Interactions

The Chou–Talalay median effect model accounts for the dose responses of the two interacting drugs to determine the combination effect.^23,24 Equation 3 describes the derivation of the combination index (CI) score:

CI = (D_{1} {/ D}_{X 1}) + (D_{2} {/ D}_{X 2})

(3)

where D₁ and D₂ denote the doses of compounds 1 and 2 required to reach an effect of X% as individual drug treatments, while D_X1 and D_X2 are the doses needed in combination to inhibit X%, respectively. DCs with CI > 1 exhibit antagonistic interactions, DCs with CI = 1 exhibit additivity, and DCs with CI < 1 exhibit synergistic interactions. The COMPUSYN freeware program was utilized to calculate CI values and evaluate DC synergy as described previously.³⁵

Isobologram Plots of Drug Interactions

Isobologram plots are simple graphic approaches to determine whether drug interactions are additive.³⁷ In a plot of equally effective dose pairs termed isoboles for a single effect, when a specific effect level such as 50% of maximum is selected, the doses of drug A and drug B alone that produce this effect are plotted as axial points in a Cartesian plot. When a combination of drugs is simply additive, then the locus of points (dose pairs) that produce this effect form a straight line connecting A to B. If the line connecting the actual dose pairs that produces the selected effect level experimentally significantly diverges from a straight line between A and B, then the DC is not additive. The isobologram plots were constructed in MATLAB.

Pharmacodynamic Drug Interaction Model

The pharmacodynamic drug interaction model has been used to describe the relation between drug effects such as %GI or CLF.³⁸ Equation 4 describes the derivation of the drug interaction effect F:

\begin{matrix} F = F_{\max} \times {(\begin{array}{l} C_{A} / {IC}_{50 A} + C_{B} / {IC}_{50 B} + α \times C_{A} / \\ {IC}_{50 A} \times C_{B} / {IC}_{50 B} \end{array})}^{n} / \\ {(\begin{array}{l} C_{A} / {IC}_{50 A} + C_{B} / {IC}_{50 B} + α \times C_{A} / \\ {IC}_{50 A} \times C_{B} / {IC}_{50 B} \end{array})}^{n} + 1 \end{matrix}

(4)

where F_max is the maximal effect of drug A and drug B, C_A and C_B are the concentrations of drug A and drug B, IC_50A and IC_50B are the individual drug concentrations that induce 50% of the max growth inhibition or cell loss, n is the slope of the response surface, and α is a parameter that characterizes the synergistic status of drug interaction. When α = 0, the drug interaction is additive. If α > 0, the drug interaction is synergistic, and if α < 0, the drug interaction is antagonistic. The pharmacodynamic drug interaction model analysis and plots were produced in MATLAB.

Z-Score Analysis

We employed a z-score statistical scoring method to analyze the distribution of DI scores >3 in the NCI-60 pilot DC HTS data grouped by either tumor tissue type, cell line, or DCM: Z score = (X_i – $\bar{X}) /$ σ,^36,39,40 where X_i is the raw measurement of the ith variable, and $\bar{X}$ and σ are the mean and standard deviation of all the sample measurements.

Results

Reformatting the NCI-60 Cell Line Growth Inhibition Assays

We needed to miniaturize the NCI-60 growth inhibition assays into 384-well plate format, and to utilize a homogeneous viability detection reagent to measure growth inhibition. In addition, we were required to increase the amount of FBS in the media from 5% to 10% and lengthen the compound exposure from 48 to 72 h. We wanted to select a cell seeding density that under the altered 384-well assay conditions would demonstrate active cell proliferation throughout the 96 h cell incubation period and minimize the tissue culture burden. We employed a process that we previously used to develop 384-well growth inhibition assays in melanoma, prostate cancer, and head and neck cancer cell lines using the homogenous CTG detection reagent to measure cellular ATP levels.^{19,20,34–36} We conducted two independent cell seeding density experiments in each of the 60 cell lines to select suitable cell seeding densities for the 96 h 384-well proliferation assays ( Fig. 1 and Table 1 ). We show the data for three representative cell lines that exhibited slow, medium, and fast growth phenotypes, respectively, in the 384-well assays: the NCI-H322M non-small-cell lung cancer cell (NSCLC) line, the Colo-205 colon cancer cell line, and the PC-3 prostate cancer cell line. Four cell seeding densities for each cell line were tested to compare the dynamic range of the T0 (24 h) to T72 (96 h) S/B ratio, and the Z′-factor coefficients, and to investigate the linearity of the CTG RLU signals at each cell seeding density over time in culture ( Table 1 and Fig. 1A ). Figure 1A illustrates that the CTG RLU signals for all three cell lines were linear (r² ⩾ 0.98) with respect to the number of viable cells seeded into the wells of the 384-well assay plates immediately after cell harvest. Similar data were observed with the other 57 cell lines. Figure 1B illustrates that at the cell seeding densities selected for the 384-well assay format the NCI-H322M, Colo-205, and PC-3 cell lines exhibited exponential growth throughout the 96 h incubation period and produced doubling times of 62.9, 33.5, and 30.2 h, respectively. Again, similar data were observed in the other 57 cell lines. Table 1 shows the HTS assay performance statistics for the assay development seeding density experiments. For the Colo-205 and PC-3 cell lines with medium- and fast-growing phenotypes, respectively, the lower cell seeding densities produced larger S/B ratios and Z′-factor coefficients >0.5, while assay performance declined at the higher seeding densities, presumably due to nutrient depletion and/or waste product buildup. For the slower-growing NCI-H322M cells, the lower cell seeding densities also produced larger S/B ratios, but in general higher seeding densities were required to achieve Z′-factor coefficients >0.5. Based on these data, we selected two cell seeding densities for further evaluation in a second experiment, which also included Dox GI₅₀ determinations. For all three cell lines, both cell seeding densities selected produced acceptable S/B ratios and Z′-factor coefficients, and high-quality curve fits and Dox GI₅₀ values ( Table 1 and Fig. 1C ). The Dox GI₅₀ values for the Colo-205 and PC-3 cell lines reformatted into 384-well format were slightly less sensitive and shifted <3-fold higher than the mean Dox GI₅₀ values for the NCI 96-well 48 h assay format performed in 5% FBS. In contrast, the Dox GI₅₀ for the NCI-H322M 384-well format assay was more sensitive and shifted 6.4-fold lower than the posted mean Dox GI₅₀ values in the 96-well NCI assay format. Based on these data ( Fig. 1 and Table 1 ), we selected cell seeding densities for the 384-well assay format of 469, 469, and 1250 cells/well for the PC-3, Colo-205, and NCI-H322M cell lines, respectively.

Figure 1.

Reformatting the NCI-60 cancer cell line growth inhibition assays into 384-well CTG format. (A) Linearity of CTG signal versus viable cell number. Linear regression analysis of CTG RLUs produced by seeding the indicated numbers of viable cells into the wells of 384-well assay plates for the following NCI-60 cell lines: PC-3 (○), Colo-205 (■), and NCI-H322M (□). The mean CTG RLUs ± SD (n = 24) of 24 replicate wells at each seeding density are presented together with the corresponding linear regression analysis of the data; r² > 0.98 for all three cell lines. Representative experimental data from one of at least two independent experiments are shown. (B) Cell growth and doubling times. Exponential cell growth analysis of the CTG RLUs produced after cells from the following NCI-60 cell lines were seeded into the wells of 384-well assay plates and cultured in an incubator for the indicated times: PC-3 (○), Colo-205 (■), and NCI-H322M (□). Cell lines were seeded at 469, 469, and 1250 cells/well for the PC-3, Colo-205 and NCI-H322M cell lines, respectively. The mean CTG RLUs ± SD (n = 24) of 24 replicate wells at each time point are presented together with the corresponding exponential cell growth analysis of the data; r² > 0.95 for all three cell lines. The NCI-H322M, Colo-205, and PC-3 cell lines produced doubling times of 62.9, 33.5, and 30.2 h, respectively. Representative data from one experiment are shown. (C) Dox growth inhibition curves. The three NCI-60 cell lines were seeded into T0 and T72 384-well assay plates and cultured in an incubator for 24 h. After 24 h, the CTG RLU signal was captured for the T0 control wells and the indicated concentrations of Dox were transferred into the test wells of the T72 assay plates, which were then returned to the incubator for 72 h before the CTG RLUs were captured. The PG was calculated using the T0 (n = 64) and T72 (n = 64) DMSO control wells and the standard NCI-60 protocol as described in the Materials and Methods section. The GI₅₀ value represents the concentration at which cell growth was inhibited by 50%, the TGI value represents the concentration at which cell growth was fully inhibited, and the LC₅₀ value represents the concentration at which 50% of the cells were killed. The mean PG ± SD (n = 4) data from quadruplicate wells for each Dox concentration are presented at two different seeding densities per cell line: PC-3 938 cells/well (○), PC-3 469 cells/well (•), Colo-205 938 cells/well (□), Colo-205 469 cells/well (■), NCI-H322M 2500 cells/well (∆), and NCI-H322M 1250 cells/well (▲). Representative data from one experiment are shown.

Table 1.

NCI-60 Cell Line 384-Well Assay Development HTS Performance Statistics.

Cell Line/Experiment No.	384-Well Seeding Density	S/B Ratio	Z-Factor Coefficient	Dox GI₅₀, µM
NCI-H322M/1	5000	1.2	−0.39	ND
NCI-H322M/1	2500	2.0	0.52	ND
NCI-H322M/1	1250	2.3	0.52	ND
NCI-H322M/1	625	2.5	0.38	ND
NCI-H322M/2	2500	1.7	0.54	0.71
NCI-H322M/2	1250	2.6	0.54	0.88
Colo-205/1	3750	2.1	0.29	ND
Colo-205/1	1875	3.2	0.58	ND
Colo-205/1	938	4.3	0.64	ND
Colo-205/1	469	5.7	0.80	ND
Colo-205/2	938	5.2	0.59	0.47
Colo-205/2	469	7.2	0.63	0.79
PC-3/1	1875	2.5	0.76	ND
PC-3/1	938	4.0	0.81	ND
PC-3/1	469	5.4	0.79	ND
PC-3/1	234	6.2	0.69	ND
PC-3/2	938	4.2	0.83	1.21
PC-3/2	469	5.8	0.81	0.97

S/B ratio = T0 control wells (n = 384) to T72 DMSO control wells (n = 64). Z-factor coefficient = T0 control wells (n = 384) to T72 DMSO control wells (n = 64). ND = no data/not determined.

Validation of the 384-Well NCI-60 Cell Line Growth Inhibition Assays

To further validate the NCI-60 384-well growth inhibition assay format, we conducted GI₅₀ determinations across the 60 cell lines with 20 drugs selected for the pilot phase of the DC HTS ( Fig. 2 and Table 2 ). This is also a critical step in the DC HTS process because the GI₅₀ drug response profiles are used to help select the individual drug concentrations for the DCMs (see below). Individual drug GI₅₀ values in the 384-well assays were determined in 5-point 10-fold serial dilution series conducted in single wells per concentration starting at a maximum concentration ~100-fold higher than the average GI₅₀ values that they produced across all 60 cell lines in the NCI format assay. The individual graph panels presented in Figure 2A show the GI₅₀ curve fits in all 60 cell lines for the 20 drugs. The drug response profiles revealed GI₅₀ curves that can be subdivided into three or four categories. In the drug-resistant category of response profiles, lenalidomide and altretamine exhibited no evidence of growth inhibition in any of the cell lines at the concentrations tested. This was almost true for megestrol and dacarbazine, although there were indications of growth inhibition in a small subset of cell lines at the two highest concentrations tested. In the mixed drug-resistant/sensitivity category, the 60 cell lines exhibited a wide range of growth inhibition response profiles ranging from very sensitive to resistant, spanning 2–4 logs of drug concentrations and eliciting a variety of effects from growth inhibition to total growth inhibition and lethality. Ten of the drugs exhibited mixed drug-resistant/sensitivity response profiles: pemetrexed, melphalan, dexrazoxane, vinorelbine tartrate, chlorambucil, uracil nitrogen mustard, daunorubicin, teniposide, etoposide, and mitoxantrone. In the drug-sensitive category, the 60 cell lines exhibited a much more consistent growth inhibition response profile, typically demonstrating sensitivity within a much narrower range ⩽2 logs of drug concentrations and tending to produce more profound effects, such as total growth inhibition and lethality. Six of the drugs exhibited drug sensitivity response profiles: 5-fluorouracil, lomustine, exemestane, raloxifene, gefitinib, and thioguanine. Figure 2B shows that there is a significant correlation between the mean GI₅₀ values across the 60 cell lines for the 20 pilot drugs and Dox in both the 384-well and NCI assay formats: Pearson r = 0.84, two-tailed p value <0.0001, and r² = 0.7. The pattern of growth inhibition across the NCI-60 cell lines has been shown to be similar for drugs that have closely related MOAs.^27,28 A two-way hierarchical clustering analysis of the individual GI₅₀ values for the 20 pilot drugs and Dox against the 60 cell lines grouped by tumor type revealed that agents with similar MOAs clustered together: etoposide, daunorubicin, mitoxantrone, Dox, and teniposide ( Fig. 2C and Table 2 ).

Figure 2.

Validation of the 384-well NCI-60 cell line growth inhibition assays. (A) Representative individual GI₅₀ curves for the pilot HTS drugs. The NCI-60 cell lines were seeded into T0 and T72 384-well assay plates and cultured in an incubator for 24 h. After 24 h, the CTG RLU signal was captured for the T0 control wells (n = 64) and the indicated concentrations of the 20 pilot-phase drugs were transferred into the test wells of the T72 assay plates, which were then returned to the incubator for 72 h before the CTG RLUs were captured. The PG was calculated using the T0 (n = 64) and T72 (n = 64) DMSO control wells and the standard NCI-60 protocol as described in the Materials and Methods section. GraphPad Prism 5 software was used to plot and fit data to curves using the sigmoidal dose–response variable slope equation. (B) Correlation between the log mean GI₅₀ values for the 384-well and historical NCI-60 assays. The correlation plot between the mean GI₅₀ values across the 60 cell lines for the 20 pilot-phase drugs and Dox in both the 384-well and NCI assay formats is presented; the Pearson correlation coefficient r was 0.84 for the 21 XY pairs, with a two-tailed p value <0.0001, and r² = 0.7 for the linear regression. (C) Hierarchical clustering analysis of the individual GI₅₀ data for the pilot HTS drugs. A two-way hierarchical clustering analysis of the individual GI₅₀ values for the 20 pilot drugs and Dox against the 60 cell lines grouped by tumor type shows drug clusters that respond with similar growth response patterns across tissue types and cell lines.

Table 2.

DC Pilot HTS: Anticancer Agents, Mode of Action, and DCM Concentrations.

NSC No.	Name	Mode of Action	High µM	Middle µM	Low µM	Human Cmax µM	Serial Dilution Factor
752	Thioguanine	Antimetabolite	10	1	0.1	0.59	10
3088	Chlorambucil	Alkylates and cross-links DNA	100	10	1	1.62	10
8806	Melphalan	Alkylates and cross-links DNA	10	1	0.1	?	10
13875	Altretamine	Alkylating agent	75	7.5	0.75	520	10
19893	5-Fluorouracil	Antimetabolite/thymidylate synthase inhibitor	100	10	1	426	10
34462	Uracil nitrogen mustard	Alkylating agent	50	5	0.5	?	10
45388	Dacarbazine	Alkylates and cross-links DNA	50	5	0.5	34	10
71423	Megestrol acetate	Progesterone mimic	30	3	0.3	0.35	10
79037	(CCNU) lomustine	Alkylates and cross-links DNA	50	5	0.5	3600	10
82151	Daunorubicin	DNA intercalator/topoisomerase inhibitor	0.2	0.02	0.002	0.16	10
122819	Teniposide	Topoisomerase II inhibitor/DNA strand breaks	1	0.1	0.01	45	10
141540	Etoposide	Topoisomerase II inhibitor/DNA strand breaks	5	0.5	0.05	33	10
169780	Dexrazoxane	Iron chelator, chemo- and cardioprotective	100	10	1	136	10
279836	Mitoxantrone	DNA intercalator/topoisomerase inhibitor	0.5	0.05	0.005	1.25	10
608210	Vinorelbine tartrate	Tubulin polymerization inhibitor	1	0.1	0.01	0.59	10
698037	Pemetrexed	Antifolate thymidylate synthase inhibitor	1	0.1	0.01	306	10
713563	Exemestane	Aromatase inhibitor	100	1	0.01	0.03	100
715055	Gefitinib	EGF-R tyrosine kinase inhibitor	10	1	0.1	0.97	10
747972	Lenalidomide	Angiogenesis inhibitor of TNFα/VEGF/bFGF	30	3	0.3	3	10
747974	Raloxifene	Selective estrogen receptor modulator	30	3	0.3	0.56	10

NSC No. = the identifying number assigned by DTP to an agent or product. High µM = the top concentration in micromoles of the drug used in the DCMs. Middle µM = the intermediate/middle concentration in micromoles of the drug used in the DCMs. Low µM = the lowest concentration in micromoles of the drug used in the DCMs. Human Cmax µM = the human peak plasma concentration in micromoles at the FDA-approved clinical dose. Serial Dilution Factor = the serial dilution factor between the top, middle, and lowest concentrations of the drug used in the DCMs.

DCM 384-Well Plate Layout

On each 384-well assay plate, 20 4 × 4 DCMs were arrayed in columns 3–22 together with DMSO (0.2%) controls in columns 1, 2, 23, and 24, and a 5-point 10-fold serial dilution Dox GI₅₀ curve arrayed in quadruplicate wells in columns 23 and 24 ( Fig. 3A ). Including the DMSO control well in each DCM, there were a total of 64× DMSO control wells on each plate. Each individual DCM included 3 × 3 DC wells (9 total) together with 3 wells (6 total) containing each of the corresponding individual drug concentrations, and 1 DMSO control well ( Fig. 3B ). The three drug concentrations in the DCMs were selected based on the response profiles produced in the individual GI₅₀ determinations and their clinical relevance. If a single-agent concentration produces too much growth inhibition by itself, then there is little or no ability to detect a DC effect. A top concentration was selected that achieved ⩽30% growth inhibition in the majority (⩾70%) of NCI-60 cell lines. Where possible, concentrations were selected to be within the range of the human peak plasma concentration (Cmax) at the FDA-approved clinical dose ( Table 2 ). In most cases, the middle and low drug concentrations represent 10-fold serial dilutions of the top concentration in the DCM, and this helped to ensure that even the most sensitive lines had at least two concentrations where growth was not highly inhibited. Table 2 shows the high, middle, and low concentrations that were selected for the 20 drugs and 190 DCMs that were screened in the pilot-phase DC HTS.

Figure 3.

384-well DC plate layout. (A) Plate controls and 20 4 × 4 DCMs. On each 384-well assay plate, 20 4 × 4 DCMs were arrayed in columns 3–22 together with DMSO (0.2%) controls in columns 1, 2, 23, and 24, and a 5-point 10-fold serial dilution Dox GI₅₀ curve arrayed in quadruplicate wells in columns 23 and 24. Including the DMSO control well in each DCM, there were a total of 64× DMSO control wells on each plate. (B) 4 × 4 DCMs. Each individual DCM included 3 × 3 DC wells (9 total) together with 3 wells (6 total) containing each of the corresponding individual drug concentrations, and 1 DMSO control well.

DCM HTS Implementation

On most days of DC HTS operations, 10 NCI-60 cell lines were screened against 10 DC replica daughter plates in 100× 384-well assay plates. The pilot-phase DC HTS was completed in 6 days of HTS operations while the first and second production phases were accomplished in 42 and 30 days of operations, respectively. In the pilot phase of the DC HTS, 190 individual 4 × 4 DCMs made from all possible pairwise combinations of 20 drugs were arrayed as 11,400 DCMs on 600× 384-well assay plates to generate 230,400 data points, including plate controls. In production phase I, the 20 pilot-phase drugs were screened in combination with 72 other drugs in 1440 individual 4 × 4 DCMs arrayed on 4320× 384-well assay plates, and the 86,400 DCMs and controls produced 1,658,880 data points. In the second production phase, 10 drugs were screened in combination with 99 drugs from the previous phases in 990 individual 4 × 4 DCMs arrayed on 3000× 384-well assay plates and the 59,400 DCMs and controls generated 1,152,000 data points.

DCM HTS QC Data Review

Figure 4 shows the HTS assay performance statistics and QC data for the three representative NCI-60 cell lines with slow, medium, and fast growth phenotypes that were used to illustrate the assay development process ( Fig. 1 and Table 1 ). Each NCI-60 cell line was screened on a total of 132× 384-well assay plates that were run in batches of 10 of 12 plates on 13 different days of HTS operations throughout the DC campaign. DC HTS plates that achieved one or more of the following QC criteria were failed and scheduled for retesting: plates where the S/B ratio was >2-fold different than cumulative means, plates where the Z′-factor coefficient was <0.25, and plates where the Dox GI₅₀ values were >3-fold different from cumulative means. Cumulative mean values for the S/B ratios and Dox GI₅₀ values were updated throughout the HTS campaign from assay development and GI₅₀ determinations, and throughout the three DC HTS phases. Although the T0 to T72 S/B ratios of the three cell lines varied from day to day, either above or below their respective cumulative average, the S/B ratios within each day of operations were consistent ( Fig. 4A ). We believe that the day-to-day variability of the S/B ratios was due to small daily variances in the trypan blue excluding viable cell counts manually obtained using a hemocytometer. On average, the S/B ratios for the PC-3, Colo-205, and NCI-H322M cell lines during the DC HTS campaign were 5.8-, 7.2-, and 3.5-fold, respectively, reflecting their relative growth rates in 384-well plates. The corresponding average T0 to T72 Z′-factor coefficients for the PC-3, Colo-205, and NCI-H322M cell lines during the DC HTS campaign were 0.63, 0.76, and 0.60, respectively ( Fig. 4B ), indicating that the assays performed very well. None of the 132× PC-3 assay plates produced Z′-factor coefficients <0.5, and although 14 (10.6%) of the Colo-205 assay plates exhibited Z′-factor coefficients <0.5, none were <0.42. Thirty (22.7%) of the NCI-H332M assay plates exhibited Z′-factor coefficients <0.5, but only two dropped below 0.25. Most of the NCI-H332M assay plates with Z′-factor coefficients <0.5 occurred during the pilot phase and early on in production phase I when the assay signal windows (S/B ratios) were also lower. The average Dox GI₅₀ values for the PC-3, Colo-205, and NCI-H322M cell lines during the DC HTS campaign were 0.556, 0.069, and 0.649 µM, respectively ( Fig. 4C ). There was <2-fold drift in the Dox GI₅₀ across the 132× NCI-H332M assay plates, and <3-fold variation in the PC-3 cell line. In the Colo-205 cell line, however, the Dox GI₅₀ values for 7 (5.3%) of the 132× assay plates were >3-fold higher than the average GI₅₀. In the pilot-phase DC HTS, only 5 of the 60 cell lines met one or more of the QC fail criteria and had to be repeated at the end of the pilot screen. During the first and second DC HTS production phases, only a small number of cell lines failed and were subsequently repeated. Only qualified DC HTS data that passed our QC review criteria were uploaded to the NCI-60 share point site.

Figure 4.

DC HTS assay performance QC statistics. The PG for the DC HTS campaign was calculated using the T0 and T72 DMSO control wells and the standard NCI-60 protocol as described in the Materials and Methods section. To monitor assay performance, for each assay plate the T0 (n = 384) and T72 (n = 64) DMSO control wells were used for calculation. (A) T0 to T72 S/B ratios. (B) T0 to T72 Z′-factor coefficients. (C) Dox GI₅₀ values. The 5-point 10-fold serial dilution Dox control curve arrayed in quadruplicate wells in columns 23 and 24 on each assay plate was used to calculate GI₅₀ values. Representative data from all three phases of the DC HTS campaign are presented for the following NCI-60 cell lines: PC-3 (○), Colo-205 (■), and NCI-H322M (□). The mean values for the respective measurements across 132 assay plates from all three phases are presented by the different dotted lines: PC-3 (-..-..-..), Colo-205 (-----), and NCI-H322M (-.-.-.-.).

Pilot DC HTS Results

Figure 5A shows a heatmap of the DI scores for the 190× DCMs in the NCI-60 pilot DC HTS. Cell lines are grouped by tumor tissue type on the Y axis, and the DCM number is on the X axis. DI scores were used to flag whether the two drugs in a DCM were interacting synergistically (DI scores >3, green), were interacting antagonistically (DI scores < –3, red), or were additive (–3 < DI < 3, blue). In the pilot DC HTS, 1366 (1.3%) out of 102,600 DC wells produced DI scores >3, and their drug interactions were flagged synergistic. The NCI-60 cell line panel comprises nine distinct cancer lineages with six to nine cell lines from each tumor type, except for prostate cancer, which is only represented by two cell lines. When grouped by tumor tissue type, the number of DI scores >3 was close to the average of 152 for five of the nine tumor lineages, with colon and melanoma exhibiting ~33% and 100% higher than the average, and ovarian and prostate producing ~50% and ~66% lower than the average ( Fig. 5B ). In the individual cell lines, the mean number of DI scores >3 was 23. A z-score analysis of the distribution of DI scores >3 in the individual cell lines indicated that 54 (90%) of the cell lines were within 1 SD of the mean (z scores between –1 and 1), and only six cell lines produced z scores between 1.4 and 3.7 ( Fig. 5C ): two colon cancer cell lines, KM12 and SW-620; one leukemia cell line, MOLT-4; and three melanoma cell lines, M14, MDA-MB-435, and UACC-62. Based on Figure 5B , C , there does not appear to be an obvious tissue type or cell line bias in the distribution of DI scores >3 in the NCI-60 DC pilot HTS. With respect to DCMs, 38 (20%) of the 190 DCMs failed to produce a single DC well with a DI score >3. A z-score analysis of the distribution of DI scores >3 in the remaining 152 individual DCMs produced an average of 9 per DCM and demonstrated that 11 (5.8%) DCMs produced z scores between 1.6 and 4.7 and accounted for 54% of DI scores >3 ( Fig. 5D ). Furthermore, the 11 DCMs were formed between a small number of interacting drugs, each participating in three or four of the DCMs: megestrol acetate, mitoxantrone, vinorelbine tartrate, raloxifene, gefitinib, and daunorubicin. These data indicate that a DCM bias exists in the distribution of DI scores >3 in the NCI-60 DC pilot HTS.

Figure 5.

DC pilot HTS results. (A) DI score heatmap. A heatmap of the DI scores for the 190 DCMs in the NCI-60 pilot DC HTS is presented. The 60 cell lines were grouped by tumor tissue type on the Y axis, and the DCM number is indicated on the X axis. DI scores calculated as described in the Materials and Methods section were used to flag whether two drugs in the 3 × 3 DC wells of a DCM were interacting synergistically or antagonistically or were additive. DI scores >3 indicated that the two drugs interacted synergistically (green), while DI scores < –3 indicated that the two drugs interacted antagonistically (red). Where −3 < DI < 3, the drug interaction was considered additive (blue). (B) Number of DI scores >3 by tumor tissue type. The average number of DI scores >3 across all nine tumor types was 152 (- - - - -). The number of DI scores >3 indicating synergy per tumor tissue type is indicated on the Y axis, and the tumor tissue types are indicated on the X axis. The number of cell lines per tumor tissue type is indicated in brackets at the top of each bar, representing the total number of DI scores >3 per tumor tissue type. (C) Number of DI scores >3 by NCI-60 cell line. The average number of DI scores >3 per NCI-60 cell line was 23, ranging from 2 on the low end to 81 on the high side. The results of a z-score analysis of the distribution of DI scores >3 versus the NCI-60 cell line is presented with the z score on the Y axis and cell lines on the X axis. (D) Number of DI scores >3 by DCMs. The average number of DI scores >3 per DCM was 9, ranging from 1 on the low end to 92 on the high side. The results of a z-score analysis of the distribution of DI scores >3 versus DCM is presented with z score on the Y axis and DCMs on the X axis.

Confirmation of Synergistic Drug Interactions Flagged in the Pilot DC HTS

We selected a single DCM example with DC wells that had produced DI scores >3 to illustrate the process to confirm synergistic drug interactions flagged in the DC HTS. Three of the DC wells in DCM 12 between the vinca-alkaloid microtubule assembly inhibitor vinorelbine (Navelbine) tartrate and the EGF-R tyrosine kinase inhibitor gefitinib (Iressa) produced DI scores >3 in the SK-MEL-5 melanoma cell line ( Fig. 6A ). Individual drug control wells from DCM 12 match up well with the replicate data from other DCMs, and the small error bars indicate the reproducibility of the HTS data. The DC wells of 0.1 µM vinorelbine tartrate in combination with all three gefitinib concentrations (A2-B1, A2-B2, or A2-B3) had a greater than additive effect on SK-MEL-5 PG. To confirm the synergistic interactions between vinorelbine and gefitinib in SK-MEL-5 cells, we prepared a more extensive 10 × 10 DCM between the two drugs and analyzed the growth inhibition/fraction of SK-MEL-5 cells affected using the COMUSYN software to calculate CI scores ( Fig. 6B ). For fixed DC ratios of gefitinib/vinorelbine of 20:1 and 10:1, most of the CI scores were <1.0, confirming that the two drugs interacted synergistically. We also plotted the data as an isobologram contour graph ( Fig. 6C ) and used the pharmacodynamic drug interaction model to fit the experimental data and construct a three-dimensional (3D) graph of the data ( Fig. 6D ). The synergistic interaction between gefitinib and vinorelbine was also confirmed by the concave contour lines of the isobologram plot, and because the fitted α parameter of the pharmacodynamic drug interaction model was 0.77 and therefore >0.

Figure 6.

Confirmation of gefitinib-vinorelbine flagged synergy. (A) Pilot HTS DCM data for vinorelbine (drug A) were screened individually and in DC wells at concentrations of 1.0 (A1), 0.1 (A2), and 0.01 µM (A3), while the gefitinib (drug B) concentrations were 10.0 (B1), 1.0 (B2), and 0.1 (B3) µM. The PG indicated on the Y axis for the three individual drug concentrations of vinorelbine (light grey bars) and gefitinib (dark grey bars), together with the nine DC wells (black and white bars) from the 4 × 4 DCM, is presented along with the means ± SD (n = 19) of the replicate data from the three individual drug concentrations screened in other DCMs. The three DC wells with DI scores >3 are indicated by the white bars (A2-B1, A2-B2, and A2-B3). Whether the bar represents an individual drug concentration of A or B or the DC (A-B) well from the DCM, or the average of 19 replicates of the individual drug concentrations, is indicated on the X axis. (B) CI analysis. The growth inhibition/fraction of SK-MEL-5 cells affected data from a more extensive 10 × 10 vinorelbine and gefitinib DCM and was analyzed in the Chou–Talalay median effect model using the COMPUSYN software to calculate CI scores and fit curves for the 20:1 and 10:1 fixed ratios of gefitinib and vinorelbine. CI scores <1.0 indicate that two drugs interact synergistically. (C) Isobologram plot. The data from the 10 × 10 vinorelbine and gefitinib DCM were plotted as an isobologram contour graph. The concave contour lines of the isobologram plot indicate that vinorelbine and gefitinib interacted synergistically. (D) Drug–drug interaction surface response. The data from the 10 × 10 vinorelbine and gefitinib DCM were analyzed using the pharmacodynamic drug–drug interaction model to fit the experimental data and construct a 3D graph of the data. The fitted α parameter of the pharmacodynamic drug–drug interaction model was 0.77, which is >0 and indicates that vinorelbine and gefitinib interacted synergistically.

Discussion

The NCI-60 panel represents nine major cancer lineages, and together with two other cancer cell line panels that have been assembled, they have been utilized for cancer drug screening in conjunction with multitiered omics profiling; the GlaxoSmithKline (GSK) panel of >300 cancer cell lines, which encompasses 24 tumor tissue types; and the Cancer Cell Line Encyclopedia (CCLE) of 947 cancer cell lines, which were derived from 21 distinct cancer types.^28,31,32 We believe that the NCI-60 DC HTS campaign described herein represents the first application of any of these large, extensively characterized tumor cell line panels to screen for synergistic and effective combinations of cancer drugs.³⁰ DC screening is more challenging than typical HTS strategies, including those that utilize compound mixtures (pooled libraries) or that are intended to find compounds that interact with a specific small molecule or RNAi to elicit synthetic lethal effects.^41–43 Relatively few anticancer DC HTS campaigns have been reported, and none were implemented for a full unbiased HTS campaign of all possible DCs.^{12–14,16,44–47} One group screened a single 9 × 9 DCM composed of a MEK inhibitor and an ERK inhibitor against 16 cancer cell lines of different lineages.¹³ Another group screened seven dose combinations of an EGFR/ERBB2 inhibitor with a PI3K/mTOR inhibitor against 65 cancer cell lines.¹⁶ A third group screened 6 × 6 DCMs formed between 459 compounds and a single drug, Bruton’s tyrosine kinase inhibitor ibrutinib, against two B-cell tumor cell lines.¹⁴ In a fourth study, 420 DCs formed between 7 primary and 60 secondary drugs were screened against 13 tumor cell lines: 7 bladder cancer, 3 head and neck cancer, and 3 prostate cancer cell lines.¹² The Combinatorix group screened 5 × 5 DCMs formed from ~600 approved drugs against the A549 NSCLC cell line to generate 100,000 DC data points.⁴⁴ A sixth group screened 583 4 × 4 DCMs formed by 22 experimental cancer drugs in all pairwise combinations, as well as in combination with 16 approved drugs in a panel of 39 cancer cell lines.⁴⁵ Heske et al. screened 946 6 × 6 DCMs formed from 679 anticancer agents against four Ewing sarcoma cell lines.⁴⁷ The 679 anticancer drugs had individually displayed concentration-dependent growth inhibitory activity against all four of the Ewing sarcoma cell lines.⁴⁷ He et al. describe an experimental-computational DC analysis pipeline that has been used to test and score the effects of DCs in cancer cells.⁴⁶ The pipeline uses the FIMMcherry computer program to produce and visualize combination assay plates that can be directly exported to the robotic system for automated dispensing, and uses the R-package SynergyFinder software to implement synergy scoring models, including highest single agent (HAS), Loewe, Bliss, and zero interaction potency (ZIP).⁴⁶ The pipeline method was implemented for screening 8 × 8 DCMs formed from 525 small-molecule cancer drugs against four cancer cell lines: two breast, one ovarian, and one acute monocytic leukemia cell line.⁴⁶ During this DC HTS campaign, we screened a total of 2620 4 × 4 DCMs formed from 100 FDA-approved drugs in the NCI-60 panel to generate 3.04 million data points that were uploaded to the NCI and used to populate the ALMANAC database ( Table 3 ).

Table 3.

NCI-60 DC HTS Campaign Summary.

DC HTS Phase	Pilot	Prod I	Prod II	Total
DCMs	190	1440	990	2620
DCM plates	9.5	72	49.5	131
DCMs tested	11,400	86,400	59,400	157,200
DC wells	102,600	777,600	534,600	1,414,800
Assay plates	600	4320	3000	7920
Data points generated	230,400	1,658,880	1,152,000	3,041,280

Pilot = pilot phase of the DC HTS campaign. Prod I = production phase I of the DC HTS campaign. Prod II = production phase II of the DC HTS campaign. Total = all three phases of the DC HTS campaign combined.

Tumor cell line panels that have been annotated both genetically and pharmacologically have provided tremendous value for the preclinical stratification and prioritization of anticancer agents.^{27–29,31,32} However, these panels also have some limitations. Tumor cell lines adapted to growth and maintained in two-dimensional (2D) culture conditions for extensive periods, some >20 years, proliferate more quickly than cells from primary tumors, exhibit different drug resistance profiles, and favor the discovery of antiproliferative agents while neglecting the self- and population-renewing tumor stem cells that contribute to recurrence and metastasis.^48–52 A genomic comparison of HNSCC cell lines to primary tumors revealed that although 51% of genetic alterations were shared at similar mutation frequencies, subsets of mutations were unique to patient tumors and cell lines, respectively, and those unique to cell lines favored immortalization and continuous maintenance in tissue culture.¹⁷ For solid tumors, it would be desirable to utilize 3D tumor cultures where extensive cell–cell and cell extracellular matrix interactions mimic the solid tumor architecture, morphology, and their ability to develop microenvironments with differential zones of proliferation due to oxygen, nutrient, and pH gradients that affect drug permeability, distribution, and efficacy.^53–55 NCI-60 cell lines cultured in 96-well ultra-low-attachment microtiter plates for 72 h exhibited four distinct morphologies: 28 formed condensed multicellular tumor spheroids (MCTSs) with smooth and even peripheries, 19 formed rounded MCTSs with uneven perimeters, 12 formed cell aggregates with irregular outer margins, and CAKI-1 cells formed irregularly shaped loose aggregates of cells.⁵⁶ Although HT29 colon cancer MCTSs were more resistant than monolayers to topotecan and Dox, A549 NSCLC monolayers and MCTSs were equally sensitive to carboplatin and MK1775.⁵⁶ In head and neck cancer MCTSs, ellipticine, idarubicin, daunorubicin, and Dox exhibit restricted drug penetration and distribution gradients, accumulating preferentially in cells in the outer layers of MCTSs, and monolayer cultures are 6-, 20-, 10-, and 16-fold more sensitive than MCTSs to growth inhibition by these drugs, respectively.⁵⁵ From a logistical standpoint, screening the NCI-60 panel in MCTSs would have increased the cell culture burden by ~5- to 10-fold, and the use of MCTSs might have significantly altered the sensitivity/resistance profiles of many compounds relative to historical 2D NCI-60 data.

To accommodate the scale of the NCI-60 DC HTS campaign, we needed to miniaturize the cell line growth inhibition assays into 384-well format and select suitable cell seeding densities where cells were actively proliferating throughout the 96 h culture period ( Fig. 1 and Table 1 ). The assay conditions were also altered from the NCI format to increase the FBS to 10%, lengthen the compound exposure period to 72 h, and incorporate the CTG homogeneous detection reagent. In each phase of the DC HTS, we first had to determine the GI₅₀ values of the individual drugs across all 60 cell lines ( Fig. 2 ), select the individual drug concentrations that would be used in the 4 × 4 DCMs ( Table 1 ), create DCM master and replica daughter sets ( Fig. 3 ), implement the screen, review the data for QC ( Fig. 4 ), and analyze the results ( Figs. 5 and 6 ). Reformatting of the NCI-60 growth inhibition assays into 384-well format and the three phases of the DC HTS campaign were conducted over a 2-year project period.

One limitation of the NCI-60 DC HTS campaign was that the 4 × 4 DCMs restricted the analysis of the drug interaction space to only 3 × 3 DCs. Another limitation was that only a co-administration treatment schedule was utilized. We anticipate that miniaturization into 1536-well format^14,45,47 and the application of acoustic compound dispensing^13,14 could greatly facilitate the implementation of future DC HTS campaigns. A variety of tumor cell line growth inhibition assays have been successfully implemented in 1536-well formats for quantitative HTS⁵⁷ and DC HTS,^14,45,47 and the use of higher-density plates would be expected to reduce both the cell culture burden and compound requirements, and might also enable larger DCMs with more DC concentrations to be implemented in the primary HTS, thereby enhancing the drug interaction space analysis. The ability to screen a larger DCM might also facilitate the selection of the drug concentration ranges for the matrices, especially for drugs that exhibited a wide range of sensitive/resistant response profiles spanning 2–4 logs of drug concentrations. Although a small number of the NCI-60 cell lines failed one or more of the QC review criteria during the DC HTS campaign and had to be repeated, overall the reformatted 384-well assays performed very well ( Figs. 1 and 3 and Table 1 ) and would likely perform equally well in 1536-well plates. In addition to reducing compound requirements, the use of acoustic compound dispensing to generate DCMs and transfer them into assay plates provides both efficiency and flexibility.^13,14 Acoustic compound dispensers also have the potential to reduce costs by eliminating the purchase of liquid handling platform pipette tips, although that will be offset by the cost of purchasing the acoustic dispensing instrument. The DC HTS field may be further enabled through the implementation of experimental-computational DC analysis pipelines that incorporate computer programs to produce and visualize combination assay plates that can be directly exported to the robotic system for automated dispensing, together with software to implement a variety of synergy scoring analysis models.⁴⁶

Several methods have been utilized to analyze drug interactions in DC HTS campaigns: HAS, Bliss and Loewe additivity models, ZIP, isobologram analysis, Chou–Talalay median effect models, and population-based lifetime modeling.^{12–14,16,44–46} In this paper, we used a DI score calculation (eq 2) that is a modification of the Bliss independence model. The numerator represents the difference between the experimentally observed %GI for each DC and the %GI for a Bliss additive effect calculated from the sample means of all measured replicates of the individual concentrations of drug A and drug B alone. In the pilot DC HTS, each individual drug concentration was measured in 19 replicate wells. The denominator uses the sample standard deviations of measured replicates of drug A and drug B alone to calculate a value for the variability associated with the individual drug measurements, which was then divided into the difference between the DC %GI and the calculated Bliss independence additivity for the individual drugs, and to classify the interactions between the two drugs; DI >3 indicates synergy, DI < –3 indicates antagonism, and for −3 < DI < 3 the interaction is additive. In the pilot DC HTS, 1.3% of the DC wells exhibited DI scores >3 and were flagged as synergistic ( Fig. 5A ). Although there did not appear to be a substantial tumor tissue type or cell line bias in the distribution of synergistic DI scores ( Fig. 5B , C ), there was an apparent DCM bias produced by six drugs that each participated in three or four DCMs ( Fig. 5D ).

Just as with any primary HTS, synergistic DC interactions flagged in the screen need to be confirmed. We have confirmed several of the DCMs flagged as synergistic in the pilot DC HTS in larger higher-density concentration coverage 10 × 10 DCMs (manuscript in preparation). To illustrate this process, we showed data for DCM 12 between the microtubule assembly inhibitor vinorelbine and the EGF-R tyrosine kinase inhibitor gefitinib because three of the DC wells in the HTS produced DI scores >3 in the SK-MEL-5 melanoma cell line ( Fig. 6A ). We analyzed the 10 × 10 DCM data using three independent methods. The Chou–Talay median effect model CI scores ( Fig. 6B ), isobologram contour plots ( Fig. 6C ), and pharmacodynamic drug interaction model ( Fig. 6D ) all confirmed the synergistic interactions between gefitinib and vinorelbine for inhibiting the growth/viability of SK-MEL-5 melanoma cells. Gefitinib and vinorelbine combinations have been investigated previously in preclinical studies using tumor cell line models of head and neck squamous cell carcinoma (HNSCC) and NSCLC.^58,59 In a panel of six HNSCC tumor cell lines, the combination of gefitinib plus vinorelbine had a supra-additive cytotoxic effect in four of six cell lines, and additive cytotoxic effects in the other two cell lines.⁵⁸ Gefitinib and vinorelbine are both metabolized by the CYP3A4 P450 isoenzyme, and when co-administered in vivo, drug–drug interactions may alter their respective exposure levels.⁵⁸ In 17 NSCLC cell lines that included 4 that overexpressed P-glycoprotein (Pgp/MDR1) and 3 with sensitizing EGF-R mutations, combinations of gefitinib with vinorelbine exhibited synergism (mean CIs <1) in cell lines without sensitizing EGF-R mutations.⁵⁹ Although the synergism between gefitinib and vinorelbine was more robust in the Pgp-overexpressing NSCLC cell lines, it was also apparent in cell lines where Pgp expression was below detection limits.⁵⁹ Since vinorelbine is a Pgp substrate and gefitinib a Pgp inhibitor, it seems likely that blocking the efflux pump active resistance mechanism contributed to the synergy apparent between these two agents in Pgp-overexpressing NSCLC cell lines, but that other mechanisms of active resistance were also affected by this DC.⁵⁹

The NCI utilized a different modification of the Bliss independence model to analyze the DC HTS data that were generated in the DC HTS campaign described herein.³⁰ Positive “ComboScores” indicated greater than additive in vitro activity, and while most DCs exhibited selective activity for a subset of cell lines, some were active across most of the NCI-60 cell lines, and except for leukemia cell lines, DC sensitivity was independent of cellular histology.³⁰ Based on the DC ComboScores in the ALMANAC database, the ability of the specific cell lines to grow as xenograft implants, and their potential clinical utility, a subset of DCs were selected for follow-up in human tumor xenograft models.³⁰ Five of 13 nonnovel DCs exhibited greater than single-agent efficacy in at least one xenograft model and have demonstrated appreciable efficacy in clinical trials.³⁰ Fourteen of 44 unique DC plus xenograft model experiments, designed to test 20 novel DCs flagged in the ALMANAC database, exhibited significant antitumor efficacy for the DCs compared with both single agents (p < 0.05), and for 7 more tumor regressions appeared superior for DCs than for single agents.³⁰ DCs of the purine nucleoside analog clofarabine and the proteasome inhibitor bortezomib that were identified in our DC HTS campaign and flagged as synergistic in the ALMANAC database were tested in five solid tumor xenograft models and exhibited significant tumor regressions in two colon cancer models (HCT116 p < 0.01, HT29 p < 0.001), and in the NCI-H522 NSCLC model (p < 0.01).³⁰ However, despite showing greater than additive efficacy in the ovarian cancer cell line OVCAR-3 and melanoma line M14 in vitro, neither the single agents nor clofarabine and bortezomib DCs were effective in the corresponding xenograft models.³⁰ Mechanistic studies provided evidence that DNA damage, cell cycle arrest, and apoptosis contributed to the efficacy of the bortezomib–clofarabine combination in the responsive HCT-116 xenograft model.³⁰ A phase I trial has been initiated for the bortezomib–clofarabine combination in patients with advanced refractory myelodysplastic syndromes and lymphomas (NCT02211755). DCs of the microtubule-stabilizing agent paclitaxel and the BCR-Abl kinase inhibitor nilotinib from the DC HTS campaign produced positive ComboScores in the ALMANAC database in all six hematopoietic tumors and three of the triple-negative breast cancer cell lines, but negative ComboScores in two estrogen receptor-positive breast cancer lines.³⁰ The nilotinib–paclitaxel DC was well tolerated, highly efficacious, and superior to both single agents in the triple-negative MDA-MB-468 breast cancer xenograft model and produced complete tumor regression with no tumor regrowth observed for >80 days after therapy ended.³⁰ Biomarker studies failed to identify any obvious positive associations between NCI-60 molecular markers and nilotinib–paclitaxel ComboScores, which suggested that their synergy does not involve nilotinib-mediated reduction in paclitaxel efflux via ABC transporter inhibition.³⁰ A phase I trial has been initiated for the nilotinib–paclitaxel combination in adults with refractory solid tumors (NCT02379416).

Almost 75% of the DC pairs in the NCI-60 HTS campaign data that were uploaded to the NCI and used to populate the ALMANAC database (https://dtp.cancer.gov/ncialmanac) have not been registered in the clinical trials database (clinicaltrials.gov).³⁰ Since only FDA-approved anticancer agents were screened, selected DCs have the potential for rapid clinical translation. The ability to confirm selected synergistic drug interactions flagged in the DC HTS both in vitro and to demonstrate that they had greater than single-agent efficacy in mouse xenograft human cancer models in vivo³⁰ indicates that the NCI ALMANAC database constitutes a valuable resource for selecting promising DCs for confirmation, mechanistic studies, and clinical translation.

Footnotes

Acknowledgements

The authors would like to thank Richard DeBiasio and Seia Comsa for their expert technical support throughout the NCI-60 DC screening project.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This project was funded in part with federal funds from the National Cancer Institute, National Institutes of Health, under contract no. HHSN2612005522002C, NCI-N01-CM-52202, ARRA-funded project Work Assignment 14, “In Vitro Screening of Combinations of Chemotherapy Drugs.” 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.

References

Al-Lazikani

Banerji

Workman

Combinatorial Drug Therapy for Cancer in the Post-Genomic Era. Nat. Biotechnol. 2012, 30, 679–692.

Dancey

Chen

H. X.

Strategies for Optimizing Combinations of Molecularly Targeted Anticancer Agents. Nat. Rev. Drug Discov. 2006, 5, 649–659.

Keith

Borisy

A. A.

Stockwell

B. R.

Multicomponent Therapeutics for Networked Systems. Nat. Rev. Drug Discov. 2005, 4, 71–78.

Kummar

Chen

H. X.

Wright

, et al. Utilizing Targeted Cancer Therapeutic Agents in Combination: Novel Approaches and Urgent Requirements. Nat. Rev. Drug Discov. 2010, 9, 843–856.

The Cancer Genome Atlas Network. Comprehensive Genomic Characterization of Head and Neck Squamous Cell Carcinomas. Nature 2015, 517, 576–582.

Ocaña

Pandiella

Personalized Therapies in the Cancer “Omics” Era. Mol. Cancer 2010, 9, 202–214.

Rodon

Perez

Kurzrock

Combining Targeted Therapies: Practical Issues to Consider at the Bench and Bedside. Oncologist 2010, 15, 37–50.

Stransky

Egloff

A. M.

Tward

A. D.

, et al. The Mutational Landscape of Head and Neck Squamous Cell Carcinoma. Science 2011, 333, 1157–1160.

Sun

Califano

J. A.

Sequencing the Head and Neck Cancer Genome: Implications for Therapy. Ann. N.Y. Acad. Sci. 2014, 1333, 33–42.

10.

Walter

Yin

Wilkerson

M. D.

, et al. Molecular Subtypes in Head and Neck Cancer Exhibit Distinct Patterns of Chromosomal Gain and Loss of Canonical Cancer Genes. PloS One 2013, 8, e56823.

11.

Holohan

Van Schaeybroeck

Longley

D. B.

, et al. Cancer Drug Resistance: An Evolving Paradigm. Nat. Rev. Cancer 2013, 13, 714–726.

12.

Axelrod

Gordon

V. L.

Conaway

, et al. Combinatorial Drug Screening Identifies Compensatory Pathway Interactions and Adaptive Resistance Mechanisms. Oncotarget 2013, 4, 622–635.

13.

Chan

Wilson

Schmidt

, et al. Unlocking the Potential of High-Throughput Drug Combination Assays Using Acoustic Dispensing. J. Lab. Autom. 2015, 21, 125–132.

14.

Mathews Griner

Guha

Shinn

, et al. High-Throughput Combinatorial Screening Identifies Drugs That Cooperate with Ibrutinib to Kill Activated B-Cell-Like Diffuse Large B-Cell Lymphoma Cells. Proc. Natl. Acad. Sci. U.S.A. 2014, 111, 2349–2354.

15.

Mayer

Janoff

A. S.

Optimizing Combination Chemotherapy by Controlling Drug Ratios. Mol. Interv. 2007, 7, 216–223.

16.

Peifer

Weiss

Sos

M. L.

, et al. Analysis of Compound Synergy in High-Throughput Cellular Screens by Population-Based Lifetime Modeling. PLoS One 2010, 5, e8919.

17.

Wawrose

J. S.

Gooding

W. E.

, et al. Genomic Analysis of Head and Neck Squamous Cell Carcinoma Cell Lines and Human Tumors: A Rational Approach to Preclinical Model Selection. Mol. Cancer Res. 2014, 12, 571–582.

18.

Johnston

P. A.

Cellular Platforms for HTS: Three Case Studies. Drug Discov. Today 2002, 7, 353–363.

19.

Johnston

Sen

Hua

, et al. High-Content pSTAT3/1 Imaging Assays to Screen for Selective Inhibitors of STAT3 Pathway Activation in Head and Neck Cancer Cell Lines. Assay Drug Dev. Technol. 2014, 12, 55–79.

20.

Johnston

Sen

Hua

, et al. HCS Campaign to Identify Selective Inhibitors of IL-6-Induced STAT3 Pathway Activation in Head and Neck Cancer Cell Lines. Assay Drug Dev. Technol. 2015, 13, 356–376.

21.

Johnston

P. A.

Shinde

S. N.

Hua

, et al. Development and Validation of a High-Content Screening Assay to Identify Inhibitors of Cytoplasmic Dynein-Mediated Transport of Glucocorticoid Receptor to the Nucleus. Assay Drug Dev. Technol. 2012, 10, 432–456.

22.

Bliss

The Toxicity of Poisons Applied Jointly. Ann. Appl. Biol. 1939, 26, 585–615.

23.

Chou

Talalay

Quantitative Analysis of Dose-Effect Relationships: The Combined Effects of Multiple Drugs or Enzyme Inhibitors. Adv. Enzyme Regul. 1984, 22, 27–55.

24.

Chou

Drug Combination Studies and Their Synergy Quantification Using the Chou-Talalay Method. Cancer Res. 2010, 70, 440–446.

25.

Loewe

The Problem of Synergism and Antagonism of Combined Drugs. Arzneimittelforschung 1953, 3, 285–290.

26.

Zhao

J. L.

Wientjes

M. G.

Comparison of Methods for Evaluating Drug-Drug Interaction. Front. Biosci. ( Elite Ed.). 2010, 2, 241–249.

27.

Holbeck

Collins

J. M.

Doroshow

J. H.

Analysis of Food and Drug Administration-Approved Anticancer Agents in the NCI-60 Panel of Human Tumor Cell Lines. Mol. Cancer Ther. 2010, 9, 1451–1460.

28.

Shoemaker

The NCI60 Human Tumour Cell Line Anticancer Drug Screen. Nat. Rev. Cancer 2006, 6, 813–823.

29.

Abaan

Polley

E. C.

Davis

S. R.

, et al. The Exomes of the NCI-60 Panel: A Genomic Resource for Cancer Biology and Systems Pharmacology. Cancer Res. 2013, 73, 4372–4382.

30.

Holbeck

Camalier

Crowell

J. A.

, et al. The National Cancer Institute ALMANAC: A Comprehensive Screening Resource for the Detection of Anticancer Drug Pairs with Enhanced Therapeutic Activity. Cancer Res. 2017, 77, 3564–3576.

31.

Barretina

Caponigro

Stransky

, et al. The Cancer Cell Line Encyclopedia Enables Predictive Modelling of Anticancer Drug Sensitivity. Nature 2012, 483, 603–607.

32.

Kim

Yoon

Cell Line Modeling for System Medicine in Cancers. Int. J. Oncol. 2014, 44, 371–376.

33.

Szakács

Annereau

J. P.

Lababidi

, et al. Predicting Drug Sensitivity and Resistance: Profiling ABC Transporter Genes in Cancer Cells. Cancer Cell. 2004, 6, 129–137.

34.

Fancher

Hua

Camarco

D. P.

, et al. Reconfiguring the AR-TIF2 Protein-Protein Interaction HCS Assay in Prostate Cancer Cells and Characterizing the Hits from a LOPAC Screen. Assay Drug Dev. Technol. 2016, 14, 453–477.

35.

Feng

Kochanek

, et al. Design and Activity of AP Endonuclease-1 Inhibitors. J. Chem. Biol. 2015, 8, 79–83.

36.

Johnston

Nguyen

M. M.

Dar

J. A.

, et al. Development and Implementation of a High-Throughput High-Content Screening Assay to Identify Inhibitors of Androgen Receptor Nuclear Localization in Castration-Resistant Prostate Cancer Cells. Assay Drug Dev. Technol. 2016, 14, 226–239.

37.

Tallarida

R. J.

Drug Synergism: Its Detection and Applications. J. Pharmacol. Exp. Ther. 2001, 298, 865–872.

38.

Greco

Bravo

Parsons

J. C.

The Search for Synergy: A Critical Review from a Response Surface Perspective. Pharmacol. Rev. 1995, 47, 331–385.

39.

Dudgeon

Shinde

S. N.

Hua

, et al. Implementation of a 220,000 Compound HCS Campaign to Identify Disruptors of the Interaction between p53 and hDM2, and Characterization of the Confirmed Hits. J. Biomol. Screen. 2010, 15, 152–174.

40.

Johnston

Sen

Hua

, et al. HCS Campaign to Identify Selective Inhibitors of IL-6-Induced STAT3 Pathway Activation in Head and Neck Cancer Cell Lines. Assay Drug Dev. Technol. 2015, 13, 356–376.

41.

Kainkaryam

Woolf

P. J.

Pooling in High-Throughput Drug Screening. Curr. Opin. Drug Discov. Dev. 2009, 12, 339–350.

42.

Willingham

Deveraux

Q. L.

Hampton

G. M.

, et al. RNAi and HTS: Exploring Cancer by Systematic Loss-of-Function. Oncogene 2004, 23, 8392–8400.

43.

Lazo

Cutting Down the Time to Identify Challenging Tumor Therapeutic Targets and Drug Combinations Using Synthetic Lethal Approaches. F1000Research 2018, 7, 308–313.

44.

Borisy

Elliott

P. J.

Hurst

N. W.

, et al. Systematic Discovery of Multicomponent Therapeutics. Proc. Natl. Acad. Sci. U.S.A. 2003, 100, 7977–7982.

45.

O’Neil

Benita

Feldman

, et al. An Unbiased Oncology Compound Screen to Identify Novel Combination Strategies. Mol. Cancer Ther. 2016, 15, 1152–1162.

46.

Kulesskiy

Saarela

, et al. Methods for High-Throughput Drug Combination Screening and Synergy Scoring. Methods Mol. Biol. 2018, 1711, 351–398.

47.

Heske

Davis

M. I.

Baumgart

J. T.

, et al. Matrix Screen Identifies Synergistic Combination of PARP Inhibitors and Nicotinamide Phosphoribosyltransferase (NAMPT) Inhibitors in Ewing Sarcoma. Clin. Cancer Res. 2017, 23, 7301–7311.

48.

Baker

Chen

C. S.

Deconstructing the Third Dimension: How 3D Culture Microenvironments Alter Cellular Cues. J. Cell Sci. 2012, 125, 3015–3024.

49.

Ekert

Johnson

Strake

, et al. Three-Dimensional Lung Tumor Microenvironment Modulates Therapeutic Compound Responsiveness In Vitro—Implication for Drug Development. PLoS One 2014, 9, e92248.

50.

Friedrich

Seidel

Ebner

, et al. Spheroid-Based Drug Screen: Considerations and Practical Approach. Nat. Protoc. 2009, 4, 309–324.

51.

Härmä

Virtanen

Mäkelä

, et al. A Comprehensive Panel of Three-Dimensional Models for Studies of Prostate Cancer Growth, Invasion and Drug Responses. PLoS One 2010, 5, e10431.

52.

Hongisto

Jernström

Fey

, et al. High-Throughput 3D Screening Reveals Differences in Drug Sensitivities Between Culture Models of JIMT1 Breast Cancer Cells. PLoS One 2013, 8, e77232.

53.

Minchinton

Tannock

I. F.

Drug Penetration in Solid Tumours. Nat. Rev. Cancer. 2006, 6, 583–592.

54.

Sant

Johnston

P. A.

The Production of 3D Tumor Spheroids for Cancer Drug Discovery. Drug Discov. Today Technol. 2017, 23, 27–36.

55.

Shan

D. A.

Camarco

D. P.

, et al. High-Content Screening Comparison of Cancer Drug Accumulation and Distribution in Two-Dimensional and Three-Dimensional Culture Models of Head and Neck Cancer. Assay Drug Dev. Technol. 2018, 16, 27–50.

56.

Selby

Delosh

Laudeman

, et al. 3D Models of the NCI60 Cell Lines for Screening Oncology Compounds. SLAS Discov. 2017, 22, 473–483.

57.

Xia

Huang

Witt

K. L.

, et al. Compound Cytotoxicity Profiling Using Quantitative High-Throughput Screening. Environ. Health Perspect. 2008, 116, 284–291.

58.

Erjala

Raitanen

Kulmala

, et al. Concurrent Use of Vinorelbine and Gefitinib Induces Supra-Additive Effect in Head and Neck Squamous Cell Carcinoma Cell Lines. J. Cancer Res. Clin. Oncol. 2007, 133, 169–176.

59.

Tsai

Chiu

C. H.

Chang

K. T.

, et al. Gefitinib Enhances Cytotoxicities of Antimicrotubule Agents in Non-Small-Cell Lung Cancer Cells Exhibiting No Sensitizing Epidermal Growth Factor Receptor Mutation. J. Thorac. Oncol. 2012, 7, 1218–1227.

Implementation of the NCI-60 Human Tumor Cell Line Panel to Screen 2260 Cancer Drug Combinations to Generate >3 Million Data Points Used to Populate a Large Matrix of Anti-Neoplastic Agent Combinations (ALMANAC) Database

Abstract

Keywords

Introduction

Materials and Methods

Reagents

FDA-Approved Cancer Drugs

Preparation of Individual Drug GI50 Master and Replica Daughter Plates

Preparation of DCM Master and Replica Daughter Plates

NCI-60 Cell Lines

Determination of Individual Drug GI50 Values in the NCI-60 Panel

DCM Screening in the NCI-60 Panel

Confirmation of DCs Scored Synergistically via HTS

Drug Interaction Score to Analyze HTS Drug Interactions

Chou–Talalay Median Effect Model to Analyze Drug Interactions

Isobologram Plots of Drug Interactions

Pharmacodynamic Drug Interaction Model

Z-Score Analysis

Results

Reformatting the NCI-60 Cell Line Growth Inhibition Assays

Validation of the 384-Well NCI-60 Cell Line Growth Inhibition Assays

DCM 384-Well Plate Layout

DCM HTS Implementation

DCM HTS QC Data Review

Pilot DC HTS Results

Confirmation of Synergistic Drug Interactions Flagged in the Pilot DC HTS

Discussion

Footnotes

Acknowledgements

Declaration of Conflicting Interests

Funding

References

Preparation of Individual Drug GI₅₀ Master and Replica Daughter Plates

Determination of Individual Drug GI₅₀ Values in the NCI-60 Panel