Sage Journals: Discover world-class research

Abstract

Cell-based phenotypic screening is a commonly used approach to discover biological pathways, novel drug targets, chemical probes, and high-quality hit-to-lead molecules. Many hits identified from high-throughput screening campaigns are ruled out through a series of follow-up potency, selectivity/specificity, and cytotoxicity assays. Prioritization of molecules with little or no cytotoxicity for downstream evaluation can influence the future direction of projects, so cytotoxicity profiling of screening libraries at an early stage is essential for increasing the likelihood of candidate success. In this study, we assessed the cell-based cytotoxicity of nearly 10,000 compounds in the National Institutes of Health, National Center for Advancing Translational Sciences annotated libraries and more than 100,000 compounds in a diversity library against four normal cell lines (HEK 293, NIH 3T3, CRL-7250, and HaCat) and one cancer cell line (KB 3-1, a HeLa subline). This large-scale library profiling was analyzed for overall screening outcomes, hit rates, pan-activity, and selectivity. For the annotated library, we also examined the primary targets and mechanistic pathways regularly associated with cell death. To our knowledge, this is the first study to use high-throughput screening to profile a large screening collection (>100,000 compounds) for cytotoxicity in both normal and cancer cell lines. The results generated here constitute a valuable resource for the scientific community and provide insight into the extent of cytotoxic compounds in screening libraries, allowing for the identification and avoidance of compounds with cytotoxicity during high-throughput screening campaigns.

Keywords

cytotoxicity ultra-high-throughput screening cell-based assays cancer and cancer drugs profiling

Introduction

The development of new chemical probes enables therapeutic target validation, hypothesis testing, and new insight into the biological role of genes and proteins.^1–3 At the National Institutes of Health (NIH) National Center for Advancing Translational Sciences (NCATS), the development of small-molecule probes for the scientific community allows understanding of rare and neglected diseases and novel targets and enables basic biological understanding of the “undrugged” genome. This is accomplished through a team science approach that begins with assay development and automated quantitative high-throughput screening (qHTS) with a small-molecule library to identify active hits.⁴ One or more chemotypes that emerge in a qHTS campaign may progress to medicinal chemistry to develop a small-molecule probe with strong biochemical and/or cell-based activity, specificity, and optimized properties to enable use in in vivo models. To accommodate unbiased qHTS discovery for medicinal chemistry, large libraries of small molecules are created and curated, containing molecules that capture diverse chemical space that are synthetically tractable. These large libraries of small molecules are generally referred to diversity libraries or collections.⁵

A second significant discovery screening strategy involves the creation of libraries of annotated small molecules. Annotated libraries contain drugs, probes, and tool molecules with one or more known mechanisms of action.⁶ They have emerged as information-rich databases to integrate both biological and chemical data. These can be screened in (primarily) cell or organism-based assays to identify targets relevant to a phenotype or for potential drug repurposing.⁷ Although compounds in a diversity library are expected to demonstrate weak biological activity, annotated libraries by definition are medicinal chemistry-optimized products with known activity and, in many cases, known mechanism of action (MOA). Both diversity collection and annotated library screening are important components of the NCATS Chemical Genomics Center (NCGC) program.⁸

Throughout a qHTS campaign, the activities of hits are confirmed in a retest, and a number of orthogonal and counterassays are performed to confirm that the observed modulatory activity of is on target. This is also to ensure that compounds demonstrating artifactual activity in an assay are triaged. For example, biochemical and cell-based assays that use firefly luciferase (fLuc) are sensitive to compounds that modulate luciferase activity.⁹ To ensure this is not the case, the screening libraries at NCATS are profiled for inhibitory activity of fLuc, allowing compounds that interfere with luciferase to be automatically triaged from hit lists without the need for additional screening. In one study, it was shown that ~5% of compounds in a qHTS library inhibit fLuc.¹⁰ Perhaps the most notorious example of the need to perform counterassays is the drug ataluren (PTC124), approved for treatment of patients with Duchenne muscular dystrophy, that may have been discovered due to fLuc inhibition rather than on-target activity.¹¹

A resurgence in cell-based screening, both target based and phenotypic, means that an increasing number of cell-based qHTS campaigns are performed. Although a number of phenotypic screening approaches to library profiling have been taken,^12–23 the profiling of a big library using direct cytotoxicity readout is less frequently reported. An early report of the assessment of compound cytotoxicity across a library of 1408 compounds appears in the literature,²⁴ and more recently, two studies examining cell killing with multiplexed assays were reported: one using a high-content assay against a screening library (diversity) of ~12,000 compounds²⁵ and the other using a multiplex assay against ~10,000 environmental toxins.²⁶ A significant collection of disease-agnostic annotated small molecules exist in libraries at NCATS, including a collection of drugs approved by American, European, and Japanese therapeutic regulatory agencies (such as the NCATS Pharmaceutical Collection²⁷) and many small molecules reported as tools, probes, or clinical/preclinical candidates. Understandably, a significant proportion of these molecules arose from oncology programs, and a portion of NCATS’s collaborations are oncology-related phenotypic screens. We were motivated to profile the activity of our annotated libraries for two reasons: first, to allow scientists to have a reference data set for discerning compounds whose activity is selective for cancer cell lines versus a set of noncancer normal cell lines and, second, to enable scientists to discriminate promiscuous/cytotoxic compounds when reviewing data from cell-based/phenotypic assays and provide valuable input for prioritizing compounds for further evaluation.

To this end, we assessed the cytotoxicity of NCATS annotated libraries of nearly 10,000 compounds against four normal cell lines (HEK 293, NIH 3T3, CRL-7250, and HaCat) and a cancer cell line (KB 3-1, a HeLa subline) and examined the hit rates and mechanistic pathways regularly associated with cell death. Furthermore, we assessed that activity of a diversity library (>100,000 compounds) against two normal cell lines (HEK 293 and NIH 3T3), assessed hit rate, and compared active compounds against a cancer cell line (KB 3-1). This study provides insight into the extent of cell-based killing activity in annotated and diversity libraries and the importance of confirming that active compounds in phenotypic screens are not cytotoxic.

Materials and Methods

Profiling Annotated/Diversity Libraries and Cherry-Picked Compounds

HEK 293, NIH 3T3, CRL-7250, HaCat, and KB 3-1 cells were seeded into white 1536-well plates using a Multidrop Combi peristaltic dispenser (ThermoFisher, Waltham, MA) at a density of 250, 400, 500, 500, and 500 cells/well in 5 µL of medium, respectively. A pin tool (Kalypsys, San Diego, CA) was used to transfer 23 nL of compound solution to the 1536-well assay plates. After a 48 h incubation at 37 °C, 5% CO, and 85% humidity, 2.5 µL of CellTiter-Glo (Promega, Madison, WI) was dispensed into each well using a dispenser (Aspect Automation, St. Paul, MN) with solenoid valves (Lee Valves, Westbrook, CT). Plates were left at room temperature for 10 min before imaging the adenosine triphosphate (ATP)–coupled luminescence using a ViewLux microplate imager (PerkinElmer, Waltham, MA).

Luciferase Assay Protocol

Assays determining firefly luciferase inhibition were performed as previously described.²⁸ Briefly, 3 µL of luciferase substrate solution (10 µM ATP, 10 µM D-luciferin, 10 mM Mg-acetate, 0.01% Tween-20, 0.05% bovine serum albumin, 50 mM Tris acetate, pH 7.6, in final 4 µL volume) was dispensed into each well of white, solid bottom, 1536-well plates using a dispenser. A pin tool (Kalypsys) was used to transfer 23 nL of compound solution to the assay plates. Following a 15-min incubation at room temperature protected from light, 1 µL of purified luciferase enzyme solution was added to a final concentration of 10 nM Photinus pyralis luciferase (Sigma). Luminescence was detected by Viewlux (PerkinElmer) using a 10 s exposure time and 2X binning.

Data Analysis and Clustering of Compounds by Activity Outcomes

To determine compound activity in the qHTS assay, the concentration-response data for each sample was plotted and modeled by a four-parameter logistic fit yielding EC₅₀ and efficacy (maximal response) values as previously described.²⁹ Raw plate reads for each titration point were first normalized relative to positive control (9.2 µM bortezomib, –100% activity, full inhibition) and DMSO-only wells (basal, 0% activity). Data normalization and curve fitting were performed using in-house informatics tools. Compounds were designated as class 1 to 4 according to the type of concentration-response curve (CRC) observed. In brief, class –1.1 and –1.2 were the highest-confidence complete CRCs containing upper and lower asymptotes with efficacies ≥80% and <80%, respectively. Class –2.1 and –2.2 were incomplete CRCs having only one asymptote with efficacy ≥80% and <80%, respectively. Class –3 CRCs showed activity at only the highest concentration or were poorly fit. Class 4 CRCs were inactive, having a curve fit of insufficient efficacy or lacking a fit altogether.

Compounds were further clustered hierarchically using TIBCO Spotfire 6.0.0 (Spotfire Inc., Cambridge, MA; https://spotfire.tibco.com/) based on their activity outcomes from the primary or follow-up screen across different cell lines. The compound’s area under the dose-response curve (AUC), calculated based on the qHTS data analysis and curve fittings, was used for clustering. In the heat map, the darker color indicates compounds that are more potent and efficacious (i.e., high-quality actives), and the lighter color indicates less potent and efficacious compounds. If a compound did not show any activity in an assay, it was highlighted as white in the heat map.

Statistical Analysis

To determine whether the hits predominantly identified from the specific therapeutic categories were overrepresented in the chemical library, an enrichment analysis was implemented against the drug library. A total of 9893 compounds in the annotated library were broken down to different therapeutic categories based on their primary mechanisms of action and pharmaceutical indications, and the enrichment was calculated from the following formula: E = a/n, given a is the number of actives and n is the total number of drugs in each therapeutic category. Fisher’s exact test was used as a measure of the consensus cytotoxicity potential of compounds in each MOA; all calculations were performed in R statistical computing software (https://www.r-project.org/).

Results

Profiling of an Annotated Library

Annotated libraries were profiled for cell viability by performing a primary screen against four normal cell lines using CellTiter-Glo (CTG) as the assay readout (screening assay protocol displayed in Supplementary Table S1 ). Three of the cell lines were immortalized (HEK 293, NIH 3T3, and HaCat), whereas CRL-7250 was a primary cell line. A total of 9893 compounds were tested in either an 8-pt concentration-response ranging from 0.6 nM to 46 µM (1:5 dilution) or an 11-pt concentration-response ranging from 0.8 nM to 46 µM (1:3 dilution) according to different compound-plating mechanisms.

Variation of sensitivity was observed across all four cell lines, as shown in Supplementary Table S2 . The HEK 293 cell line was more sensitive to compound treatment ( Fig. 1a ), with 41% of compounds eliciting a dose-response cell-killing effect (16% high-quality actives, 25% low-quality actives), whereas the three remaining cell lines demonstrated reduced sensitivity (7.9%–10.6% high-quality actives, 14.2%–18.1% low-quality actives), where the high-quality actives were defined as compounds in class –1.1, –1.2, –2.1, and –2.2, with maximal response ≤–50% and EC₅₀ values less than 10 µM. The detailed information for data analysis and curve class definition is provided in the Materials and Methods section. The comparison of compounds active against the four cell lines ( Fig. 1b , Venn diagram) revealed 496 compounds active against only HEK 293 cells (consistent with its greater sensitivity to compounds). A total of 409 compounds demonstrated pan-activity against all four cell lines, which corresponds to a consensus hit rate of 4.1% for the entire annotated library.

Figure 1.

(a) Pie chart distribution of high-quality actives (orange), low-quality actives (yellow), and inactives (green) identified from the annotated compound library quantitative high-throughput screening against the four normal cell lines for the 48 h incubation condition. High-quality actives: compounds in class –1.1, –1.2, –2.1, and –2.2; maximum response ≤–50%; and EC₅₀ value ≤10 µM. Inactives: compounds in curve class 4; low-quality actives: compounds with shallow curves, single-point activity, or inconclusive activity. (b) Compound overlapping Venn diagram for HEK 293, HaCat, NIH 3T3, and CRL-7250 cell lines. The number of high-quality actives in each cell line and the number of compounds overlapped were calculated.

Comparison of the AUC values of each compound using unbiased hierarchical clustering across four cell lines revealed some cross-correlation of activity, with the HEK 293 cells clustering away from the other cell lines ( Fig. 2a ). Given the long-term intention of using profiling data to make comparisons with activity against cancer cell lines, we also tested the library against KB 3-1 human adenocarcinoma cells (KB 3-1 cells are a HeLa subclone). Intriguingly, the KB 3-1 cell line did not demonstrate remarkably different sensitivity compared with the normal cells, but surprisingly, its sensitivity clustered close with the NIH 3T3 murine fibroblast line. The pairwise AUC correlation with R² for all actives among five cell lines was calculated and is shown in Figure 2b ; the R² ranges from 0.31 to 0.77. NIH 3T3 and KB 3-1 showed the strongest correlation, with R² = 0.77. Violin plots of the distribution of sensitivity to compounds across the libraries reinforce that the majority of compounds were not cytotoxic in cell lines ( Fig. 2c , pink area), whereas focused plots of only high-quality actives ( Fig. 2c , cyan area) show that only a small number of compounds are highly cytotoxic (AUC < −300) against cell lines.

Figure 2.

(a) Comparison of area under the dose-response curve (AUC) values of each compound in four normal cell lines and KB 3-1 human adenocarcinoma cells. In the heat map, each row corresponds to a compound and each column to a cell line. The darker red color indicates a more potent and efficacious compound. (b) Pairwise R² correlation matrix among five cell lines. Only high-quality actives were included in this analysis, and the Pearson correlation was based on AUC values. (c) Split violin plot showing the distribution of AUC values for all compounds screened and high-quality actives across four normal cell lines. The lines within each distribution area represent the 0.25, 0.5, and 0.75 quantiles.

One strength of the annotated compound libraries is the ability to integrate compound MOAs and targets into downstream analysis. Each compound that met the criteria for high-quality active in each cell line was aggregated by MOA. The tree map shown in Figure 3 contains one box for each MOA containing three or more compounds. The size of each box is representative of the number of compounds with that MOA. The MOAs with the most compounds were antibacterial agents and PI3 kinase inhibitors, with 42 compounds present in each. We also assessed the average AUC for each MOA, which is represented in Figure 3 by the color gradient for each box. Kinesin-like spindle protein (KIF11) inhibitors and proteasome inhibitors had the lowest average AUC of all MOAs.

Figure 3.

Tree map representation of the mechanisms of action (MOAs) of all high-quality actives from the annotated library compounds screened. The box size represents the total number of compounds representing each MOA (a bigger box size indicates more compounds present in the high-quality actives). Color represents the average area under the dose-response curve (AUC) from the cytotoxicity screen in four normal cell lines (darker red indicates a lower AUC, meaning a more potent and efficacious hit). 1 = lineage-specific differentiation; 2 = RNA polymerase; 3 = immunosuppressant.

There were several instances in which multiple molecules with the same target were screened. To examine which MOAs were enriched among molecules active against cell lines, the proportion of active drugs to total drugs for each MOA target was used to derive an enrichment ratio ( Fig. 4a , where ratio = 1 represents 100% active, ratio = 0 represents none active). All classes of MOA targets with at least five compounds presented in the annotated library were included in the enrichment analysis. The top five target classes that resulted in consensus cell killing against the normal cell lines were proteasome, heat shock protein 90 (HSP90), anaplastic lymphoma kinase (ALK), mammalian target of rapamycin (mTOR), and cyclin-dependent kinase (CDK), with the enrichment ratio greater than 50%. To examine the significance of the association between consensus activity outcomes with the primary MOA, Fisher’s exact test was applied on the entire data set, and the statistical significance is reported in Supplementary Table S3 . Eight of 11 highly enriched MOA classes had a p-value <0.05, and seven classes had a p-value <0.001, indicating that compounds with similar activity profiles as determined in normal cell line screens tend to share similar annotated MOAs. These highly enriched MOAs (along with many other targets shown in Fig. 4a ) are generally associated with anti–cancer cell killing by targeting essential cellular processes. For example, 9 of 10 proteasome inhibitors tested resulted in acute cytotoxicity across all four cell lines (see proteasome inhibitor delanzomib, Fig. 4b ). As a reference point, the cancer cell line KB 3-1 did not demonstrate hypersensitivity compared with the normal cell lines; this is also consistent with the results we discussed in the heat map clustering analysis ( Fig. 2a ). The global cytotoxicity of proteasome inhibitors is displayed by radar plot ( Fig. 4c ) and is extended to other mechanisms of actions ( Suppl. Fig. S1 ).

Figure 4.

(a) Enrichment analysis of active agents in each drug category. Enrichment ratio = the number of actives/the total number of drugs in each drug category. (b) Dose-response curves for delanzomib, a representative proteasome inhibitor, in cytotoxicity screens. EC₅₀ HEK 293 = 148 nM, EC₅₀ NIH 3T3 = 82 nM, EC₅₀ CRL-7250 = 205 nM, EC₅₀ HaCat = 326 nM, EC₅₀ KB 3-1 = 186nM. (c) Radar plot displaying log EC₅₀ of all active proteasome inhibitors in four normal cell lines. (d) Dose-response curves for TC-S7004, a potent and selective DYRK1A/B inhibitor, in cytotoxicity screens, including KB 3-1 cells. EC₅₀ KB 3-1 = 3.2 µM. (e) Dose-response curves for LDN-212854, an ALK2 inhibitor, in cytotoxicity screens, including KB 3-1 cells. EC₅₀ HEK 293 = 16.4 µM, EC₅₀ KB 3-1 = 4.1 µM.

As outlined, one use of cell-profiling data for mechanistically annotated screening libraries is to enable the identification of targeted agents selectively killing cancer cells. Here, we designated that a selective agent should show strong cell killing to KB 3-1 cells with EC₅₀ ≤10 µM and have no effect on normal cell lines or at least 10-fold of EC₅₀ shift in normal cell lines. Two such examples of selective killing are displayed in Figures 4d and 4e . First, the dual-specificity tyrosine phosphorylation regulated kinase 1A and 1B (DYRK1A and DYR1B) inhibitor TC-S7004 demonstrated complete killing of KB 3-1 cells (EC₅₀ = 3.2 µM), with no effect on any of the four normal cell lines. Second, the potent and selective bone morphogenetic protein receptor inhibitor LDN-212854 demonstrated complete killing of KB 3-1 cells (EC₅₀ = 4.1 µM), with no effect on NIH 3T3 and CRL-7250 cells, and only partial killing of one normal line (HEK 293, EC₅₀ = 16.4 µM).

Data for the annotated library screen was deposited in PubChem with AID 1296008. This assay can be accessed via the following link: https://pubchem.ncbi.nlm.nih.gov/assay/assay.cgi?aid=1296008.

Profiling of a Diversity Library

The diversity library was profiled for cell viability by performing a primary screen against HEK 293 and NIH 3T3 cell lines, using CTG as the assay readout (screening assay protocol displayed in Suppl. Table S1 ). A total of 102,726 compounds were tested in a 4-pt dose-response manner at varied concentrations, with nearly 80% of the compounds tested at 115, 58, 12, and 2.3 µM. Of the library, 30.0% and 34.6% was cytotoxic to cell lines NIH 3T3 and HEK 293, respectively, with 1.2% of the library classified as high-quality actives in each of the cell lines ( Fig. 5a,b ). Of these high-quality actives, 285 compounds overlapped and shared similar activity in both NIH 3T3 and HEK 293 ( Fig. 5c ).

Figure 5.

(a, b) Pie chart distribution of high-quality actives (orange), low-quality actives (yellow), and inactives (green) identified from the quantitative high-throughput screen against diversity collection compound library in the NIH 3T3 and HEK 293 cell line, respectively. (c) Compound overlapping Venn diagram for NIH 3T3 and HEK 293 cell lines. The number of high-quality actives in each cell line was calculated, and the number of overlapped compounds was labeled. (d) The comparison of area under the dose-response curve values of 588 cherry-picked compounds in HEK 293, NIH 3T3, and KB 3-1 cell lines. In the heat map, each row corresponds to a compound and each column to a cell line. The darker red color indicates a more potent and efficacious compound.

After cherry-picking 588 hits based on their potency, the maximal response in the primary qHTS assay, and structural features, confirmation of activity was performed in 12 dose-point testing for all cherry-picked compounds against HEK 293 and NIH 3T3 along with an orthogonal test against the KB 3-1 cell line. The final concentration of the compounds in the 5 µL assay volume ranged from 0.3 nM to 46 µM. To analyze the data, we generated a heat map with hierarchical clustering analysis (dendrogram) of compound activities based on their activity outcomes, which showed that NIH 3T3 clusters away from HEK 293 and KB 3-1 ( Fig. 5d ).

The Promega CTG reagent uses an engineered version of firefly luciferase to measure ATP concentrations. As such, we tested compounds that appeared cytotoxic against recombinant fLuc in a biochemical assay to triage any compounds that were inhibiting fLuc rather than reducing cell viability. A number of compounds inhibited fLuc, and two examples from a novel fLuc inhibitor chemotype are shown in Figure 6a , b . In both cases, the compounds appeared equally active against all three cell lines in the CTG assay and demonstrated potent fLuc inhibition. The Promega CTG reagent contains a proprietary engineered fLuc that claims to be resistant to inhibition compared with wild-type fLuc.³⁰ This indeed appears to be the case, as a counterassay testing hits in cell-free media with CTG and ATP added as substrate demonstrated at least a 10-fold lower sensitivity to inhibition. Bright-field microscopy of cells grown in the presence of NCGC00413522 and NCGC00413607 ( Fig. 6c ) demonstrated unaffected cell growth at 48 h at concentrations suggested to be cytotoxic by CTG dose-response data (DMSO and 500 nM bortezomib as negative and positive controls, respectively).

Figure 6.

(a) Dose-response curves for NCGC00413522 in cytotoxicity screens, luciferase inhibition assay, and CellTiter-Glo (CTG) luciferase assay. EC₅₀ HEK 293 = 8.1 µM, EC₅₀ NIH 3T3 = 9.1 µM, EC₅₀ KB 3-1 = 8.1 µM. EC₅₀ luciferase = 0.89 µM. EC₅₀ CTG luciferase = 26.6 µM. (b) Dose-response curves for NCGC00413607 in cytotoxicity screens, luciferase inhibition assay, and CTG luciferase assay. EC₅₀ HEK 293 = 3.2 µM, EC₅₀ NIH 3T3 = 2.3 µM, EC₅₀ KB 3-1 = 2.3 µM. EC₅₀ luciferase = 0.08 µM. EC₅₀ CTG luciferase = 7.3 µM. (c) Representative bright-field images of HEK293 cells after 48 h treatment.

Moreover, comparing the AUC values of high-quality actives in the three cell lines, we found that they moderately correlated: NIH 3T3 versus HEK 293, R² = 0.77; NIH 3T3 versus KB 3-1, R² = 0.59; KB 3-1 versus HEK 293, R² = 0.61 ( Suppl. Fig. S2a–c ). Sixty-one of the top-ranking hits were active across all cell lines (pan-active), whereas others were more selective toward a specific cell line or cell lines. If taking only the pan-actives that showed strong cytotoxicity against all three cell lines in the analysis, it was clearly shown that most of the compounds have very good EC₅₀ agreement with low standard deviation, except for a few outlier compounds that showed a differential cell-killing effect in three cell lines ( Suppl. Fig. 2d ). These outlier compounds had greater cell killing against two normal cell lines (HEK 293 and NIH 3T3), although the selectivity window to the cancer cell line KB-3-1 was not significant (individual dose-response curves are shown in Suppl. Fig. S3 ). Among these five compounds, NCGC00420512, NCGC00420455, and NCGC00420435 belonged to the same structural chemotype with a tetrahydropyrazolo-pyrimidine core. To examine if the observed selectivity is a chemotype-specific outcome, we further analyzed the screening data through substructure search and found 172 compounds sharing the same tetrahydropyrazolo-pyrimidine scaffold presented in our diverse library, but none of the remaining 169 compounds showed a similar differential cell-killing effect, indicating that the selectivity is irrelevant to the specific structural features.

The remaining 56 pan-actives were cytotoxic across all three cell lines and lacked any selectivity toward either normal or cancer lines. Representatives of the top-ranking pan-actives are shown in Supplementary Figure 2e . Pan-actives were further clustered using fingerprint structural keys and the Tanimoto coefficient as similarity metric in the Molecular Operating Environment software (https://www.chemcomp.com/). We identified three enriched clusters based on their structural similarity using the MACCS fingerprint and Tanimoto coefficient ( Suppl. Table S4 ). The most significant enriched scaffolds were in clusters 1 and 2. Members in cluster 1 shared the same pyrimidine-quinuclidine core, with comparable EC₅₀ values across all three cell lines ranging from 7 to 12 µM, except one compound (NCGC00421344), which was >10-fold more potent in the HEK 293 than in the other two cell lines. Cluster 2 had a spiro structural feature, and it was the most cytotoxic cluster among all four clusters discussed in Supplementary Table S4 . Compounds in this cluster showed almost identical activity across the three cell lines, with EC₅₀ values ranging from 0.5 to 3 µM for most of the compounds. For example, NCGC00419015 was the most cytotoxic agent in cluster 2, with EC₅₀ values for HEK 293, NIH 3T3, and KB 3-1 of 0.51, 0.72, and 0.64 µM, respectively. Furthermore, cluster 3 with a tetrahydro-imidazo-diazepine core was a less enriched cluster but also was a nonselective cluster in which compounds were lacking selectivity against a specific cell line. We also ran the pan-actives against multiple toxicity data sets, including the Bursi mutagenicity data set³¹ (4337 compounds with mutagenicity data) and the National Toxicology Program data set³² (503 structures with carcinogenicity data for male/female mouse and male/female rat), but found no overlapping structures, which indicated that the chemotypes identified from our diverse library profiling are new additions to the well-known toxic collections.

Data for the diverse chemical library screen were deposited in PubChem with AID 1345083 for the HEK 293 cell line and AID 1345082 for the NIH 3T3 cell line. The assay data can be accessed via the following links: https://pubchem.ncbi.nlm.nih.gov/assay/assay.cgi?aid=1345083 and https://pubchem.ncbi.nlm.nih.gov/assay/assay.cgi?aid=1345082.

Discussion

Here, we describe our effort to profile the cytotoxicity of two distinct libraries: (1) our annotated libraries (nearly 10,000 compounds of known MOAs or therapeutic indications) against four normal cell lines (HEK 293, NIH 3T3, CRL-7250, and HaCat) and a cancer cell line (KB 3-1, a HeLa subline) and (2) a diversity library (>100,000 compounds) against two normal cell lines (HEK 293 and NIH 3T3) and a cancer cell line (KB 3-1, a HeLa subline). The assays were performed in qHTS format. Hit rates and mechanistic pathways regularly associated with cell death are described. As one might reasonably anticipate, annotated (or mechanistic) libraries containing small molecules developed against targets or known to possess phenotypic activity have a high hit rate (7.91%–16.05%) against the cell lines, especially for many annotated molecules developed against oncology targets.

The large diversity collection revealed a low rate of profound cytotoxicity (1.2%), which is consistent with the fact that the majority are not optimized against targets but also reinforces the need to confirm the toxicity of such compounds through counterassays as part of the screening process for compound prioritization and triage. In our case, these profiling data are now used at NCATS to allow for rapid cross-referencing of active compounds in a screen against our data set of general cytotoxicity in normal cell lines. To facilitate the use of these data by the broader research community, we have also made a significant amount of these data available through PubChem.

What is a “normal” cell line, and what are its limitations? The four normal cell lines used in this study were selected for a number of reasons. HEK 293 and NIH 3T3 cell lines are very commonly employed in the scientific literature as control, normal, or comparator cell lines, along with conventional research uses. A common criticism of these lines is how normal they are given their capacity for uncontrolled cell growth. The HEK 293 cell line is one of the most used tool cell lines. It is a neuronal lineage line, generated in 1973 from embryonic kidney cells from an aborted fetus immortalized by transformation with adenovirus.³³ The NIH 3T3 murine fibroblast cell line was generated in 1962 from a Swiss albino mouse embryo that spontaneously immortalized after multiple passages.³⁴ The HaCat human keratinocyte cell line was spontaneously immortalized in culture and was first reported in 1988.³⁵ All three of the aforementioned immortalized cell lines are adherent, and all are nondiploid despite being noncancer, nontumorigenic cells. The CRL-7250 cell line differs from the others in that it is a primary, nonimmortalized fibroblast cell line generated from human foreskin.³⁶ The HEK 293 cell line demonstrated greater sensitivity than the other three normal cell lines in the annotated library screen.

One potential use for normal cell line–profiling data is to enable the identification of compounds selectively active against cancer cell lines based on their target biology. A well-known example is the inhibitors of MAPK/ERK kinase (MEK), which elicits acute cell killing in cells harboring somatic activating mutations of Ras but not in cells expressing wild-type Ras (such as the normal cells discussed here³⁷). As a proof-of-concept comparison, we tested the KB 3-1 adenocarcinoma cells against the annotated libraries. KB 3-1 cells are a subclone of HeLa cells,³⁸ originally called KB squamous cell carcinoma before it was identified as a HeLa contaminant,³⁹ and have subsequently been studied and acknowledged as such. KB cells do not possess activating Ras mutations, but selective activity was seen by a DYRK1A/B inhibitor that killed KB 3-1 cells without affecting the normal cells, demonstrating the utility. These profiling data have already been applied to the analysis of multiple oncology-related screens across NCATS.

Perhaps the key result (reassuring from a screening perspective) from cytotoxicity profiling of >100,000 diversity library compounds was the very low rate of cytotoxicity observed, although some chemotype-related activity was observed. Cell lines such as HEK 293 (along with cell lines such as the Chinese hamster ovary cell line) are commonly used to engineer reporter cell lines for cell target-based and high-content assays, and the relative insensitivity to diversity libraries supports their utilization. The only prior study we have identified (also from NCATS but not involving any of the current authors) tested 1408 compounds against 13 cell lines, including a number of cancer cell lines along with HEK 293 and NIH 3T3 cells.²⁴ Although the compound number was relatively modest in scale, the normal HEK 293 and NIH 3T3 cells were reported to be among the least sensitive, supporting the basis for utilization of normal cell lines rather than cancer cell lines for designing cell-based assays for discovery screening.

The data reported here are made available through PubChem, and it is hoped that these data will act as a general guide for normal cell-line sensitivity to killing and assist in guiding others in the design of counterassays for HTS of cell-based assays.

Supplemental Material

CellProfilingPaperSupplementary20190805 – Supplemental material for Cytotoxic Profiling of Annotated and Diverse Chemical Libraries Using Quantitative High-Throughput Screening

Supplemental material, CellProfilingPaperSupplementary20190805 for Cytotoxic Profiling of Annotated and Diverse Chemical Libraries Using Quantitative High-Throughput Screening by Olivia W. Lee, Shelley Austin, Madison Gamma, Dorian M. Cheff, Tobie D. Lee, Kelli M. Wilson, Joseph Johnson, Jameson Travers, John C. Braisted, Rajarshi Guha, Carleen Klumpp-Thomas, Min Shen and Matthew D. Hall in SLAS Discovery

Footnotes

Acknowledgements

We thank Dr. Aleksandra Michaloa for advice regarding data analysis.

Supplemental material is available online with this article.

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 work was supported by the NCATS Division of Pre-Clinical Innovation Intramural Program.

ORCID iD

Matthew D. Hall

References

Schreiber

S. L.

Kotz

J. D.

; et al. Advancing Biological Understanding and Therapeutics Discovery with Small-Molecule Probes. Cell 2015, 161, 1252–1265.

Moustakim

Felce

S. L.

Zaarour

; et al. Target Identification Using Chemical Probes. Methods Enzymol. 2018, 610, 27–58.

Arrowsmith

C. H.

Audia

J. E.

Austin

; et al. The Promise and Peril of Chemical Probes. Nat. Chem. Biol. 2015, 11, 536–541.

Michael

Auld

Klumpp

; et al. A Robotic Platform for Quantitative High-Throughput Screening. Assay Drug Dev. Technol. 2008, 6, 637–657.

Renner

Popov

Schuffenhauer

; et al. Recent Trends and Observations in the Design of High-Quality Screening Collections. Future Med. Chem. 2011, 3, 751–766.

Wassermann

A. M.

Camargo

L. M.

Auld

D. S.

Composition and Applications of Focus Libraries to Phenotypic Assays. Front. Pharmacol. 2014, 5, 164.

Coussens

N. P.

Braisted

J. C.

Peryea

; et al. Small-Molecule Screens: A Gateway to Cancer Therapeutic Agents with Case Studies of Food and Drug Administration-Approved Drugs. Pharmacol. Rev. 2017, 69, 479–496.

Thomas

C. J.

Auld

D. S.

Huang

; et al. The Pilot Phase of the NIH Chemical Genomics Center. Curr. Top. Med Chem. 2009, 9, 1181–1193.

Auld

D. S.

Lovell

Thorne

; et al. Molecular Basis for the High-Affinity Binding and Stabilization of Firefly Luciferase by PTC124. Proc. Natl. Acad. Sci. U.S.A. 2010, 107, 4878–4883.

10.

Auld

D. S.

Thorne

Nguyen

D. T.

; et al. A Specific Mechanism for Nonspecific Activation in Reporter-Gene Assays. ACS Chem. Biol. 2008, 3, 463–470.

11.

Auld

D. S.

Thorne

Maguire

W. F.

; et al. Mechanism of PTC124 Activity in Cell-Based Luciferase Assays of Nonsense Codon Suppression. Proc. Natl. Acad. Sci. U.S.A. 2009, 106, 3585–3590.

12.

Quereda

Hou

Madoux

; et al. A Cytotoxic Three-Dimensional-Spheroid, High-Throughput Assay Using Patient-Derived Glioma Stem Cells. SLAS Discov. 2018, 23, 842–849.

13.

Hou

Tiriac

Sridharan

B. P.

; et al. Advanced Development of Primary Pancreatic Organoid Tumor Models for High-Throughput Phenotypic Drug Screening. SLAS Discov. 2018, 23, 574–584.

14.

Joslin

Gilligan

Anderson

; et al. A Fully Automated High-Throughput Flow Cytometry Screening System Enabling Phenotypic Drug Discovery. SLAS Discov. 2018, 23, 697–707.

15.

Ding

Clark

Bardelle

; et al. Application of High-Throughput Flow Cytometry in Early Drug Discovery: An AstraZeneca Perspective. SLAS Discov. 2018, 23, 719–731.

16.

Young

Margaron

Fernandes

; et al. MyoScreen, a High-Throughput Phenotypic Screening Platform Enabling Muscle Drug Discovery. SLAS Discov. 2018, 23, 790–806.

17.

Giuliano

K. A.

Wachi

Drew

; et al. Use of a High-Throughput Phenotypic Screening Strategy to Identify Amplifiers, a Novel Pharmacological Class of Small Molecules That Exhibit Functional Synergy with Potentiators and Correctors. SLAS Discov. 2018, 23, 111–121.

18.

Collia

Bannister

T. D.

Tan

; et al. A Rapid Phenotypic Whole-Cell Screening Approach for the Identification of Small-Molecule Inhibitors That Counter beta-Lactamase Resistance in Pseudomonas aeruginosa. SLAS Discov. 2018, 23, 55–64.

19.

Madoux

Tanner

Vessels

; et al. A 1536-Well 3D Viability Assay to Assess the Cytotoxic Effect of Drugs on Spheroids. SLAS Discov. 2017, 22, 516–524.

20.

Booij

T. H.

Bange

Leonhard

W. N.

; et al. High-Throughput Phenotypic Screening of Kinase Inhibitors to Identify Drug Targets for Polycystic Kidney Disease. SLAS Discov. 2017, 22, 974–984.

21.

Wang

S. S.

Ehrlich

D. J.

Image-Based Phenotypic Screening with Human Primary T Cells Using One-Dimensional Imaging Cytometry with Self-Tuning Statistical-Gating Algorithms. SLAS Discov. 2017, 22, 985–994.

22.

Cao

Yao

; et al. Cellular Phenotypic Analysis of Macrophage Activation Unveils Kinetic Responses of Agents Targeting Phosphorylation. SLAS Discov. 2017, 22, 51–57.

23.

Boehnke

Iversen

P. W.

Schumacher

; et al. Assay Establishment and Validation of a High-Throughput Screening Platform for Three-Dimensional Patient-Derived Colon Cancer Organoid Cultures. J. Biomol. Screen. 2016, 21, 931–941.

24.

Xia

Huang

Witt

K. L.

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

25.

Chiaravalli

Glickman

J. F.

A High-Content Live-Cell Viability Assay and Its Validation on a Diverse 12K Compound Screen. SLAS Discov. 2017, 2472555217724745.

26.

Hsieh

J. H.

Huang

Lin

J. A.

; et al. Correction: Real-time cell toxicity profiling of Tox21 10K compounds reveals cytotoxicity dependent toxicity pathway linkage. PLoS One 2017, 12, e0181291.

27.

Huang

Southall

Wang

; et al. The NCGC Pharmaceutical Collection: A Comprehensive Resource of Clinically Approved Drugs Enabling Repurposing and Chemical Genomics. Sci. Transl. Med. 2011, 3, 80ps16.

28.

Auld

D. S.

Southall

N. T.

Jadhav

; et al. Characterization of Chemical Libraries for Luciferase Inhibitory Activity. J. Med. Chem. 2008, 51, 2372–2386.

29.

Inglese

Auld

D. S.

Jadhav

; et al. Quantitative High-Throughput Screening: A Titration-Based Approach That Efficiently Identifies Biological Activities in Large Chemical Libraries. Proc. Natl. Acad. Sci. U.S.A. 2006, 103, 11473–11478.

30.

Hook

Schagat

Compound Interference of CellTiter-Glo® vs PE ATPlite™ 1 Step. https://www.promega.com/-/media/files/resources/posters/compound-interference-of-celltiter-glo-vs-pe-atplite-1-step-poster.pdf. Accessed August 23, 2018.

31.

Kazius

McGuire

Bursi

Derivation and Validation of Toxicophores for Mutagenicity Prediction. J. Med. Chem. 2005, 48, 312–320.

32.

Allen

D. G.

Pearse

Haseman

J. K.

Maronpot

R. R

. Prediction of Rodent Carcinogenesis: an Evaluation of Prechronic Liver Lesions as Forecasters of Liver Tumors in NTP Carcinogenicity Studies. Toxicol. Pathol. 2004, 32, 393–401.

33.

Graham

F. L.

Smiley

Russell

W. C.

; et al. Characteristics of a Human Cell Line Transformed by DNA from Human Adenovirus Type 5. J. Gen. Virol. 1977, 36, 59–74.

34.

Todaro

G. J.

Green

Quantitative Studies of the Growth of Mouse Embryo Cells in Culture and Their Development into Established Lines. J. Cell Biol. 1963, 17, 299–313.

35.

Boukamp

Petrussevska

R. T.

Breitkreutz

; et al. Normal Keratinization in a Spontaneously Immortalized Aneuploid Human Keratinocyte Cell Line. J. Cell Biol. 1988, 106, 761–71.

36.

Skroder

H. M.

Hamadani

J. D.

Tofail

; et al. Selenium Status in Pregnancy Influences Children’s Cognitive Function at 1.5 Years of Age. Clin. Nutr. 2015, 34, 923–930.

37.

Zhao

Adjei

A. A.

The Clinical Development of MEK Inhibitors. Nat. Rev. Clin. Oncol. 2014, 11, 385–400.

38.

Cornwell

M. M.

Gottesman

M. M.

Pastan

I. H.

Increased Vinblastine Binding to Membrane Vesicles from Multidrug-Resistant KB Cells. J. Biol. Chem. 1986, 261, 7921–7928.

39.

Vaughan

Glanzel

Korch

; et al. Widespread Use of Misidentified Cell Line KB (HeLa): Incorrect Attribution and Its Impact Revealed through Mining the Scientific Literature. Cancer Res. 2017, 77, 2784–2788.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.90 MB