More than 80,000 chemical compounds are registered for use in the United States (NTP 2004b
). In addition, about 2,000 new compounds are introduced into commercial use each year that may pose hazards for human health (NTP 2004b
). Traditional toxicologic methods cannot characterize and define the toxicity of such a large number of compounds in a cost-efficient and timely manner. However, recently developed HTS technologies may help to solve this problem by identifying classes of compounds with similar activity profiles and by helping to select and prioritize which compounds should receive a comprehensive toxicologic evaluation. As a first step in its HTS initiative, the NTP selected 1,353 compounds for characterization in cell-based cytotoxicity screens. The choice of compounds for this initial screening set was based primarily on practical and operational considerations. For example, the compounds had to be readily commercially available, of adequate purity and quantity, and soluble in DMSO up to the maximum stock concentration of 10 mM. This, and the fact that the entire universe of compounds of potential toxicologic interest stretches into the tens of thousands, means that many important compounds were not included in this initial set of 1,353. Now that proof of principle has been established in this robust and reproducible cell-based cytotoxicity assay, the compound collection is being expanded to include a greater number and diversity of compounds.
The human and rodent primary cells and cell lines used in the study were selected to represent six tissues of interest in toxicology: liver, blood, kidney, nerve, lung, and skin. Nine different cell types from human and four from rodents were chosen to gather data on potentially varied organ- and tissue-specific responses to environmental agents. Because our goal with this study was to investigate the feasibility of performing high-throughput assays of toxicologic compounds, we conducted our initial experiments with cell lines commonly used in qHTS: HepG2, SH-SY5Y, HEK293, Jurkat, H-4-II-E, and N2a (Hynes et al. 2006
; Knasmuller et al. 2004
). These lines are transformed or otherwise adapted to growing in vitro
, so presumably are less representative of in vivo
responses than primary cells, but they served to establish the qHTS methods used here and provided a baseline against which future studies with primary cells will be compared. In addition, several types of nontransformed and primary cells were used in this study, including human umbilical vein endothelial capillary cells (HUVECs) (Hoshi and Mckeehan 1984
), MRC-5 normal human fetal lung fibroblasts (Jacobs et al. 1970
), BJ normal human foreskin fibroblasts (Steinert et al. 2000
), human kidney glomerular mesangial cells, and primary rat renal proximal tubule cells. The ability to use these cells in these qHTS assays suggests that future efforts in toxicologic profiling using primary cells will be feasible.
Reproducibility of qHTS data has been demonstrated previously (Inglese et al. 2006
), but given the particular importance of reliability in toxicity profiling, we performed additional studies of assay robustness and reproducibility both within an assay and between replicates of an assay over time. In addition we evaluated the concentration–response curves for the positive controls included in each plate and found them to be very consistent. IC50
values of active compounds among 55 duplicates included in the compound library exhibited excellent correlation (R2
= 0.71; R2
= 0.89 when compounds with efficacies lower than 30% were removed from the analysis), demonstrating high intra-experiment reproducibility. To assess interassay reproducibility, three independent runs of the HepG2 qHTS were compared, producing a good correlation and average R2
of 0.74. Taken together, these data demonstrate that the ATP-based cytotoxicity assay in the qHTS platform is highly reproducible.
Of the 1,353 unique compounds tested, only 428 produced a measurable response in the cytotoxicity assays. Although this activity rate is higher than the expected rate in high-throughput screens for drug discovery, these results represent the prevalence of cytotoxicity at 40 hr of exposure in these cell types, measured with this readout, under these conditions at these concentrations, in the absence of metabolism. Many of the 1,353 compounds tested have been associated only with more chronic or subtle toxicity in vivo, so only a small subset of the compounds might reasonably be expected to be positive in this measure of acute cell killing.
Within the subgroup of 428 cytotoxic compounds, we were able to identify multiple patterns of effects within and across compound types, cell types, and species. Some compounds (e.g., digitonin, phenyl mercuric acetate) were uniformly toxic across all cell types, whereas others showed selective toxicity (e.g., 2-methyl-1-nitroanthraquinone in HEK293 cells). To the degree that structurally related compounds were present in this limited collection, nascent structure–activity relationships could sometimes be detected (e.g., the organic mercurials). These results suggest that a full range of similarities and differences in compound effects are potentially detectable using cell-based qHTS.
A range of cytotoxicity response patterns was also seen among cell types. Overall, the human blood- (Jurkat) and neuron-derived (SH-SY5Y) cells and rodent cells (N2a, H-4-II-E and NIH 3T3) were most sensitive to compound-induced cytotoxicity; kidney-derived cells (HEK293, human mesangial, and rat primary proximal tubule) were inter-mediate in sensitivity; and human fibroblastic, endothelial, and skin cells (HUV-EC-C, BJ, and MRC-5) were least sensitive. However, there was no one “sentinel” cell type sensitive to all toxins, which could be used to triage compounds for screening against additional cell types and assays.
A striking finding from the current study is the lack of similarity in the patterns of compound activity in cells derived from the same tissue but from different species (e.g., human HepG2 and rat H-4-II-E hepatoma cells). Even cells of similar tissue origin from the same species sometimes showed considerable differences in compound activity profiles. For example, the human SK-N-SH and SH-SY5Y lines showed quite different response patterns across the 1,353 compounds, even though SK-N-SH is the parental line of the SH-SY5Y. Overall, SH-SY5Y cells were more sensitive to compound-induced toxicity than SK-N-SH cells; the former cells are more differentiated and express a variety of neurotransmitters and neuronal cell surface markers. Neuronal cells are sensitive to environmental insults, perhaps because of their high metabolic demands and physiologic/morphologic specialization (Jellinger 2006
). In general, rodent cells were more sensitive than the human cells in this study, and in each case of homologous cells from the two species, rodent cells were the more sensitive cell type. Cell doubling times did not correlate with sensitivity (as measured by the number of positive compounds), but future studies will investigate other possible mechanisms for the differential toxicities observed. In all, these studies demonstrate that in vitro
cytotoxicity is often cell-type specific and that cytotoxicity in one cell type does not necessarily predict cytotoxicity in another.
One important limitation of the assay used here is that cytotoxicity was measured at a single time point (40 hr) only. To address this limitation, dynamic measurements of cellular response to exposure over time were made with selected compounds and the time-courses of cytotoxicity correlated with known molecular and cellular mechanisms of compound activity. Although this system is low throughput, staged screening of all compounds in the end point cytotoxicity assay followed by selective testing in the dynamic system may allow inference of mechanism of cytotoxicity.
In this study we have shown that it is feasible to screen large numbers of compounds in a titration-based HTS format and generate robust and reproducible results that can be analyzed to detect and compare cytotoxicity of a large number of compounds rapidly in a variety of cell types. To facilitate the use of these data by others, we have deposited all data from this study into a public database (PubChem 2007b
) in advance of publication, and we will continue to do so as more compounds, conditions, cell types, and assay readouts are tested. Large qHTS data sets promise to provide a rich source of information for the development of in vitro
toxicologic profiles that may prove valuable for prioritizing compounds for more intensive toxicologic investigation, and ultimately, predicting in vivo