HCS- screening for toxicity (Arch Toxicol)

ORGAN TOXICITY AND MECHANISMS

P. J. O’Brien Æ W. Irwin Æ D. Diaz Æ E. Howard-Cofield

C. M. Krejsa Æ M. R. Slaughter Æ B. Gao

N. Kaludercic Æ A. Angeline Æ P. BernardiP. Brain Æ C. Hougham

High concordance of drug-induced human hepatotoxicity with in vitrocytotoxicity measured in a novel cell-based model using high contentscreening

Received: 30 April 2005 / Accepted: 1 March 2006 / Published online: 6 April 2006� Springer-Verlag 2006

Abstract To develop and validate a practical, in vitro,cell-based model to assess human hepatotoxicity poten-tial of drugs, we used the new technology of high contentscreening (HCS) and a novel combination of criticalmodel features, including (1) use of live, human he-patocytes with drug metabolism capability, (2) preincu-bation of cells for 3 days with drugs at a range ofconcentrations up to at least 30 times the efficaciousconcentration or 100 lM, (3) measurement of multipleparameters that were (4) morphological and biochemical,(5) indicative of prelethal cytotoxic effects, (6) represen-tative of different mechanisms of toxicity, (7) at the singlecell level and (8) amenable to rapid throughput. HCS isbased on automated epifluorescence microscopy andimage analysis of cells in a microtiter plate format. Theassay was applied to HepG2 human hepatocytes culturedin 96-well plates and loaded with four fluorescent dyesfor: calcium (Fluo-4 AM), mitochondrial membranepotential (TMRM), DNA content (Hoechst 33342) todetermine nuclear area and cell number and plasmamembrane permeability (TOTO-3). Assay results werecompared with those from 7 conventional, in vitrocytotoxicity assays that were applied to 611 compoundsand shown to have low sensitivity (<25%), althoughhigh specificity (�90%) for detection of toxic drugs. For243 drugs with varying degrees of toxicity, the HCS,sublethal, cytotoxicity assay had a sensitivity of 93% andspecificity of 98%. Drugs testing positive that did notcause hepatotoxicity produced other serious, human or-

gan toxicities. For 201 positive assay results, 86% drugsaffected cell number, 70% affected nuclear area andmitochondrial membrane potential and 45% affectedmembrane permeability and 41% intracellular calciumconcentration. Cell number was the first parameter af-fected for 56% of these drugs, nuclear area for 34% andmitochondrial membrane potential for 29% and mem-brane permeability for 7% and intracellular calcium for10%. Hormesis occurred for 48% of all drugs with po-sitive response, for 26% of mitochondrial and 34% nu-clear area changes and 12% of cell number changes.Pattern of change was dependent on the class of drug andmechanism of toxicity. The ratio of concentrations for invitro cytotoxicity to maximal efficaciousness in humanswas not different across groups (12±22). Human toxicitypotential was detected with 80% sensitivity and 90%specificity at a concentration of 30· the maximal effica-cious concentration or 100 lM when efficaciousness wasnot considered. We conclude that human hepatotoxicityis highly concordant with in vitro cytotoxicity in thisnovel model and as detected by HCS.

Keywords HCS Æ Hepatotoxicity Æ Human ÆSublethal Æ Multi-parameter

Introduction

Drug-induced liver injury is a major cause of attrition inpreclinical and clinical drug development, and when adrug reaches the market. Hepatotoxicity is the mostfrequent reason cited for labelling drugs with a blackbox warning, and for withdrawal of an approved drug(Fung et al. 2001). Hepatotoxicity accounts for one-third to one-half of cases of acute liver failure (Andradeet al. 2004; Kaplowitz 2001; Lewis 2002; Russo et al.2004), and for 15% of associated liver transplantations.Hepatic reactions occur in less than 1 per 10,000 personsexposed, in individuals using therapeutic doses after a

P. J. O’Brien (&) Æ M. R. Slaughter Æ P. Brain Æ C. HoughamSafety Sciences Europe, Pfizer Global Research and Development,Sandwich Laboratories, Sandwich, EnglandE-mail: [email protected]: +44-1304-651224

W. Irwin Æ D. Diaz Æ E. Howard-Cofield Æ C. M. Krejsa Æ B. GaoCEREP, Seattle, WA, USA

N. Kaludercic Æ A. Angeline Æ P. BernardiUniversity of Padua, Padua, Italy

Arch Toxicol (2006) 80: 580–604DOI 10.1007/s00204-006-0091-3

variable latency period, and are usually considered to beidiosyncratic (Kaplowitz 2005). They typically occur ona background of a higher rate of mild, asymptomaticand transient liver injuries, such as indicated by athreefold increase in serum alanine aminotransferase(ALT) greater than the upper limit of normal.

Human hepatotoxicity has not been predictable be-cause of its low concordance with either standard invitro cytotoxicity screening assay results (O’Brien et al.2003; Xu et al. 2004) or regulatory animal study findings(Olson et al. 1998, 2000). Whereas conventional cyto-toxicity assays frequently have more than 80% speci-ficity, they have less than 25% sensitivity for detection ofhuman hepatotoxicity potential (O’Brien et al. 2003; Xuet al. 2004). Along with hypersensitivity and cutaneousreactions, hepatotoxicity has the poorest correlationwith regulatory animal toxicity tests (Olson et al. 1998,2000). In only approximately half of the new pharma-ceuticals that produced hepatotoxicity in clinical drugdevelopment was there any concordance with animaltoxicity studies (Olson et al. 1998, 2000). This contrastedremarkably with the high correlation between animaland human toxicities affecting the cardiovascular, he-matologic and gastrointestinal systems.

Despite the poor predictivity, there is reasonableunderstanding of the general pathophysiological mech-anism of most drug-induced hepatotoxicities (Boelsterli2003a, b; Jaeschke et al. 2002; Lee 2003; Xu et al. 2004).Also, there are in vitro methods available for detectionof the various pathogenetic mechanisms. The majormechanistic classifications of hepatotoxicity includeinhibition of mitochondrial function, disruption ofintracellular calcium homeostasis, activation of apop-tosis, oxidative stress, inhibition of specific enzymes ortransporters and formations of reactive metabolites thatcause direct toxicity or immunogenicity.

Conventional cytotoxicity assays have had poor sen-sitivity for several reasons (Xu et al. 2004). Firstly, theyhave measured lethal events in late stages of toxicity,despite that serious toxicities may not be lethal bythemselves or that detection of this advanced endpointmay be not be possible with drugs of limited solubility.Secondly, in vitro cytotoxicity frequently takes severaldays to express itself (Slaughter et al 2002; Lewis et al.2003; Xu et al. 2004; Schoonen et al. 2005a, b) and mostcytotoxicity assays do not include preincubation of cellswith drugs for multiple days. Most assays evaluate onlyone endpoint, whereas there are multiple mechanisms oftoxicity that need to be tested for by different methods,including use of morphological and biochemical orfunctional parameters. Measurements need to be madedirectly at the individual cell level to minimize artefactand ensure they reflect cell effects. Cells need to be humanand with capacity to metabolise drugs. Finally, tests needto be conducted at concentrations of drugs that are rel-evant to concentrations having efficacious effects in vivo.

High content screening (HCS) may be an importantpredictive tool for application of the above mechanisticunderstanding to drug discovery for the assessment of

compound potential for human hepatotoxicity (O’Brienet al. 2003; Xu et al. 2004; Abraham et al. 2004; Giuli-ano et al. 2003), and for optimisation and prioritisationof a compound’s safety. HCS is a recent advance in theautomation of quantitative epifluorescence microscopyand image analysis, and in the application of microflu-orescent, multiprobe technology (Abraham et al. 2004;Giuliano et al. 2003; Haskins et al. 2001; Plymale et al.1999). It enables kinetic monitoring in vitro of live cellsin real time for multiple cellular biomarkers of processesthat are critically involved in the pathogenesis of toxic-ity. These include inner mitochondrial membrane po-tential, intracellular free Ca2+ membrane permeability,as well as nuclear number and size (Giuliano et al. 2003;Haskins et al. 2001; Plymale et al. 1999).

This study tested the hypothesis that clinical occur-rence of human hepatotoxicity concorded with in vitrocytotoxicity assessed in a cell-based model with a novelcombination of critical features and using HCS.

Materials and methods

Materials

Chemicals and supplies

HepG2 cells that had undergone 84 passages were ac-quired from the American type cell culture(ATCC;#HB-8065) and stored in liquid nitrogen as alocal cell bank for distribution as needed. Dulbecco’smodified eagle’s medium (DMEM;#21969–035), fetalbovine serum (FBS;#10108–165), L-glutamine (#25030–024) and penicillin–streptomycin mixture (#15140–122)were acquired from Gibco, Invitrogen. Poly-D-lysinehydrobromide (PDL, MW = 30,000–70,000,#P7280)and non-essential amino acids mixture (#M7145) wereacquired from Sigma. All dyes were from molecularprobes. Black-walled, 96-well plates with lids were ob-tained from packard (Packard ViewPlate#6005182). Allother drugs and chemicals were acquired from Sigma-Aldrich unless stated otherwise, and were of the highestpurity possible.

Cell culture

Human hepatocellular carcinoma cells (HepG2) weresubcultured less than ten times after being acquired fromthe local cell bank. Their doubling time was 29±3 h.They were grown according to ATCC instructions(http://www.lgcpromochem.com/atcc/) in flat-bottomedculture flasks (T25), with bottom surface areas of25 cm2. The flasks were manually coated with PDL andthe cells grown in DMEM supplemented with 10% heat-inactivated FBS, 1% penicillin–streptomycin, 2 mMglutamine and 1% non-essential amino acids mixture.Cells were grown in a standard, cell culture incubator at37�C and 5% CO2 with a water reservoir for humidity

581

control. Cell counts were determined by hemocytometryfor addition of cells to 96-well plates.

Drugs and controls

Drugs that have been marketed for use in man wereclassified into four categories according to the severity ofhuman hepatotoxicity they produce. Severe humanhepatotoxicity was ascribed to drugs producing morethan 1% frequency of increased serum ALT plus two ofeither (1) jaundice, (2) more than three reports of liverfailure, or (3) a black box warning. Moderate humanhepatotoxicity was ascribed to drugs producing 0.1–1%frequency of increased serum ALT plus either jaundiceor a label of occurrence of adverse effect. Non-toxic orminimally toxic drugs were defined as those with lessthan 0.1% frequency of increased ALT, and associatedwith no clinical symptoms. The fourth category con-sisted of drugs that were not known to be hepatotoxicbut were known to have other organ toxicities. Addi-tionally, a wide selection of toxic chemicals and chemi-cals known to be non-toxic were tested.

Twelve drugs were selected to represent drugs causingidiosyncratic hepatotoxicity: acetaminophen, diclofenac,felbamate, hydralazine, leflunomide, methyldopa, mino-cycline, nitrofurantoin, rifampicin, sulindac, terbinafineand valproate (Kaplowitz 2005). Eighteen drugs wereselected to represent drugs causing toxicity by virtue oftheir metabolism to reactive metabolites: acetamino-phen, chloramphenicol, danazol, diclofenac, flutamide,ibuprofen, imipramine, indomethacin, isoniazid, hydral-azine, nitrofurantoin, piroxicam, procainamide, sulpha-methoxazole, tacrine, tamoxifen, terbinafine andvalproate (Kalgutkar et al. 2005).

HCS analyser specifications and settings

Plates were analysed by epifluorescence microscopyusing an automated, microplate-reading analyser (Ki-netic-Scan� HCS Reader, KSR; Cellomics, Pittsburgh,PA). The system is equipped with an incubator tomaintain constant temperature, CO2 and humidityduring analysis. The instrument enables fully automatedmonitoring of fluorescence intensity at four wavelengthsas well as image analysis (Compartmental Analysis

Bioapplication) and data viewing (Cellomics vHCSTM:View)

The imaging system of the KSR (Carl Zeiss) wascontrolled entirely via a personal computer (Dell Preci-sion 650 Workstation), with dual (Xenon) 2.0 GHzprocessors, with 4 Gb of random access memory (RAM)and 4 Gb of virtual memory allocation. Of the morethan 600 cellular ratios calculated per field, only 10 wereselected for this study. The KSR computer RAM andvirtual memory were upgraded to 4,000 MB each toimprove performance. Image and database files werespooled to and stored on a remotely located server,which required approximately 2 GB storage space per

96-well plate. To avoid possible conflicts with the Cel-lomics software, no other software was loaded onto thecomputer excepting for computer virus scanning.

Unless stated otherwise, fluorescence was monitoredkinetically for the four dyes for 3 h with a single cell countmade at the end of the assay. The 20· objective was usedto collect images for all four fluorescence channels withan appropriate filter set (XF93). Dyes were excited andtheir fluorescence monitored at excitation and emissionwavelengths of, respectively: (1) 365±25 and515±10 nm for Hoechst 33342 on channel 1, (2) 549±4and 600±12.5 nm for TMRM on channel 2, (3) 475±20and 515±10 nm for fluo-4 on channel 3 and (4) 655±15and 730±25 nm for TOTO-3 on channel 4. The channelsfor TMRM and fluo-4 were set to auto-exposure. Withthis feature, the exposure time is automatically deter-mined based on the cellular fluorescence of negativecontrol wells, and then set constant for the whole plate.The exposure for Hoechst was fixed to 100 ms for con-venience. Exposure for TOTO-3 was set to a fixed expo-sure of 1 s. In the standard assay, each well is monitoredat four fields-of-view per well each hour for 3 h.

The KSR’s incubator was set to 37�C and maintainedat 5% carbon dioxide. Microplate lids were left on theassay plate during kinetic monitoring to prevent evap-oration. A study of evaporation occurring when lidswere not placed on plates indicated losses of 21±4% ofwell volume.

Cell counts were made after monitoring fluorescencesfor 3 h. Ten fields per well were imaged and analysedusing the 10· objective. Fluorescence from an average of41 cells using the 20· objective was measured for each ofthe four microscopic fields-of-view per well. Controlwells had more cells, whereas wells with toxic doses ofdrugs had many fewer cells measured per field,depending on the cytotoxicity of the drug.

Methods

Conventional cytotoxicity assay methods

Seven independent, conventional cytotoxicity methodswere evaluated for their sensitivity and specificity fordetection of human hepatotoxicity. For all, positive testresults were considered to be those that produce at leasta 50% effect in the assay at a concentration of 30 lM.New DNA synthesis was assayed by liquid scintillationdetection of pulse incorporation of 3H-thymidine(Meselson and Stahl 1958). New protein synthesis wasassayed by liquid scintillation of pulse-incorporation of14C-methionine (Colombo et al. 1965). Glutathionedepletion was assayed by fluorometric detection ofmonobromobimane-conjugated, buthionine sulfoxi-mine-inhibitable cytoplasmic thiols (Barhoumi et al.1995). Superoxide secretion was assayed by spectro-photometric detection of cytochrome c reduction (Lo-rico et al. 1986). Caspase-3 activity was assayed byspectrophotometric detection of DEVD-pNA substrate

582

cleavage (Gurtu et al. 1997). Membrane integrity wasassayed by fluorometric detection of ethidium homodi-mer-DNA conjugation in response to membrane dam-age (Levesque et al. 1995). Cell viability was assayed inHepG2 cells for 48 h by fluorometric detection ofreduction of resazurin (Alamar Blue) to resorufin inresponse to mitochondrial activity (Nociari et al. 1998).

HCS, sublethal, cytotoxicity assay method For pre-liminary studies, the HCS assay was conducted imme-diately after addition of drugs to cells. However, inmost cases, the HCS assay was conducted after a per-iod of preincubation of the cells with drug. Usually,this was for 3 days, although the effects of incubationfor 7 days were also determined. For studies in whichthere was no preincubation of drugs and cells, drugeffects were tested only to 100 lM. However, forpreincubation of drugs and cells, the drug effects weretested to at least 100 lM and to a minimum concen-tration of 30 times the maximum total concentration(Cmax) of drugs reported circulating at the therapeuticdose. The effects of protein binding were not consid-ered for determining the maximum concentration fortesting. Protein binding of drugs would be minimal inthe assay, because of the low content of plasma proteinin the culture medium.

Drugs were initially dissolved as concentrated stocksolutions in water or DMSO. When DMSO was used todissolve drugs, it was added to the same final concen-tration of 0.5% (volume/volume) for all wells used fordetermining the drug response.

In a preliminary experiment with 16 toxic drugs theoptimal time for treatment of cells prior to assay wasdetermined. Cells were preincubated with the drug for0, 3 or 7 days. Initial seeding densities of the cells wereadjusted so that there would be sufficient cells foranalysis but not overgrowth of cells: 5,000, 3,000 and1,000 cells per well, respectively. For all other experi-ments, cells were treated for a period of 3 days beforethe assay. For determining drug effects without prein-cubation, the assay was started at drug addition andcells were analysed every 36 min for ten determina-tions; six determinations were used for preincubatedcells. In this preliminary experiment, cells were analy-sed for only one microscopic field-of-view. The firsttime point is not included in the figures since thedilution effect of adding the drug to the cells obscuredany drug effects; thus the first time point illustrated is36 min after drug addition.

Preparation of plates for HCS, sublethal, cytotoxicityassay Depending on the duration of drug exposure,1,000–5,000 cells in 100 ul media were added to eachwell of a 96-well plate (‘‘cell plate’’) and incubated for16–24 h, which ensured attachment of the cells to thebottom of the plate before drug treatment. Prior to drugaddition, 50 ul of media were removed from each well.Drugs were added in 100-ul volumes to each well from a

separate 96-well plate (‘‘drug plate’’). This was preparedwith 1.5-fold the final drug concentration needed in thecell plate. Each row had a different drug whose con-centration halved from serial dilution, from the highestconcentration in column 12 to the lowest in column 2.Wells in column 1 had no drug added.

Incorporation of fluorescent probes for HCS, sublethal,cytotoxicity assay Cells were simultaneously loadedwith 0.8 lM Hoechst 33342, 20 nM TMRM, 1 lM fluo-4 AM and 1 lM TOTO-3. Cells were loaded in mediacontaining 10% FBS for 1 h at 37�C.

Data capture The following data were collected. Nu-clear size was defined as the area of Hoechst 33342 flu-orescence and measured as the mean object size forchannel 1 in the nuclear region. Cellular mitochondrialmembrane potential was defined as the TMRM fluo-rescence intensity in punctate cytosolic regions aroundthe nucleus and was measured as the mean ring spotaverage intensity for channel 2. To assess intracellularfree calcium concentration, Fluo-4 fluorescence intensitywas measured in a large intracellular circular regioncentered at the nucleus as the mean circular averageintensity for channel 3. To assess plasma membranepermeability, TOTO-3 fluorescence intensity in the nu-clear region was measured as the mean circular averageintensity for channel 4. Selected object count was usedfor cell counting.

Quality control For positive controls, three chemicalswith known effects were added in triplicate to each plateto confirm quality of testing for the plate and to deter-mine the maximum responses for TMRM, Fluo-4 andTOTO-3 dyes. The three chemicals used were: themitochondrial uncoupler FCCP (100 lM), the calciumionophore ionomycin (10 lM) and the membrane-per-turbing detergent, Triton X-100 (0.05%). Column 1 ofeach row contained drug solvent but no drug, and wasused as a negative control. Fluorescence values for up to3 of the 11 wells with the same drug were consideredoutliers and excluded if there were more than two cv’s(of the repeated measure of controls–see section onassessment of precision) different from those of bothadjacent wells. Tests in which more than three of thewells had outlying values, or wherever there wereequivocal results, were repeated.

Data reduction and statistical analysis The dose-re-sponse relationship for each parameter was quantifiedwhere possible using the IC50, the concentration causing50% inhibition. For most compounds, the standard doseresponse curve of drug dose was used to model the re-sponse. Data curves were generated using least-squaresfitting routines with IC50 values determined using vari-able-slope, sigmoidal, curve-fitting routines, using a logscale for the drug dose on the horizontal axis and a

583

linear scale for the response on the vertical axis. Thiswas the four-parameter logistic curve versus logarithm.However for 43% of compounds with a positive testresponse, there was hormesis (low dose enhancement). Amodified version of the Brain and Cousens (1989) curvewas used to model this response. The hormesis doseresponse equation used is presented as Eq. (1) below.

y ¼ ðC þ A�DoseÞ

1þ 1þ 2�AC � expðIC50Þ

� �DoseIC50

� �B� �þ Bottom

Both curves were fitted using non-linear regression(which uses least squares) and the compound IC50’sestimated where possible. The precision of the IC50’s wasalso found where possible, and can be presented as thepercentage coefficient of variation. The regressions werecarried out using the statistical packages Genstat andGraphpad Prism.

Assessment of predictivity Sensitivity is the proportionof toxic drugs testing positive, TP/(TP+FN), where TPis the number of toxic compounds testing positive andFN is the number of toxic drugs testing negative.Specificity is the proportion of non-toxic drugs testingnegative, TN/(TN+FP), where TN is the number ofnon-toxic drugs testing negative and FP is the number ofnon-toxic drugs testing positive.

Assessment of assay imprecision Imprecision was studiedin order to distinguish between a drug effect and randomvariation or artefact. Imprecision could occur at fourlevels: the cell, the field of view, the well and the plate.Accordingly, the imprecision in parameter measure-ments was determined: (a) from cell to cell within field-of-view; (b) from field-of-view to field-of-view withinwell; (c) from well to well within plate and (d) from plateto plate. In order to compare the different sources ofimprecision, each type was estimated based on a similarnumber of measures and this estimate was repeated asimilar number of times. For this comparison ofimprecision sources, each parameter measure was madeseven times, except when limited by the number ofmeasures made. This limitation occurred across field-of-view because only four fields-of-view were used per well.The imprecision was defined as the SD divided by themean. Each imprecision estimate was made seven timesand was expressed as the mean ± SD.

KSR parameter variance between control wells onthe same plate and between control wells on differentplates was evaluated after three days. Negative controlwells were used for assessing nuclear area, cell count andTMRM. Positive control wells were used for assessingcalcium after addition of 10 lM ionomycin and forassessing membrane permeability after addition of0.25% Tween 20. Plates used were OP17, OP18, OP19,OP20, OP21, OP22. To determine variance betweenplates, the CV of the well means for each of the six plates

was calculated. Outliers were defined as values morethan three standard deviations away from the mean ofthe others, and were excluded, resulting in some platesnot being used and a smaller n. For every control well,mean, SD and CV were calculated for each plate. Theaverage CV was then used to determine variance be-tween wells.

Definition of positive and negative test responses The re-sult of the HCS sublethal cytotoxicity assay of a com-pound is reported in terms of the (1) degree of positivity,(2) the ratio (TI) of the lowest cytotoxic concentration tothe total, maximal drug concentration (Cmax) in serumthat is associated with efficacy (where this information isavailable) and (3) the concentrations at which clear ef-fects are first observed with each parameter.

Discrimination of a positive test result in the HCSassay was based on several criteria. Firstly, drug-inducedeffects had to be greater than the variance in the measureof the parameter across wells within plate. When achange was two and four times, respectively, the cv fordetermination of that parameter in controls, the testresult was designated positive or strongly positive. Forthose compounds where biphasic changes (hormesis, seebelow) were identified in cell number, nuclear area andmitochondrial potential, the change had to be from thebaseline level of the signal and not from the maximalpoint of increase in the signal. Drugs producing astrongly positive test were considered to be markedlycytotoxicity. Secondly, at least two parameters had toshow this effect. Thirdly, there needed to be clear dem-onstration of a concentration-response relationship forthe parameter. This included occurrence of the effect inat least one successively higher concentration, and ab-sence of the effect in at least one lower concentration.Alternatively, if the effect occurred at the end of theconcentration range, then its occurrence was supportedby an effect in at least one other parameter within oneconcentration point on the dose-response curve. Bi-phasic effects were identified by a clear increase by twocv followed by a clear decrease. Finally, the magnitudeof the effect had to be considered biologically relevant.

When the parameter change was between 1 and 2 cvs,test results were considered to be equivocal and weaklypositive. A negative test result was designated to be aresult where the above were absent, there was not adose-response relationship, and all differences frombaseline and control values were less than one CV, un-less these could be unequivocally attributed to anexperimental artefact affecting the well in question.

Results

Conventional cytotoxicity assays and their predictivity

Table 1 compares the predictivity of various conven-tional cytotoxicity assays (see Materials and methods)

584

applied to 611 drugs: 42 drugs causing severe hepato-toxicity, 283 drugs causing moderate hepatoxicity and286 non-hepatotoxic drugs. Table 1 demonstrates thatthese assays have only half the sensitivity that regulatoryanimal tests have. Although the in vitro assays were stillrelatively insensitive at detecting hepatotoxicity, theywere highly specific, so that when drugs tested positive inthese assays there was high probability that the drugproduced human toxicity.

Comparison on the performance of the conventionalcytotoxicity assays with the new multiparametric assay(Table 2) on identical compounds indicated similar re-sults as when the conventional assays were applied to thelarger data set of 611 drugs.

Although none of the conventional assays for cyto-toxicity had adequate concordance with in vivo humantoxicity, there were notable differences between assays.The assay with greatest sensitivity, 19%, for detection ofhuman toxicity potential was glutathione depletion,being twice as sensitive as the next most effective assays,namely DNA synthesis, as assessed by tritiated thymi-dine incorporation, and cell viability, as assessed bymitochondrial redox cycling activity. The caspase-3induction test for apoptosis, protein synthesis, super-oxide induction test for oxidative stress and the mem-brane integrity test were simply ineffective.

Cellular effects in the HCS, sublethal, cytotoxicity assay

When cells were seeded prior to incubation with drugs,3,000 cells were added per well. This corresponds toabout 500 cells in ten fields at the 10· objective used forthe cell count. This number did not change significantlyon the following day when incubation began, indicatingthat the stresses of plating had a cytostatic effect. In thefour Channel Assay using a 20· objective this seedingdensity corresponds to 12 cells per field. This data can beused to assess whether a drug is merely halting cells intheir cell cycle, or actually killing them (i.e. cytostaticversus cytotoxic). Cell counts at Day 0 (i.e. 24 h afterplating) were slightly more than the theoretical amountthat should be counted when they are first plated.

Typical cytotoxic changes are illustrated in Fig. 1.See HCS analyser specifications and settings for detailson settings and procedures for image analysis. Fura-zolidone-induced decreases in cell number, nuclear areaand mitochondrial membrane potential are readily visi-ble, as are the increases in cell calcium and plasmamembrane permeability.

Effects of duration on preincubation of cells with drugson cytotoxicity

HCS assays for drug-induced cytotoxicity conducted on23 toxic drugs and 6 non-toxic drugs at concentrationsup to 100 lM were far more effective when cells werepreincubated with drugs prior to testing. Assays inwhich cells had not been preincubated with the drug,produced positive test results for only 17% of the toxicdrugs (Fig. 2, top). However, cytotoxicity was seen for70% of these toxic drugs after preincubation at up to100 lM for 3 days (Fig. 2, bottom). Toxic effects wereinduced after 3 days preincubation by amodiaquine,cerivastatin, diclofenac, fenfluramine, furazolidone,kanamycin, pamidronate, primaquine, pyrmethamine,rosiglitazone, tacrine and zidovudine. Cytotoxicity wasnot seen, even after 3 days of incubation with up to100 lM, for acetaminophen, cisapride, erythromycin,isoniazide, lamivudine, sulphanilamide and vidaribine.However, in subsequent studies (Table 2) in which thesesix drugs were tested to higher concentrations of up to30-fold the Cmax, cytotoxicity was also detected foracetaminophen, erythromycin and lamivudine. Therewere no effects detected for any of the non-toxic drugstested: flufenamate, flumazenil, glimepiride and zom-epirac. Effects were not found for foscarnet nor primi-done, although in later studies in which thesecompounds were tested to 30-fold Cmax, their toxicitywas revealed.

Preincubation of cells with toxic drugs for 3 daysproduced a fourfold greater frequency of change thanwhen cells were not preincubated with drugs. Toxicity ofthe following drugs was not detectable without prein-cubation with cells for 3 days: amodiaquine, cerivasta-tin, diclofenac, fenfluramine, foscarnet, furazolidone,

Table 1 Predictivity of drug-induced human hepatotoxicity byconventional, in vitro cytotoxicity assays and by regulatory animaltesting in vivo

Sl. no. Predictive test Sensitivity Specificity

1 DNA synthesis 10 922 Protein synthesis 4 973 Glutathione depletion 19 854 Superoxide induction 1 975 Caspase-3 induction 5 956 Membrane integrity 2 997 Cell viability 10 928 Combination of above tests 1, 3, 7 25 839 Regulatory animal toxicity studies 52 –

Sensitivity (percentage of hepatotoxic drugs detected) and speci-ficity (percentage of drugs testing positive that are hepatoxic) aretabulated for seven conventional, cytotoxicity in vitro assays, forthe optimal combination of these assays, and for regulatory, invivo, animal toxicity studies. The in vitro tests were applied to 611drugs, including: (a) 42 drugs causing severe human hepatotoxicity,defined as producing more than 1% frequency of increased serumALT plus at least two of either jaundice, more than three reports ofliver failure or a black box warning; (b) 283 drugs producingmoderate human hepatotoxicity, defined as producing 0.1–1%frequency of increased serum ALT, plus jaundice or a label ofoccurrence of adverse effect and (c) 286 drugs producing negligiblehuman hepatotoxicity, defined as less than 0.1% frequency of in-creased ALT and absence of clinical symptoms. Sensitivity forregulatory animal toxicity studies is based on retrospective exam-ination of outcomes in animal studies for drugs found in clinicalstudies to produce human hepatotoxicity (Olson et al. 2000)

585

Table

2Cytotoxic

effects

ofdrugscausinghumantoxicity

Sl.no

Drug

Toxicity

Old

tests

HCS

test

First

signal

Cell

number

Mitochondrial

potential

Ca

Mem

perm

Nuclear

area

Cmax

(lM)

TI

PPB

(%)

Mechanism

Severelyhepatotoxic

drugs

1i,r A

cetaminophen

X,FH,H

I,Ht

�Str

+#

500

›4,000

fl4,000

––

7,800

130

48

OS

2Aminosalycilate

0.005,W

,XJ

�+

#24

›750

fl100

––

–49

0.5

68

IM

3Amitryptilline

<10,H

,C,LD

�+

#13

50

25

50

50

0.3

43

95

4Amiodarone

0.25,BBW,H

+Str

+#

6›2

5fl

100

13

25

25

0.81

799

OP

5Amodiaquine

Wd,FH,H

I+

Str

+C

25

50

13

25

25

0.42

31

95

IM

6Cerivastatin

Wd,I,H

ND

Str

+#,M

0.1

›1.6

fl0.1

›0.8

fl0.2

0.4

1.6

0.03

399

M,Ca,A

p

7CyclosporinA

>3,C,H

t+

Str

+M

25

2›

13

13

13

0.2

10

93

OS

8Danazol

50,W

�Str

+#,N

13

50

›100

fl25

50

13

0.16

81

9Dantrolene

0.012,BBW,H

,X,LD

�Str

+#

25

100

50

50

50

7.9

385

IM

10

i,r D

iclofenac

0.04,W

,H,FH,J

�Str

+#

16

›126

fl250

–126

126

›4.2

499

OP,A

p

11

Didanosine

65,BBW,H

,FH,X

ND

+#

2›

––

––

12

0.2

5M

12

Disulfiram

H,FH,X

+Str

+#

13

25

25

25

25

5.4

250

OS

13

Efavirenz

8,H

tND

Str

+#

50

100

100

100

100

13

499.5

M

14

Etoposide

20,H

,HI,

ND

Str

+#,N

0.5

505

16

40.5

›17

0.03

96

Syn

15

i,r Felbamate

BBW,X

,FH,H

�Str

+M

315

160

›–

–315

42

424

IM

16

Fialuridine

Wd,X

,Ht

ND

Str

+#

2–

–500

125

›1

2M

17

Flutamide

<1,w,H

,C,J,X

,FH

�Str

+#,C,M

,P50

50

50

50

–6

890

M

18

Indomethacin

W,FH,X

,I�

+N

190

47

––

2›

60.3

90

IM,OS

19

Imipramine

<22,H

,LD,C

�Str

+#

0.8

›50

fl100

50

50

50

0.6

1100

IM,M

20

r Isoniazide

20,BBW,X

,FH,H

�+

M–

50

fl–

––

40

10

IM,OS

21

Itraconazole

XND

+M

20.2

–2

30.4

0.5

99.8

IM

22

Ketoconazole

40,BW,X

,H,C

+Str

+N

100

25

›100

fl25

50

13

›50

fl7

299

M

23

Ketorolac

BW,FH,H

,C�

Str

+M,N

,CP

430

›220

›220

220

220

70.25

99

IM

24

Labetalol

W,I,H

,FH,X

,C,H

I�

Str

+N

13›100

fl50

›50

100

60.4

15

50

25

Lamivudine

4,BBW,X

,�

+#

125

––

––

17

736

M

26

Mercaptopurine

40,W,

++

N–

––

–0.2

10.2

19

Syn

27

Methapyrilene

Ht

�+

M–

150

––

–115

1.4

M,O

P

28

Methotrexate

BW,H

t,I,Ci,F

ND

Str

+#,N

0.02

––

–0.02

0.02

1Syn

29

i,r M

ethyldopa

W,I,H

,FH

ND

Str

+All

330

330

330

330

330

11

30

15

30

i Minocycline

W,I,H

,B�

+#

13

25

––

50

›8

276

IM,M

31

Niacin

<5,W

,FH,C

,JND

+#,N

3800

›7500

––

3800

›126

30

32

Nim

esulide

X,H

tND

+#,N

340

680

1370

1370

340

15

23

99

M

33

i Nitrofurantoin

W,X

,H�

Str

+N

12

50

50

50

0.7

›6

0.1

62

IM,OS,M

34

Novobiocin

BBW,I

�Str

+#,P

100

400

400

100

400

1100

IM

35

Phenylbutazone

Wd

�+

#,M

,C6,600

6,600

›13,000

fl6,600

––

438

15

99

36

r Piroxicam

2,W

,H,X

�+

#,N

5160

––

5›

51

99

37

Propythiouracil

<30,W

,H,FH,J

�+

#25

––

––

42

0.6

85

IM

38

Pyrazinamide

W,I,

ND

+#

2400

›4800

›–

––

325

750

49

Stavudine

BBW,H

,FH

ND

Str

+#

250

–500

1,000

–4

63

5M

40

i Sulindac

<1,W

,H,X

�Str

+M

285

140

––

570

›19

894

IM,OS

41

i,r V

alproate

44,BBW,B,X

�Str

+#

1,000

4,000

›–

4,000

8,000

540

293

OS,M

42

Zileuton

W,H

tND

+#

100

––

––

17

694

OS

Sensitivity

20%

100%

Moderately

hepatotoxic

drugs

43

Aspirin

I,H,H

t�

Str

+M,C,N

6,250

1,300

1,300

6,250

1,300

1,650

0.8

30

OS,M

44

Azathioprine

<1,B,H

t�

+#

0.8

––

25

–0.34

230

OS

45

Bupropion

<1,W

,C,H

,LD

�HiTI�

#300

1,200

600

600

600

0.5

600

84

586

Table

2(C

ontd.)

Sl.no

Drug

Toxicity

Old

tests

HCS

test

First

signal

Cell

number

Mitochondrial

potential

Ca

Mem

perm

Nuclear

area

Cmax

(lM)

TI

PPB

(%)

Mechanism

46

Captopril

W,I,H

,B�

+N

––

––

55

›110

fl4

14

30

47

Ceftazidim

eI,C

+Str

+#

12

24

48

48

48

220

0.05

21

48

Chloramphenicol

C,J,H

I�

Str

+#

260

––

––

57

553

OS

49

Chlorpromazine

0.75,

ND

Str

+#

325

13

13

60.5

698.5

IM,M

50

Ciprofibrate

I,H,H

I�

+#

100

––

––

520

M

51

Clofibrate

I,Hm,C

�+

M14,000

7,100

›–

––

470

15

M

52

Colchicine

HI,

ND

+#,N

0.1

––

–0.1

0.016

640

53

Cyclophosphamide

J�

+#,N

2,300

––

–2,300

›143

16

13

54

Diethylcarbamazine

X,H

,HI

�Weak+

#100

––

––

1.3

77

055

Doxorubicin

<1,H

,C+

Str

+#,C,P

0.1

1›2

fl0.1

0.1

0.4

0.2

0.5

76

OS,M

56

Enalapril

W,I,H

,B,C

ND

+N,M

100

1.6

›–

–1.6

0.4

455

M

57

r Erythromycin

W,I,H

,c,J,LD

�+

#1.6

50

›–

––

11

0.15

84

OS

58.

Ethylestrenol

Ht

�+

#6

–100

–100

32

C

69

r Estradiol

W,C,J

+HiTI�

#,N

25

50

––

25

›0.0006

42,000

98

C

60

Famotidine

<1,C,J

�+

M–

3–

–25

0.3

10

17

61

Fenofibrate

0.063,W

,H,C

�Str

+#

250

–1,000

1,000

1,000

25

10

99

M

62

Furazolidone

C,H

I�

Str

+#

625

25

50

100

41.5

M

63

r Furosemide

C,J

�Str

+M

1,100

570

2,300

2,300

2,300

16

35

99

M

64

Fusidic

Acid

B,H

t,C

�+

M–

516

––

–17

30

95

65

Glyburide

I,H,C

�+

M50

2›

–100

50

›0.2

10

99.8

66

i,r H

ydralazine

0.005,H

HI,C,

ND

Str

+#,N

67

135

›270

–75

513.5

87

IM

67

Ibuprofen

<1,H

,J�

+#

50

––

––

250

0.2

99

IM,M

68

Ifosfamide

0.03,B,C

�+

#1,500

––

––

203

80

69

Isotretinoin

0.15,W

,HND

+#,M

100

100

813

70

Ketoprofen

H,C,LD

�Str

+M

47

3–

––

15

0.2

98.7

71

i,r Leflunomide

0.05,W

,B+

+#

25

––

––

340

0.07

99.3

IM

72

Lovastatin

0.019,W

,H,C

ND

+N

1.6

›25

fl3

›100

fl13

60.4

0.01

40

95

M,Ap

73

Mesalamine

I,H,B,FH,C,H

I�

+#

610

––

25

›12

0.5

50

IM

74

Methacycline

H,C

�+

#13

50

––

–5

385

75

Naproxen

<1,FH,H

I,C

ND

HiTI�

M,N

100

50

›–

–50

0.2

250

99

IM

76

Nizatidine

I,H,H

I,C

�+

N–

––

–26

47

35

77

Norfloxacin

1.6,H

�Str

+N

28

7–

–0.5

›8

0.8

18

IM

78

Paclitaxel

0.19,B,H

IND

Str

+#

0.1

–50

50

25

›4.3

0.02

94

79

Paroxetine

I,H,FH,H

I+

Str

+#

0.8

›25

fl13

6.3

13

13

0.2

495

M

80

Pravastatin

0.045,W

,H,B,H

IND

HiTI�

#100

––

––

0.1

1,000

50

Ap

81

r Procainamide

I,H,C

ND

Str

+N

630

630

›630

–320

12

27

18

IM

82

Pyrimethamine

C,J,H

IND

Str

+#

3–

10

10

10

›100

fl3.3

1Syn

83

Quinacrine

H+

Str

+C

23

0.4

33

10.4

90

84

Quinidine

H,H

I,Ht

ND

Str

+C

830

0.5

115

90.06

90

M

85

Quinine

H+

Str

+#,P,C,N

36

70

36

36

36

19

293

86

i Rifampicin

W,I,H

++

N50

–100

100

13

›9

181

87

Rosuvastatin

<0.5,

++

N0.4

›0.4

›1.6

fl3›

50

–0.2

›0.4

fl1.6

›6fl

0.01

20

M,Ca,Ap

88

Sim

vastatin

0.05,W

,HND

+#,M

,N0.2

›50

fl0.2

›0.9

fl3›5

0fl

66

0.2

fl0.9

›3fl

0.02

10

94

M,IM

99

r Sulfamethizole

W,H

�+

M–

1000

––

–222

590

90

r Sulfamethoxazole

W,H

,B,C,J,H

I�

+#

1,100

›–

––

–217

562

IM,M

91

r Sulfaphenazole

ND

+N

100

100

––

50

78

0.6

92

r Tacrine

0.29,W

,H,B,H

t�

Str

+N

25

50

›100

fl50

25

13

›50

fl0.1

130

55

OS,M

93

r Tamoxifen

W,I,H

,B,C,H

I+

Str

+N

100

›25

–25

0.4

›25

fl0.4

198

M

587

Table

2(C

ontd.)

Sl.no

Drug

Toxicity

Old

tests

HCS

test

First

signal

Cell

number

Mitochondrial

potential

Ca

Mem

perm

Nuclear

area

Cmax

(lM)

TI

PPB

(%)

Mechanism

94

i,r Terbinafine

C,J,H

t,LD

ND

Str

+#

36

66

13

40.8

99

95

Terfenadine

I,H,C,J

HiTI�

C6

63

613

0.01

300

CA

96

Tetracycline

I,Ht,C

�+

M,C,P

–560

560

560

–9

62

65

IM

97

Tolazamide

C�

+N

50

––

–2

90

0.02

94

98

Tolbutamide

X,C,J

�+

#30

––

––

590

0.05

96

99

Trifluoperazine

H,C

+HiTI�

C13

25

613

13

0.003

2,000

99

100

Warfarin

I,H,C,J

�+

#,M

,N100

100

––

100

714

99

IM

101

i Zarfirlukast

W,H

tND

+M,N

100

13

–100

13

›1

13

99

102

Zidovudine

BW,H

,HI

�Str

+#

63

–500

––

416

25

M

Sensitivity

24%

88%

Non-toxic

drugs

103

Acetylcysteine

HI

�–

––

––

––

1,900

>1

80

104

Aminobenzoate

�–

––

––

––

50

>30

105

Bambuterol

�–

––

––

––

0.015

>6,700

45

106

Betaine

�–

––

––

––

940

>30

107

Biotin

�–

––

––

––

<1

>100

108

Bisacodyl

�–

M–

100

––

0.15

667

109

Buspirone

1ND

–#,N

100

––

–100

0.01

1,000

95

110

Carbidopa

+–

#520

––

––

2260

36

111

Citicoline

�–

––

––

––

700

>30

112

Cromolyn

I(poss

Alerg)

�–

––

––

––

0.016

>6,600

68

113

Cyanocobalamin

�–

#,M

0.4

0.4

––

0.8

0.001

400

114

Dexamethasone

�–

M,N

100

50

––

50

›0.23

217

77

115

Dim

ethylsulfoxide

ND

–#

35,000

70,000

–70,000

70,000

<1,000

>30

116

Diphenhydramine

�–

#1,300

2,500

2,500

2,500

2,500

0.3

4,200

98.8

117

Flumazenil

+–

P–

––

50

–0.1

500

45

118

Eserine

ND

–M

–6

›–

––

0.22

>450

M

119

Fexofenadine

ND

––

––

––

0.57

>175

120

Folate

ND

––

––

––

–3.4

>30

121

Folinic

acid

ND

––

––

––

–3

>30

122

Glimepiride

H(1

case),C

�–

#,P

100

––

100

–0.73

>137x2

99

123

Isoproterenol

ND

–#

6›1

00

fl–

––

–0.006

1,000

68

124

Isosorbidedinitrate

�–

––

––

––

0.0008

130,000

28

125

Ketotifen

�–

#,P,C

50

100

›50

50

100

0.0001

500,000

126

Moxisylyte

ND

–M

–100

›–

––

1.65

>61

127

Myo-inositol

ND

––

––

––

22

>30

128

Oxyphenonium

ND

–M,N

–100

––

100

0.3

300

129

Pargyline

ND

–#

100

––

––

0.3

333

130

Picotamide

ND

+M

13

2100

100

6›

50

0.04

131

Pinacidil

ND

–#,M

,N100

100

––

100

›0.17

588

40

132

Pioglitazone

�–

––

––

––

2.6

>38

99

M

133

Praziquantel

�–

––

––

––

1.8

>56

80

134

Propranolol

HI

––

C,P,N

50

–25

25

25

0.1

250

87

135

Pyridoxine

ND

––

––

––

–1.1

>91

136

Rosiglitazone

ND

–#

50

100

––

80

0.67

>75x10

100

M

137

Thiamine

ND

––

––

––

–6.8

>30

588

Table

2(C

ontd.)

Sl.no

Drug

Toxicity

Old

tests

HCS

test

First

signal

Cell

number

Mitochondrial

potential

Ca

Mem

perm

Nuclear

area

Cmax

(lM)

TI

PPB

(%)

Mechanism

Specificity

88%

97%

Drugstoxic

toother

organs

138

Astem

izole

CA

+HiTI�

All

66

66

60.01

600

97

IM

139

Bezafibrate

I�

+#

50

––

––

17

395

M

140

Bupivacaine

Myo

ND

+#,C

6–

6–

–0.7

996

OS

141

Caffeine

�+

#1,250

––

2,500

2,500

›42

30

36

142

Capreomycin

I,R,O

ND

Str

+#

150

›300

––

–40

4143

Chloroquine

I,HI

+Str

+#,C

626

›50

fl6

13

13

0.48

13

61

144

Chlorpheniramine

ND

+N

50

50

›100

fl100

–0.4

›6fl

0.2

272

145

Ciglitizone

ND

+M

–3

––

6›

50.6

M

146

Cisapride

I,H,CA

�HiTI-

––

––

––

0.12

>810

98

CA

147

Clioquinol

�+

#,N

64

2,000

500

2,000

64

69

1148

Dipyrone

�+

#,P

1,000

––

1,000

–34.5

29

60

IM,Ag

149

Fenfluramine

�+

#6

25

50

50

50

0.1

60

30

Card

150

Flufenamate

�Str

+M,N

–100

––

100

›46

290

151

Flucytosine

I,HI,LD

ND

Str

+#,P

6–

–6

–780

0.008

RM,Syn

152

Fluvastatin

0.011,W

,Myo

ND

Str

+C

1.6

3›6

fl13

›50

fl0.8

1.6

13

0.45

299

M,Ap

153

Foscarnet

5,R

�+

C,N

4,400

8,700

›17000

fl2,200

8,700

2,200

580

420

R

154

Gentamycin

R,O

ND

+N

185

–185

185

95

›13

728

R

155

Halothane

ND

+N

13

3–

–2

›10

1RM,OS

156

Indoprofen

ND

+#

2,100

––

––

72

30

99

157

Isoxicam

H,D

,GI

ND

False�

––

––

––

20

97.7

IM

158

Kanamycin

HI,R,O

�+

#,M

11

11

––

25

›47

0.2

0Trans

159

Lidocaine

CNS

�+

M1,300

300

2,500

2,500

2,500

36

870

IM

160

Mem

antine

N�

Str

+#

25

50

100

100

–0.2

125

161

Metform

inPan

ND

+M

1,700

220

––

440

116

2162

Metoclopramide

I,Ht

�+

N–

50

––

13

0.4

32

65

163

Menadione

+Str

+#

613

13

13

25

51

OS

164

Mevastatin

ND

+N

30.2

›0.4

fl3›

0.4

13

0.1

›0.2

flAp

165

Mibefradil

Wd

ND

Str

+#

0.4

›6fl

66

50

61.1

0.4

Ca

166

Nialamide

CNS

ND

Str

+N

––

––

100

14

7167

Nicotine

ND

+M

27

0.4

––

100

0.09

55

168

Nocodazole

ND

+#

0.1

–25

25

0.4

0.5

0.2

169

Nomifensine

ND

+#

100

––

––

1.2

83

170

Pamidronate

R�

Str

+#

3›1

00

fl100

100

100

–11

0.3

30

Trans

171

Phenacetin

ND

+#

25

100

––

100

›12

233

R

172

Phenform

inND

+#,M

13

13

––

25

›0.63

21

20

LA

173

Pim

ozide

I+

Str

+#,C,P

613

66

13

0.11

55

99

174

Primaquine

�+

#,

350

25

25

50

0.11

27

175

Primidone

�+

#350

690

––

–23

15

20

176

Propythiouracil

�Str+

M630

313

––

1,300

42

7177

Sodium

chloride,

mM

ND

Str

+N

13

100

50

50

3›

145

NA

0.1

178

Streptomycin

R,O

�+

M1,600

780

––

–52

15

50

179

Streptozocin

ND

?N

––

––

100

›180

Sulfabenzamide

�?

#25

100

––

181

Sulfanilamide

�?

M,P

–100

–100

––

70

IM

182

Telenzipine

�+

M–

1–

–3

›0.02

50

183

Tem

ozolomide

BM

�+

#50

––

––

68

Syn

589

Table

2(C

ontd.)

Sl.no

Drug

Toxicity

Old

tests

HCS

test

First

signal

Cell

number

Mitochondrial

potential

Ca

Mem

perm

Nuclear

area

Cmax

(lM)

TI

PPB

(%)

Mechanism

184

Theophylline

I,H

�+

N–

––

–25

85

0.3

56

185

Topiramate

X,CNS

ND

+#,M

300

300

›20

15

13

186

Vancomycin

ND

+#,M

285

285

›19

15

187

Vidarabine

I,B

�+

N100

50

––

617

0.35

25

M

188

Vincamine

HI,CA

ND

HiTI�

M–

100

›–

––

0.44

230

Hem

189

Zalcitabine

ND

+#

63

––

500

–0.37

170

4M

190

Zomepirac

�False�

-–

––

––

5.6

92

R,IM

191

Zonisamide

ND

+N

––

––

624

0.25

50

Sensitivity

14%

92%

Positive

controls:chem

icals

causingtoxicity

192

Acetamidofluorene

ND

+#,M

100

100

––

–NA

NA

193

Aflatoxin

B1

ND

Str

+#

0.1

0.2

›3.2

3.2

0.2

›NA

NA

OS

194

Alloxan

ND

+M

100

NA

NA

OS

195

Allyalcohol

ND

+N

–3

––

0.4

›NA

NA

196

Benzopyrene

ND

Str

+#

0.1

1.6

›1.6

0.8

1.6

NA

NA

197

BetaineHCl(acidic)

ND

Str

+#

7,100

28,000

––

–NA

NA

198

Bromobenzene

ND

+#

100

––

––

NA

NA

OS

199

p-Bromophenol

ND

+#

630

2,500

1,300

1,300

1,300

NA

NA

200

Buthionine

ND

+M

50

0.2

–50

–NA

NA

OS

201

Butylhydroxytoluene

ND

+#

50

––

––

NA

NA

OS

202

Carbontetrachloride

ND

+N

–50

›–

–0.8

NA

NA

OS

203

Chloroform

ND

+#

100

––

––

NA

NA

OS

204

m-C

resol

ND

+M

–0.8

––

3NA

NA

205

p-C

resol

ND

+M,N

50

0.2

––

0.2

›NA

NA

206

CytochalasinB

ND

Str

+N

20.8

350

0.1

NA

NA

207

CytochalasinD

ND

Str

+#

0.4

100

100

50

6NA

NA

208

Diaminotriazole

ND

Str

+N

3,100

––

–1,560

NA

NA

209

Dichloroethylene

ND

Str

+M,N

–25

›–

–25

NA

NA

210

r Diethylm

aleate

ND

+N

50

25

––

13

NA

NA

OS

211

r Dim

ethylform

amide

ND

+#,N

313

625

3NA

NA

212

Diquat

ND

Str

+#

13

50

50

100

100

NA

NA

OS

213

Ethionine

ND

+N

––

––

3NA

NA

M

214

r Eugenol

ND

+M

–13

›–

––

NA

NA

OS

215

FCCP

ND

Str

+N

25

50

50

50

13

›NA

NA

M,O

S

216

Form

aldehyde

ND

+M,N

–25

›–

–25

NA

NA

217

Galactosamine

ND

Str

+#

780

12,500

13,000

1,600

13,000

NA

NA

Syn

218

Glucosamine

ND

Str

+N

750

––

–5

›NA

NA

219

Hydrochloricacid

ND

Str

+#

10

›–

––

–NA

NA

220

Ionomycin

ND

Str

+M

0.16

0.04

0.08

0.08

0.16

NA

NA

Ca

221

Isopropylphenol

ND

+#,M

100

100

––

–NA

NA

222

Malonic

acid

ND

+#

2,500

––

––

NA

NA

M

223

Methoxyphenol

ND

+N

50

––

–13

NA

NA

224

Monensin

ND

Str

+#

0.1

–0.8

1.6

0.8

NA

NA

M

225

Napthylisothiocyanate

ND

+#,M

,C,P

100

100

100

100

NA

NA

226

Nitrosodim

ethylamine

ND

+M,N

100

6–

–6

›NA

NA

227

Nitroproprionic

acid

ND

+#

600

––

2,500

–NA

NA

228

Pentachlorophenol

ND

+M,N

25

6–

100

6NA

NA

OS

229

Pyrazinamide

ND

Str

+#,M

19,000

19,000

––

–NA

NA

590

Table

2(C

ontd.)

Sl.no

Drug

Toxicity

Old

tests

HCS

test

First

signal

Cell

number

Mitochondrial

potential

Ca

Mem

perm

Nuclear

area

Cmax

(lM)

TI

PPB

(%)

Mechanism

230

Rotenone

ND

Str

+#,P

0.1

0.2

60.1

0.4

50

0.002

M,O

S

231

Ryanodine

ND

False�

-–

––

––

NA

NA

Ca

232

Sodium

hydroxide

ND

+#

5,000

––

––

NA

NA

233

Taurocholic

ND

+N

––

––

25

NA

NA

M

234

Taurodeoych

olic

ND

+N

1,250

2,500

––

310

NA

NA

M

235

Taurolithoch

olic

ND

+N

310

––

–156

NA

NA

M

236

Thioacetamide

ND

+M

100

6–

––

NA

NA

237

Trichloroethylene

ND

+M

100

6–

––

NA

NA

Sensitivity

ND

98%

Negative

controls:chem

icals

non-toxic

tohumans(atconcentrationupto

100

lM)

238

3-A

cetamidophenol

ND

––

––

––

–NA

NA

239

Ascorbate

�–

––

––

––

NA

NA

240

Culture

media

ND

––

––

––

–NA

NA

241

Ethanol

ND

––

––

––

–NA

NA

242

Lactose

ND

––

––

––

–NA

NA

243

Sorbitol

�–

––

––

––

NA

NA

Overall

sensitivity

(drugs)

21%

93%

Overall

specificity

(drugs)

90%

98%

About243drugsandcompoundsare

tabulated(column1)in

differentcategories

ofhumantoxicityandcomparedwithrespectto

howthey

tested

inconventionalcytotoxicityassays(column3),thetoxicitythey

produce

(column2),andhow

they

tested

intheHCS,sublethal,cytotoxicityassay(columns4–10).Drugsandcompoundsare

categorisedaccordingto

theirhumanhepatotoxicityas:

(1)severelyand(2)

moderately

hepatotoxic

drugs,(3)non-toxic

drugs,(4)chem

icalstoxicto

other

organs,(5)toxic

chem

icals,and(6)non-toxic

chem

icals.ResultsfortheHCScytototoxicityassayare

tabulatedfortestoutcome

anddegreeofpositivityin

column4,parameter

affectedatthelowestconcentrationin

column5,andin

columns6–10fortheconcentrationscausingeff

ects

oncellnumber,mitochondrialmem

branepotential,

intracellularionised

calcium

concentration,mem

branepermeabilityandnucleararea,respectively.Themaxim

alplasm

atotalconcentrationofdrugsassociatedwitheffi

cacy

(Cmax)isindicatedin

column11.The

ratiooflowestcytotoxicconcentrationto

Cmaxisindicatedin

column12(TI).Thepercentageofdrugthatisboundto

plasm

aproteinsisalsoindicated(column13),andwhereknownthecellularmechanism

of

cytotoxicity(column14).ThesensitivitiesoftheconventionalassaysandtheHCSassayare

indicatedin

therow

atthebottom

ofthelist

ofdrugsin

each

category

ND

notdetermined,NA

notapplicable,?nottested

to30

·Cmaxandtherefore

diagnosisuncertain.Mechanismsoftoxicityare

designated

asfollows:

Apapoptosis;

Cacalcium

dyshomeostasis,

Card

cardiotoxicity,Hem

hem

atologic,IM

immune-mediated,LA

lactic

acidosis,M

mitochondrial,OPoxidativephosphorylation,OSoxidativestress,SynDNA-synthesis,Transmem

branetransporter

inhibition,

Superscriptiidiosyncratichepatotoxicity,Superscriptrreactivemetabolite,

›signalincreased.Toxicities(column2and14):thenumber

inthiscolumnrefers

tothe%

incidence

ofpatients

withincreasedliver

functiontests,

Agagranulocytosis,

Bhyperbilirubinem

ia,H

hepatitis,

BBW

black

boxwarning,BM

myelotoxic,BW

boxed

warning,CA

cardiacarrhythmia;C

cholestasis,

Cicirrhosis,

Ddermatotoxic,F

fibrosis,FH

fulm

inatinghepatitis,GIgastrointestinaltoxicity,HIHepatocellularinjury

(withnecrosis),Hm

hepatomegaly,Hthepatotoxicity,Iincidence

ofelevatedliver

functiontests(but%

notgiven),J

jaundice,

LD

liver

dysfunction,Myomyotoxic,N

neurotoxic,O

ototoxic,Panpancreatic,

Rrenotoxic,Ststeatosis,W

regulatory

warning,Wdwithdrawn,X

deaths

591

pamidronate, primaquine, pyrimethamine, rosiglitazone,sulphanilamide and tacrine.

Preincubation of cells for 7 days resulted in prepa-rations from which little data could be obtained. Cell

counts obtained were highly variable and substantiallyreduced. This was attributed to the low seeding density,combined with lengthy preincubation time, resulting inthe clonal expansion of small numbers of cells intoirregular clumped colonies for which individual cellscould not be distinguished.

Based on the low frequency of detection of toxicitywithout drug preincubation and on the clumping of cellsgrown for 7 days, a three-day exposure protocol wasselected for standardisation of the assay protocol for thecollection of data reported in Tables 2 and 3 and Fig. 2(bottom), 3 and 4. In Table 2, cytotoxic effects of drugsin conventional and the new multiparametric assay aretabulated.

Time-course of cytotoxic change

Above a threshold drug concentration, cytotoxic effectsprogressively increased with time of incubation duringthe 3.4 h period that the cells were monitored for. Theseeffects occurred more rapidly, and followed a dose-re-sponse pattern. The threshold concentration producingeffect, the magnitude of the effect, the sequence in whichparameters were affected and whether they were in-creased or decreased, varied across drugs.

With the exception of TMRM, cell fluorescenceintensities were constant over time for the negativecontrols and for cells exposed to drugs not producingacute toxicity, or exposed at non-toxic concentrations oftoxic drugs. Fluorescence of TMRM decreased gradu-ally over time, possibly due to its extrusion by thep-glycoprotein transporter, and/or to photobleaching.Toxic effects were considered to occur only when therate of change of fluorescence was unmistakeably greaterthan for the negative controls.

In Fig. 3, the time course of change in fluorescenceparameters is indicated at the critical concentrationwhere each drug produced toxic effects. Effects aredemonstrated for a representative single cell and for apopulation of cells for each drug at the concentrationproducing toxicity. The patterns of change are quitesimilar for single cells and populations of cells, althoughthe former is more precise in determining the sequenceof events. For example, for amiodarone and dantrolene,individual cell data confirmed that calcium was the last

Fig. 1 Photomicrographs of furazolidione-induced changes inprelethal cytotoxicity parameters. Effects of preincubation ofHepG2 cells with 100 lM furazolidone, compared to controls,respectively, on nuclear area (a, b), mitochondrial membranepotential (c, d), calcium (e, f), membrane permeability (g, h) and cellnumber (i, j). Furazolidone produced a marked reduction in cellnumber, nuclear area, and mitochondrial membrane potential, andmarked increases in intracellular calcium and membrane perme-ability. See HCS analyser specifications and settings for details ofinstrument settings and procedures. Circular outlines indicate theareas within cells in which fluorescence intensity is measured. Cellsare viewed with the 20· objective for fluorescence intensityquantitation and the 10· objective for cell counting

b

592

parameter to change, and that mitochondrial membranepotential change substantially preceded membrane per-meability change.

As TMRM is known to be a substrate for the multi-drug resistance (MDR) p-glycoprotein that extrudescations, the drug concentration causing fluorescenceincrease was compared to the concentration at which thedrug inhibits MDR. Maximal TMRM fluorescence in-creases were 100, 200, 0 and 150% by, respectively,amiodarone, paroxetine, dantrolene and astemizole.

Although all four of these drugs inhibit MDR, there wasnot a correlation between the degree of TMRM increaseand potency of inhibition, 34, 16, 22 and 89%, respec-tively.

Antiproliferative effect could only be measured inassays in which cells were exposed to drug for multipledays. For the four drugs that produced effects with andwithout preincubation for multiple days (acetamino-phen, mibefradil, quinine and cerivastatin), the mostconspicuous difference in patterns of parameter change

10

20

30

Nuc

lear

are

a

Amiodarone

0200400600

TM

RM

0255075

100

Flu

o-4

0300

600

900

TO

TO

-3

Paroxetine

0

50

100

150

Cel

l C

ount

0.1

0.4

1.6

6.4 25 10

00 0.1

0.4

1.6

6.4 25 10

00

Dantrolene

0.1

0.4

1.6

6.4 25 10

00

Astemizole

0.1

0.4

1.6

6.4 25 10

00

0.1

0.4

1.6

6.4 25 10

00 0.1

0.4

1.6

6.4 25 10

00 0.1

0.4

1.6

6.4 25 10

00 0.1

0.4

1.6

6.4 25 10

00

Drug Concentration (uM)

10

20

30

Nuc

lear

are

a

Acetaminophen

0

200

400

600

TM

RM

0

200

400

600

Flu

o-4

0

750

1500

2250

TO

TO

-3

Mibefradil

0

50

100

150

Cel

l C

ount

Quinine Cerivastatin

Drug Concentration (uM)

Fig. 2 Cytotoxic effects ofspecific drugs before and after3-day preincubation of cellswith drugs. Combinedconcentration-response andtime-course of changes areshown in cell proliferation,mitochondrial function,intracellular calciumhomeostasis, cell membranepermeability and nuclear areafor drugs producing immediatetoxicity without preincubation(top half; amiodarone,paroxetine, dantrolene,astemizole) and for drugsproducing toxicity after 3-daypreincubation of drugs and cells(bottom half; acetaminophen,mibefradil, quinine,cerivastatin). At eachconcentration indicated, ninemeasurements onapproximately 40 cells in asingle field-of-view were madewith 36 min intervals fornuclear area (Hoechst 33342;top), mitochondrial membranepotential (TMRM; second fromtop), intracellular ionised Ca(Fluo-4; third from top),plasma membrane permeability(TOTO-3; second from bottom)and cell number (bottom). Thewells without drug additionserved as negative controls. Theeffects of FCCP for TMRM,Tween-20 for TOTO-3 andnuclear area and ionomycin forFluo-4 served as positivecontrols. Toxic effects occur atthe concentration producing adownward time-curve forTMRM (greater than thatoccurring for the negativecontrol) or nuclear area or anupward time-curve for Fluo-4or TOTO-3. Data is expressedas mean ± SEM

593

was the effect of cell number (Fig. 3). This was the leastaffected of the measured parameters with acute toxicity.With chronic toxicity, it was the parameter affected atthe lowest toxic concentration of dantrolene and aste-mizole and at intermediate toxic concentrations foramiodarone.

Sequence and direction of drug-induced change (Fig. 3)

The sequence of change in parameters was similar withthe four different drugs that caused toxicity without theneed for preincubation (Fig. 2, top). Nuclear area usu-ally changed early and marginally, then mitochondrialpotential, with or without permeability change, thenpermeability change if it had not already occurred andfinally Ca effects occurred last. Three of the four drugs,excepting dantrolene, had dual and opposing effects onTMRM fluorescence, showing an initial increase fol-lowed by a decrease. This dual biphasic effect was alsonoted for nuclear area for amiodarone and paroxetine(see Hormesis).

For amiodarone, the sequence of change in the dose-response curves (Fig. 2) was TMRM fluorescenceincrease at 0.2–0.8 lM, nuclear shrinkage at 6 lM,

mitochondrial inhibition at 13 lM, plasma membranepermeabilisation at 50 lM, then Ca rise at 100 lM.Potential was decreased by one-third at 13 lM, and bytwo-thirds at 25 lM and higher. However, at lowerconcentrations than this, potential was increased pro-gressively up to 100% at 12 lM. Nuclear area decreasedby 10% at 6 lM, 15% at 13 lM and by 30% at 100 lM.Membrane permeability was increased to about half ofthat of the positive controls at 100 lM. Cell counts werenot adversely affected.

Paroxetine was tested on the same plate as amioda-rone. The sequence of changes were: nuclear shrinkageat 3–6 lM, TMRM fluorescence increase at 6 lM,mitochondrial inhibition at 25–50 lM, permeabilisationat 25–50 lM, Ca rise at 100 lM and cell count decreaseat 50–100 lM. Nuclear area decreased 10% at 3 lM,15% at 6 lM and 30% at 25 lM. TMRM signal in-creased by 20% at 6 lM and by 80% at 25 lM. At50 lM, it rose over time by 200% to its maximum valueand then fell to half of its baseline value; at 100 lM itfell to zero. Membrane permeability started to increaseat 25 lM, reached half-maximal values at 50 lM andmaximal values at 100 lM. Calcium rose a slightamount and only at 100 lM. Cell counts apparentlydecreased mildly at 50 and 100 lM.

Table 3 Sources of imprecision in HCS, sublethal, cytotoxicity assay

Imprecision due to well and plate variation

KSR assay parameter Variation between wellswithin plate

Variation between plates

(% CV) N (% CV) N

Mitochondrial membrane potential 9.36±1.33 7 14.60 5Nuclear area 5.16±1.80 7 6.99 6Calcium 7.25±4.15 3 18.35 4Membrane permeability 9.30±7.02 3 16.64 6Cell count 16.45±4.30 7 16.10 5

Imprecision due to field-of-view and cell variations (data for representative plate OP18)

KSR assay parameter Variation between fields Variation between cells Variation betweenwells

(% CV)±SD N (% CV)±SD N (% CV) N

Mitochondrial membrane potential 8.50±4.04 4 30.23±10.44 7 9.83 7Nuclear area 2.84±1.24 4 17.13±8.14 7 6.03 7Calcium 10.28±0.02 4 41.57±20.17 7 5.67 3Membrane permeability 9.02±3.96 4 35.74±9.47 7 10.65 3Cell count 29.81±9.25 10 N/A N/A 21.47 7

The cell-to-cell, field-to-field, well-to-well and plate-to-plate imprecisions are compared for the parameters measured in the HCS assay.Imprecision due to well and plate variation. For well-to-well variance, a mean, SD, and CV were calculated for every control well for eachassay parameter. An average CV was then used to show the well-to-well variance for that parameter. Plate-to-plate variance was thenestimated using the well-to-well data for each plate. The well-to-well mean for each assay parameter for all six plates was averaged and aCV calculated. This CV value then shows the plate-to-plate variance for each parameter. Imprecision due to field-of-view and cellvariations (data for representative plate OP18): To estimate the field-to-field variance, the average value of each parameter for eachindividual field was calculated for all seven control wells on the plate. This gave a CV for each well, which when averaged gave the field-to-field variance for that plate. For nuclear area, TMRM, calcium and membrane permeability for four fields were used (as this was themaximum number of fields used for the assay), whilst ten fields were used for the cell count analysis. To estimate the cell-to-cell variancethe KSR parameter value for individual cells was recorded. Seven groups of seven cells were randomly chosen (using a random numbergenerator) within the same field of view and within the same well. The mean, SD and CV were then calculated for each group of seven cellsand an average CV value recorded

594

Dantrolene effects were in the sequence: nuclear areadecrease at 6–13 lM, potential decrease at 13 lM andmembrane effects were much less and occurred only at100 lM. There was no clear change in Ca. Cell countswere not adversely affected.

Astemizole effects were in the sequence: TMRM flu-orescence increase at 0.2 lM, nuclear shrinkage at6 lM, permeability increase at 13–25 lM and Ca rise at25 lM. The TMRM signal increased by 50% at 0.2 lM,100% at 1.6 lM and by 150% at 6 lM. However at13 lM it rose over time to this level then fell to baseline.At 25 lM it had decreased 90% from the maximum.Nuclei were apparently, transiently enlarged by 10% at6 lM and shrunk by one-third at 13 and 25 lM.

Membrane permeabilisation began at 13 lM and at25 lM was about one-quarter that of the positive con-trols. Cell counts were not adversely affected.

Hormesis

As occurred without preincubation, some drugs pro-duced dual effects, with increases followed by decreaseson TMRM fluorescence in 26% drugs. This was espe-cially conspicuous for those drugs that were stronginhibitors of the MDR pump (such as cyclosporine Aand quinacrine), where TMRM progressively increasedby approximately 100 and 150%, respectively (from 1.6

Amiodarone 50 uM

0

20

40

60

80

100

120

Hoechst TMRMFluo-4 Toto-3

Paroxetine 100 uM

0

40

80

120

160

Amiodarone 50 uM Cell 28

Paroxetine 100uM Cell 25

Dantrolene100 uM Cell 55

Rel

ativ

e F

luo

resc

ence

Inte

nsi

ty (

arb

itra

ry u

nit

s)

Astemizole 25 uM

0

30

60

90

120

Dantrolene 100 uM

0

20

40

60

80

100

120

Astemizole 25 uM Cell 4

Single Cell Effects:

t1 t2 t3 t4 t5 t6 t7 t8 t9 t1 t2 t3 t4 t5 t6 t7 t8 t9 t10

Time (0- 3 h)

Average Cell EffectsFig. 3 Kinetics of averagedcompared to individual cellularchange for drugs producingacute toxicity. See Fig. 2 fordetails of assay. The patterns ofchange in fluorescences of thedifferent dyes for a populationof cells are similar to that ofindividual cells. However, thesequence of change can be moreprecisely determined at thesingle cell level

595

to 13 lM, and 0.1 to 0.2 lM). Dual effects were alsonoted for cell number for 12% compounds showingpositive responses, and especially for nuclear area, where34% of compounds showing positive response.

Figure 4 (left) demonstrates dual, biphasic responsesfor cell number, nuclear area and mitochondrial mem-brane potential for several drugs. This phenomenon wasseen with 43% drugs and could not be attributed toartefacts of the position of wells within plates.

Hormesis could not be detected for calcium normembrane permeability, likely because measures of

these parameters in healthy cells could only be measuredin one direction.

Sources of imprecision

Cell-to-cell, field-to-field, well-to-well and plate-to-platevariances after 3 days of drug treatment are shown inTable 3. The greatest imprecision was in measurementsbetween cells within field-of-view. Next most variable wasmeasurements between field-of-views, then between

0

125

250

375

500 Amiodoarone

Cel

l Cou

nt

-9 -8 -7 -6 -5 -4 -3 -20

200

400

600

800Ketoconazole

Molar Concentration

TM

RM

Sig

nal

19

22

25

28

31Etoposide

Nuc

lear

Are

a

250

500

750

1000 Cyclosporin A

TM

RM

Sig

nal

16

20

24

28 Diquat

Nuc

lear

Are

a

800

1200

1600

2000 Sulfamethoxazole

Cel

l Cou

nt

-9 -8 -7 -6 -5 -420

40

60

80

100

IC50

= 0.42 uM

SE= 0.16 uM

r2 = 0.92

Cerivastatin (M)

Cel

ls p

er F

ield

0

150

300

450

600IC

50=0.90 uM

SE= 0.19uM

r2 = 0.88

Cyt

osol

ic C

alci

um

5

10

15

20

25

IC50

1.20 uM

SE= 0.16uM

r2= 0.97

Nuc

lear

Are

a

0

500

1000

1500

2000 IC50

=2.3 uM

SE= 0.11uM

r2 = 0.99

Mem

bran

e P

erm

eabi

lity

0

150

300

450

600IC

50= 3.0uM

SE= 0.3 uM

r2 = 0.87

TM

RM

Sig

nal

Fig. 4 Hormesis in dose-response for drug-inducedcytochemical effects. IC50 forcerivastatin-inducedhepatotoxicity. Biphasic dose-response curves occurred clearlyfor cell count, nuclear area andmitochondrial potential.Representative examples areshown. Curve-fitting to thedose-response data forcerivastatin cytotoxicity isshown for each of the fiveparameters assessed, with IC50

indicated as mean and errorestimate

596

plates. Measurements between wells within plate wereusually the most precise, varying between wells byapproximately 5% for nuclear area, 10% for mitochon-drial membrane potential, calcium and membrane per-meability and 15% for cell count. Nuclear area was by farthe most precise measurement for all types of variance.Mitochondrial membrane potential was nextmost preciseacross cells, fields-of-view and across plates. Most vari-able of parameters was calcium, especially between cells.

The high variance could be attributed in part to notcollecting data greater than a threshold value that wouldensure only viable cells were assessed (thresholding).Because of the rapid rise and fall in Ca compared to thespeed of analyses, timing of data acquisition once ion-ophore was added was neither precise nor consistent,resulting in counts including dead or dying cells that hadlost fluorescent dye. This could be overcome by ensuringthat positive controls are in wells spread across thewhole of the plate at regular intervals. Because of thetime it takes for the KSR to scan through each well,positive controls are measured approximately 40 minafter addition of ionomycin. Thresholding could also beused for exclusion of dying cells from the cell count.Dead cells are assumed to have lifted off the well surfaceand therefore not counted. Cell count variance is highboth between wells and between plates, due to clumpingof proliferating cells resulting in an uneven distribution.

IC50 Determination

Estimates of IC50 values could be made for those drugsfor which a near complete range of toxic effects could bedetermined in the concentration range studied, up to 30-fold Cmax or 100 lM (Table 4). IC50 values could beobtained for 39 of 82 toxicants (48%). The IC50 washighly correlated with the concentration at which effectswere first seen. For cell number the correlation coeffi-cient was 0.75 (P<0.0001) and the IC50 was 1.27±0.19fold this concentration.

Based on an IC50 analysis, the parameter that wasmost sensitive in detection of cytotoxicity was cell countfor 56% of toxicants. The next most sensitive parameterswere nuclear area and calcium, detecting 33 and 26% oftoxicants, respectively. The least sensitive parameter wasmitochondrial membrane potential. The frequency withwhich a parameter was the most sensitive or second mostsensitive was 73 % each for cell count and nuclear area,48% each for calcium and membrane permeability and28% for mitochondrial membrane potential.

The ranking of the sensitivities of the differentparameters for detection of cytotoxicity was somewhatdifferent when assessed based on determination of IC50

values than when based on the concentration at which aclear effect was detected. Both approaches indicated thatcell number and nuclear area were the most sensitiveparameters. However, mitochondrial membrane poten-tial was ranked as the least sensitive based on IC50 val-ues, whereas membrane permeability was ranked as the

least sensitive based on the parameter to first change.This difference likely reflects the fact that the point offirst clear effect measures low dose enhancement whichapparently delays the point of 50% loss of signal com-pared to baseline.

In Fig. 4 (right), concentration-response curves withIC50 values are shown for each of the five parametersmeasured for cerivastatin. Half-maximal inhibition oc-curred first for cell number at 0.4 lM, then intracellularCa and nuclear area at about 1 lM, then membranepermeability and mitochondrial potential at 2–3 lM.Hormesis is seen to occur for both TMRM and cellnumber resulting in these parameters being affected atdoses of 1 and 0.1 lM, respectively.

Predictivity of the HCS, sublethal, cytotoxicity assay

Comparison of effectiveness of different cytotoxicityparameters

Table 2 demonstrates that cell number was by far theparameter most frequently found to be affected by3 days exposure to toxic drugs and the most sensitiveindicator of cytotoxicity. About 86% of the 201 drugsthat produced a positive test response affected thisparameter, although it was not directly determinedwhether this was from antiproliferative or cell lytic ef-fects. In contrast, only 70% of drugs testing positiveaffected nuclear area and mitochondrial membrane po-tential, and only 45% affected membrane permeabilityand intracellular calcium concentration. Cell numberwas the first parameter affected in 56% of drugs testingpositive, with mitochondrial membrane potential andnuclear area being affected first for 30%, and cell per-meability and calcium affected first for only 10 and 7%,respectively.

Cell permeability was the least sensitive parametersfor detection of cytotoxicity, being the first parameteraffected in only 7% of drugs (Table 1). As for all theparameters, this frequency was independent of theseverity of hepatotoxicity and was the same for toxicitiesaffecting other organs.

The measurement of nuclear size was the most preciseof all parameters. It decreased at toxic concentrations by50% for compounds such as astemizole, chloroquine,chlorpromazine, danazol, disulfiram, furazolidone, ke-toconazol, mibefradil, paroxetine, primaquine, quina-crine, quinidine, quinine, tacrine, tamoxifen andvalproate. It decreased by about 25% for amodiaquine,cyclosporine A, dantrolene, fenfluramine, imipramine,labetalol, methyldopa and novobiocin. However, it in-creased by 20% for niacin, norfloxacin and vidarabine,and by 50% for acetaminophen.

Sensitivity and specificity

The sensitivity for detection of severely hepatotoxic,moderately hepatotoxic and other organ toxic drugs

597

was, respectively, 20, 24 and 14% for the conventionalcytotoxicity assays, and 100, 88 and 92% for the mul-tiparameter, HCS cell health assay. Specificity fordetection of organ toxicity was 98% for the 35 non-toxicdrugs and 100% for the 6 different negative controls.Overall sensitivity and specificity were 90 and 98%,respectively, for the multiparametric assay.

Safety margin estimation

The ratio of the cytotoxic concentration in vitro to the invivo concentration associated with efficacy was assumedto provide an indication of the safety margin or thera-peutic index for the drug. This ratio was independent ofwhether toxicity was direct or metabolite-mediated. The

ratio was 12±22 for hepatotoxic compounds testingpositive and 20±34 for compounds toxic to other or-gans.

Variation of assay sensitivity with drug concentration

Assay sensitivity was found to be dependent on theabsolute drug concentration tested and the safety margin(Fig. 5): 30%at 1 lMor aTI of 10 lM; 60%at 30 lMoraTI of 10 lMand 80%at 100 lMor aTI of 30 lM.Plotsof the false positive rate versus the false negative rate, thatis, receiver operating characteristic curves, indicate theoptimal sensitivity that the assay can provide withoutcompromising specificity (Fig. 5). Sensitivity falls offrapidly for specificity greater than about 93%.

Table 4 IC50s for concentration-response curves for drug induced effects on cell proliferation, mitochondrial function, intracellularcalcium homeostasis, cell membrane permeability and nuclear area

Drug Nuclear area Mitochondrial potentialCalcium Membrane permeabilityCell count

Amiodarone 9.8±0.33 19.2±0.42 17.2±0.04 15.8±0.29 25.6±5.2Amodiaquine 20.0±0.25 35.8 13.5 17.6±0.34 24.8Astemizole 2.6±0.49 5.6 4.1 4.1±0.14 10.1±4.9Cerivastatin 1.2±0.16 3.0±0.30 0.9±0.19 2.3±0.11 0.4±0.11Chlorpromazine 12.6 20.7 24.4±3.6 20.8±0.4 6.4±1.0Chloroquine 11.1±0.22 27.2±.02 22.4±0.11 22.9±0.21 16.0±6.3Cyclosporin A 10.1±0.3 >100 35.1 78.8±0.6 14.8±3.8Danazol 58.8±0.1 73.6 43.1±1.7 81.4±0.1 39.2±19.3Dantrolene 52.1±0.4 105.0±0.1 67.1 >100 14.5±31.6Diquat 85.8±6.6 43.2± 5.4 79.2±8.1 75.8±8.2 39.6±11.2Doxorubicin 0.7±0.34 3.1 0.2±0.23 0.4±0.20 0.2±0.03Etoposide >500 534.6 1927±553 772.7±173.3 0.14±0.05Fenfluramine 77.1±10.0 >100 261.8±25.6 195.8±22.1 62.9 ±8.7Fluvastatin 11.9±0.55 92.6±0.20 13.4±0.75 51.6±0.39 106.7Furazolidone 90.9±0.14 86.8±0.74 76.2 58.2±0.12 28.7±13.5Furosemide 2286 4285 2890±181 2360±8 1208±177Imipramine >100 84.9 35.4±0.4 76.6±0.3 153.8Ketoconazole 85.9±13.1 80.2 71.6±0.2 62.2±0.2 89.7±61.2Ketorolac 215.7±77.3 >450 >450 693.3±163.8 247.7±47.2Ketotifene 91.2±15.8 >100 70.0±0.5 86.5±0.7 27.6±6.1Labetalol 95.2±18.6 >100 81.9±1.2 106.0±8.0 80.2±9.0Lovastatin 14.9±0.78 65.5 17.9±1.11 86.3±0.49 104.2±54.1MethylDOPA 257.7±42.8 187.3±23.0 >670 >670 257.4±30.8Mevastatin 25.2±0.7 330.4 24.7±1.0 97.7±2.7 >100Mibefradil 5.8±0.17 5.6 16.9±0.60 29.5 8.8±2.4Novobiocin 397.3±89.2 202.0±49.9 458.9±18.7 227.3±52.6 183.2±19.9Pamidronate 52.9 70.1 110.0±0.1 63.3 63.2±15.7Paroxetine 7.1 4.7±0.93 14.3±0.05 7.7 17.6±4.4Picotamide >100 12.5±5.4 93.3±51.6 104.5±67.2 19.6±3.8Primaquine 28.1 27.8 58.9±0.08 55.2±0.14 31.4±11.6Propranolol 13.9±0.41 >100 57.7 61.6 49.0±10.4Pyrimethamine 212.3±128.2 >100 124.2±10.8 77.8±3.8 3.8±0.6Quinacrine 2.3±0.18 2.6±0.51 1.4±0.31 2.7±0.32 1.4±0.17Quinidine 72.8 118.8±26.2 118.3±8.2 102.5±1.5 14.7±2.4Quinine 38.3 ±2.8 65.0±26.8 >600 40.5±1.3 22.3±2.4Simvastatin 16.1±0.42 51.7 18.2±0.56 40.2 51.4±25.0Sodium chloride >100 mM 71.6 mM 106.6 mM 95.1 mM 20.0 mMTacrine 54.3 99.8 58.3±0.06 87.5 37.2±18.1Valproic acid 6546±1309 >16200 >16200 15930±915 1520Most sensitive assay (%) 33 15 26 5 562nd most sensitive (%) 38 13 20 44 181st or 2nd most sensitive (%) 72 28 46 49 74

Values are reported as mean or, when there was sufficient data, mean ± SE. The parameter with the lowest and second lowest IC50 for eachdrug is indicated in bold lettering or by italics, respectively

598

Discussion

Conventional cytotoxicity assays

This study demonstrates that conventional cytotoxicityassays have poor concordance with human toxicity. Thispoor predictivity of cell-based assays for in vivo toxicitylikely reflects at least in part on the lateness of the pointin the sequence of pathogenic events that they assess(Table 2). Assays that target late events in the process ofcell injury, when the cell is near death, are more likely tomiss toxicities that require chronic exposure or exertadverse but non-lethal effects. Many of the conventionalcytotoxicity assays (e.g. LDH release, cell rupture,membrane blebbing) are for late-stage toxicity and cel-lular events associated with a lethal apoptotic or necroticeffect (Jaeschke et al. 2002; Olson et al. 2000). Such as-says have low sensitivity (Table 1) for detection of ad-verse cellular effects and furthermore provide littlemechanistic understanding of the toxicologic effects inhumans. Conversely, assays that provide early assess-ment of specific toxicologic mechanisms in cells (Ta-ble 2), prior to the onset of the late stages of non-specificdegeneration and apoptotic or necrotic death, shouldtheoretically have greater predictive power and extrap-olatability across models and species. While these ret-

rospective analyses support the value of conventional invitro cytotoxicity screens in identification of the‘‘overtly’’ toxic compounds, it points to the need offurther refinement in the method to predict subtle orsub-lethal adverse events that account for the majorityof side effect profiles of human pharmaceuticals.

Although none of the conventional cytotoxicity as-says had adequate concordance with in vivo humantoxicity, it is important to note which of the endpointsused by these tests were most predictive. Glutathioneassay for oxidative stress was by far the most sensitive(Table 1), supporting the proposal of the important roleof oxidative stress in drug toxicity (Xu et al. 2004). Thetwo indicators that were next most sensitive were theAlamar Blue assay for mitochondrial reductive activityand the DNA assay for cell proliferation, both havingsensitivities an order of magnitude greater than the testsfor membrane integrity, protein synthesis, superoxideand caspase-3. These findings support the inclusion inthe HCS assay of parameters reflective of cell prolifer-ation, mitochondrial function and oxidative stress. Thelatter though is currently not yet incorporated into HCSscreening, although studies are under way to do this(Phillips et al. 2005).

An additional important finding in this study ofconventional cytotoxicity assays is that multiple anddifferent early measurements of toxicity mechanisms are

Frequency Distribution of CytotoxicConcentration for Toxic Drugs

0 30 60 90 120 1500.0

0.2

0.4

0.6

0.8

1.0

80% at100 to150 uM

60% at30 uM

Threshold (uM)

Sen

sitiv

ity

Frequency Distribution of TIfor Toxic Drugs

0 30 60 90 120 1500.0

0.2

0.4

0.6

0.8

1.0

90% atTi of10080% at

Ti of 30

Threshold TIS

ensi

tivity

Receiver Operating Characteristic Curvefor Optimising Threshold TI

0 2 4 6 8 10 12 140

20

40

60

80

100

TI dataLine for random chance

AUC = 0.92TI = 100

False Positive Rate(100% - Specificity)

Fal

se N

egat

ive

Rat

e (S

ensi

tivity

)

Fig. 5 Balancing assay sensitivity with specificity in definingdiagnostic cut-off cytotoxic drug concentrations relative to effica-cious concentration (TI). Receiver-operator characteristic curve forTI from 102 hepatotoxicants and the 23 non-toxic drugs for whicha TI was determined are plotted. Curve was only slightly differentwhen all toxicants were included. A cutoff TI of 100 gave 93%specificity and 90% specificity. Area under the curve (AUC) was

0.92, corresponding to a toxic drug have 92% probability of havinga higher TI than a non-toxic drug. Frequency distribution curvesshow the proportion of toxic drugs that are identified at increasingcut-off cytotoxic concentrations or TI’s (cytotoxic concentration/efficacious concentration [Cmax]) and indicate that a cut-off basedon TI has superior diagnostic value than based on concentrationalone

599

critical for detection of human toxicity potential. Pre-dictivity was additive for several of the conventionalassays when combined, e.g. mitochondrial activity, glu-tathione and cell proliferation. The data from the HCScytotoxicity assay also indicate the need for measure-ment of multiple, independent and early mechanisms ofcytotoxicity. As well, similar conclusion that multipledifferent toxicity endpoints need to be measured hasrecently been made by others (Schoonen et al. 2005a, b).

Retrospective analysis of the predictivity of testingstrategies using a set of marketed drugs that have al-ready been screened for those that were thought to betoxic, can only result in substantial underestimation.However, such approach provides a common ground forcomparisons across testing strategies. Thus, whereaspredictivity for human hepatotoxicity (Table 1 and 2) ofregulatory animal toxicity testing, conventional cyto-toxicity tests and HCS sublethal cytotoxicity tests mustnecessarily be underestimated using this approach, thisunderestimation should be proportionately the same forall testing strategies assessed.

Predictivity of the HCS, sublethal cytotoxicity assay

Hormesis

The concentration-response curves for many drugs withtoxic effects were characterised by early positive re-sponses in cell proliferation, nuclear area and mito-chondrial potential, prior to negative responses in theseparameters. Hormetic effects were seen only after 3 daysof preincubation with drug, with the exception ofmitochondrial membrane potential, for which biphasiceffects occurred acutely as well as after preincubation(see Fig. 2). The occurrence of these positive responsesprior to degenerative effects is suggestive of them beingcompensatory, adaptational and protective (Olson et al.2001; Calabrese and Baldwin 2002). The basis for thehormetic effect on nuclear area is unknown, but mayrelate to a specific inhibition of the cell cycle in whichnuclear area is increased followed by a degenerative ef-fect in which the nucleus shrinks. The hormetic effectsfor cell proliferation and mitochondrial potential re-sulted in IC50 values being higher (see Fig. 4, right) be-cause they were defined as the concentration at whichbasal activities were decreased by 50% rather than theconcentration at which maximal activities were de-creased by 50%.

Hormesis was found for approximately half thedrugs, for nuclear area in 34% of the 136 drugs affectingit, for mitochondrial membrane potential in 26% of the141 drugs affecting this parameter, but for cell number inonly 12% of the 173 drugs affecting it.

Requirement for exposure to drugs for multiple days

The current study clearly demonstrates that preincuba-tion of cells with drugs for multiple days is typically

needed for expression of their toxicity. Predictivity wasincreased by an order of magnitude by preincubatingcells with drugs for 3 days. This finding reasonably ex-plains why sensitivity of the HCS cytotoxicity assay is anorder of magnitude greater than that of conventionalcytotoxicity assays. Support of this can be found in re-cent work where cells were preincubated with drugs for3 days then subjected to conventional cytotoxicity as-says performed independently (Schoonen et al. 2005a,b). This finding of the effect of preincubation of cells anddrugs for multiple days is also consistent with our pre-vious observations (Slaughter et al. 2002; O’Brien et al.2003; Xu et al. 2004) and recent studies by Schoonenet al. (2005a, b). Whereas the preincubation time wasnot titrated, the finding that cell proliferation was themost highly sensitive (although least specific) indicatorof toxicity, indicates that preincubation should extendover multiple doubling times.

Identification of idiosyncratic hepatotoxicityand hepatotoxicity due to reactive metabolites

The HCS sublethal cytotoxicity assay correctly identifieddrugs that produce toxicity via metabolites and drugsthat produced idiosyncratic hepatotoxicity. Surprisingly,after 3 days of pre-incubation with drug, the assay wasable to detect toxicity thought to be mediated by reactivemetabolites of the parent drug equally as well as fordrugs that produced their toxicity directly. Most of thesetoxicities are thought to be caused by formation ofreactive metabolites that are frequently linked withoxidative stress (Kalgutkar et al. 2005). Also surprisingwas that the assay was able to detect most of the 12drugs that have been associated with idiosyncratic hep-atotoxicity. It is noteworthy, however, that there wererelatively few drugs that were clear in these categories.Cell count was affected by all of these compounds andmitochondrial membrane potential and nuclear area wasaffected by 11 of them. Calcium and membrane perme-ability was affected by only half of them.

Most sensitive parameters are cell proliferation,mitochondria and nuclear area; membrane permeabilityand calcium are least sensitive

Cell proliferation inhibition in most cases preceded thechanges in other parameters. The high sensitivity of thisparameter may reflect its dependency on the normalfunctioning of a wide range of physiological processes.On the other hand, this sensitivity comes at the expenseof specificity for the toxicologic mechanism. It is note-worthy that cell proliferation was assessed by measure-ment of cell number, which could be affected not only byeffects on cell proliferation but also by cell death. Thus adecrease in cell number could be attributed to a cyto-static effect or both. However, in most cases, cell pro-liferation was affected prior to any discernible effect onthe sub-lethal cytotoxicity parameters, indicating that

600

the effects on cell number were sub-lethal rather thanlethal. Results from this study support the idea thatcytostatic effects are a consistent and early effect of sub-lethal cytotoxicity. However, cytostatic effects may oc-cur in the absence of cytotoxic effects, as is shown byeffects of specific inhibitors of cell proliferation that haveno other adverse effects, such as azathioprine, colchi-cines, cyclophosphamide and methotrexate. Thus,additional measures of adverse effects on cells are nee-ded to interpret the cytostatic effect. These drugs did,however, affect the nuclear area at the same time as cellnumber.

Nuclear area was found to have comparable sensi-tivity as cell count for detection of toxicants. The leastsensitive parameter, based on occurrence of a significantchange, indicated that cell membrane permeability wasleast sensitive. However, ranking of parameters’ sensi-tivities for cytotoxicity detection based on IC50 valuesindicated that mitochondrial membrane potential wasthe least sensitive. The basis for these contradictoryfindings likely relates to the hormesis effect occurring inthe dose-response relationship for mitochondrial mem-brane potential but not detected for membrane perme-ability. Based on the FCCP control well results from arandom analysis of ten plates, the background signalcontributes to 24.4% (SD=8.0%) of the total TMRMsignal. Therefore, the useful dynamic range of the assayto measure mitochondrial uncoupling may limit sensi-tivity.

The finding that membrane permeability and intra-cellular calcium were the least sensitive parameters isexpected on the basis that membrane integrity andmaintenance of an intact barrier to the extracellularmilieu are essential for cell function and are maintainedwith highest priority by the cell. Rise in intracellularcalcium is an important final event in the live of a cell. Itis noteworthy that a more sensitive dye with greateraccuracy and precision for measurement of intracellularcalcium at basal levels, rather than at cytotoxic levels,may be a more effective biomarker for cytotoxicity. Forexample, studies with indo-1 have shown that there maybe subtle elevations in resting intracellular calcium andimpairments in calcium regulatory capacity with geneticand toxicologic cellular pathologies (O’Brien et al. 1989,1990).

Whereas intracellular calcium was the second leastsensitive parameter for detection of toxic effect, it was anearly indicator for several classes of drugs, such as thestatins (cerivastatin, fluvastatin, mevastatin), local ana-esthetics, (bupivacaine, lidocaine), antimalarials (amo-diaquine, chloroquin, quinacrine, quinidine, quinine),neuroleptics (amitryptilline, trifluoperazine, paroxetine),anthracyclines (doxorubicin), amiodarone, tacrine andcaptopril. The basis for this early involvement of cal-cium dyshomeostasis is uncertain, but may relate to themechanism of cytotoxicity.

Cell membrane permeability change was an early ef-fect in only 7% of drugs causing a positive response, e.g.doxorubicin, pimozide, dipyrone, flucytosine, foscarnet,

novobiocin, rotenone and sulphanilamide. Such disrup-tion of the barrier to the extracellular medium inevitablyproduced immediate and rapid increases in mitochon-drial permeability and in intracellular calcium. In con-trast, mitochondrial changes did not produce immediateand rapid changes in cell membrane permeability.

Cytotoxicity test results reported herein are sup-ported by the literature. Comparison of the HCS resultswith those for 7 conventional, independent, cytotoxicityassays from recent studies by Schoonen et al. (2005a, b)is possible because of overlap for 30 of the drugs andcompounds studied (acetaminophen, amiodarone, aspi-rin, benzopyrene, bromobenzene, carbon tetrachloride,chlorpromazine, colchicine, cyclophosphamide, dantro-lene, dexamethasone, diclofenac, diethylmalemide,doxorubicin, erythromycin, estradiol, flutamide, genta-mycin, imipramine, indomethacin, isoniazide, ketoco-nazole, labetalol, nitrofurantoin, quinidine, rotenone,sulphaphenazole, tacrine, tamoxifen, tetracycline).There was high correlation of results between studies forall parameters, especially Alamar Blue, glutathione andmembrane permeability (r�0.8), although lesser withDNA, ATP and NADPH (r�0.6) and least with reactiveoxygen species (r=0.46). This correlation supports thefindings of the current study, especially for use of amitochondrial function biomarker and glutathione.

Assessment of human toxicity potential: TI, PPB, AUC

The significance of the cytotoxic signals should beinterpreted in terms of the ratio of cytotoxic concen-tration to the concentration causing efficacy. The latterwas estimated by comparing with the maximal totalconcentration of the drug in human serum that is asso-ciated with administration of the drug at an efficaciousdose. However, the degree of plasma protein bindingwas not considered in this study. As there is only onetenth as much plasma protein in the in vitro systemcompared to in vivo, significant protein binding wouldbe expected to result in an overestimate of the circulatingfree drug compared to in vitro, with consequent pro-portionate underestimate of the safety margin andoverestimate of the toxicity potential. For example,rosiglitazone’s safety margin was underestimated with-out consideration of the high plasma protein binding.Thus, this ratio should be considered an estimate of theminimal safety margin. Finally, in this context, the bestestimates of safety margin for most drugs should bebased on drug exposure to free concentration per unittime (i.e. area under the concentration time curve,AUC). However, these values were not available for thein vitro studies.

The risk associated with a low safety margin needs tobe considered with respect to the indication and to thedose being used. Lower safety margins will be acceptedfor drugs intended for treatment of life-threatening dis-eases for which there are no equivalent alternatives.Lower safety margins may also be accepted for drugs in

601

which the ingestion is limited by the bulk required fortoxicity or by side-effects such as vomiting.

It may also be relevant to interpret the significance ofthe signal based on the degree of change and the numberof parameters affected, and the mechanism and thesteepness of the concentration-response curve.

Application of prelethal cytotoxicity testing

Whereas drug-induced cytotoxicity may indicate poten-tial for in vivo human hepatotoxicity, it is not predictiveof such. Cytotoxic effects in vivo may be limited or evenaggravated, compared to those occurring in vitro.Cytotoxicity models are limited by their incompletemodelling of the cell type’s structure and function as itoccurs in vivo, by their incomplete modelling of othercell types and of cell functions and interactions within atissue, organ, systems and whole body, e.g. (a) drugproperties, concentrations, protein binding and trans-port may differ in vivo; (b) pharmacokinetic character-istics of absorption, distribution, metabolism andexcretion can have a major influence on which targetorgan is affected and the severity of toxicity; (c) toxicitiesoccurring at the tissue or organ level such as cholestasis,cataractogenesis and myelotoxicity cannot be effectivelypredicted from cellular systems; (d) toxicities may occursecondarily to direct cytotoxicity and due to the inter-action of organs and systems, and other processes suchas inflammation, immune-mediated hypersensitivity,plasma volume expansion and endocrine effects.

Prelethal cytotoxicity assay as first tier of laboratorytesting for organ toxicity potential Although cytotoxicityassays cannot predict human hepatotoxicity and cannotbe unequivocally used for ‘‘go - no go’’ decision-makingin the progression of drug discovery, they neverthelesshave value. Knowledge of the cytotoxicity of drugs earlyin drug discovery can create opportunities for rankingand prioritizing, or developing alternatives with lowertoxicity potential and can also flag the need for furtherscholarship and experimental safety assessments if thecompound is to progress to preclinical development.Such assessments would need to include conduct ofmore mechanism-specific, or possibly tissue-specific, invitro assays, as well as in vivo animal studies. These invitro models may also provide valuable mechanisticunderstanding of the toxicity.

Cytotoxicity testing throughput Throughput is animportant consideration in assessment of the practicalityof an assay to be implemented for screening compoundsin drug discovery. Throughput of the assay in whichcells are exposed for 3 days to 12 concentrations of eachdrug then measured at three different time points over3 h is 14 drugs per standard workweek for a singleoperator. However, the assay can be abbreviated so thatonly one time point is measured and only one or a few

concentrations are measured (such as the minimallyacceptable multiple of the concentration at efficacy thatwould have no effect) then throughput can be increasedup to many hundreds of assays per week. Furtherthroughput enhancement would likely require adapta-tion of the assay to a 384 well-plate format.

Optimal drug concentration and safety margin for pre-dicting human toxicity potential The current study indi-cates the concentration of drug needed for assessment ofhuman toxicity potential. At a concentration of 30 lM,60% of drugs with human toxicity potential were iden-tified, whereas 100 lM identified about 80% (Fig. 5).These concentrations are considerably lower, that is theassay is more sensitive, than previous reported assays(Bugelski et al. 2000; Schoonen et al. 2005a, b).Assessment of toxicity potential was more accuratewhen the concentration of drug tested was based onmultiples of the total efficacious concentration (Cmax formarketed drugs), with 80% of cytotoxicities being de-tected at a concentration of 30 times the efficaciousconcentration, Cmax. In the current study, toxicitiesoccasionally went undetected by the HCS cytotoxicitytest when the Cmax had not been determined and thedrug was not tested up to 30-fold the efficacious con-centration.

False test results in the HCS prelethal cytotoxicityassay Examination of the false positives and negativesmay provide understanding of what toxicities the HCScytotoxicity assay is not relevant for. As described above,idiosyncratic heptoxicants and heptotoxicities attributedto reactive metabolites were detected. However, therewere several specific examples of toxicities that were notdetected in the current study. For some of these drugtoxicities, this could be attributed to the specific knowneffect of the drug on molecular targets that are not foundin heptocytes. For example, the assay applied to HepG2cells did not detect cholestatic effects of estradiol, cal-cium-channel effects of ryanodine, potassium channeleffects of astemizole and terfenadine, renal toxicity ofzomepirac, dermatotoxicity of isoxicam and hematologictoxicity of vincamine. Additionally, the human toxicitypotential was not detected at relevant concentrationsin the HCS cytotoxicity assay for bupropion, diethyl-carbamazine, naproxen, pravastatin and trifluopera-zine. Furthermore, whereas picotamide is essentiallynontoxic, it was found to be highly toxic in the cyotoxicityassay. The number of such false test results was too smallto draw definitive conclusions, and additional workwould be needed to confirm these proposals.

Limitations and need for further studies

Although the HCS sub-lethal cytotoxicity assay hadgreater than 90% concordance with human toxicities, itfailed to detect up to 10% of these. The basis for the

602

failure of the HCS cytotoxicity test for the other drugs isunknown, but may be partly attributable to eitherincomplete metabolic competence of the cell-basedmodelused or else to toxicities being caused by, or mediated by,extra-hepatocytic effects that require involvement ofsome other molecular transporter or process, cell, tissue,organ or system that cannot be represented by a hepa-tocyte cell line model. However, that the assay was posi-tive for most drugs that produce their toxicities by areactive metabolite argues against metabolic competencebeing significantly limiting. The extrahepatic toxicitiesmay involve (a) cytokine, growth factor or hormoneproduction, (b) immune-mediated toxicity and (c) extra-hepatocytic, tissue-specific transporters which are foundin hepatobiliary or renal tubular cells.

Metabolic competence Developing further metaboliccompetence, either by using extracellular incubation ofdrugs with S9 microsomal fractions, or by using cellsthat are more competent metabolically, may enhancepredictivity. Primary human hepatocytes would theo-retically provide the best model of metabolic competencefor cytotoxicity assays. However, current culture meth-ods in 96-well plates do not stabilise the phenotype ofprimary cells and prevent their rapid dedifferentiationover the several days of incubation needed for most drugtoxicities to be expressed. Furthermore, primary cells arealso less suitable for cytotoxicity assessment because oftheir low rates of cellular proliferation, which is the mostsensitive measure of hepatotoxicity potential.

Given the need for a proliferating cell model forpredictive cytotoxicity studies, the effectiveness of thechoice of HepG2 cells as a cell line to use is supported byother studies. Schoonen et al. (2005a, b) found themslightly more predictive than HeLa, ECC-1 and CHO-k1cells. Others have made similar findings (Bugelski et al.2000). However, a cell line with metabolic competenceand morphology more representative of the in vivo statewould likely further enhance the predictivity of cyto-toxicity assays.

Data processing We have identified two areas whereimprovement in analysis would result in increased effi-ciency with regard to performing larger scale cytotox-icity screens on a routine basis. First, the number ofcellular measurements generated by the image analysisalgorithms used far exceeded those relevant for this as-say. This required handling and archiving of more datathan was required. We are currently pursuing the use ofstreamlined assay algorithms that generate only the re-quired assay output. In addition, we are investigatingways to further automate data analysis of large volumesof kinetic data, especially the calculation of IC50 valuesusing user-definable mathematical models.

Assay refinement The Cellomics KSR was used suc-cessfully to apply automated HCS assays to investigate a

large set of experimental variables including cell platingdensity, cytotoxic indicators, drugs, drug concentrationsand incubation times. However, further improvementmay be obtainable by refinement and optimisation ofcell culture factors such as cell substrate, media and drugreplacement during preincubation, and assessment ofdrug transport and stability. Imprecision will likely bedecreased and diagnostic sensitivity improved by refine-ment of culture conditions to facilitate cell attachmentand more uniform spreading across the well surface,and by reducing evaporation and uneven heating andoxygenation of wells from the edges to the center of theplates.

Further studies are needed to define optimal strate-gies for cytofluorescent monitoring of relevant cytotox-icity biomarkers. The current study indicates significantroom for improvement in dye and parameter choice,especially with respect to membrane permeability andalso intracellular calcium concentration, at least asmeasured with a non-ratiometric dye with relatively lowcalcium-binding affinity compared to basal, intracellularconcentration of calcium. The effectiveness and mecha-nistic information of the HCS cytotoxicity assay may beenhanced by replacement of these dyes with others thatcould measure glutathione or reactive oxygen species,lysosomal mass, safety-signal transduction, additionalmorphological cell features, cell cycle information indi-cated by nuclear staining or the mass, reductive activityor DNA content of mitochondria. The specific need toincorporate an oxidative stress biomarker is also indi-cated by the results herein on conventional cytotoxicityassays, our preliminary studies currently underway(Phillips et al. 2005), and by the importance of oxidativestress in drug-induced toxicity (Xu et al. 2004).

Further studies are also needed to distinguish artefactfrom toxic response from adaptive responses (eg mito-chondrial potential dyes); to better model in vivo met-abolic competency; to sensitise to cytotoxicity andfurther enhance assay sensitivity; and to quantitatepredictivity and correlation with human hepatotoxicityand its mechanistic basis. Finally, more non-toxic drugsneed to be tested to more accurately assess the frequencyof false positives and the appropriate safety margin tointerpret cytotoxicity.

References

Abraham VC, Taylor L, Haskins JR (2004) High content screeningapplied to large-scale cell biology. Trends Biotechnol 22:15–72

Andrade RJ, Carmargo R, Lucena MI, Ganzalez-Grande R (2004)Causality assessment in drug-induced hepatotoxicity. ExpertOpin Drug Saf 3:329–344

Barhoumi R, Bailey RH, Burghardt RC (1995) Kinetic analysis ofglutathione in anchored cells with monochlorobimane.Cytometry 19:226–34

Bernardi P, Petronilli V, Di Lisa F, Forte M (2001) A mitochon-drial perspective on cell death. Trends Biochem Sci 26:112–117

Boelsterli UA (2003a) Disease-related determinants of susceptibil-ity to drug-induced idiosyncratic hepatotoxicity. Curr OpinDrug Discov Devel 6:81–91

603

Boelsterli UA (2003b) Idiosyncratic drug hepatotoxicity revisited:new insights from mechanistic toxicology. Toxicol MechMethods 13:3–20

Brain P, Cousens R (1989) An equation to describe dose responsewhen there is stimulation of growth at low doses. Weed Res29:93–96

Bugelski PJ, Atif U, Molton S, Toeg I, Lord PG, Morgan, DG.(2000) A strategy for primary high throughput cytotoxicityscreening in pharmaceutical toxicology. Pharm Res 17:1265–72

Calabrese EJ, Baldwin LA (2002) Applications of hormesis intoxicology, risk assessment and chemotherapeutics. TrendsPharmacol Sci 23:331–337

Cerep (2005) In vitro Pharmacology and ADME-Tox CatalogProduct No.900–59. Cerep, Redmond

Colombo B, Felicetti L, Baglioni C (1965) Inhibition of proteinsynthesis by cycloheximide in rabbit reticulocytes. BiochemBiophys Res Commun 18:389–95

Fung M, Thornton A, Mybeck K, Hsiao-Hui W, Hornbuckle K,Muniz E (2001) Evaluation of the characteristics of safetywithdrawal of prescription drugs from worldwide pharmaceu-tical markets—1960–1999. Drug Inf J 35:293 – 317

GenStat (2002) GenStat for Windows Release 6.1, 6th (edn.). VSNInternational Ltd, Oxford

Giuliano KA., Haskins JR, Taylor DL (2003) Advances in highcontent screening for drug discovery. Assay Drug Dev Technol1:565–577

Graphpad Prism (2004) Graphpad Prism version 4.01 for Win-dows. Graphpad Software Inc, San Diego

Gurtu V, Kain SR, Zhang G (1997) Fluorometric and colorimetricdetection of caspase activity associated with apoptosis. AnalBiochem 251:98–102

Haskins JR, Rowse P, Rahbari R, de La Iglesia FA (2001) Thia-zolidinedione toxicity to isolated hepatocytes revealed bycoherent multiprobe fluorescence microscopy and correlatedwith multiparameter flow cytometry of peripheral leukocytes.Arch Toxicol 75:425–438

Jaeschke H, Gores GJ, Cedeerbau AI, Hinson JA, Pessayre D,Lemasters JJ (2002) Mechanisms of hepatotoxicity. Toxicol Sci65:166–176

Kalgutkar AS, Gardner I, Obach RS, Shaffer CL, Callegari E,Henne KR, Mutlib AE, Dalvie DK, Lee JS, Nakai Y,O’Donnell JP, Boer J, Harriman SP (2005) A comprehensivelisting of bioactivation pathways of organic functional groups.Curr Drug Metab 6:1–65

Kaplowitz N (2001) Drug-induced liver disorders. Implications fordrug development and regulation. Drug Saf 24:483–490

Kaplowitz N (2005) Idiosyncratic drug hepatotoxicity. Nat Rev4:489–499

Lee WM (2003) Drug-induced hepatotoxicity. New Engl J Med349:474–85

Levesque A, Paquet A, Page M (1995) Improved fluorescent bio-assay for the detection of tumor necrosis factor activity. JImmunol Methods 178:71–6

Lewis JH (2002) Drug-induced liver disease. Curr Opin Gastro-enterol 18:307–313

Lewis W, Day BJ, Copeland WC (2003) Mitochondrial toxicity ofNRTI antiviral drugs: an integrated cellular perspective. NatRev Drug Discov 2:812–822

Lorico A, Masturzo P, Villa S, Salmona M, Semeraro N, deGaetano G (1986) Gentisic acid: an aspirin metabolite withmultiple effects on human blood polymorphonuclear leuko-cytes. Biochem Pharmacol 35:2443–5

Meselson M, Stahl FW (1958) The replication of DNA. ColdSpring Harb Symp Quant Biol 23:9–12

Nociari MM, Shalev A, Benias P, Russo C (1998) A novel one-step,highly sensitive fluorometric assay to evaluate cell-mediatedcytotoxicity. J Immunol Methods 213:157–67

O’Brien PJ, Kalow BI, Brown BD, Lumsden JH, Jacobs RM(1989) Porcine malignant hyperthermia susceptibility: halo-thane-induced increase in cytoplasmic free calcium of periph-eral blood lymphocytes. Am J Vet Res 50:131–135

O’Brien PJ, Kalow BI, Ali N, Lassaline LA, Lumsden JH (1990)Compensatory increase in calcium extrusion activity of un-treated lymphocytes from swine susceptible to malignanthyperthermia. Am J Vet Res 51:1038–1043

O’Brien PJ, Slaughter MR, Biagini C, Diaz D, Gao, B, Irwin W,Krejsa C, Hougham C, Abraham V, Haskins JR (2003) Pre-dicting drug-induced human hepatotoxicity with in vitro cyto-toxicity assays. In: Proceedings Tox ’03, London, UK

Olson H, Betton G, Stritar J, Robinson D (1998) The predictivityof the toxicity of pharmaceuticals in humans from animal da-ta—an interim assessment. Toxicol Lett 102–103:535–538

Olson H, Betton G, Robinson D, Thomas K, Monro A, Kolaja G,Lilly P, Sanders J, Sipes G, Bracken W, Dorato M, Van DeunK, Smith P, Berger B, Heller A (2000) Concordance of thetoxicity of pharmaceuticals in humans and in animals. RegulToxicol Pharmacol 32:56–67

Olson H, Kadyszewski E, Beierschmitt W (2001) Hormesis—apharmaceutical industry perspective. Crit Rev Toxicol 31:659–61

Phillips GW, Irwin WA, Howard-Cofield EJ, Randle LE, AbrahamV, Haskins J, O’Brien PJ (2005) Incorporation of an oxidativestress biomarker into high content screening (HCS) for humantoxicity potential. Proc Soc Biomol Screen

Plymale DR, Haskins JR, De La Iglesia FA (1999) Monitoring ofsimultaneous subcellular events in vitro by means of coherentmultiprobe fluorescence. Nat Med 5:351–355

Russo MW, Galanko JA, Shrestha R, Fried MW, Watkins P(2004) Liver transplantation for acute liver failure from drug-induced liver injury in the United States. Liver Transpl10:1018–1023

Schoonen WGEJ, Westerink WMA, de Roos JADM, Debiton E.(2005a) Cytotoxic effects of 100 reference compounds on HepG2 and HeLa cells and of 60 compounds on ECC-1 and CHOcells. I. Mechanistic assays on ROS, glutathione depletion andcalcein uptake. Toxicol In Vitro 19:505–516

Schoonen WGEJ, Westerink WMA, de Roos JADM, Debiton E.(2005b) Cytotoxic effects of 110 reference compounds on HepG2 and for 60 compounds on HeLa, ECC-1 and CHO cells. I.Mechanistic assays on NAD(P)H, ATP, and DNA contents.Toxicol In Vitro 19:491–503

Slaughter MR, Thakkar H, O’Brien PJ (2002) Effect of diquat onthe antioxidant system and cell growth in human neuroblas-toma cells. Toxicol Appl Pharmacol 178:63–70

Xu JJ, Diaz D, O’Brien PJ (2004). Applications of cytotoxicityassays and pre-lethal mechanistic assays for assessment of hu-man hepatotoxicity potential. Chem Biol Interact 150:115–128

604

HCS- screening for toxicity (Arch Toxicol)

Documents

nd nd nd nd nd nd nd

os os os os os

os syn

os ap ca

os hcs test

ca i myo

o cns n pan

ap r r rm