We sought to develop a high-throughput assay to study both antiproliferative potency and mechanism of action of cell cycle-targeting drugs. High-throughput microscopy enables direct counting of cells. Optimization of sample preparation conditions (fixation, permeabilization and staining) and image analysis has enabled a one-step no-wash assay that is also quantitative for DNA content, and hence cell cycle distribution. MitoTracker staining only required an extra reagent addition step, since confocal imaging practically eliminated background fluorescence. The elimination of any requirements for aspiration or washing steps will also greatly facilitate implementation of this assay in 1536-well format.
More complex detection and analysis methods (e.g. immunostaining for cell-cycle regulated proteins and phospho-epitopes, multiple nuclear morphology, texture and intensity parameters) have been used to identify cell cycle sub-populations by high-content analysis (for example 
). However we chose to use monoparametric DNA-content binning for several reasons. One of the key objectives of optimizing a no-wash protocol was to ensure that all cells, including detached and fragmented apoptotic cells, are retained, hence immunostaining is not possible. Differentiating G2 from M cells based on nuclear morphology is possible with the cell-level analysis parameter we measured (see for example 
), but in many cases compound treatment results in abnormal morphologies which do not closely correspond to any of the populations found in untreated cells.
Comparison of direct cell counting with two commonly-used ‘proliferation’ assays that are based on cellular metabolism; ATP content and MTS reductase activity, revealed a common and significant artifact in that, under conditions of cell cycle arrest, the assumed linear relationship between assay signal and cell number breaks down. The average amount of ATP (as inferred from increased Luciferase/Luciferin bioluminescence) or MTS-reducing activity per cell is significantly increased. This increase correlates with, and can be explained by, drug-induced increases in per-cell cell size and consequently mitochondrial content.
In comparison, estimation of viable cell number using total DNA fluorescence (CyQuant assay) was less prone to deviation from the actual cell number. Previous studies have also reported differences in cell number determination between DNA quantification and metabolism-based assays 
. However, treatments that significantly changed the average DNA content per cell by inducing mitotic arrest and/or endoreduplication also led to an underestimation in the percent change in cell number with the CyQuant assay.
In general, changes in cellular ATP content and MTS activity resulted in one of two types of deviations between cell number and ATP/MTS assay. First, there were cases where Emax was greatly reduced, i.e. very shallow dose-response curves. For example gemcitabine reduces ATP signal for HT29 cells by approximately 20%, when in fact cell number has been reduced by 80% relative to control. The other situation is exemplified by etoposide, where the EC50 is right-shifted but the dose-response curves converge to similar Emax at a sufficiently high concentration. This could lead to underestimation of antiproliferative potency by 10-fold or more. VX-680 presents an intermediate case, where there is a biphasic ATP curve with an intermediate plateau corresponding to an elevated ATP/cell ratio, followed by a second decline. Curves like this fit poorly, if at all, to the standard 4-parameter model and can result in very inconsistent results in terms of both EC50 and Emax.
These different response profiles can be explained by the mechanisms of action of these compounds: There is a biphasic cell cycle dose-response to many of these drugs. First, on-target antiproliferative or cytostatic responses coincide with the reduction in absolute cell number but little cell death. Under these conditions the arrested cells increase in size and mitochondrial content, and correspondingly, the amount of ATP per cell increases. At higher drug concentrations the population phenotype may become less exclusively cytostatic, depending on cell line and treatment time. With increasing late-apoptotic (sub-G1) fraction the average MitoTracker intensity and ATP and MTS per cell declines. For example, for HT-29 cells, aphidicolin and gemcitabine resulted in S or G2 arrest and elevated mitochondrial and ATP content per cell across a wide concentration range, whereas p53-wild type A375 and A549 cell lines underwent a phenotypic switch at higher gemcitabine concentrations, where a greatly increased increased apoptotic fraction correlated with less average ATP per cell. Etoposide on the other hand induced elevation of ATP and MTS activity and mitochondrial mass over a limited concentration range in all cell lines tested. This is consistent with a biphasic mechanisms of action previously observed for these drugs 
: At lower concentrations repairable DNA damage causes arrest in late S or at the G2 checkpoint with minimal apoptosis and thus accumulation of mitochondria, ATP and MTS activity per cell. At higher concentrations the DNA damage accumulates more rapidly and pervasively, causing arrest and apoptosis earlier in S-phase. A similar pattern of biphasic response explains the two-step curves observed with VX-680, where the predominant phenotype switches from cytostatic endoreduplication to predominantly 4N arrest and cell death, possibly related to off-target activities, at higher concentrations.
The behavior of the MTS assay in the case of the MEK inhibitor PD901 is unusual in that the per-cell level of MTS dehydrogenase activity is increased but the per-cell ATP amount is unchanged by drug treatment. However other kinase inhibitors; VX-680, BI-2536, and crizotinib, also caused a greater discrepancy between MTS assay and cell number than ATP. A similar observation has been reported for imatinib 
, and faslodex 
. The latter two papers also demonstrated increased mitochondrial activity and mitochondrial mass. Since there are multiple mechanisms and cellular locations of tetrazolium reductase activity 
, the observation that some treatments in this study can result in disconnects between changes in mitochondrial mass and MTS reduction (e.g. PD901) is not unexpected. The mechanism of the biphasic induction by BI-2536 of mitochondrial mass, ATP, and MTS activity at concentrations higher than the fully efficacious antiproliferative concentrations was clearly different from other kinase inhibitors, and warrants further investigation.
There have been a number of reports of chemotherapeutic agents causing increases in mitochondrial mass, including doxorubicin 
and etoposide 
. Several different mechanisms have been proposed to explain these increases. Fu et al 
proposed a direct mechanistic link where activated ATM phosphorylates and activates AMPK, thereby increasing mitochondrial biogenesis. McGowan et al 
demonstrated that induction of cell cycle arrest by enforced expression of p14ARF resulted in increased mitochondrial mass.
It is worth noting that none of the above reports examined cell size as a factor in changes in mitochondrial content, and therefore were not in a position to differentiate specific increases in mitochondrial biogenesis from baseline mitochondrial proliferation continuing in the absence of cell division. The latter mechanism seems plausible for many of the agents described in the current study. A similar observation was recently published by Kitami et al 
where a many compounds identified in a screen for increased mitochondrial mass were shown to correspondingly increase cell size.
There has also been a report of microtubule-targeting drugs affecting mitochondrial function via regulation of VDAC activity and ΔΨ by levels of free tubulin 
. In the current study we also observed an increase in ATP content despite a slight decrease in respiratory activity (both OCR and ECAR) in paclitaxel-treated cells. However we observed increases in cellular ATP levels at Emax
in response to both microtubule stabilizing and destabilizing drugs, suggesting that the level of free tubulin is not causative. Our data imply that microtubule-targeting agents increase per-cell ATP through a mechanism that is uncoupled from changes in cell size, in contrast to the DNA synthesis-targeting agents and mitotic kinase inhibitors.
While changes in respiratory function and flux clearly control the rate of ATP synthesis, it is less clear when, if at all, changes in flux result in changes in steady-state ATP concentration, which is generally under tight feedback control 
. The relationship between mitochondrial mass, membrane potential, and cellular ATP levels could also be confounded by variations in contribution of glycolysis to the intracellular ATP pool 
, however with the exception of PD901 we did not observe changes in the OCR/ECAR ratio.
In summary, it appears that there are multiple mechanisms by which different compounds can yield discrepant and misleading results in proxy assays based on energy metabolism, however investigation of specific mechanisms is beyond the scope of the current study.
This study also highlights the fact that the compound mechanisms of action and phenotypic responses frequently do not obey monotonic dose-response behavior. Correspondingly, the discrepancies between absolute cell number and ATP or MTS assay signals can vary greatly depending on the concentration tested. When non-monotonic curves are observed in proxy assays without appreciation of the underlying mechanisms of action, not only is the quality of EC50 data compromised but also valuable mechanism-of-action information discarded. There is also potential significant risk of false-negative results when using ATP or MTS assays to either screen compounds for antiproliferative activity, or cell lines for sensitivity to compounds, especially if the compound mechanisms of action and effects on cell cycle, metabolic activity, and survival are not well understood. We show that increasing compound treatment time can reduce the difference between proxy assay readout and actual cell number, by allowing more cells to proceed to apoptosis. However this comes at the cost of losing MoA information. A further rationale for the complementarity of absolute cell count and metabolic proxy assays is demonstrated by the determination of dose-, compound-, and cell line-dependent changes in the ATP/cell value, which can give extra insights in to compound mechanisms of action.
High-content imaging, even simply by DNA staining, thus not only provides more accurate information on the number of viable cells but also provides multiple parameters that offer further insight into compound mechanism of action and heterogeneity of response. The simplicity of the staining procedure and absence of wash steps also makes this approach highly amenable to high-throughput screening and compound profiling in both 384- and 1536-well format.