www.impactjournals.com/oncotarget/ Oncotarget, 2017, Vol. 8, (No.

5), pp: 8854-8866

Review

Improving anticancer drug development begins with cell culture:

misinformation perpetrated by the misuse of cytotoxicity assays
Alan Eastman1
1
Norris Cotton Cancer Center, Geisel School of Medicine at Dartmouth, Lebanon, NH, USA
Correspondence to: Alan Eastman, email: Alan.R.Eastman@Dartmouth.edu
Keywords: cytotoxicity assays, cell survival, apoptosis, target validation, synergy
Received: July 05, 2016 Accepted: October 12, 2016 Published: October 14, 2016

ABSTRACT
The high failure rate of anticancer drug discovery and development has
consumed billions of dollars annually. While many explanations have been provided,
I believe that misinformation arising from inappropriate cell-based screens has been
completely over-looked. Most cell culture experiments are irrelevant to how drugs
are subsequently administered to patients. Usually, drug development focuses on
growth inhibition rather than cell killing. Drugs are selected based on continuous
incubation of cells, then frequently administered to the patient as a bolus. Target
identification and validation is often performed by gene suppression that inevitably
mimics continuous target inhibition. Drug concentrations in vitro frequently far
exceed in vivo concentrations. Studies of drug synergy are performed at sub-optimal
concentrations. And the focus on a limited number of cell lines can misrepresent the
potential efficacy in a patient population. The intent of this review is to encourage
more appropriate experimental design and data interpretation, and to improve drug
development in the area of cell-based assays. Application of these principles should
greatly enhance the successful translation of novel drugs to the patient.

INTRODUCTION treatments were tested in these PDX models. Only a single

dose/schedule was tested for each drug or combination,
The majority of anticancer drugs that enter clinical which may explain why the majority of treatments still
trials exhibit little or no therapeutic benefit and fail to failed to induce a significant response. Importantly, every
obtain regulatory approval [1]. This high failure rate drug tested had previously been developed in cell-based
consumes billions of dollars annually, and contributes in vitro assays.
to the high cost of those few drugs that are eventually This review addresses a critical question: can we
approved. There has been much discussion over the past improve the preclinical development of drugs at an even
few years about the rate of failure of novel anticancer earlier stage, before they reach animal and human testing?
drugs in clinical trials, and many reasons have been I believe that a major problem has arisen from inadequate
proposed including poor pharmacokinetics and drug and inappropriate preclinical evaluation of drugs, and a
bioavailability, unexpected toxicity, lack of efficacy, and failure to place this development in the context of how
regulatory hurdles [2, 3]. Others have placed the blame on the drugs will be administered to patients. This is an area
the poor predictive value of preclinical models which do that the pharmaceutical industry has under-funded; it has
not effectively mimic human disease. Wilding and Bodmer been estimated that only 7% of drug development costs
[4] suggested “this has become such a widespread belief are expended on preclinical research [2], yet an increased
that it approaches dogma in the field of drug discovery and investment at this stage could reduce the exorbitant cost of
optimization, and has spurred a surge in studies devoted failed clinical trials later. But the concerns apply equally to
to the development of more sophisticated animal models.” every academic laboratory involved in target identification
As an example, it has been proposed that patient-derived and drug discovery. The rate of attrition of drugs in clinical
xenografts (PDX) better predict clinical drug response, trials could be reduced by better preclinical analysis, and
but the recently proposed use of 1000 PDX models for only advancing truly promising drugs into clinical trials.
high-throughput drug screening is financially beyond The ideal strategy to cure cancer is to kill all
the scope of almost every laboratory [5]. In fact, only 62 tumor cells while leaving enough normal cells alive that

www.impactjournals.com/oncotarget 8854 Oncotarget

 the patient survives. It is not surprising therefore that assaying the decrease in cell number (or any surrogate
cytotoxicity assays have become a mainstay of cancer drug marker of cell number) over time (Figure 2). Albeit, most
discovery. Unfortunately, there are many critical issues that treatments do not kill cells rapidly, so it is important to
are far too commonly ignored when using these assays. As extend assays for a sufficient period of time. Using this
a consequence, many inappropriate conclusions have been approach, and plotting the results as percent survivors,
made that may have little if any relevance to the treatment the potential failure of many drugs would have been
of patients with cancer. Furthermore, this misleading immediately evident.
information likely contributes to the failure of many drugs There are alternate outcomes of drug treatment
to effectively control cancer when tested in the clinical such as differentiation or senescence that can influence
setting. The intent of this review is to encourage more the apparent results of a cytotoxicity assay. In these cases,
appropriate experimental design and data interpretation. the results might be observed as 100% growth inhibition
While the focus of the discussion, and the examples with no loss of cell viability. Appropriate assays exist to
presented, is on traditional 2-dimensional cell culture, score these alternate endpoints if it is thought they might
many of the concerns would equally apply to alternate cell be occurring.
culture models such as 3-dimensional spheroids that have
been suggested as a better approximation of solid tumors. INCUBATION TIMES IN VITRO OFTEN
HAVE LITTLE RELEVANCE TO THE
MOST “CELL VIABILITY” ASSAYS DO PATIENT
NOT MEASURE CELL VIABILITY
Most in vitro cytotoxicity assays use continuous
Most cell-based drug screens use growth inhibition incubation of cells with drug, yet many drugs are
as an endpoint. The majority of studies use various then administered to a patient as a bolus. In vitro, the
commercial assays that are almost ubiquitously referred continuous incubation limits recovery whereas in the
to as viability assays even though they do not measure patient, the tumor is only exposed to the drug for a short
cell viability (Figure 1). The term “viability” implies period followed by time for recovery, a scenario that is too
a measurement of both live and dead cells, and is an frequently overlooked in vitro.
expression of the proportion that remain viable. Using The first large drug screen was performed in
these cytotoxicity assays, a reduction in signal by 50% the NCI60 panel of cell lines using a 2 day continuous
compared to control usually means there are fewer incubation with each compound followed by a
cells; it does not mean that any cells have lost viability. sulforhodamine B assessment of total protein [7]. By 1997,
Furthermore, if the rate of growth of a patient’s tumor more than 60,000 compounds had been analyzed in this
were decreased by 50%, it would still be called progressive screen [8]. A more recent screen of 481 small molecules
disease. It is critical that preclinical development define in 664 cell lines has been published, but a continuous
concentrations and schedules that result in tumor cell incubation for 72 h was used, and the endpoint was total
killing if it is to translate to tumor shrinkage in the patient. ATP level which was again misleadingly stated as being an
Tumor cell killing has alternately been measured assessment of viability [9]. A publically available database
by the ability of the cell to divide sufficiently to form of the sensitivity of ~1000 cells lines to >100 drugs also
a colony [6]. However, even colony formation assays exists (http://cancer.sanger.ac.uk/cosmic), but again relies
have limitations as they reflect the number of cells that on a 72 h continuous incubation with drug.
have proliferated to produce a colony of perhaps 50 cells Most small molecules have a short half-life in
within a given time frame (perhaps 14 days). But when patient plasma, but the consequence for target inhibition
counting the colonies, there are often smaller colonies and cell killing varies with drug. Some drugs are rapidly
and individual cells that may remain viable but whose inactivated or excreted while others may concentrate in
proliferation was slowed or temporarily blocked by the the tissues. The binding affinity for the target also varies
drug. Consequently, these assays do not truly reflect only markedly. For example, some drugs irreversibly bind to
the surviving cells, but rather those that have been able their target (gemcitabine to ribonucleotide reductase;
to grow to a defined size within a defined time period. ibrutinib to Bruton’s tyrosine kinase; afatinib to EGFR),
Cells that survive but do not form a colony within the some are pseudo-irreversible (5-fluorouracil to thymidine
time frame of the assay may mimic cells in vivo that could synthase; methotrexate to dihydrofolate reductase),
produce a relapse after initial response to therapy. while others have very high affinity for their target
While colony formation assays may be closer to (e.g., vinca alkaloid binding to tubulin). These drugs
reporting cell killing, such assays are not generally as can cause target inhibition long after any free drug has
amenable to high-throughput drug screens as the modern been cleared from the peripheral circulation. Most DNA
commercial assays. Experimentally, it is possible to damaging drugs covalently bind to DNA and require
modify a growth inhibition assay so it can detect cell checkpoint activation, cell cycle arrest and DNA repair
killing. This requires starting with more cells, and then

www.impactjournals.com/oncotarget 8855 Oncotarget

 before recovery (or death). In contrast, many new targeted Many new drugs may fail in the clinic because
therapeutics are ATP mimetics that reversibly inhibit a inadequate consideration has been given to how long a
protein kinase but rapidly dissociate from the target once target needs to be inhibited. For readily reversible drugs,
the drug concentration decreases. In this latter case, it is even daily administration to a patient may only inhibit its
usually critical to maintain chronic exposure to achieve target transiently, permitting sufficient time for recovery
continuous target inhibition. each day. Hence, transient target inhibition may kill no

Figure 1: The misuse of cytotoxicity assays. Because of its ease of application to multiple samples, and its low cost, tetrazolium
dye reduction assays such as “MTT” or MTS” (available from many companies) are frequently used. This assay measures primarily
mitochondrial dehydrogenase activity that is then extrapolated to reflect the number of cells in a culture dish. However, cells can rapidly
change the activity of these enzymes such that it may not be an accurate reflection of the cell number. The CellTiter-Glo assay (Promega)
relies on changes in ATP levels that can fluctuate rapidly with many environmental stresses so may not reflect the number of viable cells.
Alternate assays measure total cell protein yet arrested cells can markedly increase their protein content without dividing, while dead cells
still have protein. In our own experiments, we routinely quantify DNA content for high throughput assays as the possible variation per cell
is generally limited to only 2 fold (i.e., whether the cells are in G1 or G2 phase of the cell cycle) [14, 21]. However, the major problem
with all these assays it that they are almost ubiquitously referred to as viability assays when none of them measure cell viability. In an ideal
situation where mitochondrial enzymes, ATP or DNA levels do not change per cell, these assays still only measure the number of viable
cells. Consider a typical cytotoxicity experiment performed in a 96 well format. If you plate 1000 cells on day 1, the control may have 2000
cells on day 2 and 4000 cells on day 3 (blue line). If drug treatment results in 2000 cells on day 3 (green line), this is often reported as 50%
viability even though it is an increase over the starting number of cells. To express the increase, it is necessary to subtract the starting cell
number, so the increase from 1000 to 2000 actually reflects 67% growth inhibition (because the control increased by 3000). If the treated
cells have completely arrested, there are still 1000 cells on day 3 (yellow line). In this case, the results will often be reported as “25%
viability” even though there may be 100% growth inhibition and no loss of viability (albeit there is likely a mixture of dying and growing
cells at this concentration). If the drug is not killing any cells, it will not cure the tumor, so a conclusion that only 25% of the cells are viable
- implying 75% have died - would provide inappropriate optimism for a potential new therapeutic agent or strategy. If there were only 250
cells on day 3, this would indeed represent loss of viable cells (red line). However, this decrease in viable cell number can not be realized
unless the starting number of cells is subtracted from the results. Unfortunately, few people measure the signal on day 1. Examples of curves
from actual experiments of cell growth and death are presented in Figure 2.

www.impactjournals.com/oncotarget 8856 Oncotarget

 tumor cells in the patient. Similarly, many drugs may only across myeloma cell lines [15]. A short treatment may
work at a particular phase of the cell cycle. Inhibitors also provide a much greater therapeutic index if it takes
targeting mitotic kinases are an excellent example [10]. longer to kill normal cells. Hence, the time of exposure
The problem with targeting, for example, aurora kinase may provide an additional parameter that can enhance
or PLK1 is that these kinases are only needed for a brief therapeutic index, and could be just as important as the
period of time during the cell cycle. Even if such a drug administered dose. These examples highlight the need to
could quickly kill mitotic cells, the majority of cells in perform extensive preclinical investigations to develop
a tumor would still survive. Hence, such a drug would rationale schedules for effective drug administration to
need to be present for an entire cell cycle if it is to elicit the patient.
sufficient cell killing. In contrast, a drug that covalently
modifies its target may persist until a critical phase of the INCUBATION CONCENTRATIONS
cell cycle is reached. IN VITRO OFTEN HAVE LITTLE
Most of the current cytotoxicity screens camouflage
these problems by continuous exposure of cells in culture, RELEVANCE TO THE PATIENT
but as discussed, this is rarely how drugs are administered
to patients. In preclinical development, cells need to be Many publications justify their in vitro experimental
incubated with drug for various times (from a few hours protocol by stating that a drug is being used at a clinically
to days), and then the ability to recover assessed. It should relevant concentration, but in many cases this means that
also be kept in mind that cells can arrest and survive the experimental concentration approximates the peak
for long periods of time, often for much longer than a concentration achieved in patient plasma. Unfortunately,
typical cytotoxicity assay. Our experience suggests that this gives no consideration for the half-life of the drug in
recovery periods of at least 7 days should be investigated vivo, nor the amount distributed to the tumor. “Clinically
to determine whether cells die or recover (Figure 2). If relevant” has to mean much more than administration of
continuous drug exposure is required for cell killing, then a drug at a concentration that is only achieved briefly in
a continuous infusion or repetitive administration will be patient plasma.
required. For example, “metronomic therapy” is based on Consider the DNA damaging drug cisplatin as an
the concept of giving low dose therapy on a repetitive, example: the peak plasma concentration in patients is about
frequent basis. While this approach is often thought to 10 µM and its half-life is <1 hour [16]. A relevant design
target the tumor vasculature, it probably has significant for an experiment with cisplatin would be to incubate cells
impact directly on the tumor as well [11, 12]. For some with low µM concentrations for only a few hours, then
drugs such as 5-fluorouracil or cytarabine, a daily or twice remove the drug and follow the cells while they try, or
daily schedule is well established as the standard-of-care. fail, to recover. However, some in vitro experiments use
Other approaches to provide extended drug exposure concentrations as high as 100 µM and with incubations of
include liposomes that slowly release drugs over many 24-72 h [17]. At low drug concentrations, cisplatin works
days [13]. through cross-linking DNA. At these low concentrations,
We recently published on the sensitivity of 65 cell inhibition of transcription or protein function are not
lines to the Chk1 inhibitor MK-8776 [14]. The novelty of observed until cells are committed to die several days later
our screen was in using three different incubation times [18]. These effects can be observed much earlier when
(24 h, 48 h or 7 day). Following the shorter incubation, higher concentrations are used, yet these concentrations
drug was removed and cells allowed to grow until 7 days. and effects have no relevance to how the drug works in
Hence, the results after the short drug exposure reflect the patients. Experiments at these high concentrations have
ability of cells to recover. The three different incubation simply created a clinically irrelevant science. Other
times provided very different results: about 15% of the examples of inappropriate drug concentrations in vitro
cell lines were very sensitive to MK-8776 after only a 24 have been addressed in a prior commentary [19].
h incubation, 30% were sensitive after 48 h, while ~30% There are other issues that make it difficult to
were completely resistant to continuous 7-day incubation extrapolate an in vivo concentration to an in vitro assay.
with drug. These differences would have been completely For example, a drug may exhibit a long terminal half-life
missed if only a single incubation time had been used. in vivo suggesting prolonged exposure of the tumor to
These results suggest that a few tumors may be sensitive drug. However, it is possible that the concentration during
to a Chk1 inhibitor when administered as a bolus or short this terminal half-life is too low to effectively inhibit its
infusion. target. In the case of the Chk1 inhibitor MK-8776, we note
A second example is the proteasome inhibitor that the plasma concentration in patients drops below 1
bortezomib which has commonly been studied by µM by 6 h [20], and the lower concentrations thereafter
continuous incubation in cell culture, yet a recent study has are below those needed to inhibit Chk1 in cells [21].
shown that a more clinically relevant 1-h incubation with Alternately, the concentration of drug in plasma
bortezomib elicits much greater differential in sensitivity may be high, and have a long half-life but this does not
mean that the drug is bioavailable. Early studies with the

www.impactjournals.com/oncotarget 8857 Oncotarget

 Figure 2: The impact of pulsed drug treatment on long-term cell growth and death. Examples of long-term growth curves
for cells incubated with either gemcitabine or cisplatin are shown. In both cases, asynchronously growing MDA-MB-231 breast cancer cells
were incubated with drug at the indicated concentrations for 6 h, then the drug was removed, and the cells allowed to repair, grow, and/or
die over the following 8 days. The experiment was performed in a 96-well format and DNA content was assessed at each time point [14]. By
starting with sufficient cells/well, and harvesting a plate on day 0, the starting DNA content can be assessed. The growth rate of untreated
cells is limited as the wells rapidly reach high cell density, and cells whose growth is partially inhibited will eventually attain the same cell
number as controls. Cells incubated with either 150 nM gemcitabine or 20 - 40 µM cisplatin exhibit curves that would be considered “stable
disease” in a patient. Higher concentrations of both drugs clearly caused a decrease in cell number, but this was not observed until 6 or 4
days following gemcitabine or cisplatin, respectively.

www.impactjournals.com/oncotarget 8858 Oncotarget

 non-selective Chk1 inhibitor UCN-01 established plasma but whether they underwent mitotic events prior to dying
concentrations that exceeded 20 µM (far higher than the remains to be determined.
100 nM needed to inhibit Chk1) with a half-life of >200 The importance of this discussion is in highlighting
h, but it was discovered that the drug was bound avidly to that cells can arrest for a while before they die. The
the plasma protein alpha-1-acid glycoprotein and hence expectation that apoptosis, if it is going to occur, should
was not bioavailable to the tumor [22, 23]. This problem be seen within 2 or 3 days has resulted in escalation of
was not predicted from studies with cell culture or animal drug concentrations so that apoptosis occurs within a short
experiments (it does not bind avidly to murine or bovine time frame. I would recommend that one of the first assays
serum), but could have been discovered if human serum to perform with any new agent is a long term analysis to
has been added to cell cultures [24]. assess the drug concentration and time over which cells
Another issue that confounds extrapolation of die, perhaps using a simple assay such as trypan blue
in vitro concentrations to in vivo administration is exclusion or measuring a decrease in the number of viable
the impact of drug metabolism that can result in drug cells as in Figure 2. Only after it is established that cells
inactivation. Even more important perhaps are the cases die, and when, is it worth asking whether apoptosis was
where metabolism is required to activate a pro-drug; involved. There is certainly no rationale to study apoptosis
examples include cyclophosphamide irinotecan, and if cells do not die.
nucleoside analogs (e.g., gemcitabine, cytarabine) that Apoptosis provides many convenient assays, but
must be phosphorylated before they can impede DNA they are often misused or the results misrepresented.
synthesis. Many of these metabolites only occur inside Consider for example a caspase 3/7 enzyme assay
the tumor cells so their level cannot be assessed in blood. (available from many commercial sources) where the
Consequently, pharmacodynamic analysis to assess the results are obtained as optical density or fluorescence and
impact of target engagement in the tumor is required and then expressed as “fold increase in apoptosis.” The extent
this should always be included in the design of early phase of apoptosis reported is primarily dependent on the base
clinical trials. line value. If the base line apoptosis is only 1% of the
cells, then a 5-fold increase as is typical for this assay
CELLS CAN TAKE A LONG TIME TO DIE might reflect only 5% of the cells undergoing apoptosis,
leaving 95% of the cells alive [30-32]. Similarly,
Many years ago, we compared a variety of antibodies that selectively detect a caspase-cleaved
cytotoxicity assays following treatment of cells with substrate such as poly(ADP-ribose) polymerase (PARP)
cisplatin [25]. We noted a 10-fold range in GI50 values can be very sensitive, but provide no information whether
(50% growth inhibition) depending on the assay used, but the cleavage is occurring in a small percent of the cells
critical to the current review is the fact that the cells took or all of them [30, 33]. When analyzing PARP cleavage
4-6 days to die as assessed by trypan blue uptake. In a by western blotting, it is far more informative to use an
subsequent study, we observed that cells took 3-4 days to antibody that detects both the parent and the cleaved form
undergo apoptosis, consistent with the concentration and so as to be able to report the percent of cleavage [34, 35].
time for trypan blue uptake [26]. It is often feasible to kill However, a further reservation is that when 50% of PARP
cells much faster by using a higher concentration of drug, is cleaved, it is still unknown whether this represents
but this may involve a completely different mechanism 100% cleavage in 50% of the cells, or 50% cleavage
that has no clinical relevance as discussed above. in 100% of the cells. The most informative assays for
In the case of DNA damaging agents, the delay in apoptosis record the percent of cells that are apoptotic.
cell death (or apoptosis) is easy to reconcile. Cells initially The most common assay uses Annexin V to measure the
perceive the damage, activate cell cycle checkpoints, arrest relocation of phosphatidylserine to the outer surface of the
cell cycle progression, and attempt to repair the damage. cell membrane. This assay is often called an early marker
At low levels of damage, cells may recover after a few of apoptosis, yet it occurs fairly late in the apoptotic
days, which certainly contributes to the observed growth cascade as it is a consequence of caspase activity. We find
inhibition even though no cells may have died. When the that both PARP cleavage and chromatin condensation can
level of damage is too high, cells may still progress slowly occur many hours before the cells become positive for
through the cell cycle and eventually reach a decision as Annexin V [35, 36].
to whether to undergo mitosis. The damaged cells appear
to undergo a mitotic catastrophe and then apoptosis TARGET IDENTIFICATION: GENE
[27]. The mitotic catastrophe can occur prematurely KNOCKOUTS MAY NOT PHENOCOPY
in S phase-arrested cells, or after bypassing the G2 DRUG INHIBITION
arrest [27-29]. We have performed cell cycle analysis to
complement the growth inhibition curves shown in Figure Genetic approaches are used frequently to
2. Concentrations of gemcitabine that killed cells were identify and validate a target, and thereby provide the
preceded by persistent arrest in S phase (data not shown), justification for a drug development program. Yet genetic

www.impactjournals.com/oncotarget 8859 Oncotarget

 knockdown or knockout can provide a very different CYTOTOXICITY IN CELLS MAY BE A
result than transient inhibition of that protein. All proteins CONSEQUENCE OF INHIBITION OF A
participate at some point in a protein:protein interaction,
DIFFERENT TARGET THAN EXPECTED
and many proteins function as a component of a multi-
protein complex. If one protein is removed from this FROM CELL-FREE ASSAYS
complex, the function of the other components can be
unpredictable. One excellent example is the proteasome, Most anticancer drugs are identified though a cell-
wherein knockdown of one protein can cause failure of free, in vitro assay; for example, kinase inhibition or
the other proteins to form a productive proteasome and disruption of protein:protein interactions. These drugs are
consequently is lethal to the cells [37]. then added to cells and, when inhibition of proliferation
We have shown that suppression of the DNA exo/ or cytotoxicity is observed, it is assumed the compound
endonuclease Mre11 also reduces levels of Rad50 and is working through the predicted target. However,
Nbs1 because Mre11 stabilizes this protein complex [38]. this is a leap of faith that is too often not investigated
In contrast, a small molecule inhibitor of Mre11 nuclease further. In depth target validation experiments need to
activity, mirin, does not change levels of these other be performed in cell culture to ensure that the intended
proteins [39, 40]. Mre11 has several functions that are target is appropriately inhibited for the desired time. Target
either dependent or independent of its nuclease activity validation should also be confirmed in a patient once the
[41] such that a small molecule inhibitor may have a very drug enters clinical trials.
different outcome than genetic suppression. Our experience in this area relates to “BH3
A number of other examples have been reported mimetics” which target the BCL2 family of anti-apoptotic
where genetic analysis may mislead drug development. proteins. We have reported that the majority of putative
For example, it has been shown that an inhibitor of BH3 mimetics do not inhibit BCL2 proteins directly in
ATM does not phenocopy loss of the ATM protein [42]. cells, but rather induce an integrated stress response
Specifically, ATM inhibition prevented damage-induced that results in upregulation of the pro-apoptotic protein
sister chromatid exchange (SCE), whereas cells deleted NOXA [36]. The consequence is selective inhibition
of ATM exhibited normal SCE when damaged. Similarly, of MCL1, another of the anti-apoptotic BCL2 proteins.
isoform selective inhibitors of the PI3K p110-beta These putative BH3 mimetics appear to function through
can suppress tumor growth while deletion of the gene a variety of different mechanisms, and often kill cells
enhances tumorigenesis in mice [43]. This differential in a BAX/BAK-independent manner [36, 44-46]. Most
response is rationalized as a consequence of substitution recently we identified a novel pathway for the putative
by the p110-alpha isoform in the absence of p110-beta. It BH3 mimetic gossypol, demonstrating that it is an agonist
is unknown how many other examples may exist because of phospholipase A2, which as a consequence elevates
such comparisons have been rarely made. It certainly intracellular calcium, endoplasmic reticulum stress and
warrants a cautionary note, particularly as more high NOXA [47]. Several of these drugs are in clinical trials
throughput genetic approaches are being used as a starting under the erroneous expectation that they act as direct
point for drug development. inhibitors of BCL2 anti-apoptotic proteins (e.g., obatoclax,
There is a further concern for all genetic suppression gossypol). While the induction of NOXA may provide a
approaches used to identify novel targets as this strategy novel therapeutic approach, it is important to recognize
inevitably mimics continuous target inhibition and the correct mechanism as it may inform novel therapeutic
does not provide an opportunity for target recovery. As strategies and patient stratification. This problem of off-
discussed above, this may have little relevance to how target effects of putative BH3-mimetics is addressed in
a drug is administered to a patient. As a consequence, more detail in a recent review [48].
a tremendous effort may be expended on what may The ability of a compound to inhibit its target in a
eventually be deemed a poor drug target. cell-free system is often reported as IC50 (50% inhibition
Finally, a limitation of knockout cells is the inability of activity) or Ki (affinity for target) and potency in
to generate any dose-response data, a property that is the low nM range is commonly desired. An erroneous
inherent with drugs. However, one valuable use of gene assumption is often made that this concentration of drug
knockout systems is to assess the potential for off-target will also inhibit its target when added to cells. However,
activities of a drug. If a cell lacking the target gene is the intracellular concentration is usually much lower than
still viable, then administration of a drug should have no the applied drug concentration because of extracellular
impact on such a cell. protein binding, or because the cytoplasmic membrane
can be an effective barrier between drug and target.
The bioavailable concentration may also be limited by
intracellular metabolism or protein binding. Generally a
drug with nM inhibitory concentration in vitro will require
µM concentrations to inhibit its target in cells. Hence, if

www.impactjournals.com/oncotarget 8860 Oncotarget

 a drug inhibits its target at µM concentrations in a cell- which measurements are made as the cell killing is more
free system, as do many putative BH3 mimetics, and evident at longer times.
inhibits cell growth at the same concentration, then its As growth inhibition is the usual endpoint of
proposed mechanism of action in cells should normally be most cytotoxicity assays, most synergy assessments are
considered suspect. Exceptions to this concern are drugs performed at lower, sub-optimal concentrations of each
that are indeed concentrated in cells though usually into drug, so that a combination effect can be observed. If one
specific organelles such as weak acids in mitochondria or uses two drugs that each inhibit cell growth by 95%, it
weak bases in lysosomes. would not be possible to calculate a combination index.
In fact, synergy is most commonly observed at drug
OVER ENTHUSIASM FOR SYNERGY concentrations eliciting low efficacy as single agents.
This is evident in a recent investigation that combined
If the combination of two drugs works better gemcitabine and a Chk1 inhibitor [51]. The authors
than either drug alone, we begin to get excited. If the compared three different calculations of synergy (Bliss,
combination is much greater than expected, we begin to Lowe and Highest single agent) and reported that synergy
think the interaction might be synergistic. We then apply was observed at sub-GI50 concentrations of each drug.
a mathematical calculation to demonstrate that the drug Another recent example is the reported synergy between
combination is truly synergistic. But does demonstration inhibitors of Her2 and CDK4/6; the problem with this
of synergy in vitro really lead to clinically important study is that incubation with a CDK4/6 inhibitor should
advances? I believe the term synergy has been hijacked arrest the cells in G1, yet its failure to do so demonstrates
by mathematicians without understanding the critical that it was not (or minimally) inhibiting its target [52].
biological questions. An extensive review of many Hence, I question whether a combination of drugs at sub-
mathematical approaches can be found in reference [49]. optimal concentrations has any real relevance to treating
The most commonly used assessment of synergy the patient who will hopefully be administered optimal
is based on the median effect analysis whereby two doses.
drugs are tested at a constant ratio across a range of It has been stated that synergism is independent of
drug concentrations, starting with an equi-effective mechanism of action because the mass-action law-based
concentration; for example drugs are combined at a determination of synergism is mechanism independent
range of concentrations above and below the GI50. [53]. Unfortunately, the mechanism of action of two drugs
A combination index is then calculated and if it is <1, can have a critical impact on the experimental design and
the drugs are deemed synergistic [50]. While this may conclusions. This is particularly evident with drugs that
be mathematically meaningful, it may have very little perturb the cell cycle where one drug may arrest the cell
relevance to preclinical drug development. For example, cycle and thereby elicit resistance to the second drug;
a problem with medium effect analysis is that it supposes this would be deemed antagonism. There is a classic
that drugs will be administered to patients at doses that example of the cell cycle kinetic effects associated with
will have a similar (median) effect. the combination of vinblastine and cytarabine in a murine
As discussed above, most cytotoxicity assays only leukemia model [54]. When both drugs were given
measure growth inhibition, yet as shown in Figure 2, simultaneously, they were antagonistic. However, as the
higher drug concentrations may (hopefully) kill cells. An time between treatments was extended an additive or even
increase in cell death is the desired goal, but how does one synergistic effect was observed. These observations can be
express synergy when the endpoint changes? For example, explained if vinblastine treatment transiently synchronized
if one combines two drugs that are equal in efficacy to 150 the cells in mitosis, so that 16 h later, there were many
nM gemcitabine as shown in Figure 2 (i.e., 100% growth more cells progressing through S phase to be susceptible
inhibition but no apparent cell death), the additive effect to cytarabine. A similar cell kinetic association was
experimentally would be equivalent to that observed with observed with the combination of paclitaxel and cisplatin
300 nM gemcitabine. In this case the number of viable wherein greater efficacy was observed when cisplatin was
cells decreased from 100% with 150 nM gemcitabine administered after paclitaxel [55, 56]. Hence, mechanistic
to about 20% with 300 nM gemcitabine after 8 days. understanding of each drug should lead to a rationale
If these were two different drugs, each eliciting 100% schedule of administration. It should also be emphasized
survival alone but with the combination resulting in 20% that these experiments were performed in vivo in which
survival, it would be considered synergistic according the drugs were cleared by the mouse such that the tumor
to current mathematical models, yet the experimental was only exposed to each drug for a short time period.
design clearly shows only additivity. Perhaps a better Experiments using continuous treatment schedules in vitro
definition of synergy might be when the effect of the drug would have been unable to observe this effect.
combination exceeds that achieved simply by increasing CPX-351, a mixture of cytarabine and daunorubicin
the concentration of either drug alone by two fold. The in a slow-release liposomal formulation, is an interesting
magnitude of the effect also depends on the time point at exception to the apparent need for appropriate drug

www.impactjournals.com/oncotarget 8861 Oncotarget

 scheduling. The molar ratio of the two drugs (5:1) mutations in EGFR prior to prescribing such therapy, and
was derived from in vitro synergy experiments using a similar approach will need to become the standard for all
continuous drug treatments [57]. The slow release new targeted therapeutic strategies.
liposome circumvents the different clearance rates for Another example from clinical observations is a
the two drugs, and maintains this molar ratio in vivo. patient with metastatic bladder cancer whose remarkable
Improved therapeutic responses have been observed in response to the mTOR inhibitor everolimus was attributed
patients with AML [13]. As these studies did not assess to a loss-of-function mutation in TSC1, a suppressor of the
the potential role of cell cycle kinetics, it is possible that mTOR pathway [61]. It is suggested that mutated TSC1
an even greater therapeutic response might be attainable may be present in 9% of bladder cancers and so may
with different drug schedules. provide a responsive patient population. Positive response
An understanding of the mechanism of action of to everolimus has also been reported in patients with sub-
each drug in a combination is critical for another reason. ependymal giant cell astrocytomas, a disease associated
Consider a Chk1 inhibitor that inhibits its target but does with inactivation of TSC [62].
not elicit any cytotoxicity as a single agent, as we have A more extreme case is the one patient who had
shown for many cell lines [14]. If the concentration of a complete and durable response to the topoisomerase
the drug is escalated to achieve a GI50 concentration, it is inhibitor irinotecan plus the Chk1/Chk2 inhibitor
probable that the cytotoxicity is due to inhibition of some AZD7762. Genomic sequencing led to identification of a
other target. While the drug might still be inhibiting (or mutation in RAD50, a component of the ATM-dependent
killing) more cells, it is no longer relevant to the target DNA damage response pathway, and subsequent
and hypothesis being tested. If a drug inhibits its target, experiments validated this mutation as being causative in
its concentration should not be increased further, and the sensitivity of the tumor [63].
the outcome of the drug combination should perhaps Clinical trials may identify these outlier responses,
be referred to as an enhancement ratio such as the fold yet these trials would be far less expensive if potential
decrease in GI50 when the second drug is added. These responders were identified before the trial began. We
situations are perhaps much more exciting than situations discussed above our screen for cell lines sensitive to
in which both drugs are cytotoxic. Chk1 inhibitors, and discussed the availability of other
large databases that may predict sensitivity to drugs. But I
THE FOCUS ON ONLY A FEW CELL reiterate my concern that most of those screens have used
LINES CAN MISREPRESENT THE continuous drug treatments that may provide results that
are overly optimistic compared to a patient receiving bolus
EFFICACY OF A DRUG drug treatment. The detection of potential responders may
require appropriate assays across large panels of cell lines,
Despite the recent analysis of large panels of cell but that will still be far cheaper than conducting a large
lines in a few publications as discussed in Section 2, the clinical trial in which very few responders are seen.
majority of cancer biology still tends to focus in depth
on one cell line, and then only occasionally compares
outcomes in a few other lines. Such experiments miss TUMOR HETEROGENEITY AND A
the variability that exist across cell lines and that could SOLUTION FOR IT WAS RECOGNIZED
lead to the identification of novel therapeutic strategies in 50 YEARS AGO
hypersensitive subsets of patients. One relevant example
has been the extensive use of the U2OS osteosarcoma cell Tumor heterogeneity is synonymous with drug
line to study DNA damage-induced checkpoint regulation resistance. The realization that tumors contain multiple
[58, 59], yet this cell line turns out to be one of the few genetic variants, exhibit a heterogeneous phenotype and
that is hypersensitive to Chk1 inhibition [14, 21], and consequently contain clones with different responses to
consequently, the different responses to Chk1 inhibitors drug is not a new observation [64, 65]. However, even
has been missed. Put another way, the overt use of one before this realization, drug treatments had been developed
sensitive cell line may lead to the expectation that a drug to circumvent the problem. The earliest administration
should have broad therapeutic activity in patients. of chemotherapeutic drugs involved single agents, but it
The potential variation in drug responses has been rapidly became evident that drug combinations were much
primarily driven by clinical observations. For example, more effective. A series of guidelines were developed,
early clinical trials with EGFR inhibitors did not the most important of which for the current discussion
recognize that tumors with a gain-of-function mutation is that multiple drugs that do not have overlapping
in EGFR would be uniquely sensitive [60]. Additional mechanisms of resistance should be combined (certainly
preclinical analysis may have identified this sensitivity and they should avoid overlapping toxicities too) [66]. Many
saved millions of dollars in the cost of clinical trials by standard therapies involve 3-6 different drugs while
selecting only those patients with the greatest likelihood in the case of childhood acute lymphocytic leukemia,
of responding. These days all patients are screened for curative therapy has involved administration of up to 11

www.impactjournals.com/oncotarget 8862 Oncotarget

 different drugs through remission induction, consolidation combinations that has rarely been addressed. Cell culture-
and maintenance therapy [67]. The new era of targeted based research generally operates on the concept that the
molecular therapeutics has unfortunately reverted two drugs work on a single cell. In vivo, this rationale does
primarily to administration of single agents. Resistance not have to hold, and drugs may target different cells, for
to single drugs is inevitable, and the overt optimism that example tumor cells at different phases of the cell cycle, or
one new molecular targeted drug will cure cancer needs the microenvironment or vasculature. Furthermore, drugs
to be discarded. Most of these new drugs provide only do not have to be administered simultaneously but could
marginal increase in life span, and at an enormous expense be administered in alternating or sequential schedules.
[68] (imatinib and crizotinib are possible exceptions). This might reduce the potential increase in toxicity
Unfortunately, the goal of clinical trials has taken a which can result in having to decrease the dose with a
backward step to accept an “increase in life-span, ” but concomitant decrease in efficacy.
the dream of a cure is still possible if we build effective Today, we can dream of true precision medicine
multi-drug combinations. where tumor DNA is sequenced to predict which drugs
A palpable tumor has >109 tumor cells, though will be most effective against a particular tumor. Tumors
most advanced tumors present with far more cells. Given have multiple mutations, and hopefully many Achilles
a spontaneous mutation frequency at any locus of 10-6, heels, and hopefully many drugs can be identified that
there will already be at least 1000 cells resistant to target these susceptibilities. If these drugs are selective
any drug at the time of presentation or relapse (usually for the tumor, then it should be possible to combine them
many more as the mutation frequency in a tumor with without increased toxicity and thereby provide effective
genomic instability is thought to be much higher). Indeed, combination therapy.
mutations eliciting resistance to targeted therapies have
been identified in biopsies collected prior to therapy [69, SUMMARY
70]. One problem in clinical trials is that we wait until
the tumor has reappeared before initiating a second line The efficacy of a drug in a clinical trial is usually
of therapy, yet resistance to that therapy will already recorded as stable disease, partial or complete remission.
have developed. Chemotherapy (which includes targeted These endpoints are far removed from most of the
therapies) would be much more effective if administered preclinical research in which growth inhibition has been
when there is minimal disease. Debulking (surgery) is one the mainstay of cytotoxicity assays. Drugs need to induce
means to reduce tumor load, as is commonly used in breast cell killing if they are going to shrink the tumor. Drug
cancer patients, and is then followed by adjuvant therapy concentrations and exposure times in culture should
even if there is no detectable tumor. Yet surgical options reflect and guide future administration schedules. A simple
are often limited, detection of tumor margins difficult, and change in expression of data to “surviving cells” may go a
micrometastases missed. I believe we should not discard long way to emphasizing how much better our preclinical
the well-established chemotherapy agents that have been drug development needs to be.
the mainstay of therapy for many decades, and are still
the most effective drugs in use. While these drugs have ACKNOWLEDGMENTS
only cured some cancers, they are often effective initially
at reducing the tumor burden. Consequently, we should The ongoing work of the author is supported by NIH
be developing therapeutic strategies that combine surgery grant CA117874. The author also acknowledges support
and established chemotherapy (or radiation), and novel from a Cancer Center Support Grant to the Norris Cotton
targeted therapies should be administered when the tumor Cancer Center (CA23108).
burden is minimal.
I discussed the issue of whether drug combinations
should be considered synergistic. In designing drug CONFLICTS OF INTEREST
combinations to circumvent resistance, it will be important
to recognize that any drugs that are only effective The author declares no conflict of interest
in combination will increase the number of possible
resistance mechanisms, and resistance to either drug REFERENCES
would make the combination ineffective. Consequently,
it will be far more beneficial to combine two drugs that 1. DiMasi JA, Feldman L, Seckler A, Wilson A. Trends in
have completely independent actions, and independent risks associated with new drug development: success rates
mechanisms of resistance. Any drugs that are only for investigational drugs. Clin Pharmacol Therap. 2010;
effective in combination should be considered as a single 87:272-277.
drug when adding up the total number of drugs in the 2. Paul SM, Mytelka DS, Dunwiddie CT, Persinger CC,
combination. Munos BH, Lindborg SR, Schacht AL. How to improve
There is one final issue in developing drug R&D productivity: the pharmaceutical industry’s grand

www.impactjournals.com/oncotarget 8863 Oncotarget

 challenge. Nature Rev Drug Discovery. 2010; 9:203-214. Lucas MM, Eastman A, Kisselev AF. Molecular basis
3. Scannell JW, Blanckley A, Boldenn H, Warrington B. of differential sensitivity of myeloma cells to clinically
Diagnosing the decline in pharmaceutical R&D efficiency. relevant bolus treatment with bortezomib. PLoS One. 2013;
Nature Rev Drug Discovery. 2012; 11:191-200. 8:e56132.
4. Wilding JL, Bodmer WF. Cancer cell lines for drug 16. Himmelstein KJ, Patton TF, Belt RJ, Taylor S, Repta AJ,
discovery and development. Cancer Res. 2014; 74:2377- Sternson LA. Clinical kinetics of intact cisplatin and some
2384. related species. Clin Pharmacol Therap. 1981; 29:658-664.
5. Gao H, Korn JM, Ferretti S, Monahan JE, Wang Y, Singh 17. Wang C, Youle RJ. Predominant requirement of Bax
M, (and 58 other authors). High-throughput screening using for apoptosis in HCT116 cells is determined by Mcl-1’s
patient—derived tumor xenografts to predict clinical trial inhibitory effect on Bak. Oncogene. 2012; 31:3177-3189.
drug response. Nature Med. 2015; 21:1318-1325. 18. Sorenson CM, Barry MA, Eastman A. Analysis of events
6. Brown JM, Attardi LD. The role of apoptosis in cancer associated with cell cycle arrest at G2 and cell death
development and treatment response. Nature Rev Cancer. induced by cisplatin. J Nat Cancer Inst. 1990; 82:749-754.
2005; 5:231-237. 19. Smith MA, Houghton P. A proposal regarding reporting
7. Monks A, Scudiero D, Skehan P, Shoemaker R, Paull K, of in vitro testing results. Clin Cancer Res. 2013; 19:2828-
Vistica D, Hose C, Langley J, Cronise P, Vaigro-Wolff 2833.
A, Gray-Goodrich M, Campbell H, Mayo J, Boyd M. 20. Daud AI, Ashworth MT, Strosberg J, Goldman JW,
Feasibility of a high-flux anticancer drug screen using a Mendelson D, Springett G, Venook AP, Loechner S, Rosen
diverse panel of cultured tumor cell lines. J Nat Cancer Inst. L, Shanahan F, Parry D, Shumway S, Grabowsky JA, et
1991; 83:757-766. al. A Phase I dose-escalation trial of Checkpoint kinase 1
8. Weinstein JN, Myers TG, O’Connor PM, Friend SH, inhibitor MK-8776 as monotherapy and in combination
Fornace AJ, Kohn KW, Fojo T, Bates SE, Rubinstein LV, with gemcitabine in patients with advanced solid tumors. J
Anderson NL, Buolamwini JK, van Osdol WW, Monks AP, Clin Oncol. 2015; 33:1060-1066.
et al. An information-intensive approach to the molecular 21. Montano R, Chung I, Garner KM, Parry D, Eastman
pharmacology of cancer. Science. 1997; 275:343-349. A. Preclinical development of the novel Chk1 inhibitor
9. Seashore-Ludlow B, Rees MG, Cheah JH, Cokol M, SCH900776 in combination with DNA damaging agents
Price EV, Coletti ME, (and 20 other authors). Harnessing and antimetabolites. Mol Cancer Therap. 2012; 11:427-438.
connectivity in a large-scale small molecule sensitivity 22. Fuse E, Tanii H, Kurata N, Kobayashi H, Shimada Y,
dataset. Cancer Discovery. 2015; 5:1210-1223. Tamura T, Sasaki Y, Tanigawara Y, Lush RD, Headlee D,
10. Komlodi-Pasztor E, Sackett DL, Fojo AT. Inhibitors Figg WD, Arbuck SG, Senderowicz AM, et al. Unpredicted
targeting mitosis: tales of how great drugs against a clinical pharmacology of UCN-01 caused by specific
promising target were brought down by a flawed rationale. binding to human α1-acid glycoprotein. Cancer Res. 1998;
Clin Cancer Res. 2012; 18:51-63. 58:3248-3253.
11. Bocci G, Kerbel RS. Pharmacokinetics of metronomic 23. Perez RP, Lewis LD, Beelen AP, Olszanski AJ, Johnston
chemotherapy: a neglected but crucial aspect. Nature Rev N, Rhodes CH, Beaulieu B, Ernstoff MS, Eastman A.
Clin Oncol. 2016; 13: in press. Modulation of cell cycle progression in human tumors: a
12. Francia G, Shaked Y, Hashimoto K, Sun J, Yin M, Cesta pharmacokinetic and tumor molecular pharmacodynamic
C, Xu P, Man S, Hacki C, Stewart J, Uhlik M, Dantzig AH, study of cisplatin plus the Chk1 inhibitor UCN-01 (NSC
Foster FS, Kerbel RS. Low-dose metronomic oral dosing of 638850). Clin Cancer Res. 2006; 12:7079-7085.
a pro-drug of gemcitabine (LY2334737) causes antitumor 24. Eastman A, Kohn EA, Brown MK, Rathman J, Livingstone
effects in the absence of systemic vasculogenesis. Mol M, Blank DH, Gribble GW. A novel indolocarbazole,
Cancer Therap. 2012; 11:680-689. ICP-1, abrogates DNA damage-induced cell cycle arrest
13. Lancet JE, Cortes JE, Hogge DE, Tallman MS, Kovacsovics and enhances cytotoxicity. Similarities and differences to
TJ, Damon LE, Komrokji R, Solomon SR, Kolitz JE, the cell cycle checkpoint abrogator UCN-01. Mol Cancer
Cooper M, Yeager AM, Louie AC, Feldman EJ. Phase Therap. 2002; 1:1067-1078.
2 trial of CPX-351, a fixed 5:1 molar ratio of cytarabine/ 25. Sorenson CM, Eastman A. Mechanism of cis-
daunorubicin, vs cytarabine/daunorubicin in older adults diamminedichloroplatinum(II)-induced cytotoxicity: role
with untreated AML. Blood. 2014; 123:3239-3246. of G2 arrest and DNA double-strand breaks. Cancer Res.
14. Sakurikar N, Thompson R, Montano R, Eastman A. A 1988; 48:4484-4488.
subset of cancer cell lines is acutely sensitive to the Chk1 26. Barry MA, Behnke CA, Eastman A. Activation of
inhibitor MK8776 as monotherapy due to CDK2 activation programmed cell death (apoptosis) by cisplatin, other
in S phase. Oncotarget. 2016; 7:1380-1394. doi:10.18632/ anticancer drugs, toxins and hyperthermia. Biochem
oncotarget.6364. Pharmacol. 1990; 40:2353-2362.
15. Shabaneh TB, Downey SL, Goddard AL, Screen M, 27. Demarcq C, Bunch RT, Creswell D, Eastman A. The role

www.impactjournals.com/oncotarget 8864 Oncotarget

 of cell cycle progression in cisplatin-induced apoptosis in Biol. 2008; 4:119-125.
Chinese hamster ovary cells. Cell Growth Different. 1994; 40. Garner KM, Pletnev AA, Eastman A. Corrected structure of
5:983-993. mirin, a small-molecule inhibitor of the Mre11-Rad50-Nbs1
28. Schlegel R, Pardee AB. Caffeine-induced uncoupling complex. Nature Chem Biol. 2009; 5:129-130.
of mitosis from the completion of DNA replication in 41. Tsang E, Miyabe I, Iraqui I, Zheng J, Lambert SAE,
mammalian cells. Science. 1986; 232:1264-1266. Carr AM. The extent of error-prone replication restart
29. Aarts M, Sharpe R, Garcia-Murillas I, Gevensleben H, by homologous recombination is controlled by Exo1 and
Hurd MS, Shumway SD, Toniatti C, Ashworth A, Turner checkpoint proteins. J Cell Science. 2014; 127:2983-2994.
NC. Forced mitotic entry of S phase cells as a therapeutic 42. Choi S, Gamper AM, White JS, Bakkenist CJ. Inhibition
strategy induced by inhibition of WEE1. Cancer Discovery. of ATM kinase activity does not phenocopy ATM protein
2012; 2:524-539. disruption. Implications for the clinical utility of ATM
30. Escobar D, Hepp MI, Farkas C, Campos T, Sodir NM, kinase inhibitors. Cell Cycle. 2010; 9:4052-4057.
Morales M, Alvarez CI, Swigart L, Evan GI, Gutierrez JL, 43. Utermark T, Rao T, Cheng H, Wang Q, Lee SH, Wang
Nishinakamura R, Castro AF, Pincheira R. Sall2 is required ZC, Iglehart JD, Roberts TM, Muller WJ, Zhao JJ. The
for proapoptotic Noxa expression and genotoxic stress- p110α and p110β isoforms of PI3K play divergent roles in
induced apoptosis by doxorubicin. Cell Death Disease. mammary gland development and tumorigenesis. Genes
2015; 6:e1816. Dev. 2012; 26:1573-1586.
31. Sinha BK, Kumar A, Bhattacharjee S, Espey MG, Mason 44. Soderquist RS, Pletnev AA, Danilov AV, Eastman A. The
RP. Effect of nitric oxide on the anticancer activity of the putative BH3 mimetic S1 sensitizes leukemia to ABT-
topoisomerase-active drugs etoposide and adriamycin 737 by increasing reactive oxygen species, inducing the
in human melanoma cells. J Pharmacol Exp Ther. 2013; endoplasmic reticulum stress response, and up-regulating
347:607-614. the BH3-only protein NOXA. Apoptosis. 2014; 19:201-209.
32. Ham J, Costa C, Sano R, Lochmann TL, Sennott EM, Patel 45. Vogler M, Weber K, Dinsdale D, Schmitz I, Schulze-
NU, (and 26 other authors). Exploitation of the apoptosis- Osthoff K, Dyer MJS, Cohen GM. Different forms of cell
primed state of MYCN-amplified neuroblastoma to develop death induced by putative BCL2 inhibitors. Cell Death
a potent and specific targeted therapy combination. Cancer Differ. 2009; 16:1030-1039.
Cell. 2016; 29:159-172. 46. Billard C. BH3 mimetics: status of the field and new
33. Lee HH, Ye S, Li XJ, Lee KB, Park MH, Kim SM. developments. Mol Cancer Therap. 2013; 12:1691-1700.
Combination treatment with paclitaxel and doxorubicin 47. Soderquist RS, Danilov AV, Eastman A. Gossypol
inhibits growth of human esophageal squamous cancer cells increases expression of the pro-apoptotic protein through a
by inactivation of Akt. Oncol Rep. 2014; 31:183-188. novel mechanism involving phospholipase A2, cytoplasmic
34. Wolf CM, Reynolds JE, Morana SJ, Eastman A. The calcium and endoplasmic reticulum stress. J Biol Chem.
temporal relationship between protein phosphatase, ICE/ 2014; 289:16190-16199.
CED-3 proteases, intracellular acidification, and DNA 48. Soderquist RS, Eastman A. BCL2 inhibitors as anticancer
fragmentation in apoptosis. Exp Cell Res. 1997; 230:22-27. drugs: a plethora of misleading BH3 mimetics. Mol Cancer
35. Bates DJP, Danilov AV, Lowrey CH, Eastman A. Therap. 2016; 15:2011-2017.
Vinblastine rapidly induces NOXA and acutely sensitizes 49. Berenbaum MC. What is synergy? Pharm Rev. 1989; 41:93-
primary chronic lymphocytic leukemia cells to ABT-737. 141.
Mol Cancer Therap. 2013; 12:1504-1514.
50. Chou T-C, Talalay P. Quantitative analysis of dose-effect
36. Albershardt TC, Salerni BL, Soderquist RS, Bates DJP, relationships: the combined effects of multiple drugs or
Pletnev AA, Kisselev AF, Eastman A. Multiple BH3 enzyme inhibitors. Adv Enzyme Regul. 1986; 22:27-55.
mimetics antagonize anti-apoptotic MCL1 by inducing the
51. Koh SB, Courtin A, Boyce RJ, Boyle RG, Richards FM,
endoplasmic reticulum stress response and up-regulating
Jodrell DI. CHK1 inhibition synergizes with gemcitabine
BH3-only protein NOXA. J Biol Chem. 2011; 286:24882-
initially by destablizing the DNA replication apparatus.
24895.
Cancer Res. 2016; 755:3583-3595.
37. Hilt W, Wolf DH. Proteasomes of the yeast S. cerevisiae:
52. Goel S, Wang Q, Watt AC, Tolaney SM, Dillon DA, Li
genes structure and functions. Mol Biol Rep. 1995; 21:3-10.
W, Ramm S, Palmer AC, Yuzugullu H, Varadan V, Tuck
38. Garner KM, Eastman A. Variations in Mre11/Rad50/Nbs1 D, Harris LN, Wong KK, et al. Overcoming therapeutic
status and DNA damage-induced S-phase arrest in cell lines resistance in HER2-positive breast cancers using CDK4/6
of the NCI60 panel. BMC Cancer. 2011; 11:206. inhibitors. Cancer Cell. 2016; 29:255-269.
39. Dupre A, Boyer-Chatnet L, Sattler RM, Modi AP, Lee JH, 53. Chou T-C. Drug combination studies and their synergy
Nicolette ML, Kopelovich L, Jasin M, Baer R, Paull TT, quantification using the Chou-Talalay method. Cancer Res.
Gautier J. A forward chemical genetic screen reveals an 2010; 70:440-446.
inhibitor of the Mre11-Rad50-Nbs1 complex. Nature Chem
54. Goldin A, Vadlamudi S. Influence of mitotic cycle inhibitors

www.impactjournals.com/oncotarget 8865 Oncotarget

 on the antileukemic activity of cytosone arabinoside (NSC- 63. Al-Ahmadie H, Iyer G, Hohl M, Asthana S, Inagaki A,
63878) in mice bearing leukemia L1210. Cancer Chemother Schultz N, (and 18 other authors). Synthetic lethality in
Rep. 1971; 55:547-555. ATM-deficient RAD50-mutant tumors underlies outlier
55. Rowinsky EK, Citardi MJ, Noe DA, Donehower RC. response to cancer therapy. Cancer Discovery. 2014;
Sequence-dependent cytotoxic effects due to combinations 4:1014-1021.
of cisplatin and the antimicrotubule agents taxol and 64. Hakansson L, Trope C. On the presence within tumors of
vincristine. J Cancer Res Clin Oncol. 1993; 119:727-733. clones that differ in sensitivity to cytostatic drugs. Acta
56. Rowinsky EK, Gilbert MR, McGuire WP, Noe DA, Pathol Microbiol Scand A. 1974; 82:32-40.
Grochow LB, Forastiere AA, Ettinge DS, Lubejko BG, 65. Heppner GH. Tumor heterogeneity. Cancer Res. 1984;
Clark B, Sartorius SE, Cornblath DR, Hendricks CB, 44:2259-22265.
Donehower RC. Sequences of taxol and cisplatin: a phase I 66. Pratt, W.B., Ruddon, R.W., Ensminger, W.D., and
and pharmacologic study. J Clin Oncol. 1991; 9:1692-1703. Maybaum, J. The Anticancer Drugs. New York: Oxford
57. Tardi P, Johnstone S, Harasym N, Xie S, Harasym T, University Press, 1994.
Zismann N, Harvie P, Bernudes D, Mayer L. In vivo 67. Pui C-H, Boyett JM, Rivera GK, Hancock ML, Sandlund
maintenance of synergistic cytarabine:daunorubucin ratios JT, Ribeiro RC, Rubnitz JE, Behm FG, Raimondi SC,
greatly enhances therapeutic efficacy. Leukemia Res. 2009; Gajjar A, Razzouk B, Campana D, Kun LE, Relling MV,
33:129-139. Evans WE. Long-term results of Total Therapy studies 11,
58. Toledo LI, Altmeyer M, Rask M-B, Lukas C, Larsen DH, 12 and 13A for childhood acute lymphoblastic leukemia at
Povlsen LK, Bekker-Jensen S, Mailand N, Bartek J, Lukas St. Jude Children’s Research Hospital. Leukemia. 2000;
J. ATR prohibits replication catastrophe by preventing 14:2286-2294.
global exhaustion of RPA. Cell. 2013; 155:1088-1103. 68. Fojo T, Mailankody S, Lo A. Unintended consequences
59. Buisson R, Boisvert JL, Benes CH, Zou L. Distinct but of expensive cancer therapeutics - the pursuit of marginal
concerted roles of ATR, DNA-PK and Chk1 in countering indications and a me-too mentality that stifles innovation
replication stress during S phase. Mol Cell. 2015; 59:1011- and creativity. JAMA Otolaryngol Head Neck Surg. 2014;
1024. 140:1225-1236.
60. Paez JG, Janne PA, Lee JC, Tracy S, Greulich H, Gabriel 69. Roche-Lestienne C, Soenen-Cornu V, Gardel-Duflos N, Lai
S, Herman P, Kaye FJ, Lindeman N, Boggon TJ, Naoki JL, Philippe N, Facon T, Fenaux P, Preudhomme C. Several
K, Sasaki H, Fujii Y, Eck MJ, Sellers WR, Johnson BE, types of mutations of the Abl gene can be found in chronic
Meyerson M. EGFR mutations in lung cancer: correlation myeloid patients resistant to STI571, and they can pre-exist
with clinical response to gefitinib therapy. Science. 2004; to the onset of treatment . Blood. 2002; 100:1014-1018.
304:1497-1500. 70. Inukai M, Toyooka S, Ito S, Asano H, Ichihara S, Soh J,
61. Iyer G, Hanrahan AJ, Milowsky MI, Al-Ahmadie H, Suehisa H, Ouchida M, Aoe K, Aoe M, Kiura K, Shimizu
Scott SN, Janakiraman M, Pirun M, Sander C, Socci ND, N, Date H. Presence of epidermal growth factor gene
Ostrovnaya I, Viale A, Heguy A, Peng L, et al. Genome T790M mutation as a minor clone in non-small cell lung
sequencing identifies a basis for everolimus sensitivity. cancer. Cancer Res. 2006; 66:7854-7858.
Science. 2012; 338:221.
62. Curran MP. Everolimus: in patients with subependymal
giant cell astrocytoma associated with tuberous sclerosis
complex. Pediatr Drugs. 2012; 14:51-60.

www.impactjournals.com/oncotarget 8866 Oncotarget