R

U

Ja

b

a

ARRA

KCDCT

C

1h

Drug Resistance Updates 15 (2012) 249– 257

Contents lists available at SciVerse ScienceDirect

Drug Resistance Updates

journa l h omepage: www.elsev ier .com/ locate /drup

eview

nderstanding resistance to combination chemotherapy

ustin R. Pritcharda, Douglas A. Lauffenburgera,b, Michael T. Hemanna,∗

Koch Institute for Integrative Cancer Research, MIT, Cambridge, MA 02139, United StatesMIT Department of Biological Engineering, Cambridge, MA 02139, United States

r t i c l e i n f o

rticle history:eceived 10 September 2012eceived in revised form 1 October 2012ccepted 16 October 2012

eywords:ombination therapyrug resistanceytotoxic drugsargeted drugs

a b s t r a c t

The current clinical application of combination chemotherapy is guided by a historically successful setof practices that were developed by basic and clinical researchers 50–60 years ago. Thus, in order tounderstand how emerging approaches to drug development might aid the creation of new therapeuticcombinations, it is critical to understand the defining principles underlying classic combination therapyand the original experimental rationales behind them. One such principle is that the use of combinationtherapies with independent mechanisms of action can minimize the evolution of drug resistance. Anotheris that in order to kill sufficient cancer cells to cure a patient, multiple drugs must be delivered at theirmaximum tolerated dose – a condition that allows for enhanced cancer cell killing with manageabletoxicity. In light of these models, we aim to explore recent genomic evidence underlying the mecha-nisms of resistance to the combination regimens constructed on these principles. Interestingly, we findthat emerging genomic evidence contradicts some of the rationales of early practitioners in developingcommonly used drug regimens. However, we also find that the addition of recent targeted therapies has

yet to change the current principles underlying the construction of anti-cancer combinatorial regimens,nor have they made substantial inroads into the treatment of most cancers. We suggest that emergingsystems/network biology approaches have an immense opportunity to impact the rational developmentof successful drug regimens. Specifically, by examining drug combinations in multivariate ways, nextgeneration combination therapies can be constructed with a clear understanding of how mechanisms ofresistance to multi-drug regimens differ from single agent resistance.

© 2012 Elsevier Ltd. All rights reserved.

ontents

1. The origins and continued use of cancer combination chemotherapy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2502. Combination genotoxic chemotherapy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2503. The clinical efficacy of combination therapy and its acquired resistance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2504. The mutational analysis of leukemia suggests multidrug resistant cell states . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2505. Resistance to targeted therapy is pathway specific . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2516. Clinical drug resistance to conventional chemotherapies can also be pathway specific . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2527. Functional mechanisms of clinical efficacy in combination therapy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2528. Predicting phenotypic response to drug action in mammalian cells . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2529. Multivariate signatures characterizing drug action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25310. Biochemical signatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25311. Genetic signatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25312. In silico approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25413. Systems perspectives in combination therapy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25414. Toward combination drug signatures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 254

15. Final thoughts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

∗ Corresponding author.E-mail address: hemann@mit.edu (M.T. Hemann).

368-7646/$ – see front matter © 2012 Elsevier Ltd. All rights reserved.ttp://dx.doi.org/10.1016/j.drup.2012.10.003

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255

2 stance

1c

acuoaapaeicsa

istrotdhcd1ttbera

2

Dmvb11mimSapc

ofi(eScfkbttp

50 J.R. Pritchard et al. / Drug Resi

. The origins and continued use of cancer combinationhemotherapy

The overwhelming majority of cures in cancer chemother-py have come from the application of conventional cytotoxichemotherapies. While some cytotoxic agents have been the prod-ct of serendipity, and some the product of large scale screening,thers were part of the first wave of “rational” targeted ther-pies that were developed in the 1940s and were specificallyimed at targeting cancer cells on the basis of the nutritionalroperties that made them distinct from normal cells (Chabnernd Roberts, 2005; Wall and Wani, 1995). The history of thesearly successes illustrates a unique combination of serendip-ty and insight that has led to the development of potentiallyurative regimens for numerous forms of cancer. It is this earlyuccess that still guides current clinical practice and clinical tri-ls.

In 1946 Goodman and Gilman published a landmark studyn cancer chemotherapy. Nitrogen mustard agents that wereerendipitously discovered to induce drastic lymphoid cell deple-ion upon accidental or wartime exposure were shown to produceemarkable responses in human tumors from a variety of tissuerigins (Goodman et al., 1984). Two years later Sidney Farberook a more rational target based strategy for anti-cancer drugiscovery. He reasoned that if folate deficiency inhibited normalematopoiesis and the addition of folates accelerated leukemia inhildren, then anti-folates would make a promising anti-leukemiarug. In his landmark 1948 paper, Farber (Farber and Diamond,948) observed the first true remissions in a disease where theime from diagnosis to death was often measured in days. Forhe purpose of context, it is interesting to note the similaritiesetween Farber’s anti-folates and current targeted therapies. Likearly clinical trials with EGFR and BRAF inhibitors, these earlyeports documented some striking but short lasting remissions in

subset of patients.

. Combination genotoxic chemotherapy

In 1943, Luria and Delbruck’s fluctuation analysis (Luria andelbruck, 1943), which combined an experimental and mathe-atical modeling framework, showed that heritable resistance to

iruses was derived from pre-viral-exposure genetic variation in aacterial population. Later, Newcombe (Newcombe and Hawirko,949) extended this finding to chemotherapy in bacteria, and in952, Law extended it to the resistance to anti-folates in in vivoouse models of cancer (Law, 1952b). Taken together these exper-

ments suggested to early chemotherapy researchers that thereight be a benefit to giving drugs in combination (Law, 1952a;

kipper et al., 1954). If a drug provided a resistance rate of 1/mnd a second statistically independent (non-cross resistant) drugrovided resistance of 1/n then co-resistance would occur in 1/m*nells.

In 1958, citing the above rationales, and the successful creationf combination therapies for tuberculosis, Emil Frei III published therst randomized control trial of a combination therapy in cancerFrei et al., 1958). It established clear combination efficacy over thefficacy of the single substituent agents. In the mid 1960s, Howardkipper showed that in experimental mouse models, as few as oneancer cell could give rise to lethal disease, and that chemotherapyollowed a logarithmic killing model. Specifically, the same doseilled the same proportion of cells, regardless of the total disease

urden (Skipper et al., 1964). This suggested to Skipper and othershat to have any chance of curing a cancer, a physician would haveo administer as large of a tolerable drug dose as is possible to theatient.

Updates 15 (2012) 249– 257

The rapid success of Emil Frei’s 1958 trial, advances in support-ive care, and Howard Skipper’s principals for curing experimentalmouse models all led to the highly successful but not experimen-tally controlled adoption of the 4-drug VAMP regimen (Freireichand Frei, 1964). Coupled with care that was able to ameliorate theside effects of therapy and effectively dose even higher cumulativedrug doses into leukemia patients, the VAMP regimen was the firstbig step toward the large and potentially curative regimens that wehave today. It is important to note that this type of study (in whichcombinations of drugs are combined at maximally tolerated doses)was not able to specify the mechanism of increased efficacy follow-ing increased drug dose, nor the minimization of the outgrowth ofresistance, but it can and did demonstrate improved efficacy. It wasthis bold, but less carefully controlled strategy that led to the rapidand successful adoption of combination chemotherapy for manycancers, and by 1973 (DeVita and Schein, 1973) had revolutionizedthe treatment of cancer by finding cures for previously untreatablediseases. Thus, it is interesting to note that the biggest early suc-cesses of cancer therapy owe less to systematic controlled trials andmore to bold and decisive attempts to cure very sick patients.

These early and impressive successes fostered a manner ofthinking about care that still largely dictates current clinicalpractice in combination therapy, namely that randomized com-binations of chemotherapeutics can identify select combinationsthat improve the number of patients that show a durable response.Many clinical trials using a similar process are still constructedin attempts to improve the current standard of care (Conroyet al., 2011; Pirker et al., 2009). Next, we will examine currentgenomic evidence for what makes cells resistant to combinationtherapy in light of the early rationales underlying combinationregimens.

3. The clinical efficacy of combination therapy and itsacquired resistance

The initial hypotheses behind the first combination therapyregimens suggested that: (1) the maximum possible cumulativedose of drug should be given, and (2) independent drugs with non-overlapping mechanisms of action are important for minimizingthe probability of therapeutic resistance. The ideas about resistancehave only been tested in the context of cross resistance followingthe selection of single drug resistant cell lines in Luria–Delbruckfluctuation tests (Luria and Delbruck, 1943). However, the genomicevidence in relapsed cancers following clinical combination treat-ment in multiple systems suggests that this microbiology inspiredrationale, while true in the context of infectious diseases like tuber-culosis and HIV (Almeida Da Silva and Palomino, 2011; Durant et al.,1999), is very different in the context of relapsed human cancersthat are initially sensitive to combination chemotherapy. In order tothink about how systems biology might inform drug selection andthe clinical trials process, we must consider how current clinicalcombinations fail in patients (Fig. 1).

4. The mutational analysis of leukemia suggests multidrugresistant cell states

Examining pre and post treatment matched patient samples iscritical to identifying the nature of acquired drug resistance in can-cer. One of the first studies to analyze recurrent genomic alterationsin cohorts of matched pre-treatment and post relapse patients per-formed copy number analysis on acute lymphoblastic leukemia

samples (Mullighan et al., 2008). Mullighan and colleagues showedtwo striking results that speak to both the historic rationales forcombination therapy and current studies on therapeutic resis-tance. First, the majority of relapse clones following conventional

J.R. Pritchard et al. / Drug Resistance Updates 15 (2012) 249– 257 251

Antibiotic

A

Genotoxic

Drug A

Targeted

Drug A Antibiotics

A + B + C

Genotoxic

Drugs

A + B + C

Targeted

Drug A

+

Genotoxic

Drugs

B + C

a a a a, b, c !" a,!"

Single Drug Combinatorial Therapy

Tuberculosis/HIV Tuberculosi s/HIVcancer cell cancer cell cancer cell cancer cell

(efflux) (e fflux) (e fflux)

(bypass)

Fig. 1. A diagram showing the relationship between single and combinatorial therapy and the development of drug resistance. In response to single agent treatment ofTuberculosis/HIV or tumor cells, whether targeted or “cytotoxic”, drug target alterations or, perhaps, drug efflux, can mediate therapeutic resistance. Here lower case lettersi atoriw tions tr

ctdnblttosmrCwmiciclpsaTvn

5

gicfamithtc

ndicate specific mutations in drug target genes. Conversely, in response to combinhile mammalian cells evolve resistance to targeted therapeutics, reinforcing altera

esistance.

ombinatorial chemotherapy represent low frequency variants inhe pre-treatment tumor that were progenitors of the dominantiagnosis clone. Second, the majority of alterations that domi-ated at relapse did not include direct alterations in the knowniochemical targets of common therapeutics that were used for

eukemia treatment. Rather, the most common alterations tendedo affect B-cell development, changes that might be hypothesizedo promote generally drug resistant cell states or homing to devel-pmental niches which promote therapeutic resistance. In a latertudy by the same group, the sequencing of selected exons inatched pre- and post treatment samples also identified recur-

ent mutations in B-cell developmental pathways, as well as inREBBP and genes that induce transcriptional states that correlateith drug resistance (Mullighan et al., 2011). Finally, most recently,atched pre and post-treatment AML samples revealed mutations

n genes that were not related to the direct drug targets of frontlinehemotherapeutic action (Ding et al., 2012). Though these stud-es employed single measurement methodologies (sequencing oropy number analysis), taken together they suggest that relapsedeukemia treated with multi-agent regimens develop resistancerofiles that favor the development of a multi-drug resistant celltate. This state appears to be broad in its definition, but includes thelteration of apoptotic, epigenetic, and developmental cell states.hus, resistance to multi-drug regimens does not appear to occuria the combination of mutations conferring resistance to compo-ent parts of a regimen.

. Resistance to targeted therapy is pathway specific

Perhaps one of the most exciting aspects of emerging tar-eted therapies is that resistance to these agents is both easilydentifiable and qualitatively distinct from drug resistance toombination regimens. Specifically, resistance to these agentsrequently arises as a direct drug–target alteration affecting ther-peutic response. For example, following the treatment of chronicyelogenous leukemia (CML) with Abl kinase inhibitor Imatinib,

t was noted that resistant cells harbored a spectrum of resis-

ance mutations in the BCR-Abl kinase domain. Importantly, cellsarboring some of these resistance mutations remained sensi-ive to a distinct multi-kinase inhibitor, dasatinib. While stillompetitively inhibiting BCR-Abl, dasatinib was found to have

al therapy, Tuberculosis/HIV evolve specific resistance mechanisms to each agent,hat restore target activity as well as mechanisms of multi-drug (target non-specific)

improved efficacy over imatinib in treating leukemia with mostBCR-Abl kinase domain mutations (Shah et al., 2004; Talpaz et al.,2006).

But how might this discovery be used improve the develop-ment of effective drug combinations? Surprisingly, when used ina temporally distinct manner, these two kinase inhibitors, whosespectrum of resistance mutations are somewhat distinct can bealternatingly dosed over the course of the disease to stave off resis-tance, as long as compound mutations or single mutations causingresistance to both drugs are not present. Furthermore, severalrecent studies suggest that cells bearing the T315I resistance muta-tion (a mutation that confers resistance to both kinase inhibitors)can be treated with enhanced efficacy by using combination ther-apy. For example, cells expressing BCR-Abl T315I are sensitive tothe combination of dasatinib with an aurora kinase inhibitor (Shahet al., 2007). Additionally, allosteric inhibitors that bind to themyristate-binding groove of the Abl protein were recently devel-oped, that in combination with kinase domain targeted therapies,are capable of reducing T315I positive disease in tumor bearingmice (Zhang et al., 2010). Finally, combinations of conventionalcytotoxic drugs (l-asparaginase and dexamethasone) with BCR-Abl inhibitors can overcome T315I-mediated resistance in a mousemodel of BCR-Abl positive acute lymphoblastic leukemia (Bouloset al., 2011). All of these strategies illustrate attempts to create nextgeneration combination regimens that incorporate agents to resen-sitize tumors to targeted therapeutics. Thus, unlike conventionalchemotherapeutic regimens, some regimens involving targetedagents seek to restore individual component-specific activity – asituation that would presumably select again for single agent resis-tance.

While these BCR-Abl-related studies are the most advancedefforts to rationally build clinical regimens to circumvent targetedresistance, similar investigations into the modes of targeted resis-tance in the hedgehog pathway in medulloblastoma confirm theprevalence of on- target pathway mutations (Yauch et al., 2009)in response to targeted therapy. Studies in EGFR inhibitor-treatedlung cancers have identified amplifications in the Met receptor

tyrosine kinase, an example of a parallel pathway activation mech-anism inducing resistance to targeted therapy (Turke et al., 2010).Furthermore, studies in BRAFV600E treated melanoma have iden-tified downstream mutations in Mek (Wagle et al., 2011). This data

2 stance

stsgiimtotmtira

6c

atsvTncgit

ttcia22gttrs(cacmecaddtst

7c

sir(

52 J.R. Pritchard et al. / Drug Resi

uggests that these resistance mechanisms (mutations directly inhe pathways of drug action or parallel to drug action in drugensitive cancers) can be employed in response to numerous tar-eted therapeutics. With this in mind, there is increasing interestn the use of these agents in combination regimens. This rationales highly similar to the original rationale for the first combo regi-

ens, i.e. that the addition of independent drugs blocks targetedherapeutic resistance. Just like the initial regimens, in the contextf combination therapy, it will be important to consider whetherhe therapeutic combinations adopted are actually able to shift

echanisms of therapeutic resistance, or whether they increasehe effective amount of killing, or both. These data will be criticaln understanding and preventing resistance that might develop inesponse to combinatorial therapy involving new “targeted” ther-peutics.

. Clinical drug resistance to conventional chemotherapiesan also be pathway specific

The comparison between conventional regimen resistancend targeted therapeutic resistance is obscured by the facthat currently regimens contain multiple drugs that are notpecifically targeted at oncogenic pathways, and because con-entional therapeutics are almost always dosed in combination.his makes it difficult to directly compare the clinical mecha-isms of drug resistance following single cytotoxic therapy inlinical cohorts. In spite of this, some pre-clinical evidence sug-ests that many classic chemotherapeutic agents, when dosed insolation, can promote very similar modes of resistance as targetedherapies.

Traditional chemotherapeutic agents are often generallyhought of as pleiotropic; however, this is not an accurate descrip-ion of all cytotoxic chemotherapies. While nitrogen mustards andisplatin are highly chemically reactive, and nucleoside analogs canncorporate into DNA and RNA, other drugs such as methotrex-te, camptothecin, doxorubicin, and dexamethasone (Bledsoe et al.,002; Matthews et al., 1977; Pommier et al., 2010; Staker et al.,002) make direct and specific contacts with their enzymatic tar-ets. An examination of the preclinical literature suggests that inhe case of camptothecin, doxorubicin and methotrexate, resis-ance to these agents can result in direct modifications to theirespective drug targets (Hashimoto et al., 1996). Though manyelection experiments have also revealed the role of P-glycoproteinP-gp) and drug efflux in multi-drug cross resistance (a topic dis-ussed at greater length in the next section), if P-gp is inhibitednd anthracycline resistance is selected for, the resistance profilean revert back to a target dependent (topoisomerase I or topoiso-erase II alpha) mechanism of drug resistance (Beketic-Oreskovic

t al., 1995; Zander et al., 2010). This is in contrast to the clini-al picture of multidrug resistance portrayed above. Thus, in thebsence of direct evidence, it is interesting to speculate that theirect target mediated resistance to classic chemotherapeutics isifficult to select for in the face of current clinical combinationherapy regimens, and that this is due to the combination of clas-ic chemotherapeutic agents, and not the nature of the agentshemselves.

. Functional mechanisms of clinical efficacy inombination therapy

If a property of combination therapy in human cancers is that it

elects for multi-drug resistant cell states, the distinctions betweennfectious disease and cancer may be indicative of the complexegulation of drug metabolism and cell death in mammalian cellsLetai, 2008). A well-described mechanism for the development

Updates 15 (2012) 249– 257

of such a multi-drug resistant state is the overexpression of P-gp(Borst, 2012). This pump can mediate the efflux of numerous drugsused in combinatorial regimens, including anthracyclines, taxanes,and Vinca alkaloids (Szakacs et al., 2006). Surprisingly, however,while there is substantial evidence that elevated drug transport canmediate drug resistance in model systems, there is only limited evi-dence for a role of these transporters in chemoresistance in humantumors.

More recently it has been shown that in patient clinical sam-ples and primary tissues, the “proximity” to the apoptotic thresholdin cancer cells, as measured by the sensitivity of the mitochon-drial cytochrome c release to pro-apoptotic peptides, is correlatedwith therapeutic response and the size of the therapeutic windowthat a drug can achieve. Thus, the proximity of cancer cells to theapoptotic threshold may be a key determinant of the therapeuticresponsiveness of mammalian cancers (Ni Chonghaile et al., 2011).If maximizing the therapeutic dosing across the apoptotic spec-trum increases the effective dose of drug and maximally activatesapoptosis as a therapeutic response, then it suggests that intrin-sic resistance to targeted therapies may be due to the fact thatthey have not been formulated into regimens sufficiently potentto bypass the apoptotic threshold. Curiously, however, while celldeath is the ultimate objective of cancer therapies, apoptotic path-way alterations are not a common mechanism underlying theevolution of drug resistance (Borst et al., 2007). Thus, commonlypredicted mechanisms of multi-drug resistance (drug efflux, apo-ptosis, etc. . .) are rarely identified in multi-drug resistant tumors.

These data highlight a conundrum in modern therapeutics.Commonly used drug combinations lack clearly defined mecha-nisms of resistance, thus it is difficult to stratify cohorts of patientsbased on the status of individual genes (Abramson and Shipp, 2005).Thus, for many malignancies, we are committed to the use of theseregimens as first line therapies, even if they have a substantial fail-ure rate – a fact that explains the continued use of these regimens60 years after their inception. Conversely, targeted agents havedefined mechanisms of resistance, but have limited durable effi-cacy as single agents or when combined with agents that reinforcesingle drug action. Thus, given that cancer evolves resistance tomulti-drug regimens in a manner that is distinct from antibiotic orantiviral resistance, it is critical to develop strategies to understandthe genetic determinants of the response of cancer cells to combi-natorial chemotherapeutics. In other words, we need to understandmechanisms of multi-drug resistance in order to begin to deviatefrom decades-old drug regimens. Here, we will describe emergingapproaches to model multi-drug action.

8. Predicting phenotypic response to drug action inmammalian cells

Cancer cells exist in a multivariate landscape (Kreegerand Lauffenburger, 2010), involving both oncogene and tumorsuppressor signaling as well as signaling from the tumor microen-vironment. Similarly, drugs activate and inhibit cellular pathways,and there is significant complexity to drug response. A variety ofmodeling approaches, each requiring different levels of molecu-lar detail, and each uniquely suited to different types of data havebegun to make progress in predicting drug effects across differentcellular contexts. Predictive in vitro models of cellular systems maybe effective ways to predict the cellular context in which a smallmolecule or a combination of small molecules might be active.Though most often employed in the context of targeted therapeutic

inhibition, the lessons and methods of these studies can be used toexamine diverse cytotoxic combinations.

In well-studied systems with well-characterized pathways, itis possible to use differential equations of biochemical reactions

stance

tasftcasdmtiTls2tvo(t

mcamTkattdiidpteoiemit

obtama

9

wmsapsoWm

J.R. Pritchard et al. / Drug Resi

o describe the mechanisms by which pro/anti-apoptotic signalsre conveyed in biochemical networks. This approach has demon-trated that even non-oncogenic signaling proteins can be targetsor drug intervention. Schoerberl et al. showed that Erb3 inhibi-ion decreases AKT phosphorylation across a broad range of initialonditions (Schoeberl et al., 2009). The utility of Erb3 inhibitioncross a range of initial conditions demonstrates the power of inilico modeling to identify drug targets that have limited context-ependence. Importantly, from a methodological perspective, toake a useful model, all biochemical parameters do not have

o be known a priori. They can be estimated by first discover-ng highly sensitive species in the set of biochemical reactions.hen, using simulated annealing, many parameters can be fit to celline data. With these fit parameters incorporated, a novel under-tanding of signaling network function can emerge (Chen et al.,009). While these approaches have proven their effectiveness forargeted therapeutic discovery, they remain underutilized for con-entional cytotoxic therapeutics. However, similar types of modelsf DNA damage have been built to describe other phenomenaToettcher et al., 2009), and may be used in the future to informhe therapeutic efficacy for conventional therapies.

When biochemical reactions are less described, higher levels ofodel abstraction can be used to understand the effect that bio-

hemical reactions have upon cell outcome, even if the reactionsre not modeled explicitly. Using partial least squares regressionodeling in mammalian cells, Janes et al. suggested that, given

NF-alpha induced cell death and opposing growth factor stimuli,inase pathway interventions could be accurately predicted in

colon cancer cell line (Janes et al., 2005, 2004). Impressively,his method incorporated enough of the signaling network con-ext that it allowed for the prediction of apoptotic responses iniverse cell lines of epithelial origin (Miller-Jensen et al., 2007). In

sogenic models of RAS signaling, Kreeger et al. showed that a sim-lar multi-pathway model could incorporate diverse effects on celleath that were mediated by different RAS proteins to accuratelyredict apoptotic response (Kreeger et al., 2009). To examine ques-ions of kinase inhibitor specificity and off target effects, Kumart al. showed that models incorporating multivariate descriptionsf signaling responses could accurately predict cellular outcomesn the presence of promiscuous kinase inhibitor activity (Kumart al., 2008). Together these studies show that regression basedodels can incorporate cell type and oncogene specific network

nfluences to estimate drug on and off target effects that contributeo therapeutic action.

While these efforts focused on predicting specific drug effectsn particular species in multivariate models, drugs often have aroader spectrum of biochemical and genetic effects. This spec-rum requires a drug-centric approach, utilizing profiling methodsnd high level statistical modeling that can simultaneously assessultiple relevant cellular effects, and predict the most relevant

lterations for cellular phenotypes.

. Multivariate signatures characterizing drug action

In order to understand the effects of combinations of drugs,e first have to understand single drug effects in a multivariateanner. Beyond simple drug–target biochemical interactions, most

mall molecules have many biochemical effects and genetic inter-ctions; they act promiscuously on a variety of cellular enzymes androcesses (Xie et al., 2012). When considered in the context of large-

cale genetic screens, these same molecules harbor a diverse arrayf genetic interactions (Bartz et al., 2006; Doles and Hemann, 2010;hitehurst et al., 2007). Comprehensive characterization of smallolecules should help provide mechanistic information on how

Updates 15 (2012) 249– 257 253

similar two drugs effects are, and can help to identify the cellularbackgrounds in which they will show greater efficacy.

10. Biochemical signatures

Most compounds exert multiple biochemical effects in cellu-lar systems. In order to examine the pleiotropy of the enzymaticeffects induced by several families of anticancer agents, manygroups have begun to examine the systematic biochemical profilesof drug action. Using recombinant protein libraries, compound Kd’scan be systematically measured for hundreds of kinase domains.This binding information can yield profiles of affinities that broadlydescribe kinase inhibitor specificity. When plotted on phyloge-netic trees of kinase sequence similarity, selectivity profiles ofkinase inhibitors can be easily visualized (Karaman et al., 2008).Because kinase inhibitors are competitive inhibitors of ATP bind-ing, a recent effort has extended the analysis of the selectivityof kinase inhibitors to the measurement of kinase activity in thepresence of high ATP concentrations to more closely approximatecellular effects. This allows for the clustering of drugs based upontheir relative kinase activity and the discovery of new drug tar-gets/functions for known inhibitors (Anastassiadis et al., 2011).Together these approaches offer broad biochemical readouts ofkinase inhibitor action, and allow for the biochemical classificationof kinase inhibitors.

Beyond kinase activity, proteomics approaches in cell lysatescan be used to couple quantitative mass spectrometry withconventional biochemical characterization approaches. Using non-specific kinase or histone-deacetylase beads as a capture reagent,proteomic signatures that are capable of semi-comprehensivelyassessing inhibitor function at various concentrations are usedto produce inhibitor signatures of phosphorylation/de-acetylationinhibition (Bantscheff et al., 2007, 2011). Together, these biochem-ical approaches offer a variety of strategies for comprehensivelyprofiling the biochemical effects of small molecules in an attemptto understand specificity and mechanism of action.

11. Genetic signatures

While drugs have typically been characterized biochemically,and this approach can elucidate a spectrum of enzymatic effects,these biochemical characterizations often lack functional informa-tion about how a drug is actually causing cytotoxicity within acancer cell.

The first efforts toward large-scale drug characterization inmammalian cell lines were performed on the NCI-60 panel of celllines. This panel was created in an attempt to accelerate drug dis-covery for the treatment of therapeutically intractable solid tumorsthat did not benefit from the first generation of combination reg-imen building. As a consequence of work with this set of celllines, a large amount of screening data on thousands of novelcompounds was generated by the NCI in the 1980s (Paull et al.,1989). Upon characterizing and examining the profiles of activ-ity for diverse compounds across cell lines, the NCI proposed theCOMPARE approach. COMPARE sought to rank-order lists of drugswith correlated patterns of response across the entire NCI-60 panelof cell lines. Following the annotation of the NCI-60 panel withgenetic and biochemical data, wide-ranging efforts were under-taken to align and cluster the cell line response data with themolecular characterization data (Scherf et al., 2000; Weinsteinet al., 1997) using the discovery algorithm. These efforts highlighted

the potential importance of the p53 tumor suppressor and the P-glycoprotein efflux pump in predicting cellular response to manycommonly used cytotoxic compounds. Recently, utilizing the NCIdata and microarray expression data for the NCI 60 cell lines, the

2 stance

T6aHNdel

iqaTpafipbl2otlcp

ub(fbtftaTcc

1

mseiewstWcombIwtm

ocsf

54 J.R. Pritchard et al. / Drug Resi

heordescu group has suggested that it is possible to use the NCI-0 data to predict drug response in completely distinct cell linesnd cancer subtypes using the CONEX algorithm (Lee et al., 2007).owever, while thousands of compounds have been screened in theCI-60 cells, and comprehensive genomic data exists, it remainsifficult to mine and interpret the data that are generated via thesefforts. Additionally, concerns exist as to the relevance of these cellines to actual human tumors.

In 2006, a large microarray compendium, termed the connectiv-ty map, was generated to allow developers of novel compounds touery a large database of reference compounds for relationships ton investigational compound or a disease state (Lamb et al., 2006).hese searches have been suggested to not only identify a com-ounds’ mechanism of action (Hieronymus et al., 2006), but alsollow for the querying of target signatures to produce predictionsor therapeutic intervention (based on the assumption that oppos-ng drug gene expression patterns should effectively cancel outathogenic disease states) (Wei et al., 2006). Finally, this work haseen used by other groups to reposition drugs with similar molecu-

ar profiles but distinct indications (Dudley et al., 2011; Sirota et al.,011). Importantly, while these efforts have produced new proofsf concept, the signatures that are generated can be very difficulto interpret from a functional perspective. Furthermore, the pub-ished efforts to validate predictions often focus on “top-rankedompounds” and have not at this point rigorously validated theredictive depth of these signature-based queries.

Recently in mammalian cells, functional gene perturbationsing targeted RNA interference or chemical perturbations haseen used to examine the differential sensitivity of cells to drugsJiang et al., 2011; Wolpaw et al., 2011). Both of these methodsocus on targeted feature sets that are capable of discriminatingetween drugs that act upon distinct subsets of biology. Whilehe chemical approach has added resolution over non-apoptoticorms of cell death, the shRNA-based approach allows for quanti-ative boundaries and predictions to be made using an algorithmicdd-on to conventional supervised machine learning approaches.he strength of both of these approaches is that they detail spe-ific functional relationships that alter cellular responses to a givenompound.

2. In silico approaches

Large distinctions exist between efforts that seek to model drugechanisms of action in silico using datasets of clinical information,

ince these contain diverse data from clinical practice. Campillost al. started with a database of side effects from the unified med-cal language system (UMLS) and used it to build a common sideffects drug interaction network. They identified sets of compoundsith modest structural similarity, but high correlations in overall

ide effect profiles. Finally, by using in vitro biochemical assays,hey confirmed common target binding (Campillos et al., 2008).

hile this method cannot attribute a specific side effect to a spe-ific target, nor prove a functional role for the interaction, it is thenly method that uses actual clinical data to classify drugs by theirechanism of action. Furthermore, in the future, this approach may

e extended to predict the mechanisms of particular side effects.t is also interesting to speculate that common side effects profiles

ould demarcate “bad” combinations in a potential regimen, andhat early dismissal of combinations with close network proximity

ight reduce drug toxicity.The field of signature-based prediction has developed numer-

us approaches to simultaneously characterize drug action beyondell death. Signatures of drug action can tell us about inhibitorelectivity/off target effects, transcriptional response similarity,unction and toxicological action. All of these indications are

Updates 15 (2012) 249– 257

valuable, but if used together they might be combined in thedrug development process to simultaneously characterize inves-tigational compounds and promote their more rational and safeuse.

13. Systems perspectives in combination therapy

There are many innovative perspectives on how sys-tems/network biology might inform targeted cancer therapy,(Erler and Linding, 2010; Fitzgerald et al., 2006; Pawson andLinding, 2008). A major articulated goal is to identify activatedpathways that are druggable, compensatory pathways that accountfor single targeted treatment failure, or combinations of networknodes that give greater than additive benefits. Furthermore, massaction kinetic models in well-studied pathways (EGFR) have beensuggested to be potentially capable of identifying combinations ofmolecular species with synergistic effects (Fitzgerald et al., 2006).This is most often considered in the context of specific kinaseinhibitors targeted against oncogenic pathways. However, manycancers are currently treated with cytotoxic clinical regimens thathave diverse cure rates (Abeloff, 2008), and some that respondwell to therapy initially but rapidly develop resistance (Abeloff,2008). Less often considered (Tentner et al., 2012) is the valueof systems biology in examining classical chemotherapeutics.Chemotherapeutics activate downstream pathway signaling fol-lowing treatment. Oncogene and tumor suppressor networks areknown to alter similar signaling molecules as a result of oncogenictransformation. For instance: given a particularly responsivepatient population, it may be desirable to use two drugs withthe same mechanism of action but non-overlapping toxicity.Furthermore, kinase inhibitors are often neither specific nor freeof side effects (Kantarjian et al., 2012), and, as such, signatures oftheir effects and models of their action could help guide regimenbuilding – even if the assumptions present in modeling efforts(inhibition of one key node) are not valid. Applying systems andnetwork biology to characterize the global alterations affectingtherapeutic response in the face of chemotherapy should promotethe appropriate – or even optimal – use of current clinically useddrugs.

14. Toward combination drug signatures

To integrate systems/network biology into the future of combi-nation regimens, network approaches will have to delineate alteredpathways and understand how combinations of drugs will interactwith those pathways. If combination chemotherapy in cancer tendsto select for drug resistant cell states, a couple of hypotheses mightaccount for this effect. The first is that resistance to multiple drugs ina combination regimen is mediated by downstream network effectsthat are common to all drugs (a common effects/non independentaction hypothesis), and second, combinations may co-opt distinctsub-networks downstream of drug targets to create combinationspecific effects. Understanding these distinctions will be criticalto future regimen building. A good way to test these conflictinghypotheses and examine them with next generation therapeuticswill be to develop combination therapy signatures. While this isan attractive and potentially useful idea, no current methods forsignature-based drug prediction as detailed above have specificallyaddressed how, relative to single agents, combinations of drugsmight function. This may be due to technical and/or conceptuallimitations of certain approaches, but combination signatures will

be critical to regimen design.

In examining combination network “signatures”, combinationsmay be a sum of their component drug networks, they mayreinforce single component drug networks, or they may act on

J.R. Pritchard et al. / Drug Resistance

Fig. 2. A diagram showing potential computational solutions revealed by the addi-tion of two drug “signatures”. Combination treatments may reveal signaturesrepresentative of one of the component signatures. In this case, the drug combi-nation would be expected to act like one of the parental drugs – merely at a lowerdose. Alternatively, the combination signature may be an average of component sig-nar

soeksorscse

cdddnbmsopaaiatp

1

toadtot(na

atures, in which global genetic dependencies are reduced or homogenized. Finally, combination signature may deviate completely from component signatures – rep-esenting a neomorphic affect achieved by the drug combination.

ub-networks that are not utilized by single drugs (combinationff target effects). All of these effects may be desirable in differ-nt personalized medicine contexts with a diverse range of priornowledge. For example, if a combination of drugs reinforces aingle component drug network (one drug potentiates the effectf another), it is best administered to a patient population that isesponsive to that single drug mechanism. While this would requireingle drug sensitizing networks to be well described, it may revealombinations of drugs that increase the therapeutic window for aingle component compound – while also increasing the on-targetffects to which a given patient is known to be susceptible (Fig. 2).

If a combination of compounds results in an averaging of theiromponent sensitivities and resistances, will drug-specific geneticependencies be canceled out? If so, this suggests a striking para-ox: namely, could a genetic interaction in a single componentrug case be rendered unimportant in the context of a combi-ation? This possibility is an intriguing concept that will have toe explicitly investigated, because, although it would suggest theinimization of resistance to one drug, it would also minimize

ensitivity to another. This dichotomy in a combination regimen,ne drug sensitizing while the other promoting resistance to aarticular legion would have a strong benefit in the absence ofny genetic knowledge – a type of drug hedge betting. This aver-ging of sensitivities and resistances may reflect a requirementnherent in clinical trials – that efficacy must be demonstratedcross genetically diverse cohorts. Of course, such homogeniza-ion would also come at the cost of optimal regimens for individualatients.

5. Final thoughts

Understanding the network biology of single and combina-ion drugs could guide clinical practice across a diverse spectrumf knowledge concerning tumor pathology and genetics. In thebsence of any information as to the underlying genetic networksriving tumor progression in the patient, broadly acting combina-ions that independently utilize diverse sub-networks may form anptimal therapeutic strategy. Conversely, in an attempt to adminis-

er drug combinations with particularly potent combination actioni.e. synergy), it may be possible to identify the signatures of sub-etworks that are functionally important for that drug combinationnd design companion diagnostics to target the right combination

Updates 15 (2012) 249– 257 255

to the right patient. In fact, it may absolutely essential to identifysuch critical sub-networks prior to the administration of “syn-ergistic” therapies. Drug synergy has historically been a terriblepredictor of in vivo combinatorial efficacy. This could be due thestrong context-dependence of synergistic efficacy. In other words,“synergistic” therapies may absolutely require biomarkers thatstratify optimal patient populations prior to the construction ofclinical trials. Finally, in the presence of extensive pathologicalinformation, we may not only be able to pick the right combina-tion for the right patient, i.e. match drugs to information about thebulk portion of a patient’s tumor, but identify resistant subpop-ulations in heterogeneous tumors before they dominate a tumorand dose drug combinations that might minimize the outgrowth ofthese subpopulations.

Acknowledgements

We thank Peter Bruno and Boyang Zhao for helpful conversa-tions. M.T.H. is the Chang and Eisen Associate Professor of Biologyand D.A.L. is the Ford Professor of Biological Engineering. Fundingwas provided by ICBP #U54-CA112967-06 (M.T.H. and D.A.L.) andNIH RO1-CA128803-04 (M.T.H.).

References

Abeloff, M.D., 2008. Abeloff’s Clinical Oncology, 4th ed. Churchill Living-stone/Elsevier, Philadelphia.

Abramson, J.S., Shipp, M.A., 2005. Advances in the biology and therapy of diffuselarge B-cell lymphoma: moving toward a molecularly targeted approach. Blood106, 1164–1174.

Almeida Da Silva, P.E., Palomino, J.C., 2011. Molecular basis and mechanisms of drugresistance in Mycobacterium tuberculosis: classical and new drugs. Journal ofAntimicrobial Chemotherapy 66, 1417–1430.

Anastassiadis, T., Deacon, S.W., Devarajan, K., Ma, H., Peterson, J.R., 2011. Compre-hensive assay of kinase catalytic activity reveals features of kinase inhibitorselectivity. Nature Biotechnology 29, 1039–1045.

Bantscheff, M., Eberhard, D., Abraham, Y., Bastuck, S., Boesche, M., Hobson, S.,Mathieson, T., Perrin, J., Raida, M., Rau, C., et al., 2007. Quantitative chemical pro-teomics reveals mechanisms of action of clinical ABL kinase inhibitors. NatureBiotechnology 25, 1035–1044.

Bantscheff, M., Hopf, C., Savitski, M.M., Dittmann, A., Grandi, P., Michon, A.M., Schlegl,J., Abraham, Y., Becher, I., Bergamini, G., et al., 2011. Chemoproteomics profil-ing of HDAC inhibitors reveals selective targeting of HDAC complexes. NatureBiotechnology 29, 255–265.

Bartz, S.R., Zhang, Z., Burchard, J., Imakura, M., Martin, M., Palmieri, A., Need-ham, R., Guo, J., Gordon, M., Chung, N., et al., 2006. Small interfering RNAscreens reveal enhanced cisplatin cytotoxicity in tumor cells having bothBRCA network and TP53 disruptions. Molecular and Cellular Biology 26,9377–9386.

Beketic-Oreskovic, L., Duran, G.E., Chen, G., Dumontet, C., Sikic, B.I., 1995. Decreasedmutation rate for cellular resistance to doxorubicin and suppression of mdr1gene activation by the cyclosporin PSC 833. Journal of the National Cancer Insti-tute 87, 1593–1602.

Bledsoe, R.K., Montana, V.G., Stanley, T.B., Delves, C.J., Apolito, C.J., McKee, D.D., Con-sler, T.G., Parks, D.J., Stewart, E.L., Willson, T.M., et al., 2002. Crystal structureof the glucocorticoid receptor ligand binding domain reveals a novel mode ofreceptor dimerization and coactivator recognition. Cell 110, 93–105.

Borst, P., 2012. Cancer drug pan-resistance: pumps, cancer stem cells, quiescence,epithelial to mesenchymal transition, blocked cell death pathways, persisters orwhat? Open Biology 2, 120066.

Borst, P., Jonkers, J., Rottenberg, S., 2007. What makes tumors multidrug resistant?Cell Cycle 6, 2782–2787.

Boulos, N., Mulder, H.L., Calabrese, C.R., Morrison, J.B., Rehg, J.E., Relling, M.V., Sherr,C.J., Williams, R.T., 2011. Chemotherapeutic agents circumvent emergence ofdasatinib-resistant BCR-ABL kinase mutations in a precise mouse model ofPhiladelphia chromosome-positive acute lymphoblastic leukemia. Blood 117,3585–3595.

Campillos, M., Kuhn, M., Gavin, A.C., Jensen, L.J., Bork, P., 2008. Drug target identifi-cation using side-effect similarity. Science 321, 263–266.

Chabner, B.A., Roberts Jr., T.G., 2005. Timeline: chemotherapy and the war on cancer.Nature Reviews Cancer 5, 65–72.

Chen, W.W., Schoeberl, B., Jasper, P.J., Niepel, M., Nielsen, U.B., Lauffenburger, D.A.,

Sorger, P.K., 2009. Input–output behavior of ErbB signaling pathways as revealedby a mass action model trained against dynamic data. Molecular Systems Biology5, 239.

Conroy, T., Desseigne, F., Ychou, M., Bouche, O., Guimbaud, R., Becouarn, Y., Adenis, A.,Raoul, J.L., Gourgou-Bourgade, S., de la Fouchardiere, C., et al., 2011. FOLFIRINOX

2 stance

D

D

D

D

D

E

F

F

F

F

G

H

H

J

J

J

K

K

K

K

K

L

L

L

56 J.R. Pritchard et al. / Drug Resi

versus gemcitabine for metastatic pancreatic cancer. The New England Journalof Medicine 364, 1817–1825.

eVita, V.T., Schein, P.S., 1973. The use of drugs in combination for the treatmentof cancer: rationale and results. The New England Journal of Medicine 288,998–1006.

ing, L., Ley, T.J., Larson, D.E., Miller, C.A., Koboldt, D.C., Welch, J.S., Ritchey, J.K.,Young, M.A., Lamprecht, T., McLellan, M.D., et al., 2012. Clonal evolution inrelapsed acute myeloid leukaemia revealed by whole-genome sequencing.Nature 481, 506–510.

oles, J., Hemann, M.T., 2010. Nek4 status differentially alters sensitivity to distinctmicrotubule poisons. Cancer Research 70, 1033–1041.

udley, J.T., Sirota, M., Shenoy, M., Pai, R.K., Roedder, S., Chiang, A.P., Morgan, A.A.,Sarwal, M.M., Pasricha, P.J., Butte, A.J., 2011. Computational repositioning ofthe anticonvulsant topiramate for inflammatory bowel disease. Science Trans-lational Medicine 3 (96), ra76.

urant, J., Clevenbergh, P., Halfon, P., Delgiudice, P., Porsin, S., Simonet, P., Mon-tagne, N., Boucher, C.A., Schapiro, J.M., Dellamonica, P., 1999. Drug-resistancegenotyping in HIV-1 therapy: the VIRADAPT randomised controlled trial. Lancet353, 2195–2199.

rler, J.T., Linding, R., 2010. Network-based drugs and biomarkers. Journal of Pathol-ogy 220, 290–296.

arber, S., Diamond, L.K., 1948. Temporary remissions in acute leukemia in chil-dren produced by folic acid antagonist, 4-aminopteroyl-glutamic acid. The NewEngland Journal of Medicine 238, 787–793.

itzgerald, J.B., Schoeberl, B., Nielsen, U.B., Sorger, P.K., 2006. Systems biology andcombination therapy in the quest for clinical efficacy. Nature Chemical Biology2, 458–466.

rei 3rd, E., Holland, J.F., Schneiderman, M.A., Pinkel, D., Selkirk, G., Freire-ich, E.J., Silver, R.T., Gold, G.L., Regelson, W., 1958. A comparative study oftwo regimens of combination chemotherapy in acute leukemia. Blood 13,1126–1148.

reireich, E.J., Frei III, K.M.E., 1964. Quadruple combination therapy (VAMP) for acutelymphocytic leukemia of childhood. In: Proceedings of the American Associationfor Cancer Research.

oodman, L.S., Wintrobe, M.M., Dameshek, W., Goodman, M.J., Gilman, A., McLen-nan, M.T., 1984. Landmark article Sept. 21, 1946: nitrogen mustard therapy.In: Goodman, L.S., Wintrobe, M.M., Dameshek, W., Goodman, M.J., Gilman, A.,McLennan, M.T. (Eds.), Use of Methyl-Bis(Beta-Chloroethyl)Amine Hydrochlo-ride and Tris(Beta-Chloroethyl)Amine Hydrochloride for Hodgkin’s Disease,Lymphosarcoma, Leukemia and Certain Allied and Miscellaneous Disorders, vol.251, pp. 2255–2261, Jama.

ashimoto, M.W., Nikaido, O., Kobayashi, N., Chang, C.C., Trosko, J.E., Mori, T., 1996.A comparison of the propensity for gene amplification between near-tetraploidand near-diploid V79 clones resistant to 150 nM methotrexate. Carcinogenesis17, 389–394.

ieronymus, H., Lamb, J., Ross, K.N., Peng, X.P., Clement, C., Rodina, A., Nieto, M., Du,J., Stegmaier, K., Raj, S.M., et al., 2006. Gene expression signature-based chem-ical genomic prediction identifies a novel class of HSP90 pathway modulators.Cancer Cell 10, 321–330.

anes, K.A., Albeck, J.G., Gaudet, S., Sorger, P.K., Lauffenburger, D.A., Yaffe, M.B.,2005. A systems model of signaling identifies a molecular basis set for cytokine-induced apoptosis. Science 310, 1646–1653.

anes, K.A., Kelly, J.R., Gaudet, S., Albeck, J.G., Sorger, P.K., Lauffenburger, D.A., 2004.Cue-signal-response analysis of TNF-induced apoptosis by partial least squaresregression of dynamic multivariate data. Journal of Computational Biology 11,544–561.

iang, H., Pritchard, J.R., Williams, R.T., Lauffenburger, D.A., Hemann, M.T., 2011. Amammalian functional-genetic approach to characterizing cancer therapeutics.Nature Chemical Biology 7, 92–100.

antarjian, H.M., Shah, N.P., Cortes, J.E., Baccarani, M., Agarwal, M.B., Undur-raga, M.S., Wang, J., Ipina, J.J., Kim, D.W., Ogura, M., et al., 2012. Dasatinibor imatinib in newly diagnosed chronic-phase chronic myeloid leukemia:2-year follow-up from a randomized phase 3 trial (DASISION). Blood 119,1123–1129.

araman, M.W., Herrgard, S., Treiber, D.K., Gallant, P., Atteridge, C.E., Campbell, B.T.,Chan, K.W., Ciceri, P., Davis, M.I., Edeen, P.T., et al., 2008. A quantitative analysisof kinase inhibitor selectivity. Nature Biotechnology 26, 127–132.

reeger, P.K., Lauffenburger, D.A., 2010. Cancer systems biology: a network modelingperspective. Carcinogenesis 31, 2–8.

reeger, P.K., Mandhana, R., Alford, S.K., Haigis, K.M., Lauffenburger, D.A., 2009. RASmutations affect tumor necrosis factor-induced apoptosis in colon carcinomacells via ERK-modulatory negative and positive feedback circuits along withnon-ERK pathway effects. Cancer Research 69, 8191–8199.

umar, N., Afeyan, R., Kim, H.D., Lauffenburger, D.A., 2008. Multipathway modelenables prediction of kinase inhibitor cross-talk effects on migration ofHer2-overexpressing mammary epithelial cells. Molecular Pharmacology 73,1668–1678.

amb, J., Crawford, E.D., Peck, D., Modell, J.W., Blat, I.C., Wrobel, M.J., Lerner, J., Brunet,J.P., Subramanian, A., Ross, K.N., et al., 2006. The connectivity map: using gene-expression signatures to connect small molecules, genes, and disease. Science313, 1929–1935.

aw, L.W., 1952a. Effects of combinations of antileukemic agents on an acute lym-phocytic leukemia of mice. Cancer Research 12, 871–878.

aw, L.W., 1952b. Origin of the resistance of leukaemic cells to folic acid antagonists.Nature 169, 628–629.

Updates 15 (2012) 249– 257

Lee, J.K., Havaleshko, D.M., Cho, H., Weinstein, J.N., Kaldjian, E.P., Karpovich, J.,Grimshaw, A., Theodorescu, D., 2007. A strategy for predicting the chemosen-sitivity of human cancers and its application to drug discovery. Proceedingsof the National Academy of Sciences of the United States of America 104,13086–13091.

Letai, A.G., 2008. Diagnosing and exploiting cancer’s addiction to blocks in apoptosis.Nature Reviews Cancer 8, 121–132.

Luria, S.E., Delbruck, M., 1943. Mutations of bacteria from virus sensitivity to virusresistance. Genetics 28, 491–511.

Matthews, D.A., Alden, R.A., Bolin, J.T., Freer, S.T., Hamlin, R., Xuong, N., Kraut,J., Poe, M., Williams, M., Hoogsteen, K., 1977. Dihydrofolate reductase:X-ray structure of the binary complex with methotrexate. Science 197,452–455.

Miller-Jensen, K., Janes, K.A., Brugge, J.S., Lauffenburger, D.A., 2007. Common effectorprocessing mediates cell-specific responses to stimuli. Nature 448, 604–608.

Mullighan, C.G., Phillips, L.A., Su, X., Ma, J., Miller, C.B., Shurtleff, S.A., Downing, J.R.,2008. Genomic analysis of the clonal origins of relapsed acute lymphoblasticleukemia. Science 322, 1377–1380.

Mullighan, C.G., Zhang, J., Kasper, L.H., Lerach, S., Payne-Turner, D., Phillips,L.A., Heatley, S.L., Holmfeldt, L., Collins-Underwood, J.R., Ma, J., et al., 2011.CREBBP mutations in relapsed acute lymphoblastic leukaemia. Nature 471,235–239.

Newcombe, H.B., Hawirko, R., 1949. Spontaneous mutation to streptomycinresistance and dependence in Escherichia coli. Journal of Bacteriology 57,565–572.

Ni Chonghaile, T., Sarosiek, K.A., Vo, T.T., Ryan, J.A., Tammareddi, A., Moore Vdel, G.,Deng, J., Anderson, K.C., Richardson, P., Tai, Y.T., et al., 2011. Pretreatment mito-chondrial priming correlates with clinical response to cytotoxic chemotherapy.Science 334, 1129–1133.

Paull, K.D., Shoemaker, R.H., Hodes, L., Monks, A., Scudiero, D.A., Rubinstein, L.,Plowman, J., Boyd, M.R., 1989. Display and analysis of patterns of differen-tial activity of drugs against human tumor cell lines: development of meangraph and COMPARE algorithm. Journal of the National Cancer Institute 81,1088–1092.

Pawson, T., Linding, R., 2008. Network medicine. FEBS Letters 582, 1266–1270.Pirker, R., Pereira, J.R., Szczesna, A., von Pawel, J., Krzakowski, M., Ramlau, R., Vynny-

chenko, I., Park, K., Yu, C.T., Ganul, V., et al., 2009. Cetuximab plus chemotherapyin patients with advanced non-small-cell lung cancer (FLEX): an open-labelrandomised phase III trial. Lancet 373, 1525–1531.

Pommier, Y., Leo, E., Zhang, H., Marchand, C., 2010. DNA topoisomerases and theirpoisoning by anticancer and antibacterial drugs. Chemistry and Biology 17,421–433.

Scherf, U., Ross, D.T., Waltham, M., Smith, L.H., Lee, J.K., Tanabe, L., Kohn, K.W., Rein-hold, W.C., Myers, T.G., Andrews, D.T., et al., 2000. A gene expression databasefor the molecular pharmacology of cancer. Nature Genetics 24, 236–244.

Schoeberl, B., Pace, E.A., Fitzgerald, J.B., Harms, B.D., Xu, L., Nie, L., Linggi, B., Kalra, A.,Paragas, V., Bukhalid, R., et al., 2009. Therapeutically targeting ErbB3: a key nodein ligand-induced activation of the ErbB receptor-PI3K axis. Science Signaling 2,ra31.

Shah, N.P., Skaggs, B.J., Branford, S., Hughes, T.P., Nicoll, J.M., Paquette, R.L., Sawyers,C.L., 2007. Sequential ABL kinase inhibitor therapy selects for compound drug-resistant BCR-ABL mutations with altered oncogenic potency. The Journal ofClinical Investigation 117, 2562–2569.

Shah, N.P., Tran, C., Lee, F.Y., Chen, P., Norris, D., Sawyers, C.L., 2004. Over-riding imatinib resistance with a novel ABL kinase inhibitor. Science 305,399–401.

Sirota, M., Dudley, J.T., Kim, J., Chiang, A.P., Morgan, A.A., Sweet-Cordero, A., Sage, J.,Butte, A.J., 2011. Discovery and preclinical validation of drug indications usingcompendia of public gene expression data. Science Translational Medicine 3(96), ra77.

Skipper, H.E., Schabel Jr., F.M., Wilcox, W.S., 1964. Experimental evaluation ofpotential anticancer agents, Xiii. On the criteria and kinetics associated with“curability” of experimental leukemia. Cancer Chemotherapy Reports. Part 135, 1–111.

Skipper, H.E., Thomson, J.R., Bell, M., 1954. Attempts at dual blocking of biochemicalevents in cancer chemotherapy. Cancer Research 14, 503–507.

Staker, B.L., Hjerrild, K., Feese, M.D., Behnke, C.A., Burgin Jr., A.B., Stewart, L.,2002. The mechanism of topoisomerase I poisoning by a camptothecin analog.Proceedings of the National Academy of Sciences of the United States of America99, 15387–15392.

Szakacs, G., Paterson, J.K., Ludwig, J.A., Booth-Genthe, C., Gottesman, M.M., 2006.Targeting multidrug resistance in cancer. Nature Reviews Drug Discovery 5,219–234.

Talpaz, M., Shah, N.P., Kantarjian, H., Donato, N., Nicoll, J., Paquette, R., Cortes, J.,O’Brien, S., Nicaise, C., Bleickardt, E., et al., 2006. Dasatinib in imatinib-resistantPhiladelphia chromosome-positive leukemias. The New England Journal ofMedicine 354, 2531–2541.

Tentner, A.R., Lee, M.J., Ostheimer, G.J., Samson, L.D., Lauffenburger, D.A., Yaffe,M.B., 2012. Combined experimental and computational analysis of DNA damagesignaling reveals context-dependent roles for Erk in apoptosis and G1/S arrestafter genotoxic stress. Molecular Systems Biology 8, 568.

Toettcher, J.E., Loewer, A., Ostheimer, G.J., Yaffe, M.B., Tidor, B., Lahav, G., 2009.Distinct mechanisms act in concert to mediate cell cycle arrest. Proceedingsof the National Academy of Sciences of the United States of America 106,785–790.

stance

T

W

W

W

W

W

J.R. Pritchard et al. / Drug Resi

urke, A.B., Zejnullahu, K., Wu, Y.L., Song, Y., Dias-Santagata, D., Lifshits, E., Toschi,L., Rogers, A., Mok, T., Sequist, L., et al., 2010. Preexistence and clonal selectionof MET amplification in EGFR mutant NSCLC. Cancer Cell 17, 77–88.

agle, N., Emery, C., Berger, M.F., Davis, M.J., Sawyer, A., Pochanard, P., Kehoe, S.M.,Johannessen, C.M., Macconaill, L.E., Hahn, W.C., et al., 2011. Dissecting thera-peutic resistance to RAF inhibition in melanoma by tumor genomic profiling.Journal of Clinical Oncology: Official Journal of the American Society of ClinicalOncology 29, 3085–3096.

all, M.E., Wani, M.C., 1995. Camptothecin and taxol: discovery to clinic–thirteenthBruce F. Cain Memorial Award Lecture. Cancer Research 55, 753–760.

ei, G., Twomey, D., Lamb, J., Schlis, K., Agarwal, J., Stam, R.W., Opferman, J.T.,Sallan, S.E., den Boer, M.L., Pieters, R., et al., 2006. Gene expression-based chem-ical genomics identifies rapamycin as a modulator of MCL1 and glucocorticoidresistance. Cancer Cell 10, 331–342.

einstein, J.N., Myers, T.G., O’Connor, P.M., Friend, S.H., Fornace Jr., A.J., Kohn, K.W.,

Fojo, T., Bates, S.E., Rubinstein, L.V., Anderson, N.L., et al., 1997. An information-intensive approach to the molecular pharmacology of cancer. Science 275,343–349.

hitehurst, A.W., Bodemann, B.O., Cardenas, J., Ferguson, D., Girard, L., Pey-ton, M., Minna, J.D., Michnoff, C., Hao, W., Roth, M.G., et al., 2007. Synthetic

Updates 15 (2012) 249– 257 257

lethal screen identification of chemosensitizer loci in cancer cells. Nature 446,815–819.

Wolpaw, A.J., Shimada, K., Skouta, R., Welsch, M.E., Akavia, U.D., Pe’er, D., Shaik, F.,Bulinski, J.C., Stockwell, B.R., 2011. Modulatory profiling identifies mechanismsof small molecule-induced cell death. Proceedings of the National Academy ofSciences of the United States of America 108, E771–E780.

Xie, L., Kinnings, S.L., Bourne, P.E., 2012. Novel computational approaches topolypharmacology as a means to define responses to individual drugs. AnnualReview of Pharmacology and Toxicology 52, 361–379.

Yauch, R.L., Dijkgraaf, G.J., Alicke, B., Januario, T., Ahn, C.P., Holcomb, T., Pujara, K.,Stinson, J., Callahan, C.A., Tang, T., et al., 2009. Smoothened mutation confersresistance to a Hedgehog pathway inhibitor in medulloblastoma. Science 326,572–574.

Zander, S.A., Kersbergen, A., van der Burg, E., de Water, N., van Tellingen, O., Gunnars-dottir, S., Jaspers, J.E., Pajic, M., Nygren, A.O., Jonkers, J., et al., 2010. Sensitivity

and acquired resistance of BRCA1; p53-deficient mouse mammary tumors tothe topoisomerase I inhibitor topotecan. Cancer Research 70, 1700–1710.

Zhang, J., Adrian, F.J., Jahnke, W., Cowan-Jacob, S.W., Li, A.G., Iacob, R.E., Sim, T., Pow-ers, J., Dierks, C., Sun, F., et al., 2010. Targeting Bcr-Abl by combining allostericwith ATP-binding-site inhibitors. Nature 463, 501–506.