Chemotherapy and signaling

Abstract

In recent years, oncologists have begun to conclude that chemotherapy has reached a plateau of efficacy as a primary treatment modality, even if toxicity can be effectively controlled. Emerging specific inhibitors of signaling and metabolic pathways (i.e., targeted agents) contrast with traditional chemotherapy drugs in that the latter primarily interfere with the DNA biosynthesis and the cell replication machinery. In an attempt to improve on the efficacy, combination of targeted drugs with conventional chemotherapeutics has become a routine way of testing multiple new agents in early phase clinical trials. This review discusses the recent advances including integrative systematic biology and RNAi approaches to counteract the chemotherapy resistance and to buttress the selectivity, efficacy and personalization of anticancer drug therapy.

Introduction

The first uses of chemotherapy to control cancer were reported in the 1940s,¹ and in the decades since, treatment of patients with broadly toxic chemicals has represented a mainstay of medical oncology, in spite of the frequent severe side effects associated with such treatments. In recent years, oncologists have begun to conclude that chemotherapy has reached a plateau of efficacy as a primary treatment modality, even if toxicity can be effectively controlled. For example, dose escalation trials in breast cancer using conventional doses of cyclophosphamide with doxorubicin, methotrexate and fluorouracil or high doses of cyclophosphamide with carboplatin and thiotepa did not produce any improvement in survival rates, even though the dose-limiting bone marrow toxicity of chemotherapy was circumvented by restitution with autologous hematopoietic precursors.^2,3 An alternative approach that has been explored to improve the efficacy of chemotherapy is the selection of patients based on expression of chemoresistance markers in their tumors such as ERCC1, to help identify patients who would likely benefit from specific chemotherapy regimens (e.g., high expression of ERCC1 confers resistance to platinum chemotherapy agents, reviewed in ref. 4). Such chemotherapy “personalization” strategies can improve the outcomes and quality of life for patients, and have the potential to be greatly expanded over the next decade to be effective in tumors with intrinsic or acquired resistance to chemotherapies.

Emerging specific inhibitors of signaling and metabolic pathways (i.e., targeted agents) contrast with traditional chemotherapy drugs in that the latter primarily interfere with the DNA biosynthesis and the cell replication machinery. Although the cancer specificity of targeted agents compared to traditional chemotherapy drugs is not absolute, many of the intended molecular targets are specifically overexpressed or hyperactivated in cancer cells and promote cancer cell proliferation, survival, angiogenesis and invasion. Specific agents employed to inhibit the key molecules of these pathways typically are small molecule inhibitors or monoclonal antibodies against proteins such as receptor tyrosine kinases (RTKs),⁵ or other proteins with catalytic activities. Practical exploitation of over three decades of basic cellular signaling research has generated a wave of new compounds targeting cancer-relevant biomolecules. Clinical evaluation of these new agents has not only revealed in some instances exceptional activity against previously highly refractory malignancies such as chronic myeloid leukemia (CML),^6–8 but has inevitably also identified limitations to the efficacy of these drugs.⁹

In the best-case scenario, a relatively small subset of malignancies depends on a uniquely hyperactive oncogene arising from unique mutations or translocations, a phenomenon often referred to as oncogene addiction.¹⁰ Such cancers will be susceptible to blockade of a single signaling molecule with its targeted drug. For example, CML is caused by chromosomal translocation t(9;22)(q34;q11) producing the chimeric hyperactive ABL1 kinase. Gleevec (imatinib) and other agents such as nilotinib¹¹ and dasatinib¹² potently inhibit ABL1 and produced, thus far, an unprecedented clinical outcome in CML.^8,13 Even patients from the earliest trials have not reached the 50^th percentile despite nearly 10 years of followup, thereby making estimates of median survival unclear. As another example, a subset of non-small-cell lung cancers (NSCLC) carries various gain-of-function mutations in the epidermal growth factor receptor (EGFR). These “oncogene -addicted” tumors are highly sensitive to EGFR kinase inhibitors such as gefitinib and erlotinib.¹⁴ Much more typically however, and for the vast majority of cancers, use of a targeted agent as monotherapy was associated with either minimal clinical efficacy or the activity was limited by rapidly acquired resistance. Such limited efficacy is likely attributable to the multiple redundancies and branched interactions within cellular signaling networks, which make it relatively easy for cancer cells to alter signaling flow around a single “node” that is inhibited by a targeted therapeutic.^15–17 The search for oncogenic driver mutations in multiple carcinomas has been disappointing and has led to realization that the monogenic cancers are the exceptions rather than the rule.^18,19

In an attempt to improve on the efficacy of the targeted agents, and as a means to increase the utility of classic cytotoxic regimens, combining targeted drugs with conventional chemotherapeutics has become a routine way of testing multiple new agents in early phase clinical trials. While only 0.6% (3 of 497) of phase I/II trials during the period of 1998–2001 involved combinations of targeted and chemotherapeutic agents, 8.6% (99 of 1,145) trials used this approach in 2005–2009 and the trend continues to grow. If successful, such combination approaches can benefit thousands of patients with metastatic cancers, whose lives now extend into their second and third years post-diagnosis, and who increasingly populate the pool of phase I trial participants.

It is possible to argue for the combination of chemotherapy and targeted therapy on the empirical and pragmatic grounds. However, to fully enable and exploit any therapeutic gains, it is important to understand the underlying biology. Based on a nuanced understanding of cellular signaling networks, what is the soundest approach to develop effective new combinations of targeted agents and chemotherapy drugs? There are a number of key issues to consider. For example, cancer cells can activate multiple pro-survival signaling cascades to maintain their viability in the context of the DNA damage and oxidative stress damage induced by chemotherapy and radiation; using a panel of targeted agents that disrupts hubs in these pro-survival signaling cascades may potentiate tumor cell killing. At the same time, such cooperative cytotoxicity of chemotherapy and targeted agents may come at the cost of increased myelotoxicity or other unwanted effects on normal organs and tissues, thus forcing a reduction in the dosing of potentially life-saving agents. Such inadvertent effects recently confounded two major adjuvant clinical trials for colon cancer, when the addition of bevacizumab or cetuximab (VEGF- and EGFR-targeting antibodies, respectively) to chemotherapy produced inferior results.^20,21 Here, we will review some successful strategies used for combination therapy, and suggest possible reasons for the failure of other strategies. We will also discuss prospects for improvements in cancer management based on the increasingly comprehensive knowledge of the complex interdependencies in cancer cell signaling networks.

Mechanisms of Chemotherapy-Induced Cell Death and Resistance

Cell killing induced by chemotherapeutic agents: many mechanisms.

Historically, drugs used for chemotherapy were developed based on the rationale that rapidly dividing cells are particularly sensitive to the damage of their DNA (and to a lesser degree, RNA) synthesis and of the mitotic spindle. The vast majority of normal cells are quiescent, resting in the G₁/G₀ phase of the cell cycle; hence, drugs selectively damaging dividing cells in S or M phases have a selective bias for tumor cells. Generation of DNA adducts by platinoids (cisplatin, carboplatin, oxaliplatin) and alkylating agents (cyclophosphamide, nitrosourea, temozolomide, thiotepa) stalls DNA replication forks in S phase,^22–24 and blocks transcription by RNA polymerases.²⁵ Structural analogs of nucleotide precursors (5-fluorouracil, citarabine, gemcitabine) and topoisomerase inhibitors (etoposide, doxorubicin, irinotecan (CPT-11)) also introduce single- and double-strand breaks in the DNA. Drug-induced DNA damage can activate checkpoints for removal of DNA lesions to provide time for DNA repair. However, prolonged or repeated exposure of cells to chemotherapy agents or unsustainable damages can lead to apoptosis in rapidly dividing cells, resulting in the reduction in tumor mass, while inducing some “collateral damage” to frequently dividing normal cells such as hair follicles, myelopoietic bone marrow precursor cells and intestinal epithelial cells.

DNA damage responses.

Aberrant activation of DNA repair in response to chemotherapy-induced DNA damage is responsible for important mechanisms of drug resistance in tumors. The ATM (ataxia telangiectasia mutated) and ATR (ataxia telangiectasia and Rad3-related) kinases have been identified (Fig. 1) as important cellular triggers for checkpoints contributing to chemotherapy resistance. ATR and ATM are required for cell cycle arrest and DNA repair in response to DNA damage, based on their ability to regulate a series of checkpoint proteins operating at G₁/S that prevent initiation of DNA replication. These proteins include CHEK2, p53 and BRCA1. A later S-phase checkpoint is activated by phosphorylation of CHEK1, NBS and SMC1. Finally, a third G₂/M checkpoint involving the proteins CHEK1/CHEK2, BRCA1 and Rad17 prevents the entry of cells with damaged DNA into mitosis.^26–28 ATR/ATM kinases also activate the Fanconi anemia (FA) group of genes, which includes BRCA2 (FANCD1). The FA proteins form a multi-protein complex that recognizes damaged DNA and mediates homologous recombination to repair double-stranded DNA breaks. Mutations in FA proteins in tumors lead to genomic instability,²⁹ but also render cells highly sensitive to anticancer drugs.^30,31

Figure 1.

Canonical signaling pathways involved in chemotherapy-induced cell death. See text for details.

Targeting specific signaling components of the DNA damage response can thus increase the efficacy of chemotherapy. This approach is being actively pursued in selecting targets for drug development. For example, AZD7762, an inhibitor of CHEK1,²⁶ synergized with the nucleoside analog gemcitabine to kill tumor cells with defects in DNA repair proteins.³² p53 is frequently inactivated either by mutations or by increased degradation in cancer cells. Nutlins, small molecule inhibitors of a p53-specific ubiquitin-conjugating E3 ligase, MDM2, enhance p53-dependent apoptosis³³ by preventing the degradation and stabilizing p53 function. Nutlins also exhibit activity in cancer cells with mutated p53 via stabilization of the transcriptional factor E2F1, which is also an MDM2 target;³⁴ stabilized E2F1 then transcriptionally activates pro-apoptotic genes p73 and Noxa.^34,35

Beyond the ATM/ATR core pathway, DNA damage also induces the expression of the GADD45 family of proteins regulating DNA repair, maintenance of genomic stability and apoptosis.^36,37 As shown in bone marrow cells from GADD45^-/- mice, deficiency of these proteins sensitized the cells to apoptosis induced by the chemotherapeutic agents daunorubicin and VP-16.³⁸ The proteins of the nucleotide excision repair (NER) pathway, including xeroderma pigmentosum (XP) proteins and ERCC1 (excision repair cross-complementing 1) are responsible for drug resistance when overexpressed in cancer cells^39,40 due to increased protection against chemotherapy-induced DNA damage. These proteins are promising targets for inhibitory drugs.

Poly-ADP-ribose polymerase (PARP) is an important regulator of DNA damage repair.^41,42 Together with its interacting proteins ERCC6, XRCC1, XRCC5 and XRCC6, PARP mediates base excision and homologous recombination repair of single and double-stranded DNA breaks, respectively.^43–45 This pathway becomes the dominant DNA damage controlling mechanism in cancers deficient in homologous recombination DNA repair because of loss or haploinsufficiency of the breast cancer 1 (BRCA1) and breast cancer 2 (BRCA2) genes.^46,47 Blockade of PARP in the setting of the BRCA-deficiency has lead to transformative changes in the care of patients.^48,49 Recent data from a clinical trial in patients with triple negative metastatic breast cancer (lacking expression of hormonal receptors and normal copy number of HER2)⁵⁰ demonstrated strong synergy between chemotherapy and the PARP inhibitor, olaparib, due to the epigenetically silenced expression of BRCA genes in this subgroup of breast cancers. Homologous recombination DNA repair is also functionally deficient in cancers with loss of PTEN and hence a concomitantly hyperactive PI3K pathway.⁵¹ Thus, loss of PTEN may sensitize cancers to PARP inhibitors.

Microtubule poisons.

Taxanes (paclitaxel, docetaxel), epothilones and vinca alkaloids (vinblastine, vincristine, vinorelbine) interfere with the assembly of tubulin subunits and the stability of microtubules.^52–54 These defects are particularly important during mitosis, resulting in the failure of spindle attachments to the kinetochores and centrosome, and hence inducing aberrant spindle tension, triggering a second set of checkpoint proteins. Spindle assembly checkpoint proteins activated in response to microtubule poisons include MAD1, MAD2, hBUBR1, hBUB1, hMPS1.^55,56 These proteins temporarily halt cell cycle progression to allow spindle readjustment and to prevent the aberrant segregation of chromosomes. The spindle assembly checkpoint also depends on the function of the Aurora-A and Aurora-B kinases.⁵⁷ Mitotic catastrophe, a mechanism of cell death occurring during or after aberrant mitosis, is caused by a defective mitotic spindle assembly checkpoint and accumulation of aberrant chromosome segregations.^58–60 If cells manage to complete mitosis in spite of DNA damage and aneuploidy, a p53-dependent postmitotic checkpoint is activated.⁶¹ In this case, p53 induces transcription of Bcl-2 family genes (Bax, Bak, Bid, Puma) that target mitochondrial permeability, causing release of cytochrome c and activation of the effector caspase 3, triggering apoptosis (Fig. 1).

Apoptotic machinery.

Apoptosis is a common source of tumor cell clearance after drug treatments that have induced mitotic or DNA damage, in cases where the checkpoints are absent or overwhelmed. While discussion of apoptosis modulation for cancer treatment extends beyond the primary focus of this review (reviewed in ref. 62), several strategies to augment chemotherapy using either the extrinsic or intrinsic apoptotic pathways warrant specific mentioning. Inhibitors of Bcl-2 have been actively pursued by the pharmaceutical industry. A recent trial of the Bcl-2 inhibitor oblimersen in combination with the DNA-damaging agent, dacarbazine, in patients with advanced melanoma⁶³ has reported improved progression-free survival and overall response rates, with results most pronounced in patients with less aggressive tumors. Similarly, a phase III trial of oblimersen with the purine analog, fludarabine and cyclophosphamide in chronic lymphocytic leukemia showed significantly improved response rates and survival compared to chemotherapy alone.⁶⁴ The fact that Bcl-2 inactivation was most striking in patients with chemotherapy-sensitive tumors suggested chemosensitizing rather than direct anti-tumor activity of this Bcl-2 antagonist.

Activation of the extrinsic apoptosis machinery has also been exploited recently in the clinical trials of DR4 and DR5 death receptors agonists.^65,66 While the efforts to obtain evidence for the safety and the efficacy of combined therapy with extrinsic apoptosis pathways activators and chemotherapy are ongoing (reviewed in ref. 67), recombinant human Apo2L/TRAIL (dulanermin, Genentech, San Francisco, CA) showed provocative single agent activity in refractory soft tissue sarcomas. This is clearly a promising path for further work.

Alternative roads to tumor destruction.

Although apoptosis is most studied as a source of irreversible control of tumor growth, other mechanisms also are relevant in some cases, and may be exploitable in the clinic. For example, DNA damage induced by the topoisomerase I inhibitor, camptothecin, and the DNA-intercalating agent, doxorubicin, accelerates p53- and p21-dependent senescence in cancer cell lines.^68,69 Camptothecin induces senescence in CPT11-sensitive colorectal cancer cell lines, but also triggers senescence in CPT11-resistant cells exposed to both exogenous and endogenous SPARC (secreted protein acidic and rich in cysteine), due to upregulation of p16 and phosphorylation of p53 at Ser15.⁷⁰

Cytotoxic drugs including alkylating agents can also induce necrosis.^71–73 A key mediator of this process is PARP, which activates DNA repair following moderate genotoxic stress. However, excessive DNA damage causes hyperactivation of PARP. This leads to energy loss and necrotic cell death.⁷⁴ The underlying cellular mechanisms of PARP-induced necrosis are not fully elucidated but may include ATP and NADH depletion, mitochondrial dysfunction and activation of extrinsic apoptosis pathway proteins, including receptor-interacting serine-threonine kinase 1 (RIPK1), tumor necrosis factor receptor-associated factor 2 (TRAF2) and JNK1.^75–77

Depending on circumstances, autophagy, a process of lysosomal degradation of cellular constituents⁷⁸ can play positive or negative roles in cancer development, progression and therapy. It can suppress tumor formation⁷⁹ by reducing chromosomal instability⁸⁰ and by lowering levels of the scaffold protein p62, which is associated with activation of the DNA damage response and activation of the NF-κB pathway.⁸¹ Conversely, in established tumors autophagy can be cytoprotective via utilization of auto-digested proteins: this can facilitate development of drug resistance, by providing a proliferation benefit.^82–85 Autophagy frequently occurs in response to chemotherapy and parallels chemotherapy-induced cell death.⁸⁶ In some cases, targeting autophagy is emerging as an effective anticancer treatment strategy. For example, an inhibitor of heat shock protein 70, phenylethynesulfonamide (PES), suppressed autophagy and was directly cytotoxic to cancer cells.⁸⁷ Blockade of PI3K-mTOR signaling interferes with normal protein synthesis, inducing autophagy as a compensatory response to signals indicating nutrient deprivation.⁸⁸ Concomitant treatment of glioblastoma cells with temozolomide, a proautophagic drug, and dasatinib, a dual BCR/ABL and Src inhibitor, resulted in synergistic cell death by autophagy in a PI3K-dependent manner.⁸⁹ A number of clinical trials are currently pursuing this approach by testing the combination of mTOR inhibitors with chemotherapy.

Beyond resistance involving cell cycle checkpoints and the cell death machinery, intrinsic and post-treatment acquired drug resistance have also been linked to other categorical changes in tumor cells. These changes can be generally categorized as those that reduce the efficacy of drugs in reaching tumor cells, those that induce changes in cell signaling that provide robust pro-survival cues. and those that fundamentally alter the “identity” of the tumor cells, affecting differentiation status and proliferation potential.

Reducing exposure to chemotherapy agents.

Besides changes in the function of the core cell cycle checkpoint and cell death machinery, other distinct mechanisms have been identified that influence tumor vulnerability to cytotoxics and suggest drug targets for co-inhibition during administration of chemotherapy. For example, the intrinsic permeability of the tumor tissue for a clinical compound can become compromised due to high interstitial fluid pressures arising from the “leaky” newly formed blood vessels in the tumor bed.⁹⁰ This fenestrated, newly formed endothelium of the tumor vessels is induced by tumor-derived angiogenic factors, especially the vascular endothelial growth factor (VEGF). Although blocking formation of new tumor blood vessels to starve tumor growth is a primary action of VEGF inhibitors, application of these agents to tumors with existing vasculature produces the collateral benefit of improving drug delivery through “normalization” of tumor vessel. Antibodies and small molecule inhibitors of the VEGF signaling pathway⁹¹ are now in routine use for colorectal, non-small-cell lung and ovarian carcinomas and resulted in dramatic improvement of the chemotherapy efficacy (reviewed in ref. 92).

One of the most-studied mechanisms of tumor drug resistance involves changes in the expression and the activity of transporter proteins that pump chemotherapy drugs across the cell membranes. The ATP Binding Cassette (ABC) transporters function as drug pumps. ABC transporter group members, classified in subfamilies ABCA-ABCG based on the homology in their ATP-binding domains and other criteria,⁹³ are widely expressed in tissues. Four subfamilies of ABC transporters, designated ABCA, ABCB, ABCG and ABCC, efflux somewhat overlapping substrates including vinca alkaloids, taxanes, epipodophyllotoxins and camptothecins.^94,95 Some appear to be the sole physiological transporter for specific drugs of clinical relevance: for example, ABCC10 is the only transporter found to date that can efflux epothilone B, a microtubule-targeting agent currently in clinical trials for ovarian, melanoma and metastatic breast cancer.^96–98

Overexpression of ABC transporters can impart chemotherapy resistance by reducing the exposure of the cellular targets to the cytotoxic agents. Acting extrinsically to the tumor in the liver, drug efflux pumps can speed up the serum clearance and catabolic conversion of drugs, promoting their excretion with bile or urine.^99,100 In cell-intrinsic action, in vitro and xenograft experiments have found that elevated expression of ABC transporters in cancer cells contributes to resistance to chemotherapeutic agents by pumping drugs out of the cancer cell.⁹⁵ Future work using mouse knockout technology will clarify the issue of the functional requirement for individual pump proteins in cancer resistance.^101,102 In attempting to improve chemotherapy, it is arguably a much better strategy to focus on limiting tumor-intrinsic rather than extrinsic transporter activity, as the latter approach might cause an unwanted increase in toxicity towards to normal tissue.

Although most of the early work on ABC transporters involved study of their regulation of classic chemotherapeutics, it is becoming apparent that ABC transporters also efflux a wide range of targeted therapeutic agents. These include the EGFR family-targeted inhibitors gefitinib, erlotinib and lapatinib; the BCR-ABL1-targeted inhibitors nilotinib and imatinib; the VEGFR-targeted inhibitor vandetanib; and others.^102–105 For these targeted agents as for chemotherapy drugs, there is considerable but incomplete overlap in drug specificity among the different ABC family members.¹⁰⁶ The fact that specific groups of transporters limit the amount of active chemotherapy and targeted drugs entering into specific tissues can greatly complicate accurate a priori estimation of the in vivo availability of a specific signaling inhibitor used in combination with a classic chemotherapeutic. Clinical exploitation of these observations is still in early stages.

Growth factor receptors.

An additional factor leading to drug resistance in tumors arises from the nature of many common oncogenic lesions, which activate downstream effectors that limit cell death as they promote cell division. For example, mutational activation or overexpression of receptor tyrosine kinase (RTK) growth factor receptors is common in many kinds of cancer.¹⁰⁷ Taking the ErbB family of receptors as an example, overexpression of ErbB1, encoding the epidermal growth factor receptor (EGFR), induces downstream signaling effectors including STAT3, Ras and its effectors Ral, ERK and PI3K/Akt/mTOR and NF-κB. These proteins and their targets (e.g., BCL-2) act as survival factors inactivating the execution steps of apoptosis. For example, PI3K, Akt and mTOR interactions lead to phosphorylation-dependent inactivation of the mitochondrial apoptosis pathway protein, Bad.¹⁰⁸

Suppression of ErbB1/EGFR prosurvival signaling has been demonstrated in the clinic to reverse resistance to irinotecan in patients with refractory metastatic colorectal cancer.¹⁰⁹ Another ErbB-family receptor, ErbB2/HER2, is amplified in 25–30% of breast cancers and confers highly pro-metastatic, treatment-refractory behavior on these tumors.¹¹⁰ Inhibition of ErbB2 by the use of monoclonal antibody, trastuzumab, in HER-2-overexpressing metastatic breast cancer synergistically enhanced chemotherapy with anthracyclins and cyclophosphamide or paclitaxel,¹¹¹ and markedly improved survival of patients with localized breast cancer.¹¹² Finally, beyond classic chemotherapy, RTK expression and activity also influences tumor cell survival in the context of other treatment modalities, including the genotoxic stresses induced by ionizing radiation. Early work by Balaban et al.¹¹³ that documented the protective effects of EGF on tumor cells, led to the rationale of using drugs to cause EGFR blockade in combination with radiation therapy.^113,114 These approaches (reviewed in ref. 94) produced marked improvements in the cure rates in patients with head and neck cancers treated with definitive radiation and the anti-EGFR antibody cetuximab.¹¹⁵

Epithelial-to-mesenchymal transition (EMT).

Tumor progression and invasion often incorporates a process known as the epithelial-to-mesenchymal transition (EMT). During EMT, epithelial cells lose cell polarity and cell-to-cell contacts and gain motility and invasiveness.^116,117 The EMT program plays an important role during embryonic development and in wound healing; however, its aberrant activation in cancer is associated with acquisition of pro-metastatic and drug-resistant phenotypes,¹¹⁸ and poor clinical outcome.¹¹⁹ On a molecular level, the EMT is accompanied by downregulation of epithelial cytokeratins and cell-to-cell adhesion proteins, including E-cadherin and upregulation of mesenchymal proteins including vimentin, N-cadherin, fibronectin and matrix metalloproteinases (MMPs). The EMT can be triggered by many growth factors and other extracellular stimuli produced either in an autocrine manner by the tumor, or from stromal or immune cells in the tumor microenvironment. Because of its association with resistance to chemotherapy,¹²⁰ angiogenesis,¹²¹ and metastasis,^122,123 the EMT has become a major focus of study, with the goal of limiting autocrine and paracrine processes.

The signaling proteins TGFβ, Wnt/β-catenin, Notch, Hedgehog, RTKs (including EGFR family members¹²⁴ and MET¹²⁵) and Axl,¹²⁶ have the capacity to activate EMT. One key step in the process is the activation of effector transcription factors including Twist, Snail, Slug and ZEB family members,^120,127 which repress E-cadherin and other proteins associated with a differentiated epithelial cell identity.¹²⁸ In vivo, knockdown of Slug markedly reduced metastasis formation with only a slight effect on primary tumor growth.¹²⁹ Suppression of Twist rendered lung carcinoma cells sensitive to chemotherapy.¹³⁰ EMT-associated cell signaling pathways communicate with core apoptosis control pathways. For example, overexpression of Twist upregulates AKT2 to induce resistance to paclitaxel in breast cancer cells.¹³¹ Elevated expression of Snail and Slug causes resistance to DNA-damaging agents via downregulation of apoptosis-related genes, including p53.¹³²

Stem cells.

In development, stem cells occupy a protected niche, and are typically drug-resistant.¹³³ One provocative idea emerging to explain the link between EMT and drug resistance is based on the observation that cancer cells in the mesenchymal state are imparted with many properties of stem cells (reviewed in ref. 134). Although “cancer stem cells” have been reported in solid tumors of the breast, colon and brain,^135–137 the idea of cancer stem cells (CSC) remains somewhat controversial (reviewed in ref. 138), as does the question of whether such cells represent a discrete, irreplaceable cell population, or whether factors including microenvironmental niche and intrinsic cellular heterogeneity can induce both an EMT transition and stem-like properties, including drug resistance.^139,140 However, given the growing evidence that CSC exist¹⁴¹ and possess intrinsic resistance to chemotherapy compared to more differentiated cells,¹⁴² better understanding of the relation between essential EMT and CSC signaling cascades is a high priority research need in order to improve therapeutic responses.

Gupta et al.¹⁴³ examined the selectivity of cytotoxic agents in the context of mesenchymal versus epithelial state of the tumorigenic breast epithelial cell line. Regulated expression of E-cadherin and acquisition of the migratory stem cell-like phenotype was associated with loss of tight junctions and nuclear liberation of beta-catenin and accompanied by a >100-fold increase in resistance to paclitaxel and doxorubicin. By contrast, the coccidiostatic agent, salinomycin, was selectively cytotoxic to the stem cell progenitors and counteracted development of metastases in vivo. This work illustrates that conventionally developed chemotherapy agents have selective effects on the non-stem cell component of a mixed tumor cell population, sparing the most aggressive cancer cells. In this context, it is tempting to speculate that adjuvant chemotherapy should be specifically designed to selectively target the resting clonogenic population of cancer cells. Previous failures in attempts to directly extrapolate effective metastatic chemotherapy regimens^144,145 to an adjuvant setting should be critically reviewed through the prism of the activity of such therapies towards the mesenchymal (migratory) sub-population of cancer cells.

Although drug transporters, RTKs and stem cells are each discussed separately here, in practice the boundaries between these resistance methods are likely fluid. For example, one of the most intriguing connections between specific ABC transporters and drug resistance is their recently described association with stem cells.¹⁴⁶ A so-called “side population” (SP) enriched for stem cells is easily identifiable by flow cytometry in clinical samples and in cell lines.^146,147 SP cells are characterized by their accelerated efflux of Hoechst 33342 and Rhodamine 123 (Rh123) fluorescent dyes. Several drug efflux pumps such as P-GP/MDR1/ABCB1 and BCRP/ABCG2, are highly expressed by these SP cells. Suggestively, expression of BCRP/ABCG2 is directly required for the maintenance of the SP in the breast cancer cell line MCF7, although the reasons for this are as yet unclear.¹⁴⁸ Similarly, the BCRP/ABCG2 inhibitor, fumitremorgin C, selectively affected the cancer stem cell population in an NSCLC cell line.¹⁴⁷ Interestingly, the ErbB1 inhibitor, erlotinib, inhibits BCRP/ABCG2-mediated resistance to flavopiridol, SN38 and mitoxantrone, raising the possibility of a potential role of EGFR signaling in regulating BCRP/ABCG2 activities related to drug resistance in stem cells.¹⁴⁹ Thus, overexpression of ErbB proteins in many solid tumors may be accompanied by hyperactivation of ABC transporters in the context of stem cells, simultaneously modifying drug exposure, activating survival pathways and altering core cell identity.

As a further example of the integration of resistance mechanisms, activation of the developmentally important Hedgehog pathway in cancer is common, and may be linked to an increase in the stem cell compartment, based on defined roles for Hedgehog in controlling cell differentiation.¹⁵⁰ Upregulation of Hedgehog signaling has been shown to promote multidrug resistance by increasing the expression of P-GP/MDR1/ABCB1 and BCRP/ABCG2 in vitro, causing increased resistance to docetaxel, methotrexate and VP-16.¹⁵¹ Furthermore, Hedgehog pathway inhibitors such as cyclopamine reduce expression levels of P-GP/MDR1/ABCB1 and BCRP/ABCG2, providing another interesting link between efflux pumps and stem cell drug resistance mechanisms.¹⁵¹ The intriguing implication of these findings is that effective targeting of tumor-specific ABC-transporters or the mechanisms that drive their expression (EMT or acquisition of a stem cell phenotype) may lead to significant clinical benefit.

Systems Biology Approaches Towards Improving the Efficacy of Chemotherapy

The pathways to drug resistance summarized above were elucidated over decades of study, based on the analysis of genes overexpressed in tumor cell lines selected for drug resistance, or based on targeted evaluation of known oncogenes, tumor suppressors and their effectors for roles in response to drugs. An alternative strategy that has become a powerful source of insight into new resistance mechanisms has been the integration of RNA interference (RNAi) screens with other genomic resources. In contrast to other high-throughput techniques such as gene expression profiling, which produce correlative but not causative associations,¹⁵² use of RNAi screens enables researchers to directly establish novel drug-gene interactions and identify new targets for cancer therapy. RNAi screens have begun to elucidate the role of general survival genes in the drug resistance, and to probe functional genetic interactions with specific chemotherapy agents (e.g., cisplatin, gemcitabine and paclitaxel).

The concept of synthetic lethal genetic interactions (in which a lethal phenotype is only obtained when two independent genes are simultaneously inactivated) was developed in studies of lower eukaryotic models such as yeast.¹⁵³ A growing number of studies use “synthetic lethal” RNAi screens (i.e., in which a siRNA selectively reduces cell growth when co-administered with a targeted drug of interest) to identify previously unidentified components of the drug resistance machinery active in cancer cells. The initial reported siRNA screen seeking to improve therapeutic usage was performed with a library specifically targeting kinases and phosphatases.¹⁵⁴ Surprisingly, 11% of the assayed kinases and a full 32% of the phosphatases were found to be essential for intrinsic cell survival in the absence of treatment. Depletion of a subset of the kinases (SGK, CDK6, CDK8, FER, JIK and PLK2) additionally enhanced apoptosis in the presence of paclitaxel, cisplatin or etoposide. Similarly, a limited number of phosphatases, including MK-STYX, PPP3CB, ACP6, PPP4R1L and a number of dual specificity phosphatases (DUSPs) specifically were required for chemotherapy resistance.¹⁵⁴ Although the mechanism of drug selective action is not yet understood, empirically, inhibitors of these kinases may be broadly advantageous for chemotherapy, assuming a suitable selective window of activity in tumor versus normal cells. Although phosphatases have been less exploited as therapeutic targets, some specific phosphatase inhibitors (e.g., (E)-2-benzylidene-3-(cyclohexylamino)-2,3-dihydro-1H-inden-1-one (BCI), NSC357756, NSC45382 and NSC295642 targeting DUSP6 and NU-126 targeting DUSP1) have demonstrated activity and acceptable specificity in vivo,^155,156 and may yield valuable drugs in the future. A subsequent genomewide screen¹⁵⁷ demonstrated that disruptions in either the BRCA pathway or RAD6/RAD18 DNA repair genes result in enhanced cisplatin cytotoxicity specifically in the presence of p53 deficiency in tumor cells.

Studies of synthetic lethal interactions in lower eukaryotes have established that these interactions commonly involve genes that operate in closely related and/or partially redundant pathways, or target sequential steps of a signaling cascade.¹⁵⁸ Consistent with these findings, genes conferring synthetic lethalilty to paclitaxel in a genome-wide screen for paclitaxel sensitizers¹⁵⁹ have been identified in functional modules related to microtubules, the direct target of the screen drug. These included microtubule-associated proteins regulating critical cell cycle transitions in mitosis such as tubulin subunits TUBGCP2 and TUBA8, dynein subunits DNAH10 and 1 and MAGUK family member, MPP7. An independent kinome-wide RNAi screen¹⁶⁰ similarly showed a notable enrichment for mitotic checkpoint regulators among the genes important for paclitaxel, again consistent with the principle that chemotherapy-modifying targets frequently are found among genes with related functions. Cementing the link to the mitotic checkpoint apparatus, depletion of paclitaxel-antagonizing RNAi targets produced polyploidy, multinucleation, centrosomal abnormalities and increased chromosomal instability in the absence of the drug.¹⁶⁰

An additional subset of the hits found by Whitehurst et al. included a number of components of the proteasome.¹⁵⁹ The proteasome is a critical regulator of mitosis, with activity required during key transitions (reviewed in ref. 161). However, in this case, the situation is more complicated. In addition to its roles in mitosis, the proteasome is an important effector of the core apoptosis and checkpoint pathways described earlier. For example, the proteasome positively regulates NF-κB survival signaling by degrading the NF-κB inhibitor IκB, an important target of proteasome antagonists such as bortezomib, in multiple myeloma and mantle cell lymphoma;¹⁶² although this does not fully explain the clinical activity of bortezomib on malignant myeloma cells.¹⁶³ Compatible with the screen findings, preclinical studies have shown that bortezomib enhances the activities of paclitaxel, gemcitabine and CPT11, based on actions including via downregulation of Bcl-2, NF-κB activation and stabilization of p21, p27 and p53.^164–166 The idea of chemotherapy-sensitization with bortezomib is now being tested in clinical trials in myeloma, AML and solid tumors (www.clinicaltrials.gov, NCT00516100).^167–169 Independently, among the paclitaxel-sensitizing hits identified by Swanton et al. COL4A3BP/CERT is a regulator of ceramide metabolism. COL4A3BP was proposed to act by augmenting the ER stress induced by paclitaxel, accompanied by enhancing Akt inactivation;¹⁶⁰ another connection to core cell survival machinery. Because of such pleiotropic actions, it is difficult to assign positive empirical results to discrete signaling pathways, as two or more are likely relevant.

Most siRNA screens performed to date have employed either the full genome or the kinome/phosphatome. Beyond the examples described above, noteworthy screens include identification of genes, which, when depleted, sensitize cancer cells to gemcitabine,¹⁷⁰ a PARP inhibitor KU0058948,¹⁷¹ or modulate activity of an anti-estrogen tamoxifen.¹⁷² As an alternative approach, we have begun to exploit another paradigm of cell signaling established largely in lower eukaryotic genetic models. This signaling model (reviewed in ref. 173) replaces the idea of linear or branching signaling pathways with the idea of a densely interconnected network of proteins, dispensing with the idea of “borders” for canonical signaling pathways. In such a model, pleiotropic actions are expected, because critical proteins interact with multiple neighbors to coordinately control processes such as drug response (reviewed in ref. 174). This shift from an idea of constrained pathways to full connectivity has practical corollaries for screening in the context of synthetic lethal analysisbased siRNA screening. One strong prediction is that the pool of proteins functioning “close” to an initial target protein in a network is likely to be enriched for proteins able to compensate for impaired function of the target: but that such close functioning proteins may extend well beyond the normal list of proteins in the target's “pathway.” If so, focused RNAi libraries for synthetic lethality screens, constructed using bioinformatics-guided selection of candidate genes centered on the putative target of the screen drug, may have a number of advantages over genome-wide or functionally constrained (e.g., kinome/phosphatome) libraries in regard to both the thoroughness and the cost of the screen, eliminating the batch-to-batch variations and prohibitively high number of false positives inherent to the genome-wide screens.¹⁷⁵

We adapted this approach to map elements conferring resistance to EGFR antagonists and chemotherapy agents. Constructed through mining of the multiple databases, our network-based library consisted of ∼600 genes with strongest evidence of functional interactions with EGFR, identified by analysis of protein-protein interactions, conventional pathway maps, transcriptional response to EGFR inhibition/stimulation and genetic interactions in model organisms. Screening this library identified resistance genes to both EGFR-targeting and standard chemotherapy, and confirmed the initiating hypothesis that hits would cluster among proteins highly connected to the drug target, EGFR.¹⁷⁶ Among these hits, NEDD9, BCAR1 and SH2D3C were highly effective sensitizers that all physically interact with each other, predicting an important core in the drug resistance network that was active in multiple cell lines. Extrapolation from these results allowed the prediction of new drug-drug synergies, between drugs targeting the NEDD9-interacting AURKA¹⁷⁷ and those targeting EGFR.

Although the initial results of these variant screens have great potential to yield new insights into resistance mechanisms, the field of RNAi-based screening is still in its early stages. In a comparative analysis of the results of the seminal screens described above, at first glance we found relatively little overlap between the hits from most of the sensitization screens discussed above (Fig. 2A), even in some cases when different research groups used the same drugs as screening agents. This could be explained based on technological reasons (e.g., differences in the readout and hit selection criteria used), or based on biological reasons (e.g., if there is a considerable difference in which genes are sensitizing in different tumor cell lines). Future work resolving this issue will be important, given the different implications of the two different interpretations for the clinical application of these findings. Intriguingly, in spite of the limited overlap among the specific genes identified as drug-sensitizing primary hits, there is evidence that RNAi screens are already revealing conserved core mechanisms of sensitizing activity.

Figure 2.

Analysis of RNAi screens intended to identify chemotherapy-potentiating genes. (A) A schematic representation of published seven RNAi sensitizing screens showed only slight overlap between the hits. Multiple synthetic lethal hits ATR (from three screens) and CHEK1 (from four screens) are represented by asterisk and pound signs, respectively. See text for details. (B) A protein-protein interaction network was assembled using hits belonging to the kinome (red nodes) as seeds to retrieve their first degree (direct) interactors from a variety of databases using MiMi plugin¹⁹⁸ to Cytoscape.¹⁹⁹ Hits do not cluster in any particular region of the network, nor have they significantly denser connections to each other. However, many hits are connected to cell cycle-related proteins (based on GO functions, see also below). Cell cycle regulators are shown in light blue, edges connecting hits and between hits and defined cell cycle regulators are shown in dark blue. Hits that are themselves cell cycle regulators are depicted in purple. (C) Topology and gene ontology function analysis of the assembled network. Left, degree and betweenness centrality is greater for the hits in comparison to the remainder of the kinome. Right, GO comparison between the set of screen hits and their nearest neighbors vs. the average of several similarly expanded sets of randomly selected kinases. Biological process categories with the biggest difference in p values (where the difference was >ten orders of magnitude between the hits and the kinases in any of the random sets) all represent cell cycle/mitosis (5 top categories shown), whereas the programmed cell death category (right column) did not produce a significant difference in p values. Error bars: 95% confidence interval. (D) A close-up expansion of (B) illustrates the dense protein-protein interactions between the hits (red) and the cell cycle genes (light blue). Hits that are themselves cell cycle regulators themselves are depicted in the purple inner circle.

We have explored the functional relatedness of the drug-sensitizing genes by constructing a protein interaction network between the hits identified in each of the screens involving or including the kinome (Table 1). Protein-protein interactions within the kinome, as well as the kinases' nearest neighbors were retrieved from the multiple databases using the MiMi plugin for Cytoscape, resulting in an extended network with 5,977 nodes (Fig. 2B). Suggestively, the topology of the hits revealed a more central position in the network compared to the remainder of kinases in the library, as measured by the “degree” (reflecting the number of neighbors) and “betweenness centrality” (which reflects the amount of control that this node exerts over the interactions of other nodes in the network¹⁷⁸ (Fig. 2C, left).

Table 1.

RNAi drug-sensitizing screens used for the network analysis

Drug	Cell line	Criteria for hits	Number of hits	Description of screen (siRNA library)	Reference
Cisplatin, gemcitabine, paclitaxel	HeLa (cervical)	mean log2 (% viability plus drug/% viability no drug) − 2 SD.	37 hits with cisplatin, 24 hits with gemcitabine, 40 hits with paclitaxel	2,400 genes with each drug; genome-wide screen targeting 22,887 genes with Cisplatin	Bartz, et al. 157
Paclitaxel	NCI-H1155 (lung cancer)	5% FDR (false discovery rate) and the 2.5-centile rank of the viability ratios	87 hits	Genome-wide screen targeting 21,127 genes	Whitehurst, et al. 159
Paclitaxel, 5-FU, doxorubicin, cisplatin	HCT-116 (colon), A549 (lung), MDA-MB-231 (breast cancer)	SI > 0.1 for sensitizing hits, SI < −0.15 for antagonistic drugs	43 hits with paclitaxel antagonistic in 2 or more cell lines; 30 sensitizing hits in 2 or more cell lines/drugs	Kinome screen targeting 779 kinases	Swanton, et al. 160
Tamoxifen	MCF7 (breast cancer)	Z-score >3	20 resistance hits	Kinome screen targeting 779 kinases	Iorns, et al. 172
PARP inhibitor KU0058948	CAL51 (breast cancer)	Z-score <−3	24 hits for Z-score <−3, 57 hits for Z-score <−2	Kinome screen targeting 779 kinases	Turner, et al. 171
Gemcitabine	MIAPaCa-2 (pancreatic)	Log2 ratios of viability siRNA + gemcitabine vs. siRNA + vehicle were 1.65 SD below the mean ratio level	25 hits at 5 nM and 62 hits at 10 nM	Kinome screen targeting 572 kinases	Azorsa, et al. 170

Of further interest, gene ontology (GO) analysis did not reveal striking differences between subsets of hits and random sets, with only an insignificant enrichment in GO functions related to mitotic cell cycle spindle checkpoint. However, these differences became highly significant when we included in our analysis the nearest neighbors of the initial subsets. Surprisingly, no enrichment was observed in apoptosis-related functions (Fig. 2C, right), suggesting that synthetic lethality is more likely to be found among genes directly interacting with the cell cycle-relevant target of the drug, i.e., the DNA. Either a global network overview (Fig. 2B) or a close-up (Fig. 2D) of the drug-sensitizing hits illustrates a high degree of connectivity with genes involved in cell cycle and mitosis regulation. Functional clustering of the chemosensitizing hits using the Ingenuity algorithm identified a number of functional modules of tightly interacting genes (Fig. 3). Consistent with the concept of synthetic lethality stated above, the hits were enriched for well-characterized regulators of cell cycle and DNA repair which are in direct proximity to the chemotherapy targets (DNA, cell cycle). As an example (Fig. 3A), a module of the essential DNA damage response regulators such as BRCA1, ATM/ATR, CHEK1 and CHEK2 kinases, produced multiple hits. A number of inhibitors of these genes are presently being tested in clinical trials: PF-477736, an inhibitor of CHEK1; MK1775, an antagonist of WEE1; and BMS-387032, an antagonist of CDK2 (www.clinicaltrials.gov, NCT01047007, NCT00648648).^179,180 Notably, several elements of the cyclin-dependent protein kinase complex (CDK1, CDK7, CDKN3, CDKN2C) were antagonizing, which suggests that cell cycle arrest following the effect of DNA-damaging agents may be protective and lead to cancer survival.

Figure 3.

(A) A network of hits highly enriched in cell cycle-related genes. Solid lines, physical interactions; dashed lines, indirect interactions. (B) A network of physically interacting (solid lines) hits clustered around protein kinase A and calmodulin. ER K1/2 is drawn as a central component based on numerous indirect links between constituents of these pathways and ER K activation. (C) An example of a cluster identifying a new target candidate, the transcription factor HNF4A, a known master regulator of multiple targets relevant to tumor aggressiveness, as a hub connected to numerous resistance genes. Although this gene has not itself been found among the hits in any analyzed screens, it can directly transcriptionally regulate multiple sensitizing hits from different screens, and may be worth exploration as a possible therapeutic target. Pathway-centered network modules clustered around subsets of hits pooled from different screens were assembled using Ingenuity Pathway Analysis Software, then redrawn for image clarity. Protein complexes retrieved by Ingenuity are colored according to the status of the constituents: sensitizing hits are shown in red and antagonistic hits in green. Functionally related non-hit proteins have been provided to connect the hit network, and are indicated in white.

Consistent with their established role in regulating cancer cell proliferation and survival,¹⁸¹ a number of chemosensitizing hits formed modules of proteins densely interacting with protein kinase A, S6K and calmodulin (Fig. 3B). The activity of S6K is directly controlled by mTOR; multiple clinical trials are under way to test inhibitors of this pathway in combination with chemotherapy. Subunits of protein kinase A, a tetrameric cAMP-dependent kinase, were identified in multiple screens as chemoresistance genes. One of the reasons for the repeated hits among these proteins is likely linked to their role in positively regulating NF-κB signaling via phosphorylation of p65.¹⁸² Three hits represent A-kinase anchoring proteins (AKAPs), directing cAMP-dependent protein kinase activity towards a range of cellular targets.¹⁸³ Another hub in this network includes a group of calcium-binding protein family members including calmodulin and a cluster of interacting kinases that are responsible for resistance to chemotherapy via multiple interactions including with the apoptosis-executing caspases¹⁸⁴ and death-associated protein kinases (DAPK).¹⁸⁵ Interestingly, calmodulin has been shown to be an important target in influencing multi-drug resistance, based on its activity in influencing a number of relevant cellular processes, including microtubule dynamics (reviewed in ref. 186). A third sub-network is organized around the master transcriptional regulator HNF4A (Fig. 3C), which controls the expression of a number of chemotherapy sensitizing targets, including its co-activator, HNF4G,¹⁸⁷ and physically associates with a large number of sensitizing hits. Pharmacological¹⁸⁸ or siRNA inactivation of these central transcriptional regulators may prove to be a highly efficient strategy to undermine cancer cell resistance to commonly used anti-neoplastic agents.

Summary

The field of cancer treatment is in active transition to the knowledge-based design of therapies. The combination of chemotherapy with targeted agents blocking critical hubs in the resistance machinery promises to improve the selectivity, efficacy and personalization of drug selection based on the specific genetic makeup and vulnerabilities of each individual's cancer. Initial candidate-driven steps in this direction are now being enhanced by insights emerging from unbiased screens. For example, activating mutations in Ras are commonly seen in human carcinomas.¹⁸⁹ Hyperactivity of the Ras oncogene has been demonstrated to produce increased demands for glucose uptake and increased glycolysis: pharmacological targeting of this process specifically suppressed the growth of K-Ras mutant cancers.¹⁹⁰ K-Ras mutant cancers are also specifically dependent on the non-canonical IkappaB kinase TBK1, offering another target for inhibition.¹⁹¹ Excitingly, siRNA screening approaches have also been successful in identifying specific dependencies of K-Ras mutated cell lines on STK33 kinases.¹⁹² En masse and extended to other cancer-associated lesions, these findings offer multiple conceptually novel approaches for combination therapies that may bear fruit in a clinical setting.

Integrative systems biology approaches have only begun to demonstrate their power in the basic science realm over the past few years.¹⁹³ In this work, mRNA profiling, protein-protein interaction and genetic screens are combined to establish complex signaling networks that govern cellular processes. Given the complexity of intragenic interactions, and the very divergent platforms required for their analysis, this data cannot yet be fully exploited for clinical benefit, until very large-scale datasets become better integrated. For example, a recent study by Yeger-Lotem et al. developed computational algorithms to better connect genetic hits obtained by screening with existing transcriptome data.¹⁹⁴ Other work has used computational methods to mine transcriptional profiling data, revealing that a small number of master transcriptional factors, such as C/EBP and STAT3, are consistently activated in malignant tumors and that modulation of the activity of these factors was sufficient to regulate mesenchymal transition and invasive behavior of the high grade brain gliomas.¹⁹⁵ Another recent study has used prior knowledge about transcription factor-promoter target relationships to analyze microarrays of primary tumor specimens to identify signatures of expression change predictive of treatment response.¹⁹⁶ Such inferential methods, which combine robust existing knowledge databases and ongoing screening to build and test networks, should help define the full set of signaling changes critical for tumor phenotype.¹⁹⁷

In conclusion, we are already at the point at which fundamental knowledge of signaling pathways relevant to DNA and microtubule damage responses allows us to begin to predict how to modify those responses selectively within tumor cells. This knowledge is continuing to deepen at a rapid pace and is currently providing a rich source for nominations of proteins to target therapeutically. Looking forward, new issues begin to emerge. Clearly, multiple classes of proteins—growth factor receptors, ABC transporters, lineage control factors, and others—interact in a complex network, with each contributing to some degree to chemotherapy resistance. In this context, it would be desirable to combine two or even three targeted agents in a therapeutic regimen designed to simultaneously incapacitate multiple resistance-promoting factors. Towards this end, issues that move to the fore include identifying ways to provide incentive for pharmaceutical companies and biotechnology companies to allow their proprietary targeted agents to be used in combinations, and modifying the trial process to more readily allow the assessment of complex combinations of targeted inhibitors. The financial and regulatory changes required for this will take some time to fully develop. If all goes well, application of designer cocktails of targeted agents as adjuvants to low dose chemotherapies may ultimately yield a potent tumor-focused response and minimal side-toxicity, greatly reducing the burden of cancer.