Cancer Systems Biology: Embracing Complexity to Develop Better Anticancer Therapeutic Strategies

REVIEW Cancer systems biology: embracing complexity to develop better anticancer therapeutic strategies

W Du and O Elemento

The transformation of normal cells into cancer cells and maintenance of the malignant state and phenotypes are associated with genetic and epigenetic deregulations, altered cellular signaling responses and aberrant interactions with the microenvironment. These alterations are constantly evolving as tumor cells face changing selective pressures induced by the cells themselves, the microenvironment and drug treatments. Tumors are also complex ecosystems where different, sometime heterogeneous, subclonal tumor populations and a variety of nontumor cells coexist in a constantly evolving manner. The interactions between molecules and between cells that arise as a result of these alterations and ecosystems are even more complex. The cancer research community is increasingly embracing this complexity and adopting a combination of systems biology methods and integrated analyses to understand and predictively model the activity of cancer cells. Systems biology approaches are helping to understand the mechanisms of tumor progression and design more effective cancer therapies. These approaches work in tandem with rapid technological advancements that enable data acquisition on a broader scale, with finer accuracy, higher dimensionality and higher throughput than ever. Using such data, computational and mathematical models help identify key deregulated functions and processes, establish predictive biomarkers and optimize therapeutic strategies. Moving forward, implementing patient-specific computational and mathematical models of cancer will significantly improve the specificity and efficacy of targeted therapy, and will accelerate the adoption of personalized and precision cancer medicine.

Oncogene (2015) 34, 3215–3225; doi:10.1038/onc.2014.291; published online 15 September 2014

INTRODUCTION approaches, the clinical failure rate of novel anticancer therapeutic 7 As a disease caused by corruption of normal biological circuits and molecules has been high; moreover, resistance to targeted processes to sustain uncontrolled proliferative growth, cancer is anticancer therapies is a ubiquitous theme, often occurring almost always characterized by a complex spectrum of alterations through a variety of mechanisms, many of which are linked to 8–10 that affect multiple scales ranging from molecular activity within complex signaling pathways and networks. It is our view that cells onto communication between cells and tissues (Figure 1). systems biology approaches that can manage and model this At each of these scales, highly complex patterns of interactions are complexity will be increasingly needed to identify and validate observed and this complexity leads to great challenges in targets, biomarkers and discover more effective and less toxic understanding cancer progression and designing effective cancer therapeutic strategies. Such approaches will complement and therapy. At the genomic level, frequent disruption of the DNA extend more traditional reductionist approaches. maintenance machinery1 and epigenetic modifiers2 result in We define systems biology as the study of complex interactions thousands of sequence alterations3 and global epigenetic in biological systems and the emergent properties that arise from reprogramming,4 greatly complicating the discovery of underlying such interactions. In the field of cancer, systems biology aims at drivers of tumor progression. At the protein level, the frequent developing an increasingly holistic view of cancer development involvement of complex signaling networks makes it difficult to and progression. Systems biology approaches help understand anticipate the influences of oncogenic perturbations and to how complex cancer-associated deregulations coordinately shape predict how to effectively reverse those influences with pharma- malignant states and phenotypes, sometimes across multiple cological agents. At the tissue level, interactions between a tumor biological scales. In parallel, rapid technological advances now and its local environment consisting of a multitude of cell types enable high-throughput and systematic profiling of cancer that dynamically coevolve critically influence tumor growth and genomes, transcriptomes, proteomes, metabolomes and of the invasion.5 The complexity of the networks involved, the large tumor microenvironment. Together with clinical information, number of cells and cell types, the patterns of coordination these data are critical for the establishment of integrated and between molecules, cells and tissues and the constantly changing predictive cancer models. Even though computational and and evolving nature of the disease dictated by a Darwinian mathematical approaches used in systems biology are highly process6 cannot be fully addressed by traditional reductionist versatile, a few categories of general methodologies have approaches. Perhaps as a result of relying solely on reductionist emerged for specific purposes in cancer research. One class is

Laboratory of Cancer Systems Biology, Sandra and Edward Meyer Cancer Center, Department of Physiology and Biophysics, Institute for Computational Biomedicine and Institute for Precision Medicine, Weill Cornell Medical College, New York, NY, USA. Correspondence: Dr O Elemento, Laboratory of Cancer Systems Biology, Sandra and Edward Meyer Cancer Center, Department of Physiology and Biophysics, Institute for Computational Biomedicine and Institute for Precision Medicine, Weill Cornell Medical College, 1305 York Avenue, New York, NY 10021, USA. E-mail: [email protected] Received 6 June 2014; revised 11 August 2014; accepted 11 August 2014; published online 15 September 2014 Cancer systems biology W Du and O Elemento 3216

Figure 1. Cancer alterations that affect complex molecular interactions across multiple biological scales. Within a cancer cell, thousands of genetic and epigenetic alterations result in aberrant protein expression or expression of proteins with abnormal functions. Consequently, the homeostasis of cellular signaling networks is frequently disrupted, leading to uncontrolled activation of survival and proliferation factors that drive tumor progression. At the tissue level, cancer cells closely interact with multiple cell types in the local environment. Through these interactions, the tumor microenvironment is educated by cancer cells to promote tumor growth and invasion. In this ﬁgure, we use activated B-cell-like diffuse large B-cell lymphoma (ABC-DLBCL) featuring aberrant B-cell receptor (BCR) signaling as an example to illustrate cancer alterations across multiple biological scales.

integrative statistical analysis of large-scale cancer multi-omics proved important in profiling the activity and states of cancer cells and clinical data. These unbiased data-driven analyses and cell populations. Finally, we discuss how cancer systems have identified key biological processes underlying cancer biology will facilitate the adoption and implementation of pathogenesis,11 prognostic biomarkers12 and predictive signatures personalized and precision cancer medicine. for drug response.13,14 Another class is mathematical modeling of interaction networks such as intracellular signaling pathways or extracellular crosstalks between tumor and the microenvironment. PROBING CANCER COMPLEXITY AT THE GENOMIC LEVEL These models have proved useful at unraveling mechanisms of Tumors are characterized by a broad array of genetic and drug resistance and in optimizing combinatorial targeted epigenetic disruptions 15–17 therapy. Furthermore, evolutionary models that simulate Cancer is a disease driven by the accumulation of genomic and tumor growth and progression have provided important insights epigenomic alterations. Genomic alterations such as point into the evolution dynamics of tumor and have led to the 18,19 mutations, translocations and copy number variations result in discoveries of more effective dosing schedules. Overall, the aberrant gene expression or expression of mutated genes with application of systems biology approaches have led to substantial abnormal function.3 Genomic abnormalities are frequently accom- improvements in our understanding of cancer initiation and panied by epigenomic alterations,4,20,21 for example, hyper- or progression and to the discovery and implementation of more hypomethylation of specific regions and genes or changes in effective anticancer therapeutic strategies (Figure 2). levels of histone modifications.22,23 Like genomic alterations, As observed in many cancer types, substantial heterogeneity of epigenomic alterations are thought to affect expression levels of molecular alterations in patient population yields highly variable certain genes.4 It is generally thought that epigenetic alterations clinical responses to the same treatment. Consequently, persona- are brought about by mutations or aberrant expression of lized and precision cancer medicine, in which treatment of chromatin-modifying enzymes, a hypothesis supported by the fi patients is selected according to their speci c molecular dereg- ubiquitous presence of mutations in epigenetic enzymes across ulations, is urgently needed to improve therapeutic specificity and nearly all tumor types.2 An alternative model is that epigenetic efficacy. We believe that systems biology approaches that alterations occur by selection of specific epigenetic patterns in an integrate multidimensional data into predictive, patient-specific initially heterogeneous cell population.24 Although a cancer cell cancer models will also play a central role in the implementation can harbor thousands of somatic genetic and epigenetic of personalized and precision cancer medicine. alterations, only a few are ‘drivers’ that confer selective growth In this review, we sought to review our current knowledge of advantage and are thus causal for malignant transformation and cancer complexity at specific levels—genomic, signaling and tumor progression. The rest are ‘passengers’ that accumulate in tissue—and summarize recent studies that adopted systems the course of cell divisions and contribute little to malignant biology approaches to help address such complexity. In each phenotypes.3 A major effort in cancer genomics has been to section, we also provide a brief overview of technologies that have identify ‘driver’ alterations, as these pathogenic disruptions are

Figure 2. A summary of cancer systems biology approaches to guide cancer therapy. Systematic measurements are crucial for implementing cancer systems biology approaches. Integrated and predictive cancer models established using these approaches are iteratively trained and validated with experimental observations. These models could assist various areas in cancer therapeutics, such as predicting prognosis and optimizing effective combinatorial therapy. likely to be effective drug targets or biomarkers.25 Of note, Cancer cells also display extensive intratumoral and intertu- systems-level analyses have made creative use of passenger moral heterogeneity.3,34 Although the majority of tumor cells may mutations to uncover potential therapeutic strategies. For contain a limited number of ‘driver’ alterations acquired in the example, Muller et al.27 showed that specific passenger mutations early stage of tumorigenesis, continuous cell division can keep may effectively expose weaknesses via synthetic lethal producing new mutations in individual progenies. Consequently, interaction26 and thus create cancer-specific, targetable at the genomic level, tumors are often not homogeneous and vulnerabilities.27 Specifically, they showed that glioblastomas with frequently display complex clonal architectures. This is illustrated passenger deletion of glycolytic gene enolase 1 (ENO1), which by observations in solid tumors that spatially separated carcinoma encodes an essential enzyme in glycolysis, are sensitive to cells within a single tumor exhibited distinct subclonal driver inhibition of the functionally redundant homolog ENO2 that mutations and gene expression patterns.35–38 Similar genetic compensates for ENO1 loss in the tumors. On a more theoretical diversity was also reported in leukemia based on single-cell note, evolutionary modeling demonstrated that tumor regression mutation analysis by in situ hybridization, where up to 10 could be induced by increasing the collective burden of genetically heterogeneous subclones were observed in the moderately deleterious passenger mutations by either elevating majority of samples analyzed.39 The existence of intratumoral mutation rates or exaggerating the deleterious effects of these heterogeneity might be of functional importance, as ecological mutations. The latter in particular might be therapeutically interactions between various subclones can affect global fitness achievable by targeting molecular machineries that alleviate and thus influence cancer progression.40 This is illustrated by the disruptions of protein structure or abundance such as chaperones recent finding that the maintenance of Wnt-driven breast cancer 28 or proteases. requires cooperation of two genetically diverse subclones. Although the HRAS-mutated subclones are essential for tumor Complexity of the cancer genome: unstable, heterogeneous and progression, their proliferation critically relies on signaling evolving molecule Wnt1 secreted by the HRAS-wild-type subclones.41 The development and progression of cancer is a dynamic interplay Intratumor heterogeneity can further seed intertumoral hetero- between randomly occurring genomic alterations and selection- geneity among different metastases, if different metastases arise —hallmarks of a Darwinian process.6,29 The rate of genomic from distinct founder cells that have escaped from the primary alteration occurrence plays a central role in this evolutionary tumor. Comparative analysis between metastases in pancreatic process. In normal cells the rate of mutation and chromosomal cancer supports this notion, in that genotype diversification was abnormalities is kept low by reliable DNA damage response already observed in the initiating cells of different metastases.42 pathways. In cancer cells, oncogene-induced DNA replication Intratumoral heterogeneity is of great clinical importance as stress and/or direct disruptions of DNA repair genes frequently subclonal drivers have been found to embed sources of drug – occur, leading to widespread genomic instability.1 For example, resistance.43 45 One such example was recently found in multiple p53, an essential protein for DNA damage response, is frequently myeloma, where BRAF inhibitors killed BRAF-mutated cells but mutated in cancer across many tumor types.30 Moreover, were ineffective against subclones harboring KRAS or NRAS environmental insults, such as toxic smoke,31 ultraviolet light32 mutation.43 Taking a Darwinian point of view, therapeutic and other mutagens,33 frequently contribute to increased DNA intervention can act as an imposed selective pressure that damage observed in tumors. As a result, the cancer genome is potently and rapidly alters subclonal constitution.46 Specifically, always significantly more chaotic than the genome of normal cells. targeting the predominant genotype may conversely promote

© 2015 Macmillan Publishers Limited Oncogene (2015) 3215 – 3225 Cancer systems biology W Du and O Elemento 3218 expansion of resistant subclones47 that remains a major obstacle more frequently used to uncover pathogenic and potentially in curing cancer.45,48 Therefore, a more systematic understanding targetable processes in tumors. For example, the HotNet algorithm of clonal structure and the evolutionary dynamics of cancer cells is developed by Vandin et al.72 evaluates the influence of mutations required to optimize therapeutic effect at the subpopulation level. in genome-wide gene interaction networks to identify subnet- We discuss below how systems biology approaches can help works that are significantly disrupted in a group of patients. achieve this. Another algorithm called PARADIGM can infer activities of curated pathways in cancer patients by incorporating multiple data types including mutation, copy number variation, mRNA and protein Interrogating the cancer genome using next-generation 73 sequencing expression levels. Application of these algorithms to the analysis of the Cancer Genome Atlas (TCGA) clear cell renal carcinoma data The complex and evolving tumor architecture and subclonal revealed chromatin remodeling, glycolytic shift and phosphatidy- structure are frequently investigated using systematic genome linositol-4,5-bisphosphate 3-kinase (PI3K)/AKT signaling as key profiling technologies. Next-generation sequencing provides ultra- pathogenic processes.11 fast, high-throughput and affordable sequencing via massively Integrated, systems-level analysis across multiple tumor types parallel assessment of sequence fragments.49 Whole-genome has provided insights into common mechanisms underlying sequencing of tumor samples with paired normal tissue is now cancer pathogenesis. For example, a systematic investigation of routinely used to analyze the entire complement of somatic gene expression changes in metabolic network across 22 tumor coding region mutations, chromosomal rearrangements, translo- – types combining various statistical approaches revealed a number cations and noncoding mutations in a tumor cell population.50 52 of metabolic pathways frequently perturbed across tumors, such Whole-exome sequencing provides a more cost-effective and as glycolysis and oxidative phosphorylation. This analysis also higher coverage alternative to whole-genome sequencing and discovered significant alteration in hundreds of isoenzymes that detects aberrations that directly alter protein functions with high may be potential therapeutic targets.74 In a TCGA pan-cancer sensitivity.53,54 Smaller targeted panels covering a dozen to a few analysis, genetic and epigenetic profiles of thousands of tumors hundred genes are also frequently used to detect highly subclonal across 12 cancer types were integrated to derive oncogenic events in a clinical setting.55 Meanwhile, various other next- signature classes associated with targetable functional events. This generation sequencing-based assays are used to analyze the oncogenic signature-based stratification may help link tumors activity and function of the cancer genome, either in cell lines or with class-specific combination therapy.75 In another study, directly in primary tumor samples. For example, DNA sequencing 21 mutational signatures were extracted from thousands of tumor coupled with bisulfite treatment has enabled genome-wide samples and associated with a diverse spectrum of underlying profiling of methylation status and has revealed global epigenetic – mutational processes including exposure to specific mutagens, reprogramming in a variety of cancer types.20,56 60 Furthermore, disruption of DNA repair machinery and immunoglobulin gene oxidative bisulfite sequencing can specifically detect hypermutation.76 5-hydroxymethlcytosine61 that has been found to be associated The stratification of cancer patients into clinically relevant with tumor progression.62 Comprehensive transcriptome profiling cancer subtypes, disease progression stages, drug response via RNA sequencing is routinely used to detect gene expression groups as well as prognosis of survival and relapse is instrumental changes in tumors versus normal tissues, thus identifying to improving cancer clinical care. Molecular profiling of tumors is deregulated pathways and functions in tumor cells.63 RNA widely seen as a promising way to identify biomarkers that can sequencing has also revealed novel expressed variants, gene – guide this stratification together with other clinical variables such fusions and splicing alterations in cancers.64 66 The recent as age and demographics. However, given that a broad spectrum development of single-cell sequencing and its application to of molecular alterations collectively shape the malignant state in cancer genomic analysis is worth special notice. By analyzing a individual tumors, it is unlikely that such stratification can be cancer cell population at single-cell resolution, valuable informa- achieved only based on single mutation or single gene tion on tumor heterogeneity and clonal evolution history can be biomarkers. Instead, integrated analysis of large-scale genomic extracted that would otherwise be buried in bulk assessment of – and transcriptomic data using systems biology approaches hold cell mixtures.67 70 For example, a recent single-cell DNA sequen- promise for establishing robust molecular predictors of cancer cing study constructed an evolutionary tree based on sequence clinical phenotypes. In many cases, these promises have been data from 100 single cells. The tree revealed the existence of already realized. For example, implementation of clustering several cell groups (clades). Each of these groups was directly algorithms on integrated data of somatic nucleotide substitutions, divergent from the root. Lack of observable intermediate microsatellite instability and somatic copy number alterations branching suggests that tumor progression might happen in a stratified endometrial carcinomas into four novel categories.77 In punctuated manner.67 This idea of punctuated tumor evolution is diffuse large B-cell lymphoma (DLBCL), hierarchical clustering supported by other sequencing studies that revealed that large of gene expression distinguished two subtypes, the activated number of chromosomal rearrangements could be induced in one B-cell-like DLBCL and germinal center B-cell-like DLBCL, featuring or few events (termed chromoplexy and chromothripsis), thus significantly different overall survival.78 A Bayesian probabilistic driving abrupt tumor progression.52,71 classifier developed by Wright et al.79 established a gene expression signature that is now routinely used to stratify DLBCL Systems approaches for addressing cancer genome complexity subtypes. Another machine learning approach, Support Vector The identification of functions and processes disrupted by specific Machines, combines expression levels of four microRNAs to genomic alterations is key to understanding cancer pathogenesis predict whether a thyroid nodule is malignant or benign with and a prerequisite for the discovery of targeted therapeutics. 490% accuracy.80 Of note, the development of novel computa- However, this is challenged by the huge diversity of genomic tional algorithms and models for predicting cancer prognosis alterations observed in various patients, even within patients based on molecular profiles has been enhanced by a systems sharing the same tumor phenotypes. An emerging principle biology community effort called the DREAM challenges.81,82 In the underlying this diversity is that even though specific genomic breast cancer prognosis prediction challenge, the best-performing alterations vary from patient to patient, they usually result in model was demonstrated to outperform previous prediction disruption of a few key pathways and processes governing cancer approaches including the 70-gene risk signature.83 phenotypes. Consequently, integrated analysis of genomics data The highly unstable and evolutionary nature of individual in the context of pathways and interaction networks is more and cancer genomes and the vast heterogeneity in subclonal genomic

Oncogene (2015) 3215 – 3225 © 2015 Macmillan Publishers Limited Cancer systems biology W Du and O Elemento 3219 composition underlie most drug resistance and cancer relapse. crucial in mediating platelet-derived growth factor receptor-driven Mathematical modeling that integrates this complexity has been cancer cell survival. A newly introduced mass cytometry approach used to analyze and predict the evolutionary dynamics of termed CyTOF (DVS Sciences, Fluidigm Sciences Inc., Sunnyvale, heterogeneous tumor populations. By modeling the growth of a CA, USA) allows simultaneous profiling of up to 34 parameters in glioblastoma clone with dynamically acquired radioresistance, single immune cells, and has been used to probe altered Leder et al.18 predicted two optimal radiation strategies that intracellular signaling in hematological malignancies.93 Another substantially enhanced survival in mouse models. In another cost-effective protein profiling technique is reverse phase protein study, a mathematical model of dynamic tumor growth based on lysate microarray. This high-throughput antibody-based approach heterogeneous genetic and phenotypic features predicted how can quantify total and phosphorylated levels of hundreds of tumor subclones evolve upon chemotherapy.19 Based on evolu- proteins across thousands of samples with high accuracy.94 TCGA tionary game theory, Basanta et al.84 constructed a mathematical profiled 171 cancer-related proteins via reverse phase protein model to investigate evolutionary double-bind therapy, in which lysate microarray and discovered two novel proteomics-defined two different treatments are applied sequentially and iteratively subtypes in breast cancer, featuring signatures of cancer-activated to repress the effect of acquired resistance to single therapy. tumor microenvironment.95 Their simulations successfully recapitulated the experimentally observed synergistic effect of combined p53 cancer vaccine and Modeling the dynamics of signaling networks chemotherapy and confirmed potential clinical benefits of the 84 Owing to the intricacy of signaling networks, the design of double-bind strategy. Overall, these studies demonstrated the effective targeted therapies that durably shut down aberrant great potential of using evolutionary models to maximize fi pathway activations remains a challenging task. This is important therapeutic ef cacy on a cell population level to achieve more because resistance against single agents can easily arise from durable response. actions of compensatory circuits and feedback regulations if a pathway is not effectively shut down. For example, in some PROBING CANCER COMPLEXITY AT THE PROTEIN AND HER2-amplified breast cancers, HER2-targeted inhibition induces POSTTRANSLATIONAL LEVELS FOXO- and MYC-mediated HER3 overexpression that restores MAPK and PI3K/AKT pathway activity and therefore establishes Signaling network homeostasis is frequently perturbed in tumors resistance to HER2 inhibitors.15,16 Drug combinations that work Basic cellular activities and actions are directly controlled by a cooperatively to suppress signal transduction can improve complex system of intracellular signaling. Signaling networks response rate and prevent emergence of drug resistance.96–98 respond to environmental stimulus, transduce and integrate However, the design of effective drug combination regimens is information through protein–protein interactions and post- not trivial. Predictive computational models are urgently needed translational modifications and eventually induce transcriptional to guide more effective therapeutic designs. Kinetic modeling of programs that lead to critical cellular processes such as protein–protein interactions is an appealing approach as it can proliferation and differentiation. reproduce signaling dynamics in silico and make quantitative In normal tissues, the homeostasis of signaling networks is predictions of the effect of various network alterations. In the past, precisely maintained by complex crosstalks and feedbacks, such parameterization of these in silico models has been a major that the timing and amplitude of each signaling response is fine- challenge. With the application of quantitative protein activity 85 tuned and kept within a proper range. In cancer cells, genomic assays, such as reverse phase protein lysate microarray, large data and epigenetic alterations leading to aberrant protein expression sets directly revealing the activation status of signaling networks or function disrupt this homeostasis.86 For instance, genomic can be generated to extensively train and validate in silico models. amplification that results in increased protein level or the A number of recent studies have developed kinetic or semikinetic expression of aberrant fusions that constitutively activate models of cancer signaling networks parameterized by protein normal protein function can induce exaggerated signaling. HER2 activity assays in cell lines upon receptor stimulations and inhibitor amplification and BRAF/KRAS mutations are examples of such perturbations.15–17 These models demonstrated great potential in disruptions that sustain PI3K/AKT and/or mitogen-activated identifying drug response biomarkers, resistance mechanisms, and protein kinase (MAPK) signaling in several cancer types.87,88 Loss synergistic drug combinations. For example, Kirouac et al.15 of negative regulator expression in signaling networks because of established a mathematical model of the ERBB signaling network genomic alterations or downregulated protein expression can that jointly modeled multitime scale events including signaling remove important brakes on signaling, resulting in uncontrolled responses, transcriptional feedbacks and drug actions. Using this response. For example, the tumor repressor PTEN (phosphatase model, they predicted that combinatorial therapy against HER2 and and tensin homolog), which deactivates PI3K/AKT signaling, is HER3 would outperform a dual mitogen-activated protein kinase frequently lost in cancers.89 kinase (MEK) and AKT inhibition strategy. They also predicted biomarkers for sensitivity and resistance against various single drug Experimental probing of aberrant function in signaling networks and drug combination treatments. These predictions were con- fi 15 Genomic and epigenomic profiling have identified a great number rmed in vitro and in vivo. of alterations affecting known signaling pathway components in cancer. However, the nonlinearity and complexity of signaling Data-driven systems biology analysis for dissecting aberrant networks complicates interpretation regarding the global impact signaling functions that these alterations have on signaling network output and The analysis of proteome-wide profiles has provided important function. To address this problem, large-scale protein profiling is insights into cancer signaling networks and into how to target increasingly used to directly and systematically monitor signaling their aberrant functions. Recently, Casado et al. developed a pathway activities in cancer cells.90 One widely adopted computational approach termed kinase-substrate enrichment technology is mass spectrometry-based phosphoproteomics that analysis that can systematically infer the activation status of globally identifies phosphorylated proteins.91 Using this method, signaling kinases from phosphoproteomic data. Applying this Moritz et al.92 systematically uncovered 4300 phosphorylation method to acute myeloid leukemia, they identified distinct substrates of receptor tyrosine kinases (RTKs) in cancer cell lines, substrate groups that are predictive of sensitivity or resistance among which the protein chaperone SGTA (small glutamine-rich to PI3K/AKT inhibition.13 Another group performed partial least- tetratricopeptide repeat-containing protein α) was found to be square regression on protein activity data assayed by enzyme-

© 2015 Macmillan Publishers Limited Oncogene (2015) 3215 – 3225 Cancer systems biology W Du and O Elemento 3220 linked immunosorbent assay from basal and stimulated signaling systems-level understanding of its dynamics are needed to networks and successfully predicted drug response in a panel of optimize TME-targeted therapies. breast cancer cell lines.14 Based on network models inferred from dynamic protein phosphorylation profiles obtained upon RNA 99 Experimental approaches for analyzing the tumor interference perturbations, Wagner et al. classified six RTKs into microenvironment three distinct classes. They demonstrated that resistance of As the tumor microenvironment is a mixed ensemble of various individual RTK inhibitor arises from increased abundance of other cell types and signaling molecules, numerous experimental RTKs within the same class, suggesting simultaneous targeting of approaches are needed for probing the TME, each one focusing RTKs within a class may prevent resistance. In another study on a specific compartment. To characterize molecular activity in adopting the similar methodology, a regulatory network of AKT tumor and stromal cells, mRNA and protein arrays can be signaling was uncovered, featuring mutual inhibition between conducted on microdissected tumor biopsy samples or xeno- fi 100 MAPK and AKT mediated by newly identi ed regulator loops. grafted tumor lesions. Such analyses have identified key factors These studies provide important examples of how systematic and pathways involved in tumor–stroma crosstalks as well as analysis of proteomics data can guide effective targeted therapy TME-related molecular signatures predictive of clinical by enabling prediction of drug response, revealing important outcome.12,107,108 Taking advantages of genomic differences regulatory circuits and uncovering mechanisms of resistance. between tumor cells and host cells in xenografts models, a novel experimental approach termed cross-species hybridization of PROBING CANCER COMPLEXITY AT THE TISSUE LEVEL microarrays was developed that enables simultaneous assessment of gene expression in both tumor and host cells via separate Bidirectional regulation between tumor and its local environment hybridization of xenograft tumors to human and mouse Tumor transformation, growth and invasion take place in a microarray.109 Using this approach to examine gene expression complex local environment composed of fibroblast cells, endothe- in cell lines xenografted into different organ sites, Park et al.109 lial cells, immune cells, signaling molecules and the extracellular revealed a strong neuronal cell-like transcriptional reprogramming matrix. This highly interactive ecological system, termed the tumor in brain microenvironment. microenvironment (TME), also plays an active role in tumor In vivo imaging has been extensively used to examine tumor- progression.101 Cancer cells benefit from the microenvironment –stroma morphology and interactions or assay specific molecular by acquiring nutrients and various biochemical factors. For markers such as pH, hypoxia or proteases with fluorescence example, hepatocyte growth/scatter factor secreted by fibroblasts probes.110 For example, Nakasone et al.111 applied in vivo was shown to promote tumor growth and invasion.101 Cancer cells microscopy to monitor real-time response to chemotherapy in also actively elicit tumor-promoting properties of the micro- mouse mammary carcinomas. They observed that Ccr2 knockout environment by releasing various cytokines and growth factors. mice with impaired vascular permeability or immune infiltration For example, vascular endothelial growth factor and basic displayed improved response to chemotherapy, highlighting the fibroblast growth factor stimulate endothelial proliferation and role of tumor microenvironment in chemoresistance.111 An drive angiogenesis,5,101 whereas the colony-stimulating factor-1 ultrasensitive pH fluorescent nanoprobe was developed, enabling cytokine fosters the differentiation of pro-tumorigenic M2 rapid detection of microenvironmental signals linked with tumor macrophages.5 In short, numerous two-way interactions occur angiogenesis and aerobic glycolysis.112 Moreover, advanced between tumors and their microenvironment, thus leading to optical technologies such as optical frequency domain imaging complex emergent behaviors that ultimately benefit the tumors can now provide three-dimensional assessment of tumor micro- themselves. environment profiles in vivo and in a rapid and repeated 113 It has long been proposed that targeting the TME may yield manner. efficient and perhaps less toxic anticancer therapy compared with In vitro culture models of tumor cells with key components of strategies aimed at directly targeting cancer cells. A major the microenvironment can also be used in high-throughput challenge of TME-targeted therapy is that the effects of various screens to discover important interactions between stroma and microenvironment components can be highly context specific, tumor cells. By co-culturing 45 cancer cell lines with 23 stromal cell 114 and that cancer cells by secreting various growth factors and types, respectively, Straussman et al. discovered that hepato- cytokines can educate the TME to assist tumor progression. For cyte growth factor secreted by stromal cells can activate MAPK example, immune cells such as macrophages can display either and PI3K/AKT signaling and thus mediate resistance to RAF antitumor or protumor functions depending on their polarization inhibitor. Using a three-dimensional culture model engineered 102 with mesothelial cells, fibroblasts and extracellular matrix, Kenny status. Cancer cells can promote the development of protumor 115 macrophages by secreting various cytokines such as colony- et al. found that the attachment and invasion of ovarian cancer stimulating factor-1.103 Although transforming growth factor-β cells are suppressed by omental mesothelial cells but enhanced by fi normally inhibits epithelial proliferation, tumor cells can escape omental broblasts and the extracellular matrix. this suppression by acquiring genetic or epigenetic alterations in Overall, comprehensive characterization of tumor microenvir- the transforming growth factor-β signaling pathway. After onment is likely to require multiple experimental approaches in acquiring these alterations, transforming growth factor-β posi- combination, thus encompassing both cellular-level information of tively contributes to tumor progression via its additional role in signaling interactions and tissue-level features such as spatial promoting angiogenesis and antagonizing tumor-suppressive organization and morphology of TME compartments. inflammatory responses.104 The plasticity of tumor microenvironment combined with a tumor’s ability to educate its surroundings Modeling the TME have led to highly variable treatment response as well as the A number of mathematical models have been developed to emergence of drug resistance in various TME-targeted therapies. reconstruct the TME in silico. By simulating the dynamic interplay For example, anti-angiogenesis therapy such as vascular endothe- between different components of the microenvironment, these lial growth factor inhibition has been shown to elicit a transient models have provided important insights into how tumor and the tumor and vascular shrinkage followed by enhanced tumor microenvironment coevolve. Moreover, such models have also invasion and revascularization.105,106 Owing to the complexity of provided valuable hypotheses about how therapeutic interven- the dynamic interplay between tumor and its microenvironment, tions might be applied to achieve maximum clinical benefits, for systematic evaluation of the global TME status and predictive, example, by predicting the most effective drug targets or the

Figure 3. An integrative, iterative and model-based strategy for personalized cancer medicine. At initial diagnosis, patients undergo multidimensional experimental profiling to comprehensively characterize their specific cancer alterations. These profiles are used to build patient-specific computational models to predict optimal short-term therapeutic strategy. This process is conducted iteratively to quickly adapt to potential resistance acquired because of continuous cancer evolution, until cancer is fully eradicated. optimal time point of treatment. For example, by integrating an share malignant hallmarks and cell of origin,122 can be driven by intercellular signaling network of glioma microenvironment vastly different molecular alterations. Tumors arising from the consisting of 5 types of cells, 15 cytokines and 69 signaling same organ or tissue are found to consist of multiple molecular pathways, Wu et al.116 predicted a three-phase development of subtypes, often associated with distinct therapeutic response and glioma and tested the efficacy of various treatment strategies. clinical outcome.121 Consequently, conventional cancer therapy They found that a microglia depletion strategy only acts at the guided by tumor cell of origins should evolve to accommodate early stage of glioma development, whereas cytokine combination more personalized therapeutic strategies, in which treatment is therapy yields more effective and robust response in the assigned to each patient based on their specific molecular intermediate and later stage.116 The critical role of immune alterations.123 Such is the central goal of precision and persona- response in tumor invasion was highlighted in another model, lized medicine. where immune cells were found to either restrict or promote Precision cancer medicine has so far mostly focused on the tumor expansion depending on the metastatic status. Interest- identification of individual molecular biomarkers that are indica- ingly, optimal tumor-suppressive effect was achieved under tive of responsiveness to specific targeted therapies. The majority moderate metastasis according to model simulations.117 By of the biomarkers validated and used in clinical practice are comprehensive profiling of 91 tumor-infiltrating lymphocytes, single genetic alterations such as oncogenic mutations, gene Oved et al.118 demonstrated that the reactivity of tumor- amplifications or deletions and translocations. Well-known associated immune system could be predicted by subpopulation examples include Trastuzumab for HER2 overexpression in breast composition and that antitumor reactivity could be restored via cancer,124 Imatinib for BCR-ABL gene fusion in chronic myelogen- selective enrichment or depletion of specific immune compo- ous leukemia,125 Gefitinib and Erlotinib for epidermal growth nents. By integrating dynamic magnetic resonance imaging data factor receptor mutations in lung cancer126 and Vemurafenib for into a comprehensive TME model that simulates tumor growth BRAF mutation in melanoma.127 Clinical trials of targeted therapy and angiogenesis, Venkatasubramanian et al.119 investigated the against patients with matching biomarkers demonstrated sig- influence of drug transport properties on therapeutic efficacy. nificant improvement in both response rate and probability of Another model built upon in vivo imaging data of tumor and survival compared with generalized cancer treatment such associated vascular status successfully predicted tumor growth as chemotherapy.127–129 Recently, a computational platform dynamics and therapeutic response.120 In summary, these was established for identifying clinically relevant alterations examples demonstrate the capability of mathematical models to based on whole-exome sequencing data from formalin-fixed, capture the highly interactive and dynamically evolving nature of paraffin-embedded tumor samples. Applying this method to TME, enabling them to investigate novel potential clinical patients of various tumor types successfully identified potential strategies targeting the microenvironment. clinical targets in 15 out of 16 patients.130 Although traditionally the connections between drug and specific molecular alterations were largely investigated in isolation, the application of high- CANCER SYSTEMS BIOLOGY AND PERSONALIZED CANCER throughput drug screenings against a large panel of cancer cell MEDICINE lines with comprehensive molecular profiles has enabled more Biomarker-based personalized cancer medicine systematic and efficient discoveries. For example, the Cancer Cell Over the past decade, extensive molecular profiling of the Line Encyclopedia performed pharmacological screening of genome, transcriptome, proteome and metabolome of tumors 24 drugs against ∼ 500 cancer cell lines. Joint statistical modeling and their microenvironment has tremendously advanced our of drug response and genomics data identified novel molecular knowledge of the molecular mechanisms underlying cancer predictors of drug sensitivity, such as AHR expression for MEK pathology.121 It is now clear that individual tumors, even if they inhibitor efficacy in NRAS-mutant cell lines.131 Furthermore, the

© 2015 Macmillan Publishers Limited Oncogene (2015) 3215 – 3225 Cancer systems biology W Du and O Elemento 3222 Connectivity Map (CMAP), which links disease states with drug prediction and treatment iteratively (updating the model with actions via gene expression profiles, provides a natural computa- newly observed molecular alterations) until a tumor is fully tional framework for drug repositioning.132 Viewing gene expres- eradicated (Figure 3). sion profiles as representative of biological states, drugs are likely to act against certain disease states if their gene expression profiles show strong negative correlation. Thus, querying gene CONCLUSION expression signatures of given disease states against the CMAP Decades of research have revealed that cancer is an extremely database, which contains ∼ 7000 gene expression signatures of complex disease featuring systems-level disruptions. Continuous 1300 compounds, can potentially identify effective treatment efforts in experimental profiling of tumors at multiple levels have strategies. This approach successfully discovered mammalian unveiled a large spectrum of molecular alterations, yet their target of rapamycin inhibitor sirolimus as effective treatment functional impacts remain to be uncovered. Embracing the for dexamethasone-resistant acute lymphoblastic leukemia complexity of cancer, cancer systems biology operates on large patients.132 A similar gene expression signature-based computa- multiscale data set to develop integrative and predictive models tional drug repositioning approach identified tricyclic antidepres- that provide systems-level understanding of how molecular sants as inhibitors of small-cell lung cancers.133 alterations coordinately drive tumor progression. The integrative and dynamic nature of these models makes them extremely Model-based personalized cancer medicine powerful in addressing some of the major challenges in cancer biology such as tumor heterogeneity and tumor evolution. Studies Although there are many examples of single biomarkers guiding implementing systems biology approaches have already made the selection of targeted therapy, the design of optimal significant contribution to facilitating cancer therapy, such as personalized treatments based on a single biomarker is likely to identifying predictive biomarkers to improve response rate or be limited to a small number of drugs and patients. Indeed, a optimizing drug combinations and dosing schedules to overcome cancer cell usually bears multiple genetic and epigenetic emergent drug resistance. As cancer treatment approaches a alterations, resulting in a unique combination of phenotypes more personalized era, we envision that cancer systems biology and response to treatment. Moreover even among patients will play a central role in the development of integrative model- carrying the same biomarker, interpatient heterogeneity because based cancer medicine. Iterative molecular profiling, computa- of a multitude of somatic and germline genetic events and a tional modeling and treatment in individual patients is likely to stochastic mutational process can still result in remarkably variable deliver maximal therapeutic effect that may help control and drug responses. It has indeed been found that resistance to perhaps eventually cure cancer. specific targeted therapy could arise from vastly distinct mechanisms such as secondary mutations attenuating drug binding, additional mutations in the same pathway or activation of CONFLICT OF INTEREST compensatory circuits.134 Therefore, integration of multidimen- The authors declare no conflict of interest. sional molecular profiles is required to establish a full portrait of the malignant defects in individual patients. A pilot study of integrative molecular profiling on individual patients successfully ACKNOWLEDGEMENTS fi fi identi ed actionable targets among their speci c oncogenic We gratefully acknowledge insightful comments from Dr Yanwen Jiang (Elemento alterations in all four patients enrolled, demonstrating the lab) and our anonymous reviewer. We also thank other members in the laboratory 135 feasibility and potential of integrative personalized medicine. and Physiology, Biophysics and Systems Biology program of Weill Cornell Graduate The NCI-DREAM drug sensitivity prediction challenge has further School for helpful discussions. This work was supported by NIH, CAREER Grant from promoted the development of computational algorithms for National Science Foundation, Starr Cancer Consortium and Hirschl Trust. predicting drug response from multidimensional omics data.136 It is important to note that because of regulatory interactions REFERENCES between genomic, transcriptional and signaling layers, information from one layer can be used to infer alterations in another 1 Negrini S, Gorgoulis VG, Halazonetis TD. Genomic instability--an evolving 11 – layer. For example, oncogenic mutations are found to generate hallmark of cancer. Nat Rev Mol Cell Biol 2010; : 220 228. 2 Shih AH, Abdel-Wahab O, Patel JP, Levine RL. The role of mutations in epigenetic unique gene expression signatures that can be predictive of 12 – 137 regulators in myeloid malignancies. Nat Rev Cancer 2012; :599 612. responses to targeted therapy. Moreover, a computational 3 Vogelstein B, Papadopoulos N, Velculescu VE, Zhou S, Diaz LA Jr., Kinzler KW. algorithm has been developed to infer post-transcriptional Cancer genome landscapes. Science 2013; 339:1546–1558. modifications that affect the activity of oncogenic or tumor- 4 Suva ML, Riggi N, Bernstein BE. Epigenetic reprogramming in cancer. Science suppressive transcription factors via expression changes in their 2013; 339: 1567–1570. target genes.138 Therefore, the benefit of integrating multiple data 5 Quail DF, Joyce JA. Microenvironmental regulation of tumor progression and types can be far beyond the ‘sum of parts’. metastasis. Nat Med 2013; 19: 1423–1437. The key to successful personalized therapeutic design lies in the 6 Greaves M, Maley CC. Clonal evolution in cancer. Nature 2012; 481:306–313. development of integrated and predictive models that can 7 Hay M, Thomas DW, Craighead JL, Economides C, Rosenthal J. Clinical devel- fi opment success rates for investigational drugs. Nat Biotechnol 2014; 32:40–51. translate patient-speci c molecular alterations into functional 8 Johannessen CM, Boehm JS, Kim SY, Thomas SR, Wardwell L, Johnson LA et al. impacts and phenotypic behaviors. As discussed above, models COT drives resistance to RAF inhibition through MAP kinase pathway reactiva- of cancer signaling pathways, interaction between tumor and the tion. Nature 2010; 468:968–972. microenvironment and evolution of heterogeneous tumor popu- 9 Poulikakos PI, Persaud Y, Janakiraman M, Kong X, Ng C, Moriceau G et al. RAF lation have been established separately. Encouraged by the inhibitor resistance is mediated by dimerization of aberrantly spliced BRAF successful construction of a whole-cell computational model able (V600E). Nature 2011; 480: 387–390. to predict phenotype from genotype,139 we may boldly anticipate 10 Emery CM, Vijayendran KG, Zipser MC, Sawyer AM, Niu L, Kim JJ et al. MEK1 mutations confer resistance to MEK and B-RAF inhibition. Proc Natl Acad Sci USA the development of sophisticated cancer cell models that, upon 106 – inputting patient’s molecule profiles, can predict the optimal 2009; : 20411 20416. 11 The Cancer Genome Atlas Network. Comprehensive molecular characterization treatment strategy. Owing to the intrinsic randomness of tumor of clear cell renal cell carcinoma. Nature 2013; 499:43–49. evolution, long-term behavior of tumor progression may be hard 12 Finak G, Bertos N, Pepin F, Sadekova S, Souleimanova M, Zhao H et al. Stromal to forecast via computational models deterministically. A potential gene expression predicts clinical outcome in breast cancer. Nat Med 2008; 14: solution is to perform molecular profiling, computational 518–527.

Oncogene (2015) 3215 – 3225 © 2015 Macmillan Publishers Limited Cancer systems biology W Du and O Elemento 3223 13 Casado P, Rodriguez-Prados JC, Cosulich SC, Guichard S, Vanhaesebroeck B, 39 Anderson K, Lutz C, van Delft FW, Bateman CM, Guo Y, Colman SM et al. Genetic Joel S et al. Kinase-substrate enrichment analysis provides insights into the variegation of clonal architecture and propagating cells in leukaemia. Nature heterogeneity of signaling pathway activation in leukemia cells. Sci Signal 2013; 2011; 469:356–361. 6: rs6. 40 Korolev KS, Xavier JB, Gore J. Turning ecology and evolution against cancer. Nat 14 Niepel M, Hafner M, Pace EA, Chung M, Chai DH, Zhou L et al. Profiles of basal Rev Cancer 2014; 14: 371–380. and stimulated receptor signaling networks predict drug response in breast 41 Cleary AS, Leonard TL, Gestl SA, Gunther EJ. Tumour cell heterogeneity main- cancer lines. Sci Signal 2013; 6: ra84. tained by cooperating subclones in Wnt-driven mammary cancers. Nature 2014; 15 Kirouac DC, Du JY, Lahdenranta J, Overland R, Yarar D, Paragas V et al. Com- 508:113–117. putational modeling of ERBB2-amplified breast cancer identifies combined 42 Campbell PJ, Yachida S, Mudie LJ, Stephens PJ, Pleasance ED, Stebbings LA et al. ErbB2/3 blockade as superior to the combination of MEK and AKT inhibitors. Sci The patterns and dynamics of genomic instability in metastatic pancreatic Signal 2013; 6: ra68. cancer. Nature 2010; 467: 1109–1113. 16 Klinger B, Sieber A, Fritsche-Guenther R, Witzel F, Berry L, Schumacher D et al. 43 Lohr JG, Stojanov P, Carter SL, Cruz-Gordillo P, Lawrence MS, Auclair D et al. Network quantification of EGFR signaling unveils potential for targeted combi- Widespread genetic heterogeneity in multiple myeloma: implications for nation therapy. Mol Syst Biol 2013; 9: 673. targeted therapy. Cancer Cell 2014; 25:91–101. 17 Faratian D, Goltsov A, Lebedeva G, Sorokin A, Moodie S, Mullen P et al. 44 Landau DA, Carter SL, Stojanov P, McKenna A, Stevenson K, Lawrence MS et al. Systems biology reveals new strategies for personalizing cancer medicine and Evolution and impact of subclonal mutations in chronic lymphocytic leukemia. confirms the role of PTEN in resistance to trastuzumab. Cancer Res 2009; 69: Cell 2013; 152: 714–726. 6713–6720. 45 Johnson BE, Mazor T, Hong C, Barnes M, Aihara K, McLean CY et al. Mutational 18 Leder K, Pitter K, Laplant Q, Hambardzumyan D, Ross BD, Chan TA et al. analysis reveals the origin and therapy-driven evolution of recurrent glioma. Mathematical modeling of PDGF-driven glioblastoma reveals optimized radia- Science 2014; 343:189–193. tion dosing schedules. Cell 2014; 156:603–616. 46 Kreso A, O'Brien CA, van Galen P, Gan OI, Notta F, Brown AM et al. Variable clonal 19 Almendro V, Cheng YK, Randles A, Itzkovitz S, Marusyk A, Ametller E et al. repopulation dynamics influence chemotherapy response in colorectal cancer. Inference of tumor evolution during chemotherapy by computational modeling Science 2013; 339:543–548. and in situ analysis of genetic and phenotypic cellular diversity. Cell Rep 2014; 6: 47 Wargo AR, Huijben S, de Roode JC, Shepherd J, Read AF. Competitive 514–527. release and facilitation of drug-resistant parasites after therapeutic 20 Landan G, Cohen NM, Mukamel Z, Bar A, Molchadsky A, Brosh R et al. Epigenetic chemotherapy in a rodent malaria model. Proc Natl Acad Sci USA 2007; 104: polymorphism and the stochastic formation of differentially methylated regions 19914–19919. in normal and cancerous tissues. Nat Genet 2012; 44: 1207–1214. 48 Ding L, Ley TJ, Larson DE, Miller CA, Koboldt DC, Welch JS et al. Clonal evolution 21 Choi JD, Lee JS. Interplay between epigenetics and genetics in cancer. Genomics in relapsed acute myeloid leukaemia revealed by whole-genome sequencing. Inform 2013; 11: 164–173. Nature 2012; 481: 506–510. 22 Ehrlich M. DNA methylation in cancer: too much, but also too little. Oncogene 49 Metzker ML. Sequencing technologies - the next generation. Nat Rev Genet 2010; 2002; 21: 5400–5413. 11:31–46. 23 Seligson DB, Horvath S, Shi T, Yu H, Tze S, Grunstein M et al. Global histone 50 Ley TJ, Mardis ER, Ding L, Fulton B, McLellan MD, Chen K et al. DNA sequencing modification patterns predict risk of prostate cancer recurrence. Nature 2005; of a cytogenetically normal acute myeloid leukaemia genome. Nature 2008; 456: 435: 1262–1266. 66–72. 24 Pan H, Jiang Y, Redmond D, Nie K, Cerchietti L, Shaknovich R et al. 51 Ojesina AI, Lichtenstein L, Freeman SS, Pedamallu CS, Imaz-Rosshandler I, Epigenomic evolution in diffuse large B-cell lymphomas. Blood 2013; 122: Pugh TJ et al. Landscape of genomic alterations in cervical carcinomas. Nature 634–634. 2014; 506:371–375. 25 Tamborero D, Gonzalez-Perez A, Perez-Llamas C, Deu-Pons J, Kandoth C, 52 Baca SC, Prandi D, Lawrence MS, Mosquera JM, Romanel A, Drier Y et al. Reimand J et al. Comprehensive identification of mutational cancer driver genes Punctuated evolution of prostate cancer genomes. Cell 2013; 153:666–677. across 12 tumor types. Sci Rep 2013; 3: 2650. 53 Sjoblom T, Jones S, Wood LD, Parsons DW, Lin J, Barber TD et al. The consensus 26 Ashworth A, Lord CJ, Reis-Filho JS. Genetic interactions in cancer progression coding sequences of human breast and colorectal cancers. Science 2006; 314: and treatment. Cell 2011; 145:30–38. 268–274. 27 Muller FL, Colla S, Aquilanti E, Manzo VE, Genovese G, Lee J et al. Passenger 54 Jones S, Hruban RH, Kamiyama M, Borges M, Zhang X, Parsons DW et al. Exomic deletions generate therapeutic vulnerabilities in cancer. Nature 2012; 488: sequencing identifies PALB2 as a pancreatic cancer susceptibility gene. Science 337–342. 2009; 324: 217. 28 McFarland CD, Korolev KS, Kryukov GV, Sunyaev SR, Mirny LA. Impact of 55 Frampton GM, Fichtenholtz A, Otto GA, Wang K, Downing SR, He J et al. deleterious passenger mutations on cancer progression. Proc Natl Acad Sci USA Development and validation of a clinical cancer genomic profiling test 2013; 110: 2910–2915. based on massively parallel DNA sequencing. Nat Biotechnol 2013; 31: 29 Nowell PC. The clonal evolution of tumor cell populations. Science 1976; 194: 1023–1031. 23–28. 56 Pike BL, Greiner TC, Wang X, Weisenburger DD, Hsu YH, Renaud G et al. DNA 30 Muller PA, Vousden KH. p53 mutations in cancer. Nat Cell Biol 2013; 15:2–8. methylation profiles in diffuse large B-cell lymphoma and their relationship to 31 Lee W, Jiang Z, Liu J, Haverty PM, Guan Y, Stinson J et al. The mutation spectrum gene expression status. Leukemia 2008; 22: 1035–1043. revealed by paired genome sequences from a lung cancer patient. Nature 2010; 57 Kim JH, Dhanasekaran SM, Prensner JR, Cao X, Robinson D, Kalyana-Sundaram S 465: 473–477. et al. Deep sequencing reveals distinct patterns of DNA methylation in 32 Berger MF, Hodis E, Heffernan TP, Deribe YL, Lawrence MS, Protopopov A et al. prostate cancer. Genome Res 2011; 21: 1028–1041. Melanoma genome sequencing reveals frequent PREX2 mutations. Nature 2012; 58 Berman BP, Weisenberger DJ, Aman JF, Hinoue T, Ramjan Z, Liu Y et al. Regions 485: 502–506. of focal DNA hypermethylation and long-range hypomethylation in colorectal 33 Poon SL, Pang ST, McPherson JR, Yu W, Huang KK, Guan P et al. Genome-wide cancer coincide with nuclear lamina-associated domains. Nat Genet 2012; 44: mutational signatures of aristolochic acid and its application as a screening tool. 40–46. Sci Transl Med 2013; 5: 197ra101. 59 Hon GC, Hawkins RD, Caballero OL, Lo C, Lister R, Pelizzola M et al. Global DNA 34 Marusyk A, Almendro V, Polyak K. Intra-tumour heterogeneity: a looking glass for hypomethylation coupled to repressive chromatin domain formation and gene cancer? Nat Rev Cancer 2012; 12: 323–334. silencing in breast cancer. Genome Res 2012; 22: 246–258. 35 Gerlinger M, Rowan AJ, Horswell S, Larkin J, Endesfelder D, Gronroos E et al. 60 Hansen KD, Timp W, Bravo HC, Sabunciyan S, Langmead B, McDonald OG et al. Intratumor heterogeneity and branched evolution revealed by multiregion Increased methylation variation in epigenetic domains across cancer types. Nat sequencing. N Engl J Med 2012; 366: 883–892. Genet 2011; 43: 768–775. 36 Sottoriva A, Spiteri I, Piccirillo SG, Touloumis A, Collins VP, Marioni JC et al. 61 Booth MJ, Branco MR, Ficz G, Oxley D, Krueger F, Reik W et al. Quantitative Intratumor heterogeneity in human glioblastoma reflects cancer evolutionary sequencing of 5-methylcytosine and 5-hydroxymethylcytosine at single-base dynamics. Proc Natl Acad Sci USA 2013; 110: 4009–4014. resolution. Science 2012; 336:934–937. 37 Bashashati A, Ha G, Tone A, Ding J, Prentice LM, Roth A et al. Distinct evolu- 62 Yang H, Liu Y, Bai F, Zhang JY, Ma SH, Liu J et al. Tumor development is tionary trajectories of primary high-grade serous ovarian cancers revealed associated with decrease of TET gene expression and 5-methylcytosine hydro- through spatial mutational profiling. J Pathol 2013; 231:21–34. xylation. Oncogene 2013; 32: 663–669. 38 Gerlinger M, Horswell S, Larkin J, Rowan AJ, Salm MP, Varela I et al. Genomic 63 Eswaran J, Cyanam D, Mudvari P, Reddy SD, Pakala SB, Nair SS et al. architecture and evolution of clear cell renal cell carcinomas defined by multi- Transcriptomic landscape of breast cancers through mRNA sequencing. Sci Rep region sequencing. Nat Genet 2014; 46: 225–233. 2012; 2: 264.

© 2015 Macmillan Publishers Limited Oncogene (2015) 3215 – 3225 Cancer systems biology W Du and O Elemento 3224 64 Horvath A, Pakala SB, Mudvari P, Reddy SD, Ohshiro K, Casimiro S et al. Novel 92 Moritz A, Li Y, Guo A, Villen J, Wang Y, MacNeill J et al. Akt-RSK-S6 kinase insights into breast cancer genetic variance through RNA sequencing. Sci Rep signaling networks activated by oncogenic receptor tyrosine kinases. Sci Signal 2013; 3: 2256. 2010; 3: ra64. 65 Maher CA, Kumar-Sinha C, Cao X, Kalyana-Sundaram S, Han B, Jing X et al. 93 Bendall SC, Simonds EF, Qiu P, Amir el AD, Krutzik PO, Finck R et al. Single-cell Transcriptome sequencing to detect gene fusions in cancer. Nature 2009; 458: mass cytometry of differential immune and drug responses across a human 97–101. hematopoietic continuum. Science 2011; 332:687–696. 66 Eswaran J, Horvath A, Godbole S, Reddy SD, Mudvari P, Ohshiro K et al. RNA 94 Spurrier B, Ramalingam S, Nishizuka S. Reverse-phase protein lysate microarrays sequencing of cancer reveals novel splicing alterations. Sci Rep 2013; 3: 1689. for cell signaling analysis. Nat Protoc 2008; 3: 1796–1808. 67 Navin N, Kendall J, Troge J, Andrews P, Rodgers L, McIndoo J et al. Tumour 95 The Cancer Genome Atlas Network. Comprehensive molecular portraits of evolution inferred by single-cell sequencing. Nature 2011; 472:90–94. human breast tumours. Nature 2012; 490:61–70. 68 Xu X, Hou Y, Yin X, Bao L, Tang A, Song L et al. Single-cell exome sequencing 96 Mao M, Tian F, Mariadason JM, Tsao CC, Lemos R Jr. Dayyani F et al. Resistance reveals single-nucleotide mutation characteristics of a kidney tumor. Cell 2012; to BRAF inhibition in BRAF-mutant colon cancer can be overcome with PI3K 148: 886–895. inhibition or demethylating agents. Clin Cancer Res 2013; 19:657–667. 69 Hou Y, Song L, Zhu P, Zhang B, Tao Y, Xu X et al. Single-cell exome sequencing 97 Greger JG, Eastman SD, Zhang V, Bleam MR, Hughes AM, Smitheman KN et al. and monoclonal evolution of a JAK2-negative myeloproliferative neoplasm. Cell Combinations of BRAF, MEK, and PI3K/mTOR inhibitors overcome acquired 2012; 148: 873–885. resistance to the BRAF inhibitor GSK2118436 dabrafenib, mediated by NRAS or 70 Dalerba P, Kalisky T, Sahoo D, Rajendran PS, Rothenberg ME, Leyrat AA et al. MEK mutations. Mol Cancer Ther 2012; 11: 909–920. Single-cell dissection of transcriptional heterogeneity in human colon tumors. 98 Flaherty KT, Infante JR, Daud A, Gonzalez R, Kefford RF, Sosman J et al. Combined Nat Biotechnol 2011; 29: 1120–1127. BRAF and MEK inhibition in melanoma with BRAF V600 mutations. N Engl J Med 71 Stephens PJ, Greenman CD, Fu B, Yang F, Bignell GR, Mudie LJ et al. Massive 2012; 367: 1694–1703. genomic rearrangement acquired in a single catastrophic event during cancer 99 Wagner JP, Wolf-Yadlin A, Sevecka M, Grenier JK, Root DE, Lauffenburger DA development. Cell 2011; 144:27–40. et al. Receptor tyrosine kinases fall into distinct classes based on their inferred 72 Vandin F, Upfal E, Raphael BJ. Algorithms for detecting significantly mutated signaling networks. Sci Signal 2013; 6: ra58. pathways in cancer. J Comput Biol 2011; 18:507–522. 100 Lu Y, Muller M, Smith D, Dutta B, Komurov K, Iadevaia S et al. Kinome 73 Vaske CJ, Benz SC, Sanborn JZ, Earl D, Szeto C, Zhu J et al. Inference of patient- siRNA-phosphoproteomic screen identifies networks regulating AKT signaling. specific pathway activities from multi-dimensional cancer genomics data using Oncogene 2011; 30: 4567–4577. PARADIGM. Bioinformatics 2010; 26: i237–i245. 101 Liotta LA, Kohn EC. The microenvironment of the tumour-host interface. Nature 74 Hu J, Locasale JW, Bielas JH, O'Sullivan J, Sheahan K, Cantley LC et al. 2001; 411: 375–379. Heterogeneity of tumor-induced gene expression changes in the human 102 De Palma M, Lewis CE. Macrophage regulation of tumor responses to anticancer metabolic network. Nat Biotechnol 2013; 31:522–529. therapies. Cancer Cell 2013; 23: 277–286. 75 Ciriello G, Miller ML, Aksoy BA, Senbabaoglu Y, Schultz N, Sander C. Emerging 103 Pollard JW. Tumour-educated macrophages promote tumour progression and landscape of oncogenic signatures across human cancers. Nat Genet 2013; 45: metastasis. Nat Rev Cancer 2004; 4:71–78. 1127–1133. 104 Derynck R, Akhurst RJ, Balmain A. TGF-beta signaling in tumor suppression and 76 Alexandrov LB, Nik-Zainal S, Wedge DC, Aparicio SA, Behjati S, Biankin AV et al. cancer progression. Nat Genet 2001; 29:117–129. Signatures of mutational processes in human cancer. Nature 2013; 500:415–421. 105 Paez-Ribes M, Allen E, Hudock J, Takeda T, Okuyama H, Vinals F et al. 77 Kandoth C, Schultz N, Cherniack AD, Akbani R, Liu Y, Shen H et al. Integrated Antiangiogenic therapy elicits malignant progression of tumors to increased genomic characterization of endometrial carcinoma. Nature 2013; 497:67–73. local invasion and distant metastasis. Cancer Cell 2009; 15: 220–231. 78 Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Rosenwald A et al. Distinct 106 Lu KV, Chang JP, Parachoniak CA, Pandika MM, Aghi MK, Meyronet D et al. VEGF types of diffuse large B-cell lymphoma identified by gene expression profiling. inhibits tumor cell invasion and mesenchymal transition through a MET/VEGFR2 Nature 2000; 403: 503–511. complex. Cancer Cell 2012; 22:21–35. 79 Wright G, Tan B, Rosenwald A, Hurt EH, Wiestner A, Staudt LM. A gene 107 Sheehan KM, Gulmann C, Eichler GS, Weinstein JN, Barrett HL, Kay EW et al. expression-based method to diagnose clinically distinct subgroups of diffuse Signal pathway profiling of epithelial and stromal compartments of colonic large B cell lymphoma. Proc Natl Acad Sci USA 2003; 100: 9991–9996. carcinoma reveals epithelial-mesenchymal transition. Oncogene 2008; 27: 80 Keutgen XM, Filicori F, Crowley MJ, Wang Y, Scognamiglio T, Hoda R et al. 323–331. A panel of four miRNAs accurately differentiates malignant from benign 108 Kahlert C, Pecqueux M, Halama N, Dienemann H, Muley T, Pfannschmidt J et al. indeterminate thyroid lesions on fine needle aspiration. Clin Cancer Res 2012; 18: Tumour-site-dependent expression profile of angiogenic factors in tumour- 2032–2038. associated stroma of primary colorectal cancer and metastases. Br J Cancer 2014; 81 Cheng WY, Ou Yang TH, Anastassiou D. Development of a prognostic model for 110:441–449. breast cancer survival in an open challenge environment. Sci Transl Med 2013; 5: 109 Park ES, Kim SJ, Kim SW, Yoon SL, Leem SH, Kim SB et al. Cross-species hybri- 181ra150. dization of microarrays for studying tumor transcriptome of brain metastasis. 82 Margolin AA, Bilal E, Huang E, Norman TC, Ottestad L, Mecham BH et al. Proc Natl Acad Sci USA 2011; 108: 17456–17461. Systematic analysis of challenge-driven improvements in molecular prognostic 110 Wu Y, Zhang W, Li J, Zhang Y. Optical imaging of tumor microenvironment. Am J models for breast cancer. Sci Transl Med 2013; 5: 181re181. Nucl Med Mol Imaging 2013; 3:1–15. 83 van 't Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AA, Mao M et al. Gene 111 Nakasone ES, Askautrud HA, Kees T, Park JH, Plaks V, Ewald AJ et al. Imaging expression profiling predicts clinical outcome of breast cancer. Nature 2002; 415: tumor-stroma interactions during chemotherapy reveals contributions of the 530–536. microenvironment to resistance. Cancer Cell 2012; 21: 488–503. 84 Basanta D, Gatenby RA, Anderson AR. Exploiting evolution to treat drug resis- 112 Wang Y, Zhou K, Huang G, Hensley C, Huang X, Ma X et al. A nanoparticle-based tance: combination therapy and the double bind. Mol Pharm 2012; 9:914–921. strategy for the imaging of a broad range of tumours by nonlinear amplification 85 Reth M, Brummer T. Feedback regulation of lymphocyte signalling. Nat Rev of microenvironment signals. Nat Mater 2014; 13:204–212. Immunol 2004; 4:269–277. 113 Vakoc BJ, Lanning RM, Tyrrell JA, Padera TP, Bartlett LA, Stylianopoulos T et al. 86 Vogelstein B, Kinzler KW. Cancer genes and the pathways they control. Nat Med Three-dimensional microscopy of the tumor microenvironment in vivo using 2004; 10: 789–799. optical frequency domain imaging. Nat Med 2009; 15: 1219–1223. 87 Burstein HJ. The distinctive nature of HER2-positive breast cancers. N Engl J Med 114 Straussman R, Morikawa T, Shee K, Barzily-Rokni M, Qian ZR, Du J et al. Tumour 2005; 353: 1652–1654. micro-environment elicits innate resistance to RAF inhibitors through HGF 88 Davies H, Bignell GR, Cox C, Stephens P, Edkins S, Clegg S et al. Mutations of the secretion. Nature 2012; 487:500–504. BRAF gene in human cancer. Nature 2002; 417:949–954. 115 Kenny HA, Krausz T, Yamada SD, Lengyel E. Use of a novel 3D culture model to 89 Li J, Yen C, Liaw D, Podsypanina K, Bose S, Wang SI et al. PTEN, a putative protein elucidate the role of mesothelial cells, fibroblasts and extra-cellular matrices on tyrosine phosphatase gene mutated in human brain, breast, and adhesion and invasion of ovarian cancer cells to the omentum. Int J Cancer 2007; prostate cancer. Science 1997; 275:1943–1947. 121: 1463–1472. 90 Pierobon M, Wulfkuhle J, Liotta L, Petricoin E. Application of molecular tech- 116 Wu Y, Lu Y, Chen W, Fu J, Fan R. In silico experimentation of glioma micro- nologies for phosphoproteomic analysis of clinical samples. Oncogene 2015; 34: environment development and anti-tumor therapy. PLoS Comput Biol 2012; 8: 805–814. e1002355. 91 Olsen JV, Blagoev B, Gnad F, Macek B, Kumar C, Mortensen P et al. Global, in vivo, 117 Eikenberry S, Thalhauser C, Kuang Y. Tumor-immune interaction, surgical treat- and site-specific phosphorylation dynamics in signaling networks. Cell 2006; 127: ment, and cancer recurrence in a mathematical model of melanoma. PLoS 635–648. Comput Biol 2009; 5: e1000362.

Oncogene (2015) 3215 – 3225 © 2015 Macmillan Publishers Limited Cancer systems biology W Du and O Elemento 3225 118 Oved K, Eden E, Akerman M, Noy R, Wolchinsky R, Izhaki O et al. Predicting and metastatic breast cancer that overexpresses HER2. N Engl J Med 2001; 344: controlling the reactivity of immune cell populations against cancer. Mol Syst Biol 783–792. 2009; 5: 265. 130 Van Allen EM, Wagle N, Stojanov P, Perrin DL, Cibulskis K, Marlow S et al. 119 Venkatasubramanian R, Arenas RB, Henson MA, Forbes NS. Mechanistic Whole-exome sequencing and clinical interpretation of formalin-fixed, paraffin- modelling of dynamic MRI data predicts that tumour heterogeneity decreases embedded tumor samples to guide precision cancer medicine. Nat Med 2014; therapeutic response. Br J Cancer 2010; 103:486–497. 20:682–688. 120 Choe SC, Zhao G, Zhao Z, Rosenblatt JD, Cho HM, Shin SU et al. Model for in vivo 131 Barretina J, Caponigro G, Stransky N, Venkatesan K, Margolin AA, Kim S et al. The progression of tumors based on co-evolving cell population and vasculature. Sci Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug Rep 2011; 1: 31. sensitivity. Nature 2012; 483:603–607. 121 Harris TJ, McCormick F. The molecular pathology of cancer. Nat Rev Clin Oncol 132 Lamb J, Crawford ED, Peck D, Modell JW, Blat IC, Wrobel MJ et al. The Con- 2010; 7: 251–265. nectivity Map: using gene-expression signatures to connect small molecules, 122 Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell 2011; genes, and disease. Science 2006; 313: 1929–1935. 144:646–674. 133 Jahchan NS, Dudley JT, Mazur PK, Flores N, Yang D, Palmerton A et al. Adrug 123 Wistuba II, Gelovani JG, Jacoby JJ, Davis SE, Herbst RS. Methodological and repositioning approach identifies tricyclic antidepressants as inhibitors of small practical challenges for personalized cancer therapies. Nat Rev Clin Oncol 2011; cell lung cancer and other neuroendocrine tumors. Cancer Discov 2013; 3: 8:135–141. 1364–1377. 124 Piccart-Gebhart MJ, Procter M, Leyland-Jones B, Goldhirsch A, Untch M, Smith I 134 Chong CR, Janne PA. The quest to overcome resistance to EGFR-targeted et al. Trastuzumab after adjuvant chemotherapy in HER2-positive breast cancer. therapies in cancer. Nat Med 2013; 19: 1389–1400. N Engl J Med 2005; 353: 1659–1672. 135 Roychowdhury S, Iyer MK, Robinson DR, Lonigro RJ, Wu YM, Cao X et al. 125 Druker BJ, Talpaz M, Resta DJ, Peng B, Buchdunger E, Ford JM et al. Efficacy and Personalized oncology through integrative high-throughput sequencing: a safety of a specific inhibitor of the BCR-ABL tyrosine kinase in chronic myeloid pilot study. Sci Transl Med 2011; 3: 111ra121. leukemia. N Engl J Med 2001; 344:1031–1037. 136 Costello JC, Heiser LM, Georgii E, Gonen M, Menden MP, Wang NJ et al. 126 Maemondo M, Inoue A, Kobayashi K, Sugawara S, Oizumi S, Isobe H et al. A community effort to assess and improve drug sensitivity prediction algo- Gefitinib or chemotherapy for non-small-cell lung cancer with mutated EGFR. rithms. Nat Biotechnol 2014. N Engl J Med 2010; 362: 2380–2388. 137 Bild AH, Yao G, Chang JT, Wang Q, Potti A, Chasse D et al. Oncogenic pathway 127 Chapman PB, Hauschild A, Robert C, Haanen JB, Ascierto P, Larkin J et al. signatures in human cancers as a guide to targeted therapies. Nature 2006; 439: Improved survival with vemurafenib in melanoma with BRAF V600E mutation. 353–357. N Engl J Med 2011; 364: 2507–2516. 138 Wang K, Saito M, Bisikirska BC, Alvarez MJ, Lim WK, Rajbhandari P et al. Genome- 128 Druker BJ, Guilhot F, O'Brien SG, Gathmann I, Kantarjian H, Gattermann N et al. wide identification of post-translational modulators of transcription factor Five-year follow-up of patients receiving imatinib for chronic myeloid leukemia. activity in human B cells. Nat Biotechnol 2009; 27:829–839. N Engl J Med 2006; 355: 2408–2417. 139 Karr JR, Sanghvi JC, Macklin DN, Gutschow MV, Jacobs JM, Bolival B Jr. et al. A 129 Slamon DJ, Leyland-Jones B, Shak S, Fuchs H, Paton V, Bajamonde A et al. whole-cell computational model predicts phenotype from genotype. Cell 2012; Use of chemotherapy plus a monoclonal antibody against HER2 for 150: 389–401.