Systems Approaches to Cancer Biology

Cancer systems biology (CSB) recognizes that many individual disciplines and data types can be usefully brought to bear, alone or in combination, to systematically study cancer. The diversity of interactions among cancer systems biologists, who come from fields such as cancer biology, biochemistry, bioinformatics, engineering, mathematics, physics, and computer science, can lead to novel approaches to the fundamental challenges within the field of cancer research. The inaugural Systems Approaches to Cancer Biology Conference highlighted the important biological insights gained through these synergistic interactions and discussed the unique challenges faced by multidisciplinary investigators. The meeting was organized and purposefully populated by mostly early career tenure-track and junior investigators in training, offering a unique perspective on the emerging field of cancer systems biology. The conference was held in Woods Hole, MA on April 3–6, 2016, and was chaired by field pioneers Elaine Mardis (Washington University, St. Louis, MO), Joe Gray (Oregon Health Sciences University, Portland, OR), and Douglas Lauffenburger [Massachusetts Institute of Technology (MIT), Cambridge, MA].

Central to the meeting was the question of what exactly constitutes CSB, especially in comparison to quantitative or computational work within other fields. Through the diversity of approaches discussed, it became clear that the field is not defined by a particular set of methods, nor simply by the application of computational methods to cancer data. Rather, the field is defined by a recognition that cancer is a dynamic, multifactorial, and complex process that must be understood using both experimental methods and analytical approaches that bridge traditional disciplinary boundaries to directly address these concomitant challenges.

In his opening remarks to the meeting, Douglas Lauffenburger made the observation that CSB is a field for those who actively embrace the complexity of biology. A central component of this complexity, the heterogeneity of cancer within and across patients, is undeniably a barrier to accurate diagnosis and treatment. CSB is uniquely poised to address this heterogeneity through application of computational analyses that glean insights from preclinical and clinical datasets. For example, Andrea Bild (University of Utah, Salt Lake City, UT) presented on the striking temporal evolution of tumor heterogeneity measured by whole-genome sequencing in an individual breast cancer patient during the course of their various treatment regimens. Her study underscored the necessity of understanding subclonal evolution within a single patient and how this might influence treatment options and decisions. In a related longitudinal genomic and transcriptomic analysis of patients with glioblastoma, Jiguang Wang (Columbia University, New York, NY) reported that upon treatment, many patients experience a loss of the dominant driver mutation but gain many novel mutations at relapse that change their disease subtype. Wendy Fantl (Stanford University, Stanford, CA) showed that intra- and intertumoral heterogeneity is also reflected at the protein level and that increased tumor cell diversity measured with mass cytometry (CyTOF) is related to aggressive disease in ovarian cancer. As this and other single-cell studies are finding, the persistence of intratumor heterogeneity makes it important to develop analytical and experimental tools that can predict the best treatment options. John Paul Shen (University of California, San Diego, La Jolla, CA) demonstrated that yeast can be used as a model system for identifying synthetic lethalities between inactivated tumor suppressor genes and genes that encode druggable proteins. Laura Heiser compared -omics data from a panel of breast cancer cell lines against publicly available molecular profiling data from primary tumors in The Cancer Genome Atlas to identify conserved pathways and gene sets recurrently aberrant in different breast cancer subtypes. She described how the cell lines recapitulate major subtypes of breast cancer and can be used to identify distinct therapeutic vulnerabilities across subsets of the cell lines (1, 2). Coupling the knowledge gained from profiling tumor heterogeneity with an arsenal of computationally predicted treatment options may lead to the realization of precision cancer treatments.

CSB is driving numerous technological developments to better study the complex and dynamic nature of cancer. Frank Stegmeier (KSQ Therapeutics) compared large-scale shRNA-mediated knockdown to CRISPR-CAS9 knockout technology, highlighting the benefits and drawbacks of both approaches. CRISPR-based knockout screens identified more synthetic lethal genes compared with RNAi across a panel of cancer cell lines, implying that the identification of cellular dependencies may require full gene inactivation. On a related topic, Kevin Janes (University of Virginia, Charlottesville, VA) presented an elegant mathematical model illustrating that the network perturbation effects of gene knockdown versus chemical inhibition of protein activity encoded by the same gene can be vastly different. The difference in effect depended intimately on the network context of the molecule and was nonlinearly related to the extent of gene knockdown (3). New experimental systems were described by Yvonne Chen (University of California, Los Angeles, Los Angeles, CA), who described the systems-level design and synthetic biology implementation of an ‘OR-gate' T-cell receptor intended to target heterogeneous tumor populations and prevent drug resistance caused by tumor cell antigen loss (4), and Shelly Peyton (University of Massachusetts, Amherst, Amherst, MA), who presented an in vitro biomaterial platform that enabled systematic comparison of various microenvironments as sites for potential metastases (5). These novel experimental approaches will fuel the collection of robust and reproducible data sets that can then be coupled with computational analysis to provide an essential step forward toward systems-level understanding of the complex tumor ecosystem.

Computational tools can aid in understanding the complexity of cancer by integrating new datasets with prior knowledge. Ben Raphael (Brown University, Providence, RI) described the HotNet2 algorithm, which integrates cancer mutational data with known protein interactions (6). His group's analysis demonstrated that while there are only a relatively small number of bona fide cancer driver mutations, many lower incidence mutations cluster within signaling pathways that may provide mechanistic insight to their role in progression. Indeed, the phenotypic consequence of most detectable mutations is unknown. To this end, Pau Creixell (MIT) suggested that in the case of protein kinases, our understanding of individual cancers might be aided by characterizing the specific functional consequences of somatic mutations. By analyzing the exomes and phospho proteomes of ovarian cancer cell lines using the ReKINect computational platform, he demonstrated that individual kinase mutations in cancer variously affect the activity and specificity of important mutated kinases (7). Taken together, the Raphael and Creixell studies suggested that integrating protein-level data with mutational information can be a powerful approach but that caution must be taken when assuming the consequence of a mutation with respect to protein activity. For example, in both studies, mutations were identified that were neither recurrent nor simply activating/inactivating, yet were functional through their network effects.

In some cases, the scale and complexity of systems-level studies make reproducibility a challenge, especially when connecting preclinical and clinical research (8). Executing reproducible research requires the ability to recapitulate data analysis and laboratory protocols. The development of software protocols has provided the means to automatically document analyses for purposes of reproducibility (9). For example, Ben Raphael showcased his network algorithms by providing direct links to the source code, available on GitHub, and making all data available on Synapse (10). Trey Ideker (University of California, San Diego) described NDEx, a new resource to facilitate the sharing of networks derived from biologic data to make these studies more reproducible (11). Data sharing is an essential part of recapitulating and building on prior work and thus is a critical aspect of reproducible research. Building new models from existing data can additionally lead to novel findings, as showcased in the meeting by Stacey Finley (University of Southern California, Los Angeles, CA), who presented a mathematical model to predict levels of angiogenic factors during tumor treatment, and Jorge Zanudo (Penn State University, State College, PA) who presented a mathematical model to predict activated signaling pathways during the epithelial-to-mesenchymal transition in cancer. Each of these models relied on parameters from open data in publications that span years of research, scientific disciplines, and experimental systems (12–14). Many of the present works took advantage of open-access datasets such as ICGC or TCGA as well as more modestly sized datasets collected through collaborations or gleaned from the literature tailored to their studies. With this in mind, closing keynote speaker Gordon Mills dubbed the conference community the “Society for Data Parasites” in support of data sharing as a means to accelerate the study of cancer.

During his opening comments, Douglas Lauffenburger pointed out that the results of CSB must ultimately improve patient care. Toward this goal, the Systems Approaches to Cancer Biology meeting included a number of studies with a strong translational focus. Galit Lahav (Harvard Medical School, Cambridge, MA), the opening keynote speaker, showed how oscillating behavior in p53, a key tumor suppressor, can give rise to dynamic transcriptional and phenotypic behavior at the single-cell level that is often masked in bulk-population analyses (15). This has directly led to recommendations for timing of treatments involving DNA-damaging therapies such as cisplatin (16) and radiation (15). Jiyang Yu (Pfizer) used an shRNA screen to identify HDAC6 as a potential target for inflammatory breast cancer (IBC; ref. 17). HDAC6 is not itself an oncogene in IBC, but Yu identified it as a key regulator of other genes involved in IBC proliferation. Mohammed Shahrok Espahani (Stanford University) described cancer personalized profiling by deep sequencing (CAPP-seq), a method for targeted sequencing of circulating tumor DNA (ctDNA) that may enable “liquid biopsies” to track cancer progression. Raghu Kalluri (The University of Texas MD Anderson Cancer Center, Houston, TX) showed that exosomes in fact can contain DNA and can provide ctDNA with reduced contamination from nontumor cells. The systems approaches taken in these studies yield clinical impact that go beyond what traditional studies based on single genomic mutations or amplification events can uncover.

Progress in translational research often depends on collecting multiple types of molecular data from the same tumor samples to enable comprehensive, systems-level studies. Early efforts to understand breast cancer at a systems level applied hierarchical clustering methods to gene expression data to identify clinically relevant subtypes (18). Anne-Lise Børresen-Dale (Oslo University Hospital, Oslo, Norway) described her team's efforts to extend this work, which went beyond clustering of transcriptomic data to also include other types of molecular data and to provide additional biologic insights about these breast cancer subtypes (19). Interestingly, this analysis identified different subtypes, depending on the measurement (e.g. mRNA levels, somatic mutations, etc.) used in the analysis. This observation that subtype clusters are not uniquely defined was further supported by Kevin Brennan (Stanford University), who found that DNA methylation profiles segregate head and neck squamous cell carcinoma by etiologic factors.

To collect, analyze, and interpret multidimensional datasets, multi-institutional collaborations are often crucial. Over the past several decades, Børresen-Dale has helped assemble a large group of international collaborators to derive insights about the interplay among genomic, transcriptomic, proteomic, and metabolomic features in breast cancer. For example, they examined the transcriptional consequences of somatic mutations and found that tumor cells express these mutations more actively than stromal cells and that this relationship depends on estrogen receptor status (20). As noted by Gordon Mills, greater collaboration across institutions and between academia and industry should help accelerate clinical translation of CSB studies.

There has been significant progress in characterizing the genetics of cancer and the contributions of individual intracellular pathways to tumor initiation and progression. Future progress in understanding fundamental cancer biology will continue to extend the range of hallmarks available for systems-level interrogation. For example, Isaac Harris (Harvard Medical School) described efforts to identify therapeutic sensitivities tied to the altered redox and metabolic state of breast carcinoma cells. The challenges faced when interpreting single-parameter studies highlight the need to understand cancer as an integrated system. The emerging success, and limitations, of cancer immunotherapy illustrate this point well, where simultaneous understanding of the communication networks within and between cell types in the tumor microenvironment will be essential for extending the success of this treatment modality. As an example of how to begin addressing this multifactorial environment, Molly Carroll (University of Wisconsin-Madison, Madison, WI) applied machine learning to identify ovarian cancer–released cytokines that modulate macrophage M1/M2 phenotype. On the other hand, Annelien Zweemer (MIT) showed that factors such as apoptotic cell debris can act as microenvironmental cues through tumor-associated macrophage receptor signaling. Clearly, the complexity of cancer requires a multipronged approach that may only be fully realized when bringing together disparate academic disciplines as is common in systems biology.

At the same time, technological and methodological advancements will place systems-level approaches at the forefront of translating these for clinical benefit. Single-cell technologies are providing ever-finer quantitation of tumor heterogeneity demanding improved and alternative methods to visualize and therapeutically address this complexity. On the scale of individuals, genome- and proteome-scale clinical diagnostic capabilities are presenting new challenges to prognostically identify effective therapies for individual patients. Realizing this goal, beginning to be addressed clinically through precision trials such as NCI-MATCH and ASCO TAPUR, will require new approaches integrating outside data and knowledge as well as new data such as more longitudinal measurements. The spirited discussions about the scientific challenges in cancer, the importance of building and maintaining strong collaborations, the need to improve data access, and the importance of reproducibility in systems analysis that occurred at the Systems Approaches to Cancer Biology meeting made it clear that participants in the interdisciplinary field of cancer systems biology are well-poised to improve our understanding of cancer and how to overcome it.

The program committee consisted of the co-chairs Joe Gray (Oregon Health Sciences University), Douglas Lauffenburger (MIT), and Elaine Mardis (Washington University). The meeting planning committee of the Association of Early Career Cancer Systems Biologists (AECCSB) organized the SACB meeting and included Sara Gosline (Sage Bionetworks), Marc Hafner (Harvard Medical School), Shannon Hughes (NCI, NIH), Brian Joughin (MIT), Aaron Meyer (MIT), Stephen Piccolo (Brigham Young University), and Erin Wetzel (NCI, NIH). In addition to the program committee members, invited participants included Andrea Bild (University of Utah), Anne-Lise Børresen-Dale (Oslo University Hospital), Yvonne Chen (University of California, Los Angeles), Dan Gallahan (NCI, NIH), Trey Ideker (University of California, San Diego), Kevin Janes (University of Virginia), Raghu Kalluri (The University of Texas MD Anderson Cancer Center), Galit Lahav (Harvard Medical School), Gordon B. Mills (University of Texas), Shelly Peyton (University of Massachusetts, Amherst), Ben Raphael (Brown University), and Frank Stegmeier (KSQ Therapeutics).

No potential conflicts of interest were disclosed.