Despite many technological and therapeutic advances, cancer death rates are still considered high, mainly due to limitations of early diagnosis of many tumor types and low efficacy of available treatment options. Carcinogenesis is a stepwise process characterized by the accumulation of genetic and epigenetic aberrations that favor the initial formation of premalignant lesions that gradually progress to invasive carcinomas. Signaling molecules and network perturbations are the mediators of these events leading to carcinogenic features, such as uncontrolled proliferation, apoptosis escape, neoangiogenesis, migration, and metastasis. In the last decade, it has been shown that a wide gamut of signal transduction molecules and oncogenic transcription factors are implicated in human tumorigenesis. Most of these proteins contribute both to initiation and progression of carcinogenesis and to the development of resistance to currently used therapeutic strategies (eg, chemotherapy, hormonal therapy, molecularly-targeted agents, and radiotherapy). One major characteristic of cancer signaling cascades is the complex intracellular and extracellular cross-talk among different proteins and their interactions throughout space and time. Additional cellular events, such as posttranslational modifications (eg, phosphorylation, methylation, acetylation), single-nucleotide polymorphisms, and microRNA signatures profoundly augment the levels of complexity in the carcinogenesis-related networking “labyrinth.”[3-5]
The “chaotic” nature of signaling cascades implicated in various stages of carcinogenesis has necessitated the development of new techniques that will provide concrete clues to identify crucial molecules in each network, as well as the construction of mathematical models based on experimental results to achieve a quantification of the accumulated data regarding cellular changes and defects that characterize cancer cells.[6-8] Theoretically, such a combinatorial approach could facilitate effective and reliable diagnostic algorithms, identification of patient groups with increased cancer risk, categorization of molecularly-driven tumor subtypes, prediction of the clinical outcome after treatment with specific molecularly-targeted agents, and early detection of disease progression.
At present, the shortage of independent cancer research funding and typically high cancer treatment health care costs are emerging as major problems. The scientific cooperation of disciplines such as biochemistry, mathematics, bioinformatics, genetics, molecular biology, and pharmacology seems to be a promising venue in better understanding the cellular circuitry alterations that govern carcinogenesis as well as in identifying optimal molecular targets in a cost-effective way. The majority of defects in complex biological systems, such as malignant tumors, result not only from interactions within a specific level (eg, DNA) but also from crosslinking interconnections among different levels (eg, epigenetic defects, protein expression levels, transcriptional control). Therefore, the data collected by the highly sophisticated techniques of genomics, proteomics, and cellomics combined with biostatistics and bioinformatics analyses could create computational models that will enable the identification of novel therapeutic anticancer opportunities.
The Evolution of Proteomics in the Research of Signaling Cascades
Cellular signaling networks comprise proteins, substrates, and ligands that cooperate in a complex and well-orchestrated way. Accumulated aberrations in these molecules are being reported in many tumors and represent the most attractive candidates for novel anticancer regimens. To date, the available molecularly targeted agents in cancer therapeutics still lack the desired activity and selectivity, because the understanding of cellular events is limited. A major drawback is the lack of experimental techniques that can analyze these complex networks in real time. The development of antibodies that can target a large fraction of the proteome and detect posttranslational modifications has allowed the identification of a broad variety of signaling spots. Medium- and high-throughput screening methods specify many parallel signaling events, such as analyses based on phospho-flow cytometry and mass spectrometry, which have contributed to the assessment of whole signaling networks. The recent advances in chemical biology and chemical proteomics also provide a plethora of small-molecule compounds that can be used to study signaling pathway perturbations and further explore the changes that characterize a single cell as well as a cellular system level.
Among various posttranslational modifications engaged in cellular signaling, phosphorylation and dephosphorylation are likely the most prevalent and are vital in healthy and cancer cells; it has been estimated that approximately 30% to 40% of all human proteins are phosphorylated or dephosphorylated in a balanced way at any given time. Up to now, available biochemical methods have failed to provide adaptable and reliable measurements of these enzymatic activities throughout the evolution of signaling events. New-generation techniques are being developed to further illuminate in real time these dynamic processes.[9, 12, 13] For example, the incorporation of synthetic phospho-amino-acid mimics in predefined phosphorylation sites of target proteins allows the evaluation of certain molecules in crucial cellular networks. The design of novel, activity-based chemical probes can help to analyze the kinase/phosphatase activity equilibrium in complex biological systems. A major challenge is the elucidation of phosphorylation-driven signaling cascades in a continuous manner. Synthetic peptide-based biosensors as well as allele-specific kinase mutants are being generated and will provide a powerful tool in reporting phosphorylation status during signal transduction, after the application of a ligand and/or a potential inhibitor of molecular pathways in vitro and in vivo. New-generation microarray platforms using low-molecular-mass compounds have been developed, which allow the assessment of many enzymatic activities at one time and identify signaling and/or docking proteins that control the enhancement of the signal from cellular membrane to the nucleus.[14, 15] All these new techniques may offer powerful tools to better understand interactions in parallel signaling cascades as well as feedback loops that are built over time in the cellular proteome.
Protein functions in a cell or tissue can be affected by many factors, such as expression, localization, affinity, stability, interactions, catalytic activity, and posttranslational modifications. Therefore, proteomics represents a scientific challenge, because the available experimental tools are considered inadequate to tackle signaling molecule alterations in normal and abnormal (eg, cancer) situations. The enormous data acquired from novel techniques are entered into chemical bioinformatics databases and are used for the identification of novel targets and synthetic “targeted” compounds.[16, 17] However, the spatial and temporal plasticity of signaling cascades seems an inevitable obstacle, necessitating the creation of mathematical models as a prerequisite for better understanding and therefore real targeting of cellular networks.
The Additive Value of Signaling Cascade Computational Modeling
Mathematical modeling can have a pivotal role in cancer research by providing an independent evaluation of the compatibility of a generated hypothesis, and improving experimental design by depicting the steps and additional data that are needed. Computational models are governed by the simple notion that in living cells, not all combinations of molecular alterations are equal through time and space. Thus, computational techniques attempt to elucidate complex molecular interplays and also to comprehend their interactions in a structural and functional manner. Tumorigenesis is considered a multilevel dynamic process during which the accumulation of genetic and epigenetic defects results in changes of signaling cascades. Multiscale models of carcinogenesis include many biological parameters, such as signaling aberrations in tumor cells, protein–protein interactions, cell-to-cell interactions, changes in tumor microenvironment, and neoangiogenesis.[20-22] A wide variety of techniques have been developed for the accurate identification of molecular aberrations. Microarrays represent a major advance in cancer research, but the analysis of the obtained data is only feasible with computational techniques. Mathematical models are increasingly used to evaluate massive quantities of data and to study signaling cascades in cancer cells before and after the application of promising “targeted” agents.[24, 25] The major perspective of computational models seems to be their potential contribution to the personalized cancer treatment approach. Each patient could receive individualized treatment based on a specific profile of genetic/epigenetic changes and accompanied protein perturbations in signaling cascades.[12, 26]
Statistical models are usually divided into analytical and stochastic ones. Analytical models use different assumptions and experimental data in order to better identify tumor-cell kinetics. The building of such a mathematical model facilitates the analysis of a molecular pathway's response to various kinds of ligands and/or potential inhibitors. By contrast, stochastic models exploit random parameters and probability distributions to study the evolution of a complex biological system based on a set of initial conditions. The in silico simulation of the effect of various molecules on cellular signaling dynamics mainly depends on the applied affecting conditions. Statistical tools are usually employed in large biomolecular databases in order to create models that will further elucidate the complex relationships between molecules and signaling networks in healthy versus cancerous cells and/or tissues before and after the application of various agents. Mechanistic modeling is a different computational approach that enables the study of the physical interaction among molecules, change of their concentrations, biochemical nature and rate of their interaction, parameters that control this interaction, and how their concentrations change over time in a predefined cellular scenery. At the signaling network level, the number of states in mechanistic models is typically smaller than in statistical ones. This is because it can be difficult to measure certain states and/or reaction rates experimentally, therefore it is necessary to make assumptions about which biological processes and components should be included in the mechanistic models. Cancer is deemed a multiscale disease, and mechanistic models taking into account most of these scales are being developed and applied to predict tumor-cell behavior.
Results generated from microarray and other high-throughput technologies have been inserted in large databases and used to identify network-based signatures. Nonetheless, proceeding from a number of implicated molecules to a certain cancer-related pivotal signature requires further annotation and simulation tools. Proteins that have been captured as possibly important are searched in pathway databases, and a signaling network is then constructed and analyzed. Computational modeling allows the quantitative assessment of the qualitative process, the interpretation of the results from in vivo experiments and early-phase clinical trials, and the reduction of the cost of setting up very sophisticated, and therefore expensive, techniques. However, at least for the moment, these models can evaluate a limited number of biological parameters, and their results should be validated in the laboratory before they can be considered valuable for clinical testing. Malignant tumors are characterized by a variety of defects in signaling pathways that can be detected and quantified through next-generation technologies.[30-33] Novel “targeted” agents have been produced and used in everyday clinical practice, on the basis of rather simple, albeit quite sensitive, predictive assumptions with improved clinical outcomes compared to the sole use of traditional treatment options. Although simplicity does not necessarily limit performance, the clinical use of “targeted” agents revealed the complex questions that need to be addressed in order to optimize the cost-effectiveness of the currently used, as well as upcoming “targeted” anticancer compounds. The best example to date is epidermal growth factor receptor (EGFR) protein targeting in cancer therapeutics.
EGFR Protein Signaling Cascades as a Model
The EGFR family represents a transmembrane receptor protein group with cytoplasmic tyrosine kinase (TK) activity. It consists of 4 structurally related members (ERBB1 through ERBB4) and, with the exception of ERBB2 and/or truncated receptors, ligands must be attached to their extracellular domain in order to be functional. Upon ligand binding, conformational changes lead to dimerization and potentiation of their TK activity. ERBB receptors trigger downstream intracellular pathways, including phosphatidylinositide-3 kinase (PI3K)/Akt, Ras/mitogen-activated protein kinase (MAPK), phospholipase C (PLC)γ1/protein kinase C (PKC), signal transducer and activator of transcription (STAT), and Par6-atypical PKC pathways. These pathways are involved in cellular functions such as inhibition of apoptosis, progression of proliferation, differentiation, angiogenesis, metastasis, epithelial–mesenchymal transition, and cell motility. Proper regulation of these signaling networks is a prerequisite for cell homeostasis. Deregulation and subsequent aberrant signaling due to mutation, amplification, and/or presence of autocrine loops contributes to the development of carcinomas. Several strategies are being developed to disrupt EGFR-initiated signal transduction pathways. Among them, anti-EGFR monoclonal antibodies (mAbs) as well as EGFR tyrosine kinase inhibitors (TKIs) have already been used in cancer therapeutics.[35-37] Somatic mutations and other molecular parameters are being evaluated for treatment individualization and optimal patient selection for anti-EGFR therapy (eg, K-Ras mutations in colorectal carcinomas, EGFR mutations in non–small cell lung carcinomas).[38, 39] In addition, many of the downstream elements in EGFR protein family networks are commonly mutated in solid tumors. These genetic alterations are thought to be responsible for eliciting aberrant, constitutive growth signals and are currently “targeted” with various inhibitors. For example, Raf and MAPK kinase (MEK) inhibitors represent one of the latest advances in the treatment of patients with metastatic melanoma. Moreover, several inherent or acquired mechanisms of resistance to EGFR protein family receptor inhibitors have been identified based on the concept of signal transduction cross-talk and feedback loop development, either at the cellular membrane or at downstream levels.[41-43] However, the activation of a protein is more complicated than the predicted alterations in standard network diagrams, because it develops many intracellular interactions through space and time. For instance, targeting Raf kinases has been shown to potentiate upstream K-Ras signaling, possibly through the activation of autocrine loops.[44-46]
It is becoming increasingly clear that the response assessment of current “targeted” therapies in cancer patients and the identification of novel optimal targets should always consider signal transduction networks as multiscale systems. Therefore, mathematical and computational approaches in combination with advances in biochemical techniques could offer new opportunities for the mechanistic understanding of signaling cascades that are important in certain carcinomas as well as the alterations of the participating molecules under different conditions (Fig. 1). These tools can simulate and track changes in concentrations and functional state of many proteins concurrently in response to various stimuli of EGFR protein family pathways (Fig. 1). In silico modeling of cell signaling and transcriptional control has been already applied to the EGFR protein family and represents a rational approach for next-generation cancer therapeutics.[14, 47]
Based on ERBB2 overexpression, patients with breast cancer currently follow treatment with anti-ERBB2 agents, such as trastuzumab (a mAb against the extracellular domain), pertuzumab (a mAb that binds to the dimerization domain of ERBB2 with ERBB3), or lapatinib, which is a TKI. Many patients with breast cancer develop resistance to ERBB2-directed therapies, and growing evidence supports the participation of ERBB3 and its ligand heregulin (HRG) in this process. Dimerization of ERBB2 with ERBB3 and/or ERBB4 can activate the TK domain of ERBB2 and downstream signaling cascades that promote proliferation and inhibit apoptosis, such as PI3K/AKT and MAPK pathways. Thus, the dimerization pattern rather than the expression of each member of EGFR protein family seems to be of greater importance in further revealing the downstream events in breast carcinomas.[50, 51] This is also a critical point in cross-talk interactions among EGFR protein family members and other membrane-initiated signaling cascades (eg, insulin-like growth factor-1 receptor, Met) that represent well-recognized resistance mechanisms to ERBB-directed agents.[52, 53]
Recently, a kinetic model of the entire ERBB signaling network has been developed, and sensitivity analysis was used to identify optimal targeting molecules for ERBB3. Using this information, a novel, fully human mAb (MM-121) was found to inhibit with high affinity this EGFR protein family member. Preclinical and early clinical activity of this computationally designed compound is promising in ERBB2-dependent breast carcinomas, but also in other carcinomas.[55, 56] Computational modeling in conjunction with next-generation proteomics can also be used to identify the dose–response activity of the combination of EGFR protein family receptor inhibitors and radiotherapy, as well as to better delineate the defects that characterize or impair the downstream kinome after the application of certain TKIs.[58-60]