How implementation of systems biology into clinical trials accelerates understanding of diseases

Abstract

Systems biology comprises a series of concepts and approaches that have been used successfully both to delineate novel biological mechanisms and to drive translational advances. The goal of systems biology is to re-integrate putatively critical elements extracted from multi-modality datasets in order to understand how interactions among multiple components form functional networks at the organism/patient-level, and how dysfunction of these networks underlies a particular disease. Due to the genetic and environmental diversity of human subjects, identification of critical elements related to a particular disease process from cross-sectional studies requires prohibitively large cohorts. Alternatively, implementation of systems biology principles to interventional clinical trials represents a unique opportunity to gain predictive understanding of complex diseases in comparatively small cohorts of patients. This paper reviews systems biology principles applicable to translational research, focusing on lessons from systems approaches to inflammation applied to multiple sclerosis. We suggest that employing systems biology methods in the design and execution of biomarker-supported, proof-of-principle clinical trials provides a singular opportunity to merge therapeutic development with a basic understanding of disease processes. The ultimate goal is to develop predictive computational models of the disease, which will revolutionize diagnostic process and provide mechanistic understanding necessary for personalized therapeutic approaches. Added, biologically meaningful information can be derived from diagnostic tests, if they are interpreted in functional relationships, rather than as independent measurements. Such systems biology based diagnostics will transform disease taxonomies from phenotypical to molecular and will allow physicians to select optimal therapeutic regimens for individual patients.

Keywords: clinical trials; clinical trials methodology; multiple sclerosis; polygenic diseases; systems biology.

Figure 1

Simplified animal models versus complex biological systems exemplified by human polygenic diseases. (A) Reductionist (linear) research model (e.g., experimental autoimmune encephalomyelitis): current animal studies are almost exclusively performed in a single animal species of a single genetic strain (usually the one that is susceptible to induction of the disease). Furthermore, the animals are housed in the same (often pathogen-free) environment; they are exposed to identical food and identical environmental stimuli, which leads to synchronization of circadian rhythms, similar levels of activity, etc. Disease is induced by identical regimens applied in a highly synchronized manner to animals of the same age and often only of single sex. Therefore, animal experiments utilize highly simplified input (input 1). Despite standardized input, the outcome is usually somewhat heterogeneous (outcome 1 and 2), but an application of traditional statistical methods leads to clear conclusions. These conclusions are often readily generalized across species and across diverse environmental inputs and disease triggers. (B) Humans and other non-artificial complex biological systems: measurements in the complex biological system exemplified by a human being affected by a disease are the results of multiple different inputs (i.e., an outbred genetic background, many environmental influences such as type, dose, and virulence of an infectious agent, diverse food, premorbidities, and drug regimens influencing both the metabolome and the microbiome, varied endocrine regulations resulting from circadian and reproductive rhythms and aging). The organism processes these varied inputs, utilizing complex decision-making mechanisms and the outcomes are also diverse (e.g., maintenance of heath or development of the disease of varied severity). Furthermore, the outcomes are processed by the organism as additional inputs through constant ubiquitous feedback loops, leading to dynamic changes of behavior. Current statistical methods are largely inadequate for analysis of such complex datasets. As a consequence, frequently no reproducible conclusions are reached. (C) Linearity assumption of reductionist approach: reductionist research methods are based on assumption that if an element (e.g., gene, its transcript or protein; Element A) is linked to a disease process (output X), then an observed or induced change in the element has to be reflected in the change in the output. (D) Non-linear behavior of biological systems: the existence of functional networks consisting of interacting elements with partially overlapping functions allows biological systems to retain a given output in a normal (non-disease) state if only one or few elements have been altered (this is called robustness of the system). For example, an alteration in Element A was compensated by a reciprocal change in Element B (State 2). However, despite maintenance of the same output, this new state is different from the original state (healthy; State 1), even though phenotypically there is no evidence of a disease. Such an altered state of the system is more fragile, insofar as any subsequent change in Element B may now cause a dramatic change in the output X (i.e., disease state; State 3).

Figure 2

Organizational hierarchy of the human body affected by the disease (example of MS) and its simplified model. Schematic illustration of the organizational layers (depicted by different colors) of human body affected by MS. X axis depicts relative ease (left) versus difficulty (right) of obtaining high-quality quantifiable data within each organizational plane. The panel to the left of the organizational chart represents different methodologies utilized by systems biology: “bottom-up approach” uses genomic data to predict disease occurrence or severity. “top-down” divides patients into diagnostic categories and analyzes disease-specific differences in the transcriptome by expression profiling of accessible tissues, such as blood; “middle-out” approach (probably the least utilized, while most useful) collects and analyzes data within one organizational plane and then expands data gathering vertically in both directions: e.g., one can start by analyzing data on the cellular level and expand these analyses downward by expression profiling of studied cells and upward by pathological studies of the affected tissues. The panel to the right of the organizational chart represents an idealized model of the studied system that emerges from a systems biology approach. In this simplified model, each organizational level is represented by limited number of essential elements, which are derived from multiple different studies (e.g., simplified in vitro models, including those derived from iPS cells, organotypic cultures, animal models and finally, human observational, and interventional studies) that provide information that is relevant to the particular disease entity, in our case MS. The interactions (protocols) among the essential elements are schematically depicted as different types of connecting lines (with the understanding that protocols represent complex behaviors that go beyond currently utilized positive or negative interactions). For example, on the genomic (yellow) level, essential elements are known MS susceptibility alleles, as well as yet unknown genetic variants that modulate MS disease activity or phenotype. On cellular/tissue level (green), essential elements are all cells of the immune system that mediate damage (e.g., cytotoxic T cells) or promote repair (e.g., regulatory immune cells, alternatively activated macrophages), as well as cells of the CNS that are the target of inflammation (e.g., oligodendrocytes and myelin) or the source of healing (e.g., oligodendroglial precursors in remyelination and neural stem cells in adult neurogenesis). Thus, each organizational level can be said to represent a “system.” There may be further subsystems within organizational levels, such as, e.g., myelination within cellular/tissue system. Yet, there are also vertical interactions among individual organizational levels, making a disease like MS a “complex system” as defined by Mesarovic et al. (24). For example, the genetic background of the individual together with acquired epigenetic changes and microbial influences present at the time of disease induction determine the severity of intrathecal inflammation, the ability of CNS tissue to survive inflammatory insults and the forms and levels of repair. The essential elements at all hierarchical levels form an “in vivo functional network” and the interactions among these elements capture the essence of the information processing and output of this functional network [i.e., different protocols or scripts regulating behavior of each subsystem and the interactions between them (36)]. The model itself is constructed by integration of data analyses from different studies of the disease process, as well as varied physiological processes (e.g., MS in its entire evolution from early relapsing-remitting disease to late secondary-progressive MS, but also studies of developmental myelination or basic immune regulation). Until a complete understanding of a disease process is achieved, the corresponding model always represents a “work in-progress.” Models develop as each new study adds or validates information about essential elements or about protocols within individual subsystems or about interactions between subsystems that affect behavior of the system. The value of this simplified model resides in its ability to predict the behavior of the biological system based on the input data. The validity of the model has to be determined by repeated confirmation of model-based predictions in the studied biological system in vivo (upper red arrow). Paradoxically, discordant observations are the most valuable for guiding further efforts to adjust the computational or conceptual model (lower red arrow), leading to an increasingly better understanding of the system.

Figure 3

Utilization of clinical trials with associated biomarker studies to increase understanding of the biological system. Clinical trials represent a unique opportunity for employing systems biology research methods. Data are obtained on multiple organizational levels before and after treatment from the same set of patients. Subtraction analysis (i.e., comparing within each subject data collected before and at specified time-intervals after initiation of treatment and considering those markers that change synchronously in subjects sharing the same treatment allocation) then identifies those changes that are direct, or, more often, indirect consequences of applied therapy. While applied therapy affects some elements directly (for example rituximab directly depletes B cells), this primary effect induces multiple secondary effects, as the system “adjusts” to the induced change. For example, inhibition of certain T cell functions under rituximab therapy would “alert” the investigators to the fact that T and B cells normally interact in vivo in mediating antigen-specific immunity, because the lack of B cells has to directly or indirectly explain the observed functional deficit in T cells, as T cells do not express CD20, the target of rituximab. Of course, it was known before initiation of rituximab trials that T and B cells interact together, but by analyzing samples from clinical trials of daclizumab we gained new insights into elements of the immune system that had previously been obscure such as innate lymphoid cells and their unexpected role in the MS disease process. Elements that are changing synchronously by applied therapeutic perturbation are part of the same “in vivo functional network” that has been disturbed by treatment. Identification of gene transcripts, proteins, and cell types that interact with each other in vivo then greatly facilitates investigation of interactions between these elements in simplified models, such as in vitro functional assays, ex vivo signaling assays, or in vitro pharmacology studies. Mechanistic insight (i.e., identified protocols) gained from these simplified models has to be verified in vivo, either by studying a new set of patients or by applying new therapies.