Translational systems biology of inflammation and healing

Authors

  • Yoram Vodovotz PhD

    1. Department of Surgery; Center for Inflammation and Regenerative Modeling, McGowan Institute for Regenerative Medicine, University of Pittsburgh, Pittsburgh, Pennsylvania
    Search for more papers by this author

  • This material was presented as part of the Keynote Address given by Dr. Vodovotz at the 2008 Annual Meeting of the Wound Healing Society, Symposium for Advanced Wound Care-Wound Healing Society Joint Meeting, San Diego, CA, April 24–27, 2008.

Reprint requests:
Yoram Vodovotz, PhD, Department of Surgery, University of Pittsburgh, W944 Starzl Biomedical Sciences Tower, 200 Lothrop St., Pittsburgh, PA 15213. Tel: +1 412 647 5609;
Fax: +1 412 383 5946;
Email: vodovotzy@upmc.edu

ABSTRACT

Personalized medicine is a major goal for the future of healthcare, and we suggest that computational simulations are necessary in order to achieve it. Inflammatory diseases, both acute and chronic, represent an area in which personalized medicine is especially needed, given the high level of individual variability that characterizes these diseases. We have created such simulations, and have used them to gain basic insights into the inflammatory response under baseline, gene-knockout, and drug-treated experimental animals; for in silico experiments and clinical trials in sepsis, trauma, and wound healing; and to create patient-specific simulations in polytrauma, traumatic brain injury, and vocal fold inflammation. Since they include both circulating and tissue-level inflammatory mediators, these simulations transcend typical cytokine networks by associating inflammatory processes with tissue/organ damage via tissue damage/dysfunction. We suggest that computational simulations are the cornerstone of Translational Systems Biology approaches for inflammatory diseases.

SYSTEMS BIOLOGY IN INFLAMMATION AND WOUND HEALING: READY FOR PRIME TIME

The constellation of diseases that we face in modern industrialized societies (atherosclerosis, metabolic syndrome, rheumatoid arthritis, sepsis, cancer, chronic nonhealing wounds, and many others) appears bewildering at first glance. Is there anything at all that connects these diseases besides the fact that they often occur together in the patients that populate our healthcare system? This article will make the case that inflammation is the unifying factor that binds these diseases together. And yet, inflammation also underlies the beneficial effects of exercise and rehabilitation.

Indeed, inflammation is different things to different people. It may be argued that inflammation is a communication network, not an inherently detrimental process or set of processes. It is the way the body tells itself that something has changed, either from without or from within. Inflammatory pathways are evolutionarily conserved, and it is not surprising that these pathways would be complex, redundant and interconnected, which is what makes it difficult to try to intervene in any disease having inflammation at its core.1 Thus, though a well-regulated inflammatory response is necessary for proper healing and regeneration, deranged and self-sustaining inflammation may drive the pathology seen in settings such as trauma, sepsis, and chronic, nonhealing wounds. A central thesis of this paper is that these processes are interrelated, and systems biology approaches can highlight those interrelations.

There has been much promise in the systems biology field as applied to health and disease.2–7 These processes are, if studied to the nth degree, hopelessly complex.7 However, what is needed is not just the systems approach (e.g., mechanistic computational simulations and “omic” data), but also the discovery of mechanisms that result in the patterns observed in such “omic” data. Moreover, what is sorely lacking is a clear translational focus, defined as the following:

  • Facing up to the messy clinical reality.
  • Using existing systems biology methods and developing new ones, based on clinically obtainable data and with clinically relevant goals.
  • Answering the question: “Is a computational model good enough to improve some aspect of clinical practice?”
  • Structuring systems biology studies in animals and cells based on the clinical reality, and with the goal of near-term clinical validation.

This manuscript will present the case that to some degree or another, various groups are addressing the above issues, and that progress in this nascent field has reached a point that can be described as “ready for prime time.”

A SYSTEMS APPROACH TO INFLAMMATION

There are multiple pieces of the inflammation puzzle (e.g., innate immunity, coagulation, etc.), some of which have been studied in relation to wound healing or other chronic processes. The problem lies in the fact that there is nothing inherently systems-oriented or that conveys any-sense of process from this kind of a static picture. To go from this potentially daunting static picture of complexity to a systems perspective, one needs to determine how the various elements interact.

Inflammation is set in motion due to various possible insults, which in modeling parlance are referred to as “initial conditions”; these different initial conditions can, in many settings, lead to diverse outcomes even if all the remaining components in a system do not change any of their other interrelationships.8 The response to infection is probably where inflammation is triggered to the largest degree in the shortest period of time. When the initial condition is a bacterial infection, the ensuing response is commonly known as a proinflammatory response or, more correctly, a T helper type-1 (Th1) response. In this setting, inflammatory mediators typically act to kill the invading bacteria. However, the organism pays a price for that bactericidal effect, and the price is bystander tissue damage, or at times dysfunction in the absence of histologically overt cell damage. These damaged or dysfunctional tissues will release so-called “alarm/danger signals” (also known as “Damage-associated Molecular Pattern” [DAMP] molecules), which play housekeeping roles during normal cellular physiology but upon the onset of cellular stress are released from normal tissues and restimulate inflammation.9,10 Therefore, a patient with bacterial sepsis may exist in a state in which most or all of the infecting bacteria have been killed, but there is still continuous and ongoing inflammation maintained by a vicious cycle of damage→inflammation→damage. In complex systems parlance, this cycle is known as a “positive feedback loop.” Simultaneously with this Th1 response, antiinflammatory (or, more properly, alternative-inflammatory Th2 influences such as those that characterize M2 macrophages) are set in motion. Separately, Th1 cytokines can stimulate Th2 cytokines. Th2 cytokines often act to suppress Th1 cytokines, forming a negative feedback loop. Importantly, some of these same Th2 cytokines are also the ones the body uses for healing. The cytokine transforming growth factor-β1 (TGF-β1) is a prototype for this process, but many other cytokines can do so as well. The problem for sepsis patients in the intensive care unit is that the Th1 and Th2 influences are not in balance. If they were in balance, the system would resolve and return back to health. However, because these influences are not in balance, the patient remains in state of long-term damage/dysfunction, maintained by the positive feedback loop of inflammation→damage→inflammation.8

The response to trauma is similar. Physical injury clearly damages tissues directly. It is now appreciated that DAMPs are secreted from this damaged tissue. Catecholamines and other agents can set in motion the Th2 response, and the same kind of interrelationship as described above for infection-induced inflammation occurs in this setting as well. Again, there is an attempt to heal and there is typically a shift to Th2 cytokines following trauma. However, as in the case of severe infection illustrated previously, often the response is out of balance, and the patient remains in a state of continuously maintained inflammation that is driven by DAMPs. In the setting of chronic injury (or low-level, so-called “para-inflammation”),11 as well as chronic, non-healing wounds, the same types of interactions occur. In the setting of allergy, asthma, and related chronic inflammatory diseases, Th2 cytokines predominate. Yet, the disease does not resolve because again these forces are out of balance. They are either out of balance because of the relative quantities of Th1 vs. Th2 cytokines or because of peculiarities in the time evolution of the various components. In the case of these types of diseases, fibrosis (viewed from this perspective as a chronic, unsuccessful attempt to heal) ensues.

This same set of pathways can take place during beneficial exercise or rehabilitation, resulting in a state of tissue healing or muscle growth. Exercise maintained within some individually determined range is well-tolerated and beneficial, and can result in improved physiological parameters. However, overtraining can cause ongoing damage, and may predispose to infections (e.g., in competitive cyclists who become susceptible to bronchitis during multiday stage races).

The organizing principles described above may be utilized to gain translational insights into the inflammatory response. Many groups of scientists have already created simulations of inflammation in wound healing at various scales, from cell culture studies to the whole organism.12–17 However, what is needed is a framework for Translational Systems Biology in the setting of inflammation and healing.1,8,18,19 The process of preclinical studies (Domain 1)→clinical trials (Domain 2)→in-hospital care (Domain 3)→ultimate long-term use of any therapy (Domain 4) may be termed a “fragmented continuum,” in which there are four microdomains.8 These domains result in the fragmentation of the healthcare delivery process, because there are no mechanisms by which one can extrapolate mechanistically and directly from the results of early preclinical studies to what will happen at a much later time.8 The National Institute of Health's Roadmap initiative stresses the use of system approaches to try to get past these bench-to-bedside gaps. The Food and Drug Administration's “Critical Path” document stresses the same point but is targeted toward increasing the number of available therapies for a given complex disease. Given the assumption that many, and perhaps most, current diseases involve derangements of inflammation and healing, the goal should therefore be to create mechanistic computational simulations that span the entire range from the basic biological processes to chronic and rehabilitative care.8

The vision for Translational Systems Biology has been articulated elsewhere,1,19 but is summarized below. In classical systems biology, basic insights are the primary focus, often necessitating experimentation in very simple model organisms or in cell culture, settings in which variables can be controlled relatively easily. In the classical systems biology paradigm, simulations are driven by a search for organizing principles,7 whereas Translational Systems Biology is aimed at rapid clinical validation. Classical systems biology stresses the creation of models structured for the greatest and most precise basic insights. The translational framework is oriented toward models structured for clinical and translational utility. Some of the earliest work applying Translational Systems Biology to inflammation focused on simulated clinical trials, but this nascent field has progressed toward personalized diagnostics, personalized medicine, and the rational design of drugs or devices.1,19 Finally, so-called “-omics” studies that often define classical systems biology are typically used in very controlled settings. These techniques have been and continue to be used in very clinically relevant situations. However, there is currently no framework by which to actually judge the outcome of these studies beyond a form of pattern recognition: pattern A is associated with outcome X. However, it is very difficult to obtain mechanistic information that can guide future studies. In the Translational Systems Biology framework, mechanistic simulation should lead to an understanding of the origin of patterns in “-omic” data in order to facilitate the design of novel therapies.1,19

Translational Systems Biology of inflammation follows an iterative process common to mechanistic simulation studies.20 The process is initiated by researching relevant biological mechanisms for a given inflammatory disease. Investigators typically create influence diagrams in various forms, whose goal is to represent the way that cells, cytokines, etc. interact with each other.21 The level at which models are created is fully dependent upon the desired use of the model. If the desired outcome is a whole-animal simulation, the focus is on the dynamics of cell populations, cytokines, etc. rather than molecular processes. The next step involves the collection of as much relevant data as possible, which in the clinical setting especially may be a limiting factor due to medical, regulatory, or ethical concerns.22 Therefore, such models are based on sampling the biofluid nearest to where the inflammatory response occurs in a given disease or preclinical model (e.g., serum/plasma,21,23–25 urine,26 cerebrospinal fluid, and, in a recent study, laryngeal secretions27). Clearly, it is impractical to say that if the disease process occurs in the liver then one should sample liver biopsies very frequently; hence, Translational Systems Biology models are structured with that limitation in mind. Parameter estimation is then performed in order to fit the models to the available data.8 This is actually where the entire exercise can fail or succeed, because given large mathematical models and numerous “knobs” (i.e., the parameters in these models), the likeliest outcome is that model predictions will become worse rather than better. The next step is validation of model predictions. The creation of a mechanistic model that can plausibly describe the time courses of various cells and cytokines in and of itself is a major accomplishment, but is insufficient: a mechanistic model must be used to make predictions outside of the dataset that was used to calibrate it. Alternatively, if a model reproduces the inflammatory response of a single individual, varying some parameters of that model should produce a group of simulated individuals that might be vetted against data from a clinical trial that has already occurred.28 Where possible, the investigator should collect relevant biomarker data, and then if it is determined that the model predictions are incorrect then further parameter estimation should be carried out in an attempt to improve the fit to calibration data. If that fails, then the model must be examined critically to determine if any elements are missing. In essence, this process leads to a testable hypothesis. The advantage of this framework is that it is a rational means by which to hone in on gaps in knowledge about a given biological system.

COMPUTATIONAL SIMULATIONS: FILLING THE GAPS IN THE FRAGMENTED CONTINUUM OF HEALTHCARE DELIVERY

A large number of Translational Systems Biology studies have been focused on enhancing the preclinical stage (referred to as Domain 1 above) of the fragmented continuum of healthcare delivery in the setting of inflammatory diseases.1,8 The main finding from these studies is that DAMPs are central integrators of the inflammatory response.29 Based on these simulations, it may be hypothesized that the body integrates the signals communicated via inflammatory mediators via DAMPs derived from damaged or dysfunctional tissues, and to a certain degree the dynamics of these DAMPs reflect the tissue or organism's health.8 Therefore, DAMPs may be surrogates for an individual's health status. Thus, computational simulations calibrated on the dynamics of classical cytokines but predicting the dynamics of DAMPs may be capable of predicting outcomes in a given inflammatory disease even absent the identification of the DAMPs particular to that disease.29 Additional achievements within the realm of preclinical studies include the use of computational simulations to predict the inflammatory features and outcomes of gene knockout mice at the whole-animal level,23 as well as the combination of computational simulations with functional genomics in an attempt to unify these two aspects of systems biology.24

The next step in the continuum of care is the clinical trial stage (Domain 2).8 The anti-tumor necrosis factor (TNF) trials in sepsis that failed to show benefit were simulated, using knowledge that the sepsis field had at the time the trials were carried out.28 These simulation studies showed that the trials would have failed if they were carried out as they indeed were, but that these trials could have succeeded if they had been performed using guidance from computational simulations with regard to inclusion/exclusion criteria.28 Similar methods were utilized to examine combined vaccination and antibiotic administration to first responders in the setting of a bioterror attack.30 This approach is increasingly being used by the biosimulation industry in partnership with the pharmaceutical industry, for example in a setting of sepsis.1 In that setting historical trials were simulated and used to predict the outcome of new trials using Xigris.1

The next step in the continuum is in-hospital care (Domain 3).8 A recent manuscript in Wound Repair and Regeneration describes agent-based simulations used to gain insights into forces driving the pathology of diabetic foot ulcers.31 Following from a simulation of the inflammation healing cascade in normal skin, the authors demonstrated how single changes in the production of individual cytokines (e.g., TNF-α and TGF-β1) can lead to many hallmarks of a diabetic foot ulcer. This study also simulated existing therapies such as debridement and the use of platelet-derived growth factor (PDGF), and simulated novel strategies such as TGF-β1 modulation or neutralizing anti-TNF antibodies.31

A central part of translational systems approaches for in-hospital care involves the creation of patient-specific computational simulations as one means of achieving personalized medicine.1 For example, studies were carried out in volunteers who were subjected to phonotrauma for an hour, followed by the collection of the most proximal biofluid for this particular injury (laryngeal secretion). This procedure was carried out at baseline, immediately after the phonotrauma, and then at 4 and 24 hours after phonotrauma.27 Patient-specific agent-based models were created by calibrating a generic agent-based model with laryngeal secretion data from individual patients, with the goal of identifying individually optimal treatments.27 These patient-specific simulations reproduced trajectories of inflammatory mediators in laryngeal secretions of individuals subjected to experimental phonotrauma up to 4 hours postinjury, and predicted the levels of inflammatory mediators 24 hours postinjury. Subject-specific simulations also predicted the effects of behavioral treatment regimens to which subjects had not been exposed.27 Similar methods were used to create patient-specific EBM in the setting of polytrauma. Human trauma patients were recruited into an observational study in which blood samples were obtained daily up to 1 week postadmission, then weekly thereafter. Plasma was assessed for TNF, interleukin (IL)-6, IL-10, and NO2/NO3. Trauma was modeled as an exponentially decaying function the EBM developed originally for mice.21 The coefficient of the trauma function was scaled from 1 to 2 with 1 corresponding the lowest Injury Severity Score (ISS, an established clinical scoring system) and two highest ISS for any patient. The rate constants of the EBM that relate to generation of TNF, IL-6, IL-10, and NO2/NO3 were estimated to fit the time-course data of individual patients. Using this methodology, the resultant patient-specific models accurately predicted patient survival when ISS alone could not (Sarkar et al., unpublished observations).29 Most recently, this approach was applied to inflammation induced by traumatic brain injury. Multiple inflammatory cytokines were determined in serial cerebrospinal fluid samples from traumatic brain injury patients. Principal Component Analysis was used to suggest the cytokines involved primarily in this response, and accordingly multiple patient-specific models were fit to these data.26

The last stage of the continuum of healthcare is chronic or rehabilitative care (Domain 4).8 Given that inflammation plays both beneficial and detrimental roles in the setting of exercise and rehabilitation,32 it would seem appropriate to utilize computational simulations to discern among and predict possible outcomes in these more benign contexts. In Domain 4, it is important to consider the effects on inflammation and healing of factors such as age and environment. Perhaps not as intuitively, it is also important to include the interplay between damaged tissue and damaged nerve, given that the nervous system exerts a potent antiinflammatory effect,33 while inflammatory agents can affect the viability of neurons.34 For the wound healing community, this type of integration may allow for improved long-term care for conditions such as chronic, nonhealing diabetic foot ulcers, a setting in which loss of sensation is intertwined with infection and inflammation.35–39

CONCLUSIONS AND FUTURE PROSPECTS

The field of wound healing is rife with reductionist studies that describe various aspects of the process in at times exquisite mechanistic detail, along with a relatively small number of studies utilizing systems approaches such as computational simulation. Clearly, a synthesis of the wound healing process in a manner that will link reductionist data to improve clinical translation is a worthwhile goal, and Translational Systems Biology provides the necessary framework by which to accomplish this goal. Using this approach, the future holds both novel insights into and rationally designed therapies for wound pathobiology.

ACKNOWLEDGMENTS

This work was supported in part by the National Institutes of Health grants R01-GM-67240, P50-GM-53789, R33-HL-089082, R01-HL080926, R01-AI080799, and R01-HL-76157; National Institute on Disability and Rehabilitation Research grant H133E070024; as well as grants from the Commonwealth of Pennsylvania, the Pittsburgh Lifesciences Greenhouse, and the Pittsburgh Tissue Engineering Initiative.

Disclosure: Dr. Vodovotz is a co-founder and stakeholder in Immunetrics, Inc., which is commercializing aspects of the computational modeling work described herein.

Ancillary