One hundred and twenty-five years and counting: into an era of systems dermatology

Publication of a journal for 125 years indicates both longevity and success. Such milestones provide an opportunity not only to reflect back but also to look forward. During this period dermatology has moved from being a largely descriptive specialty to one embracing experimental and evidence-based medicine and incorporating advances in skin biology as well as other scientific fields. In this editorial, which is one of a series in celebration of the 125-year anniversary of the British Journal of Dermatology (BJD), I will consider the rising areas of systems biology and informatics, and how these may be further applied to dermatological and skin biology research. Further, I will argue that the U.K. is well placed to take advantage of these developments, whilst at the same time highlighting the challenges to the dermatological community in engaging in such new initiatives, in terms of scale, the infrastructure and skill sets required. Definitions of systems biology vary widely depending in part on the scale of the system being considered. For example, this approach can be applied at the level of a signalling pathway, cell, tissue or organism. Systems biology can be considered an experimental approach, designed to provide a holistic working model of a biological system, while systems medicine (dermatology) can consider perturbations to the model by specific interventions or under disease conditions. These approaches generally involve a computer model that integrates a variety of experimental data. Such models are particularly applicable to the present day in view of the exponential growth of ‘omic’ informatics data, which are being generated at many layers relevant to the ‘dermatological system’. Indeed, it could be argued that the ability to generate data, using next-generation sequencing technologies for example, is gathering pace at such a rate that it will potentially outstrip the ability of systems biological approaches to integrate and optimally interpret it. Thus, acquisition of data per se is useful only if relevant bioinformatics tools and systems biological methods are available. A key feature of a systems approach is the production of a computational model that integrates, according to specific hypotheses, a variety of experimental data and thereby allows simulation and prediction of biological outcomes. By running such simulations, specific knowledge gaps can be identified leading to design of further experiments and an iterative cycle with relevant feedback loops. The systems approach is distinct from the experimental reductionist approach that has driven much of science over the last several decades, in which model systems have been broken down to their individual components. However, although we have learned much, it is becoming increasingly clear that the complexity of biological systems and skin disease cannot fully be explained by such an approach.1 This is further underscored by the genetic diversity among humans that contrasts with outputs from experiments conducted in genetically identical cell lines or genetically highly homogeneous mouse strains. Some of the benefits of computational models are that they allow simulation and prediction of biological outcomes and thereby permit testing of specific hypotheses that would otherwise be difficult, ethically challenging or extremely time consuming to investigate fully experimentally. As mentioned, such models also facilitate the identification of knowledge gaps leading to the design of further experiments. This concept was highlighted in a recent review by Tanaka and Ono that considered the application of mathematical models to skin disease.2 Classically, computational models have been classified as deterministic and stochastic. In brief, deterministic models are compartmentalized, principally cover large populations and describe average reproducible results of populations and subpopulations. They often involve differential equations that predict and help to explain experimental data. Feedback loops around neural networks fall into this category. On the other hand, stochastic models encompass random events that may occur in biological situations. Each run leads to a slightly different result. This kind of model allows testing of a range of possibilities, and multiple simulations help to define the stability and boundaries of the model. Recently, the potential for integrating these kinds of models has been seen. To illustrate this concept, I will consider some models that have been developed for psoriasis. Interestingly, mathematical considerations of psoriasis epidermal hyperplasia were evaluated in the 1970s, before the term ‘systems biology’ was coined.3, 4 Some of these important papers were published in the BJD.5, 6 Modelling has also been applied to the study of how psoriatic epidermis develops its characteristic papillomatous shape.7 Subsequently, deterministic models have been utilised in the study of immune system signalling in human disease and have provided insight into the role of positive and negative feedback loops in specific signalling pathways.8, 9 For example, Valeyev et al.10 utilized this approach to study immune-cell signalling in human skin and psoriasis, and developed a series of differential equations based on experimental data that describes multiple interactions between immune-cell populations. The model predicts that modified feedback-loop interactions can result in altered homeostatic balance and the appearance of alternative steady states. Importantly, running simulations of the model showed that only certain states are stable. Moreover, the model suggested that even minor alterations in cytokine production, as initiated for example through an interaction between specific genetic polymorphisms and environmental factors, and subsequently amplified through feedback loops, could lead to pathological inflammation.10 On the other hand, Weatherhead et al.11 developed an agent-based, stochastic, spatial model of psoriatic epidermis. This model was used to study the effects of ultraviolet-induced apoptosis and growth arrest on epidermal homeostasis in psoriasis. A systems approach integrated experimental data generated from human psoriatic epidermis in vivo with the model using experimental data generated over many decades.3-6 The model was stable under a range of experimental parameters.11 Moreover, the model allowed simulations to be run under different irradiation conditions and permitted the comparison of outcomes to known time courses of psoriasis clearance.12 Such experiments would have taken years and the involvement and study of many patients. Importantly, the running of simulations generated specific questions and gaps of quantitative knowledge that needed to be addressed to permit accurate predictions. For example, the time it takes for epidermal keratinocytes to undergo apoptosis was required in the model, but such information was lacking from the literature. Specific experiments were conducted to address this knowledge gap, and these data allowed a quantitative model to be built that addressed the role of apoptosis in epidermal remodelling and plaque clearance.11, 12 The U.K. is well placed to take advantage of the revolution in informatics particularly as applied to health care. Thus, for example, the U.K. government's Medical Research Council (MRC), in partnership with nine other government departments, research councils and charities, has recently invested over £90 m in several initiatives aimed at supporting informatics research, infrastructure and scientists.13 The challenge for dermatology is to find ways to establish links with such initiatives. One mechanism may be through large consortia that have recently been established, including a stratified medicine initiative. In 2012, the MRC pledged £60 m in a 4-year investment in stratified medicine research, and U.K. dermatology was successful in securing £5.2 m through the Psoriasis Stratification to Optimise Relevant Therapy (PSORT) consortium (http://www.psort.org.uk). This partnership involves five U.K. universities (Manchester, Newcastle, King's College London, Queen Mary and Liverpool), 10 pharmaceutical and diagnostics companies, the Psoriasis Association and National Health Service partners representing patients. The principal aim of PSORT is to understand better the determinants of response of psoriasis to biological therapies. Specifically, two parallel, interlinked work strands will collect detailed clinical, adherence, pharmacological, genetic, transcriptomic and immune data in blood and skin from patients with psoriasis treated with biologics. These datasets will be integrated using cutting-edge bioinformatics, and statistical, machine learning and systems biology techniques. By considering these complex datasets together, we aim to build informatics models that will predict the relative contributions of individual factors to, for example, primary and secondary therapeutic failure. This research may yield scalable stratification biomarkers that will require further testing and validation. In this context, the British Association of Dermatologists’ Biologic Intervention Register (BADBIR), which now comprises 151 centres across the U.K. and Ireland, will prove an invaluable resource (http://www.badbir.org). Moreover, it could be argued that the financial success of the BJD has facilitated investment by the British Association of Dermatologists in such large-scale projects that include the recently established U.K. Translational Research Network in Dermatology (U.K. TREND; http://www.uktrend.org). The first paper emanating from BADBIR was recently published in the BJD,14 and this data-rich pharmacovigilance register should itself yield important information over the next few years. Ultimately, PSORT is designed to produce an integrated algorithm that predicts treatment response to individual biological agents. At the same time, we predict that this research will lead to greater understanding of psoriasis pathogenesis, mechanisms of psoriatic plaque clearance and remission. The data generated will facilitate further development and integration of spatial11 and cytokine map models of psoriasis. By building on previous successful work in the cancer field on rule-based modelling and verification of signalling pathways,15 the predictive power and accuracy of mechanistic psoriasis models should be increased. In summary, the dermatological community is indeed well placed to take advantage of the significant progress being made in systems biology, data integration and health informatics. The accessibility of skin for clinical observation, experimental medicine and sequential sampling provides unique opportunities for generation of large datasets linked to specific clinical and disease phenotypes. The vision and initial experience so far suggests that this approach may be of significant benefit to the community and ultimately patients. Given the multidisciplinary nature of this research, the primary research outputs may not be published in the BJD. Nevertheless, we trust the editorial team will keep the BJD readership appraised of relevant advances through review articles. In the end though, the success or otherwise of ‘systems dermatology’ will depend on our ability to integrate this new approach and technology with clinical and experimental dermatology, and to manage multicentre and multidisciplinary collaborations effectively. We envisage that platforms such as U.K. TREND will facilitate this process. None declared.