Eating differently and exercising more is the general advice for metabolic syndrome, but this may not be the whole story.
About a quarter of all adults have metabolic syndrome; a condition whose most well-known symptoms are obesity, high blood pressure, impaired blood glucose and poor cholesterol levels.
Metabolic syndrome is used as clinical definition of overweight individuals at increased risk of a large number of comorbidities such as type 2 diabetes (T2D), cardiovascular disease (CVD) and non-alcoholic fatty liver disease (NAFLD). It is expected that by 2030 33% of the total population (200 million individuals) in the 27 countries of the EU will be obese.
Many of them will have one or more of the above co-morbidities. At the current prevalence at least 20 million individuals will suffer from diabetes, 60 million have NAFLD and ten million CVD. Without successful interventions, besides the toll on lives of EU citizens, costs of treating the comorbidities will increase to more than €100bn per year beyond 2030. Clearly, there is an urgent clinical as well as economic need to conquer the sequelae and comorbidities of metabolic syndrome.
Modelling metabolism to understand human disease
In principle, metabolic syndrome is a condition that is difficult to study. It often takes ten years or more to develop, and many organs and tissues play a role, making it hard to create a complete picture. Yet that is exactly what the European research project RESOLVE did.1
RESOLVE implemented a systems biology approach for integrative and quantitative life science research which includes computational biology and modelling to study molecular processes fundamental in biological systems. The implementation of systems biology approaches in medical concepts, research and practice is called systems medicine. Systems medicine enables the implementation of personalised medicine (or P4 medicine: predictive, preventive, personalised, and participatory).
To study metabolic syndrome and associated metabolic diseases, RESOLVE combined human studies, in vivo modelling in mice, and computational modelling. In genetically modified lab mice, metabolic syndrome occurs in a similar way to that in humans, only much faster: the entire development takes only a few months with the mice, making it easier to examine the whole process.
A team of computational biologists, under supervision of Professor Natal van Riel (Eindhoven University of Technology (TU/e), the Netherlands), made a mathematical model, called MINGLeD, that describes the sugar and fat balance of the entire body. The TU/e scientists applied a dedicated simulation method, allowing for accurate prediction of gradual, long-term development of the disease. When the model is run with this method it takes data from real-world experiments to iteratively learn and calibrate the model.
In the end, the model yields an integrated, systems-level description of all the different data and, importantly, provides information on biological variables and parameters that cannot be directly observed from the data. This modelling approach correctly predicted progression of metabolic syndrome in the mice, as well as development of comorbidities such as fatty liver disease.
The model also uncovered the unexpected existence of subgroups of mice with metabolic syndrome. Animals that became very sick – with the most severely disrupted fat balance – had livers with higher metabolic activity. This resulted in a higher production of bile, a substance that promotes the absorption of fat from the diet.
Additionally, the sick livers also produced more cholesterol and fat themselves. Thus, the group with more active livers ended up with more cholesterol and fat in the blood than the group with less active livers.2
Bile is composed of dozens of different bioactive molecules called bile acids. The important role played by bile acids in metabolic syndrome was confirmed in a validation study, which included a larger group of the laboratory mice that were followed for a longer period of time. It is important to know that bile acid molecules present in blood are to a large extent produced by bacteria in our gut.
One could think of bile acids in plasma as a metabolic ‘window’ on the gut microbiome (the collection of all micro-organisms present) and other digestive processes in the gastrointestinal tract. When travelling through the body, bile acids exert effects similar to hormones; they regulate metabolic processes in all kinds of different cells and tissues.
Ongoing research links the gut microbiome to metabolic, immune and neurologic conditions.
To investigate the role of bile acids in humans with metabolic syndrome a different systems biology model was developed.3 In particular, the effect of gastric bypass surgery was investigated. A gastric bypass is a certain type of bariatric surgery for patients to combat metabolic syndrome. After this operation, food travels a shorter route through the digestive system, skipping most of the stomach and a piece of intestine. A special feature of this successful type of procedure is that sugar levels in the blood improve quickly, while substantial weight loss often takes many months.
This means that if a patient had T2D before the surgery, the treatment is likely to improve or even resolve this condition. The data of the operations shows that, although the release of bile acids remains the same, more of them remain, causing a higher concentration in the blood. The fact that more remains is caused by the bypass, the mathematical model says. As the food passes through the small intestine differently, more bile acids can be absorbed at the end of it. It is likely that the increased bile acids play an important role in the beneficiary metabolic effect of bariatric surgery.
Construction of the systems biology models requires large amounts of quantitative molecular data and a thorough understanding of the biology. As much as possible of existing knowledge in the physiological, biomedical and clinical domains is used. To describe changes over time (system dynamics), the models make use of differential equations, essentially implementing an advanced type of ‘bookkeeping’ to trace how molecules travel through the body, accumulate in certain organs or tissues, and are produced and degraded by biochemical processes.
A main advantage of mechanistic models built using domain knowledge (domain intelligible models) is their interpretability, which is critical in biomedical applications.
One of the applications of the systems biology models is to explore novel avenues for the development of therapeutic interventions (systems pharmacology). Recently, pharmaceutical companies have become interested in the potential of drugs that mimic the effect of bile acids as a novel treatment for metabolic diseases. In particular, agonists that bind to and activate FXR (a receptor protein in the cell nucleus) are currently being tested in clinical trials. The bile acid model is a stepping stone towards future models to determine which combination of drugs, operations and lifestyle changes is the best approach for individual patients.
The highly detailed computer model of bile acid metabolism, made by TU/e scientists together with collaborators at the Amsterdam University Medical Centers (Amsterdam UMC), turned out to work so well that personalised variants for individual patients could be made.3 The input for this includes the bile acid molecules measured in the blood of individual patients after they have eaten. From studying many meal responses in different individuals but also within the same person, it becomes clear how largely different these responses can be.
For the implementation of personalised health and medicine it is important to recognise the impact of metabolic individuality. Working with the bile acid model it became possible to detangle inter-individual variation from intra-individual heterogeneity in meal responses and to causally link metabolic phenotypes with differences in activity of metabolic processes. Through this approach differences between obese individuals with and without T2D could be identified.
Using metabolomics it becomes possible to measure hundreds of different metabolites in a single sample of blood or other bodily fluids. When combined with experiments in which a person receives a standardised test meal and blood is taken at multiple consecutive time points after the meal, it becomes possible to obtain incredibly detailed information about an individual’s metabolic health. Building personalised metabolic models provides a way to interpret all this data. If such tests are repeated over time, changes in the dynamic response can indicate onset of disease. Likewise, this protocol can be used to monitor the effect of (novel) treatments.
Bringing metabolic models to the patient’s home
In addition to supporting the diagnosis and treatment of disease, computational modelling can also implement personalised approaches for prevention (personalised health). In case of patients with diabetes, adequate control of blood glucose levels is key to live a healthy life and prevent serious cardiovascular complications that develop when sugar levels are too high all day, every day. e-DES (the Eindhoven Diabetes Education Simulator) is a physiology-based dynamic model that predicts glucose and insulin levels throughout the human body in response to food, medication, physical activity and mental stress.4
The model has been embedded as the ‘engine’ of a serious game developed as a means to educate patients with diabetes (and their families). This is an example of how a complex simulation model is brought to the home of the patient, with direct benefit for healthcare.
In ongoing research, algorithms are being developed to use data from the patient to make the computer model personalised and patient-specific. Data is collected under daily-life conditions using wearable technologies; in particular, participants in this research are equipped with devices for continuous glucose monitoring. A personalised model makes the game more realistic for the patient, which is expected to improve the educational value, resulting in patient empowerment and better self-management, with fewer complications.
In silico models of inborn errors of metabolism
Another disease area where mathematical modelling of human metabolism is relevant is in inborn errors of metabolism. IEMs are a group of genetic disorders in humans characterised by dysregulation of the metabolic networks that underlie the development and homeostasis.5 IEMs represent examples of complex gene-environment interactions and, more specifically, gene–nutrient interactions that lead to complex disease. Newborn screening for IEMs is common in European countries.
Using dried blood spots obtained from a heel prick, IEMs that can be treated with medication or change in diet is diagnosed. Through the development of novel techniques for genetic testing and metabolomics, new hereditary metabolic disorders continue to be discovered and underlying molecular mechanisms of disease are revealed.
The genetic metabolic disease very long-chain Acyl-CoA dehydrogenase deficiency (VLCADD) was studied using mechanistic modelling.6 Muscle energy homeostasis is compromised in this disease. The model created by TU/e scientists generated testable hypotheses about underlying metabolic mechanisms and provided new clues about the risk of rhabdomyolysis in these patients. Rhabdomyolysis is a serious condition in which damaged skeletal muscle breaks down rapidly and is common in VLCADD and other metabolic myopathies following exercise.
So far, applications of dynamic modelling of human metabolism have been discussed. A different type of mathematical modelling of metabolism also offers a powerful approach to study metabolic health. Genome-scale metabolic network (GSMN) models are used to study IEMs, lifestyle-related disorders and the metabolic basis of cancer.
Future computational models of human metabolism should include the brain as a key regulator of metabolism. The human brain consumes about 20% of the total energy used in rest. Moreover, there are strong indications that obesity and metabolic syndrome are associated with differences in the dopamine reward system in the brain.
Finally, obesity and metabolic syndrome are associated with an increased risk for neurological disorders such as depression. In fact, many molecular precursors of neurotransmitters originate from the gut, partly produced by microbiota (representing an important path for communication in the body known as the gut-brain axis). Especially because it is difficult to measure molecular processes in the human brain, mathematical models are expected to shine new light on the role of our brain in health and disease, and provide better insight into differences between people.
Dynamic systems theory and other complex systems approaches can be applied to identify sources of robustness and resilience in the gut-brain system. Detailed, well-validated metabolic models can support identification of potential markers or metrics to predict trajectories and transitions in dynamic behaviour, like blood bile acids offer a window to the gut-liver axis.
For application in healthcare and prevention, these models need to be transformed into algorithms that can take data from patients or healthy individuals as input, which brings us to a second challenge in the field of computational systems medicine.
Technology, including biosensors, data engineering and artificial intelligence, is rapidly changing medicine. We are entering the era of personalised and consumerised healthcare. Health-related technologies (such as wearables) are quickly adopted by consumers.
Patient-gathered health data are distinct from data generated in clinical settings. Patients, not healthcare providers, are responsible for recording these data, and patients decide how to share these data. The data typically is of lower quality than can be collected in the hospital.
However, it is collected frequently and in a daily-life setting. Intelligent, model-based data engineering algorithms need to be developed to transform low-quality data into high-quality health information. The use of patient-gathered health data supplements clinical data, providing a comprehensive picture of patient health.
Important biomedical insights and innovations are expected to arrive from the application of computational models and algorithms that combine domain knowledge with data collected from the individual to make personalised predictions about health upon which healthcare providers and the individual citizen can act.
Biomedical systems biology at Eindhoven University of Technology
At Eindhoven University of Technology, a team led by Van Riel develops computational techniques for modelling and data engineering and applications in health and healthcare. Important in the approach is the combination of data-driven methods with mechanistic models.
Engineering approaches are developed and applied to integrate high-dimensional molecular data and link it to physiological signals and other data collected in the hospital and/or in daily life. Hereto techniques from machine learning and artificial intelligence are deployed in combination with mathematical modelling. Mathematical models are built to study human metabolism and its multi-level regulation in health and disease, and predict effects of interventions over time.
A particular challenge is the development of predictive models that combine static biomedical data (e.g. the DNA of a patient) with variable and dynamic molecular markers (e.g. blood glucose) and physiological signals (e.g. heart rate).
The research is driven by societal and clinical needs, focusing on interrelated cardio-metabolic disorders such as T2D and fatty liver disease. To develop innovative ideas and meaningful innovations for personalised medicine and health, the team of computational biologists collaborates with hospitals and private companies and other partners outside the university.
Metabolome refers to the complete set of small-molecule chemicals (metabolites) in a biological cell, tissue, organ or organism, which are the products of cellular processes. The study of the metabolome is called metabolomics. To understand the metabolome, different technologies and data need to be integrated, including genomics, proteomics and physiological information.
The metabolome reflects the interaction between an organism’s genome and its environment; metabolomics provides a direct ‘functional readout of the physiological state’ of an organism. Whereas the genome tells you what might happen, the metabolome tells you what is happening.
Metabolic flux is defined as the production or elimination of a quantity of metabolite per mass of cell, organ or organism over a specific timeframe (mole/kg/hr).
Mathematical models of human metabolism have the ability to causally link phenotypes with changes in metabolic fluxes. Effective therapy of metabolic disorders requires the alteration of metabolite flux.
Genome-scale metabolic network models
A GSMN model includes all genes encoding for proteins (enzymes) with metabolic activity. Metabolites and biochemical reactions form a complex interconnected network. GSMNs are implemented as large, graph-based network models and simulated and analysed using so-called ‘constraint-based approaches’. Current human GSMN models include thousands of reactions and metabolites.
Also for GSMNs, methods are under construction to go from generic models describing the ‘average’ human to implementation of patient-derived GSMN models. Molecular data, such as transcriptomics and proteomics, obtained from patients’ samples (such as tissue biopsies) are used to tailor the generic metabolic network to represent metabolism in a specific tissue of an individual patient.
Comparing network models for different patients provides insight into different manifestations of a disease and might suggest personalised treatments. GSMN models of human metabolism are also being applied to study inborn errors of metabolism7 and weight-related metabolic diseases.8
- RESOLVE, Grant agreement ID: 305707, funded under FP7 call topic HEALTH.2012.2.1.2-2 – Systems medicine: Applying systems biology approaches for understanding multifactorial human diseases and their co-morbidities, https://cordis.europa.eu/project/rcn/106225/en
- Rozendaal YJW et al. In vivo and in silico dynamics of the development of Metabolic Syndrome. PLOS Computational Biology 14, e1006145 (2018)
- Sips FLP et al. In Silico Analysis Identifies Intestinal Transit as a Key Determinant of Systemic Bile Acid Metabolism. Front Physiol 9, 631 (2018)
- Maas AH et al. A Physiology-Based Model Describing Heterogeneity in Glucose Metabolism: The Core of the Eindhoven Diabetes Education Simulator (E-DES). J Diabetes Sci Technol (2014) doi:10.1177/1932296814562607
- Lanpher B, Brunetti-Pierri N & Lee B. Inborn errors of metabolism: the flux from Mendelian to complex diseases. Nat Rev Genet 7, 449-460 (2006)
- Diekman EF et al. Altered Energetics of Exercise Explain Risk of Rhabdomyolysis in Very Long-Chain Acyl-CoA Dehydrogenase Deficiency. PLoS ONE 11, e0147818 (2016)
- Sahoo S, Franzson L, Jonsson JJ & Thiele I. A compendium of inborn errors of metabolism mapped onto the human metabolic network. Mol BioSyst 8, 2545-2558 (2012)
- Mardinoglu A et al. Personal model‐assisted identification of NAD+ and glutathione metabolism as intervention target in NAFLD. Molecular Systems Biology 13, 916 (2017)
Natal van Riel
Professor in Biomedical Systems Biology
Eindhoven University of Technology
Professor in Computational Modelling
+31 40 247 5506
Please note, this article will appear in issue 9 of Health Europa Quarterly, which will be available to read in April 2019.