Systems biology of energy metabolism in skeletal muscle

The primary function of skeletal muscle tissue is to produce force or cause motion. To perform this task chemical energy stored in nutrients (glucose and fatty acids) has to be converted into an energy currency that can drive muscle contraction (adenosine-tri-phosphate, ATP). This process is known as the energy metabolism of skeletal muscle and consists of a large number of chemical reactions that are organized in metabolic pathways. Unraveling this complex network is important from a fundamental biological perspective, but also essential to understand how a disturbance of muscle bioenergetics can cause metabolic disorders. ?? 31P magnetic resonance spectroscopy (MRS) has emerged as one of the premier methods to study skeletal muscle energy metabolism in vivo. It, however, remains challenging to relate the observed metabolite dynamics to an understanding of the underlying processes at the level of the metabolic pathways. A possible solution for bridging this gap between macroscopic measurements and mechanistic understanding at pathway level is the application of mechanistic computational modeling. This dissertation describes a series of studies in which a mechanistic model of ATP metabolism was developed and applied in the analysis of skeletal muscle bioenergetics. Skeletal muscle cells contain two primary processes that are responsible for the conversion of glucose and fatty acids into ATP. These processes are known as glycolysis and oxidative phosphorylation in mitochondria. The initial mathematical models of these processes were obtained by integration of known enzyme kinetics and thermodynamics. Testing of these models, however, showed that they failed to reproduce many of the in vivo observed metabolite dynamics, as has been described in chapter 1 and 2. These results indicated that the models might be missing essential regulatory mechanisms or that the model parameterization required changes. First, the physiological implications of necessary model adaptations were investigated in a series of studies described in chapters 2 – 5. ?? Numerical analysis of the initial glycolysis model revealed that the experimentally observed slow turnover rate of phosphorylated sugars post exercise could only be explained by rapid deactivation of phosphofructokinase (PFK) and pyruvate kinase (PK) in non-contracting muscle. In particular the deactivation of PFK was crucial for adequate control of pathway flux. Therefore, in a follow-up study, it was tested if the missing regulation at the level of PFK could be explained by calcium – calmodulin mediated activation of this enzyme. To this end, pathway behavior, represented by phosphocreatine (PCr) and pH dynamics, was measured in ischemic skeletal muscle for a wide variety of muscle excitation frequencies (0 – 80 Hz). Next, it was shown that addition of the calcium – calmodulin mediated activation of PFK was necessary to accurately reproduce these data. These results provided important new quantitative support for the hypothesis that this particular mechanism has a key role in the regulation of glycolytic flux in skeletal muscle.?? The initial model of oxidative phosphorylation was first tested against empirically determined mitochondrial input – output relations, i.e., [ADP] – mitochondrial ATP synthesis flux (Jp) and phosphate potential (¿Gp) – Jp. These empirically determined relations were derived from 31P MRS measurements of metabolite dynamics post-exercise. They capture key features of the regulation of oxidative phosphorylation in vivo and are therefore considered relevant for testing the quality of the mathematical model. Numerical model analysis (i.e., parameter sensitivity analysis) was applied to investigate which components significantly influenced predictions of these input – output relations. Based on these results it was concluded that the adenine nucleotide transporter (which facilitates the exchange of ATP and ADP across the inner mitochondrial membrane) has a dominant role in controlling the ADP sensitivity of mitochondria. Furthermore, we identified that Pi feedback control of respiratory chain activity was essential to explain measurements of ¿Gp at low metabolic rates. These insights were used to improve the predictive power of the model, as described in chapters 4 and 5. ?? In the studies described in chapters 2 - 5 the glycolytic and mitochondrial model components were tested for conditions in which only one of the two processes was active (ischemia and post exercise recovery, respectively). It remained therefore unknown if the control mechanisms included in these models could also explain the contribution of mitochondrial versus glycolytic ATP synthesis for conditions in which both processes are active (aerobic exercise). In an attempt to answer this question, dynamics of ATP metabolism were recorded during a full rest – exercise – recovery protocol under aerobic conditions and subsequently used for testing of the integrated mitochondrial + glycolytic model. The results presented in chapter 8 showed that the integrated model could accurately reproduce the observed metabolite and pH dynamics for varying exercise intensities. The main physiological implications of these results were that, substrate feedback control (ADP + Pi) of oxidative phosphorylation combined with substrate feedback control (ADP + AMP + Pi) and control by parallel activation (calcium – calmodulin mediated activation of PFK) of glycolysis, provides a set of key control mechanisms that can explain the regulation of ATP metabolism in skeletal muscle in vivo for a wide range of physiological conditions. By application of several cycles of model development it was possible to improve the models performance to the point it was consistent with 31P MRS measurements of muscle bioenergetics in both healthy humans and animals. As described in chapters 6 and 7, it is was investigated if the model could be applied to analyze the adaptations of muscle physiology that underlie changes in mitochondrial capacity that occur in for instance type 2 diabetes patients or with aging. A decrease of mitochondrial capacity in these subjects can be diagnosed accurately by determining the rate of PCr recovery post exercise. However, the changes in muscle physiology responsible for any observed difference in oxidative capacity cannot be deduced from these measurements. Therefore additional muscle biopsy samples are collected and analyzed for in vitro markers of oxidative capacity. State-of-the-art analyses of these data are typically limited to statistical or intuitive approaches. We investigated if the insight obtained from the combined in vivo + in vitro data sets could be increased by application of our mathematical model. To this end, first, the model was extended from a single uniform cell type model to a three types cell model (type I, IIA, and IIX), capturing the microscopic heterogeneity of muscle tissue. In addition, several key validation tests were conducted, as described in chapter 6. Subsequently, we demonstrated that the model could explain the prolongation of PCr recovery period observed in type 2 diabetes patients by integrating available literature data of in vitro markers of mitochondrial function. Although this result was already very promising, it was also concluded that the approach could be tested more rigorously by obtaining all data (in vivo + in vitro) in a single study. Therefore, the method was further tested in an animal model of decreased mitochondrial function: 8 versus 25 week old Wistar rats. The first main result of this study was that the mathematical model could accurately reproduce the delayed PCr recovery kinetics in 25 week old animals based on in vitro determined changes in muscle physiology. In addition, model predictions provided quantitative insight in the individual contribution of the different factors responsible for the decreased oxidative capacity. This type of information is considered very relevant for the design of (pharmaceutical) therapies aimed at improving mitochondrial function. For example, model predictions of the physiological changes that contribute the most to the decrease in oxidative capacity provide potentially promising targets for therapy design. Based on these considerations it was concluded that application of the mathematical model provides new promising opportunities for future studies of mitochondrial (dys)function in skeletal muscle. ?? In conclusion, through application of a series of iterative cycles of model development combined with multiple new experimental studies it was possible to develop a detailed mechanistic model of ATP metabolism that was consistent with in vivo observations of skeletal muscle bioenergetics for a wide range of physiological conditions. This process provided new insight in the key control mechanisms embedded in the metabolic pathways that have a dominant role in regulating ATP metabolism in skeletal muscle in vivo. In addition, we successfully demonstrated the feasibility and added value of application of the model for integration of in vivo and in vitro measurements of oxidative capacity in future studies of mitochondrial (dys)function in, for example, type 2 diabetes, aging or mitochondrial myopathy.