Lessons from the Growth Plate: Computational Modelling of Chondrogenic Differentiation

Introduction: The pathways and genes/proteins orchestrating postnatal fracture repair and those involved in embryonic bone development are deemed to be highly similar. As a result of this similarity, combining insights from the wealth of studies focusing on bone developmental biology may provide useful information on the driving mechanisms of differentiation in fracture repair. In this study, we develop a computational approach that evaluates how the individual interactions between genes/proteins can give rise to behaviour observed in the growth plate and what influence individual perturbations can have on cell fate decisions. Materials and Methods: Through manual literature curation a network consisting of 46 genes/proteins, playing a primary role in endochondral ossification, was developed. The network model focuses on cell fate decisions governed by Sox9 and Runx2, transcription factors that play instrumental roles in both chondro- and osteogenic differentiation. In the context of the growth plate, Sox9 and Runx2 are respectively associated with proliferative and hypertrophic chondrocytes. The dynamics of this network were simulated using a simple additive framework. Through a Monte Carlo canalisation analysis the stability of various attractors, associated with a specific cell fate was assessed. Lastly, a genetic algorithm was employed to generate an ensemble of models that capture expression profiles (based on 27 genes/proteins) observed in proliferative and hypertrophic chondrocytes. Results: By means of perturbations (i.e. loss of function and gain of function of individual genes/proteins) this model can be used to perform an in silico screening to determine the relative contribution of each signal to different cell fates. This in silico screening hence highlights which genes/proteins have a large, positive or negative, impact on determining cell fate. In repeating this screening for the generated ensemble of models, trained to mimic growth plate expression profiles and differing only in parameter values, a measure for the uncertainty of the screening results was obtained. Most effects were consistent throughout the ensemble, though some pathways (e.g. the effects of BMPs on Sox9) showed ambivalent results. Discussion: The development of the network model allows for a more holistic assessment of the effect of individual genes/proteins on cell fate decisions. In addition, ensemble modelling can show where more input is needed to determine the effect of a particular perturbation. For tissue engineering, the analysis provides an overview of possible interventions that decrease or increase the propensity for attaining stable (Sox9) or transient (Runx2) cartilage. Conclusion: In integrating interactions from a wide variety of studies the developed model comprises a compendium of knowledge on growth plate signalling. Through dynamical simulations, this information can be effectively harnessed to instruct tissue engineering approaches.status: publishe