A10 - Using computational models to dissect relationships between gene dosage sensitivity and kinase signaling architectures

Introduction: : Biological systems are well-buffered against large perturbations by redundancies and robust alternative pathways in many circumstances. In contrast, a host of neurodevelopmental disorders are caused by both genomic duplications and deletions in specific genes. Why brain development and function are so unusually sensitive to the genetic dosage of these genes remains unknown. Human genomic studies have identified 1292 genes that are intolerant to copy number loss and gain, of which 17.9% belong to the kinase protein family. Kinase proteins are basic signal carriers along a myriad of intracellular pathways from growth and differentiation to activation and apoptosis. Interestingly, bi-directional expression changes are predicted to disrupt biological function in only ~13% of all human kinases. For example, DYRK1A is a kinase that is sensitive to both high and low dosage changes and considered a major causative gene for Down syndrome neurophenotype, but the downstream kinase CDK5 is tolerant of downward expression change s. How kinase cascades are impacted by dosage sensitivity has not been extensively explored. We hypothesize that dosage sensitivity and abnormal function due to a small change in protein levels arises as an emergent network property of the architecture of the signaling cascade. Given gene dosage aberrations cause a subtle 50% change in expression levels, we propose to use quantitative, computational models to dissect the relationship between gene dosage and pathway architectures.

Materials and

Methods: : A computational model of a 3-tiered kinase cascade has been previously developed to investigate the impact of pharmacokinetic or genetic inhibition on pathway output. The pathway output is derived from the last kinase and dynamics are simulated using Michaelis-Menten kinetics and ordinary differential equations. The model is generalized to include all possible combinations of additional activation and inhibition interactions between the 3 kinase members, including feedforward, feedback, autoactivation, and autoinhibition interactions. Genetic dosage was modeled by virtually increasing and decreasing each kinase in the cascade across all architectures, and the output signal was integrated and normalized for comparison across perturbations. Dimensionality reduction with principal component analysis and clustering with k-means identified network motifs with similar output dynamics between architectures. Analysis was done with fitted parameters and a large set of randomly sampled parameters to assess robustness of the observed dynamics to parameter choice.

Results, Conclusions, and Discussions:: We first used clustering analysis of the integrated network outputs and found that 61% of architectures fall into a large regime of stable network dynamics that are relatively insensitive to dosage across parameter sets. Members of this “linear” regime are characterized by smooth and mostly linear dosage-output responses that reach a maximum saturation point early in overexpression of each of the kinase species. Kinase architectures with multiple inhibitory rules all fell into this regime, indicating that inhibitory feedforward and feedback can act as stabilizing interactions against dosage changes. A smaller subset of architectures exists in a “plateau” regime with a stabilizing region centered around normal dosage that quickly changes at critical values of increased and decreased dosage. These abnormal outputs represent outcomes we would predict from dosage sensitive genes, but we have not identified a unifying network motif that leads to this property. Generally, all networks in this regime involved some component of positive feedback or autoactivation paired with some other interaction. However, this component combination is not unique to this regime and is not sufficient to cause the observed dynamics. A third regime we identified consisted of architectures that remain stable at a high level of activation across all dosage levels except the very lowest dosages. This dynamic appears when the output kinase autoactivates, leading to a “switch” behavior where maximum output is triggered by any input level of the other kinases. Finally, some architectures have individually unique properties, including biphasic behavior or oscillatory instabilities that are introduced at only certain dosage levels. Taken together, this classification of pathway architectures into relevant dosage-response regimes suggests that kinase architecture is a driving component of dosage sensitivity in kinase genes. Our future directions are to experimentally perturb kinases in pathways that match architectures prone to destabilization to validate our model predictions. For example, the unstable architecture 57 matches the activation pathway of DYRK1A, which would be a promising avenue of experimental exploration.

Acknowledgements and/or References (Optional):: Jensen, K. J., Moyer, C. B., & Janes, K. A. (2016). Network architecture predisposes an enzyme to either pharmacologic or genetic targeting. Cell Systems, 2(2), 112–121. https://doi.org/10.1016/j.cels.2016.01.012
Chen, B., McCuaig-Walton, D., Tan, S., Montgomery, A. P., Day, B. W., Kassiou, M., Munoz, L., & Recasens, A. (2021). Dyrk1a negatively regulates cdk5-sox2 pathway and self-renewal of glioblastoma stem cells. International Journal of Molecular Sciences, 22(8). https://doi.org/10.3390/ijms22084011
Arron, J. R., Winslow, M. M., Polleri, A., Chang, C. P., Wu, H., Gao, X., Neilson, J. R., Chen, L., Heit, J. J., Kim, S. K., Yamasaki, N., Miyakawa, T., Francke, U., Graef, I. A., & Crabtree, G. R. (2006). NFAT dysregulation by increased dosage of DSCR1 and DYRK1A on chromosome 21. Nature, 441(7093), 595–600. https://doi.org/10.1038/nature04678