(600d) Using Uncertainty to Assess Feedback Mechanisms in the Innate Immune DNA Sensing Pathway

Quantitative understanding of the DNA sensing pathway—a branch of the innate immune system that detects the accumulation of double stranded DNA (dsDNA) in the cytosol—is required to understand disease detection and immune system activation. This pathway is commonly described in the context of DNA encoded viruses where their genomic material is detected by the key sensor molecule cGAS through nonspecific binding. This narrow context has been expanded to include other viruses like HIV[1] that stimulates this pathway through a DNA intermediate, and even bacteria like Mycobacterium tuberculosis (Mtb)[2] whose DNA can enter the cell and stimulate cGAS. Because cGAS possesses a nonspecific binding domain, it can detect both pathogenic DNA and self-DNA that leaks from the nucleus of damaged cells. This DNA leakage is an early indicator for tumorigenesis in cancer and is negatively regulated by the DNA sensing pathway. Interestingly, there is a fine balance in regulation as sustained activation can lead to inflammation-mediated tumor growth[3]. Dysregulation of the DNA sensing pathway is associated with Aicardi–Goutières syndrome (AGS)[4], an autoimmune disorder in which a DNA degradative enzyme is missing, leading to chronic inflammatory state. The DNA sensing pathway has also been associated with more common autoimmune diseases, e.g. lupus erythematosus[5], and age-related diseases such as atherosclerosis and neurodegenerative diseases[6]. This expansive variety of diseases and disorders demonstrate a key aspect of the DNA sensing pathway: the pathway must be appropriately regulated to ensure a robust and healthy immune response to disease while limiting dangerous secondary processes, e.g. inflammation, tumor growth, and autoimmunity. A systems-level understanding of the pathway dynamics can define the performance capabilities of the DNA sensing pathway, reveal possible weaknesses in the pathway design, identify conditions that lead to prolonged or chronic inflammation, and provide an in-silico platform to optimize pharmaceutical interventions.

Here, we used a Markov Chain Monte Carlo (MCMC) approach to generate an ensemble of pathway models to determine how parameter uncertainty impacted the regulatory feedback mechanisms of the DNA sensing pathway. The pathway is initiated by the binding of DNA to cGAS. This activates a series of chemical reactions which include several common signaling molecules (cGAS, cGAMP, STING, IRF3, IRF7) and ultimately results in the production of type I interferons (primarily IFNβ). We developed an Ordinary Differential Equation (ODE) model of the signaling pathway using mass action, Michaelis Menten, and competitive inhibition kinetics to characterize the interactions of the model species (states; 13 states total). While the DNA sensing pathway is critical to a large variety of diseases, there remains little data on the pathway dynamics. We collected data from 15 experiments published in the literature and used the collected data to constrain parameter training. A multi-chain parallel tempering MCMC approach was implemented to effectively explore a large parameter space for collections of models that suitably characterized the data. This approach allowed us to simultaneously address over fitting issues and to evaluate the robustness of the pathway dynamics to parameter uncertainty. We ran the MCMC algorithm for 10⁷ iterations (instances) and then evaluated the resulting parameter distributions and goodness of fit. We found that a wide range of parameter values fit the data; suggesting that the system dynamics are not strongly dependent on precise parameter values.

Next, we selected the top 1000 models that possessed a wide variation in their parameters values but approximately equivalent goodness of fits and performed a series of in silico knock-down experiments. We observed that the feedback loop responsible for degrading cytosolic DNA and ultimately abrogating the pathway signal was highly robust to changes in the initial concentration of the key effector protein, TREX1. Lack of a functional TREX1 gene has been identified in AGS patients. Our simulation experiments suggest that an inefficient, but working, TREX1 protein does not lead to this chronic inflammatory state. Moreover, as we exploited an ensemble of 1000 models with different parameterization, these results suggest that the observed robust regulation of the pathway via TREX1 feedback is not an artifact of an over fitted model.

References

[1] D. Gao, J. Wu, Y.-T. Wu, F. Du, C. Aroh, N. Yan, L. Sun, and Z. J. Chen, “Cyclic GMP-AMP synthase is an innate immune sensor of HIV and other retroviruses.,” Science, vol. 341, no. 6148, pp. 903–6, Aug. 2013.

[2] R. Wassermann, M. F. Gulen, C. Sala, S. G. Perin, Y. Lou, J. Rybniker, J. L. Schmid-Burgk, T. Schmidt, V. Hornung, S. T. Cole, and A. Ablasser, “Mycobacterium tuberculosis Differentially Activates cGAS- and Inflammasome-Dependent Intracellular Immune Responses through ESX-1.,” Cell Host Microbe, vol. 17, no. 6, pp. 799–810, Jun. 2015.

[3] K. W. Ng, E. A. Marshall, J. C. Bell, and W. L. Lam, “cGAS–STING and Cancer: Dichotomous Roles in Tumor Immunity and Development,” Trends Immunol., vol. 39, no. 1, pp. 44–54, Jan. 2018.

[4] E. E. Gray, P. M. Treuting, J. J. Woodward, and D. B. Stetson, “Cutting Edge: cGAS Is Required for Lethal Autoimmune Disease in the Trex1-Deficient Mouse Model of Aicardi–Goutières Syndrome,” J. Immunol., vol. 195, no. 5, 2015.

[5] J. An, L. Durcan, R. M. Karr, T. A. Briggs, G. I. Rice, T. H. Teal, J. J. Woodward, and K. B. Elkon, “Expression of Cyclic GMP-AMP Synthase in Patients With Systemic Lupus Erythematosus,” Arthritis Rheumatol., vol. 69, no. 4, pp. 800–807, Apr. 2017.

[6] H. Yang, H. Wang, J. Ren, Q. Chen, and Z. J. Chen, “cGAS is essential for cellular senescence.,” Proc. Natl. Acad. Sci. U. S. A., vol. 114, no. 23, pp. E4612–E4620, Jun. 2017.