(709d) Computational Methods at the Intersection of Tuberculosis Immunology and Therapy

Tuberculosis (TB) poses a global public health threat, causing 9.6 million new infections and 1.5 million deaths in 2014 (WHO Global tuberculosis report 2015). While most drug-susceptible TB is treatable with a combination of four antibiotics, the treatment lasts 6-9 months and has failed to prevent drug-resistant TB worldwide. Three of the challenges facing new TB treatment strategies are: 1) complex disease pathology (i.e. spherical lesions called granulomas form in the lung), 2) variable bacterial susceptibility to antibiotics, and 3) combination therapy (necessary to avoid drug resistance). It is difficult to address these challenges in a single experimental system; the costly non-human primate model of TB is the only animal model that recapitulates human-like pathology, and in vitro systems do not capture the spatial or long-term temporal aspects of TB pathology.

We use a computational approach to integrate spatial and temporal data from multiple in vitro and in vivo experimental systems. The experimental data encompass a large variety of temporal scales, spatial scales, biological compartments and experimental methods. The biology, experimental data type and model structure determine the model calibration workflow. Quantitative approaches in complex biological systems require well-defined strategies to incorporate a variety of experimental data. We illustrate one such strategy for TB immunology and therapy. Our computational model captures host immunity, bacterial dynamics (including differential antibiotic susceptibility), and antibiotic dynamics in plasma, lung tissue and TB granulomas. The model comprises a combination of agent-based, ODE and PDE models. We calibrate each of the model components to non-human primate microbiological and immunological data, rabbit and human pharmacokinetic (PK) data, and in vitro and animal pharmacodynamic (PD) data. In some instances multiple experimental data types are available for a single model component. E.g. antibiotic distribution in granulomas is measured using LCMS as well as MALDI-MS. In such situations the model is calibrated to both data sets in parallel, weighting the contribution of each data set based on known experimental measurement constraints.

Our model bridges and connects multiple experimental data sets, and predicts efficacy of multiple antibiotics in the context of host immune dynamics. This model is applied to efficiently and systematically evaluate potential TB treatment regimens containing five anti-TB antibiotics (isoniazid, rifampin, moxifloxacin, gatifloxacin and levofloxacin), to complement and inform future animal or human trials.