Using the OSG to investigate the human immune response to infection with M. tuberculosis

Tuberculosis (TB) remains the world’s leading cause of death from infectious disease. Despite the availability of antibiotic treatment, TB is responsible for around 2 million deaths per year. Together, TB and HIV-1 co-infections account for a dramatic level of morbidity and mortality in the world.

Dr. Denise Kirschner, a professor in the Department of Microbiology and Immunology at the University of Michigan Medical School, and Dr. Jennifer Linderman, a professor in Michigan’s Chemical Engineering department, create predictive models of the human immune response to the pathogen that causes TB, Mycobacterium tuberculosis. Their goal is to use a systems biology approach to directly and significantly impact our understanding and treatment of TB.

The human immune response is a complex biological system that encompasses multiple organs (lymph nodes and particular sites of infection such as the lung), length scales, and time scales. However, while effective in fighting many pathogens, its success rate in clearing M. tuberculosis is not very good. Five to ten percent of initial infections with M. tuberculosis result in primary TB, and 90% result in latent infection in 2 billion individuals worldwide. Of those with latent infections, around 10% will later progress to active disease (reactivation TB). Drug resistant forms of the bacteria are becoming more common, risking a global increase in the disease unless more effective prevention and therapeutic methods are developed.

The host factors that control the outcome of infection—whether primary TB, latent infection, or reactivation TB—are not well understood. The particularly slow progression of TB in humans and non-human primates (the best animal model for the disease) makes comprehensive data collection particularly difficult and expensive for scientists.

To overcome these limitations, over the past decade Drs. Kirschner and Linderman and their students have developed a series of computational models to qualitatively and quantitatively characterize the immune response to M. tuberculosis infection in the lung and lymph node. Their goals are to predict biological mechanisms that lead to different infection outcomes, to generate new systems-level hypotheses as to how the immune response fails, to achieve a predictive and systems-level understanding of the immune system’s interactions with mycobacteria, and to provide a platform to design and test potential therapeutics via computer simulation. Their computational models are informed by data from multiple animal models, including mice and non-human primates.

Resources available from the Open Science Grid (OSG) have helped the Kirschner-Linderman group overcome computational limitations through access to a large number of compute cores. If not for the OSG, the cost of the computational resources they need would exceed the funding available via standard research grants. The studies they perform involve computer simulations of events occurring at molecular, cellular, and tissue length scales and time scales of seconds to years. These require high throughput computational resources like those available through the OSG.

Their analysis requires performing uncertainty and sensitivity analysis, which involves large numbers of parameter searches. This calls for computer systems with data processing and storage that far exceed the performance and data storage capabilities of standard computer systems or local clusters. Using the OSG, parameter sweeps with thousands of runs can be performed (which would otherwise be impractical). This, in turn, enables the transition from two-dimensional to three-dimensional models, a necessity for the deeper understanding of the immune response to infection that will be required to achieve the ultimate goals of better prevention and treatment of TB.

For example, the in silico (i.e., computer simulated) models shed some light on mechanisms governing granuloma formation and function, as well as on the spatial distribution and balance of key pro- and anti-inflammatory molecules during infection (measures that can’t be performed in vivo, i.e., in living organisms—see figures 1 and 2). The models also provide immunotherapy strategies to deal with latent TB and reactivation.

Figure 1:

Figure 1: Comparison between a necrotic granuloma from a non-human primate (Panel A) and an in silico granuloma generated by our Agent-Based Model (ABM, Panel B). Panel A: cells in the image are labeled as phosphorylated-STAT3 (red, anti-inflammatory cells), phosphorylated-STAT1 (green, pro-inflammatory cells) and CD163 (blue, anti-inflammatory cells). Panel B: ABM snapshot at day 200 post infection. Pro-inflammatory cells are in green, anti-inflammatory cells are in red.

Figure 2:

Figure 2: A: Changes of total number of Mycobacterium tuberculosis (Mtb) with time for simulation of containment baseline, a case of Mtb clearance, and deletion (or knock-out) of two key molecules (tumor necrosis factor [TNF] and interferon γ[IFN-γ]). B-E: Granuloma structures for a case of containment, clearance of Mtb infection in less than five weeks as a result of an efficient immune response, a TNF deletion, and an IFN-γ deletion, respectively.

Currently, TB treatment is lengthy (six to nine months) and occasionally ineffective. By incorporating antibiotic, bacterial, and immunological dynamics to explore events during antibiotic treatment, the Kirschner-Linderman group aims to answer questions like: Why does TB treatment take so long? Why are some infections resistant to treatment? Their in silico models can perform research not possible experimentally, such as a true side-by-side comparison of outcomes using different antibiotics and dosing schedules in the same granuloma (see figure 3).

Figure 3:

Figure 3: Antibiotic distribution within a simulated granuloma. Panel A: simulation snapshot of a granuloma at day 100 post infection. Panels B and C: Isoniazid (B) and Rifampin (C) are two first line antibiotics used to treat TB and distribute differently within the granuloma. Heatmaps indicated spatial concentrations of each antibiotic.

Because these models are large and require complex computer codes with very long running times, there were a number of technical challenges to overcome. OSG’s excellent technical support has been essential in overcoming those challenges.

~ Paul Wolberg