World TB day, March 24th, is an annual commemoration of the day in 1882 when Robert Koch announced his discovery that Mycobacterium tuberculosis is the bacterium responsible for tuberculosis (TB). M. tuberculosis (Mtb) is, according to the World Health Organization, “a globally established priority for which innovative new treatments are urgently needed.” Nearly one-third of the world population is infected with Mtb and and in 2015 there were 1.8 million deaths due to TB worldwide. On a positive note, the TB infection rate has declined by an average of 1.5% per year since 2000, but 4-5% per year decline is needed to achieve benchmarks set by the “End TB Strategy.” Fittingly, the theme for the 2017 World TB Day is “Unite to End TB.”
Slow to grow, slow to treat
Figure 1. Colonies of Mycobacterium tuberculosis on an agar plate appear cream-colored and rough. Source
One of the conundrums facing TB research is the fact that Mtb grows incredibly slowly. Slow growth confounds both research and treatment since it can take days to weeks to grow a culture for study, a month or more to determine drug resistance properties of a patient sample, and months of antibiotic therapy once an infection is confirmed. At the University of Michigan, two labs have united to develop research strategies that bypass this limitation to provide new insights about TB. By using computer models, labs run by Denise Kirschner, Ph.D. and Jennifer Linderman, Ph.D., MSE are generating important insights into TB treatment, vaccines, and disease progression that might speed up the process of ending TB.
Computational modeling compiles the lab measurements of immune response or bacterial properties that occur during an infection in order to simulate potential outcomes. By altering the conditions of the modeled infections (like antibiotic treatments), computer models can perform hundreds or thousands of complex, hypothetical experiments in moments. Thus identifying new mechanisms for researchers at the bench to explore further. (Kirschner lab alum, Hayley Warsinske, Ph.D., explains the complexities here.)
The tuberculosis timeline
Modeling an event as complex as a Mtb infection and the corresponding immune response requires a nuanced understanding of the events that occur. TB infection begins with inhalation of Mtb bacilli from an infected individual by an uninfected person. In the lungs, macrophages, the early responders of the innate immune response, engulf the bacteria, attempting to destroy the invaders. However, Mtb is quite durable, employing many strategies to avoid killing by macrophages that allow survival and continued growth of the bacilli, both within the lungs and even within the hostile intracellular environment of macrophages. The death of those infected macrophages causes the release of messenger molecules called cytokines that recruit more innate immune cells such as monocytes (immature macrophages) to continue battling Mtb.
At the same time, scout cells such as dendritic cells collect Mtb samples and carry them to the nearest lymph node, a command center for the adaptive immune response. The dendritic cells display degraded Mtb proteins on their cell surfaces, looking for T cells that recognize the peptides. When a T cell match is found, cytokines direct the T cell to mount either an inflammatory or anti-inflammatory immune response. The types of cytokines released by the dendritic cells and the proportion of different cytokines to the others are key in directing the T cell response. The newly activated anti-Mtb T cells then migrate to the site of infection. The process up to this point takes about 2 months and results in a Mtb-specific immune response.
Figure 2. Tuberculosis granulomas from a nonhuman primate (top) and computer modeled (bottom). These are about 2mm across, about the width of a nickel. The granuloma wall is mostly composed of inactive macrophages (bright green, bottom). Since Mtb doesn’t stain readily, the bacteria (olive green, bottom) aren’t readily visible in the stained granuloma (top). Other cell types present include: activated macrophages (blue), infected macrophages (orange, red), and T cells (pink, purple, light blue). Source, Fig. 3c
When the Mtb-specific T cells respond, they produce more cytokines that activate the macrophages, enhancing their ability to kill intracellular Mtb. This leads to the formation of a granuloma, a cocoon of macrophages (active and inactive) around surviving bacteria that prevents them from spreading any further. In some cases, the bacteria are eventually cleared from the granulomas (sterilized). In other cases of latent Mtb, the bacteria survive within the granulomas but the infection remains asymptomatic. Such infections can later reactivate, a process where Mtb escapes the granuloma to cause further damage to the lungs and re-initiate the immune response.
Translating the biology of infection into a computational model
According to Joey Cicchese, a graduate student in the Linderman lab, the crux of computational modeling is “translating the mechanisms that govern biological processes into computer code.” To translate the timeline of events during a Mtb infection into code, Kirschner and Linderman need numbers: how many macrophages are present and how many are active, how many bacteria are alive, what cytokines are produced and to what extent, how many T cells are recruited, how many granulomas are formed, how many are sterilized, how many latent granulomas are reactivated, etc. And all of this from different time points during an infection. They obtain this data from collaborators studying TB in animal models such as nonhuman primates (JoAnne Flynn, Ph.D.) and rabbits (Véronique Dartois, Ph.D.).
The next step is using the code as a framework to simulate the biological events and their eventual outcomes. There is a vast number of events, molecules, and cells to take into account, so the models can be built at different degrees of interaction: the cellular level (between macrophage and bacilli or dendritic cell and T cell), the tissue level (granulomas in the lung, lymph nodes, T cell migration or drug dynamics in the blood), the body level (a single virtual patient), and the population level (many virtual patients). Results from one model can, in turn, be scaled up to populate the model for the next degree of interaction or patched together with another model at a similar interaction level (e.g., lymph node to blood).
The real power behind computational modeling, Cicchese notes, is that “during our granuloma simulations, we can keep track of bacteria and immune cell counts at any time throughout the simulation.” He goes on to add that “because computational modeling is typically faster than most wet lab experiments (e.g., from Flynn and Dartois), it can be a way to guide experimentalists to studies that are more likely to succeed or produce interesting results.”
From models to identifying new hypotheses
Just in the last couple of years, Kirschner and Linderman have published at least three intriguing models or hypotheses for experimentalists to follow up on. TB treatment regimens last months and require the use of multiple antibiotics to prevent drug resistance, problems that confound research into the efficacy of new therapeutics. A 2015 study published a combined model system that provides the ability to look at drug dynamics at many different levels (blood, tissue, and granuloma). This methodology could predict the odds of an infection developing drug resistance under different dosages and frequencies. Additionally, the model could predict the efficacy of treatment strategies like immune modulation and therefore save time, money, and lives during clinical trials.
Another recently published multilevel model identified biomarkers capable of predicting the spectrum of outcomes between fully active TB infection and latent TB. The spectrum is influenced by many factors, including the immune response, the bacterial strain, and the treatment strategy. Being able to predict disease outcome in a disease of long duration like TB would allow clinicians earlier opportunities to alter treatment and improve outcomes. Previous attempts to identify biomarkers have been unsuccessful, in part because the best biomarkers should be easily accessed (e.g., via blood draw) but the cytokines or cell types present in the blood don’t necessarily represent what’s occurring in the lungs. Using nonhuman primate data, their model correlated Mtb-specific T cells with positive disease outcomes. The more Mtb-specific T cells in the blood, the more likely the infection is active versus latent. This gives experimental researchers a direction (identifying Mtb-specific T cells) to pursue in their search for a reliable biomarker.
Activating such Mtb-specific T cells is essential to contain a TB infection, and if this can be achieved by vaccination, infection could be avoided. However, eliciting a strong T cell response through vaccination is tricky and only one vaccine, BCG, is currently used to prevent TB. Administered in infancy, the vaccine protects children but is largely ineffective by the time adolescence is reached. A granuloma model reported in Frontiers in Microbiology by Linderman and Kirschner tests the efficacy of various vaccine types, routes, and doses. While they caution that a larger system model is needed to better understand what happens at the body and population level, this model would be valuable for virtual trials of new TB vaccines and to predict vaccine components that will activate Mtb-specific T cells.
Collectively, computational research offers experimental researchers a way to preview expensive and complex hypotheses. But as Cicchese pointed out “the simulated data are only as good as the experimental data.” Collaborations between computer scientists, wet lab researchers, and clinicians are essential to TB research. The work pioneered by Linderman, Kirschner, Flynn, and Dartois is just one example of how we can unite to end TB.