Systems-level computational translation of mouse and human transcriptomics reveals a role for the unfolded protein response in Mycobacterium tuberculosis infection

Numerous studies have identified similarities in blood transcriptomic signatures of tuberculosis (TB) disease phenotypes between mouse and human systems, such as type 1 interferon (IFN) production and innate immune cell activation, yet the pathophysiology observed in murine infection does not recapitulate some of the hallmarks of human TB disease. We hypothesized that some of these differences relate to quantitative differences in the importance of different biological pathways in the two systems, rather that completely distinct responses to infection. To test this hypothesis and address the critical animal-to-human gap, we applied a systems biology machine learning framework, Translatable Components Regression, which identifies axes of variation in the preclinical study that correspond with numerous biological pathways and processes and estimates their relative contributions to TB disease state in humans. Our modeling confirmed the presence of biological axes of variation that differed in their relative importance between species. Prominent among the pathways most predictive of human TB phenotype, beyond the previously established common signatures, is the infection-induced Unfolded Protein Response (UPR). To validate that this mechanism can be uncovered even in the murine context, wherein it has not previously been a major avenue of investigation, we show experimentally that this cellular stress pathway controls a variety of immune-related functions in Mycobacterium tuberculosis-infected mouse macrophages. Thus, our work here demonstrates how systems-level computational models enhance the value of animal studies for purpose of elucidating complex human pathophysiology.