Identification of host gene expression biomarkers for tuberculosis

The presence of disease, including infectious disease, has been observed to give rise to specific patterns of gene expression in peripheral whole blood, regardless of disease site. These gene expression signatures allow for distinction between diseases and have the potential to reform diagnostics, particularly in diseases and patient groups for whom current diagnostics are unreliable, like Tuberculosis (TB). Although TB is a treatable infectious disease, it has high morbidity and mortality, especially in low resource countries and HIV infected patients. In this thesis, I propose a bioinformatics toolbox that derives minimal transcriptomic signatures from microarray datasets acquired from heterogeneous groups regardless of underlying co-infections and geographic locations. The transcripts’ expression values are then aggregated into a single value disease risk score (DRS) for every patient, that allows for classification between the disease groups in a binary manner. The toolbox was employed to analyse an adult and a paediatric TB transcriptomic study, comprising HIV infected and uninfected patients from sub-Saharan Africa. In the adult study, the DRS based on a 27-transcript signature distinguished culture confirmed TB from latent TB infection (LTBI), while 44 transcripts distinguished TB from other diseases phenotypically similar to TB (OD), with high sensitivity and specificity. Out-of-sample validation was performed using a publicly available dataset. In the paediatric study, a 51-transcript signature distinguished TB from OD and a 42-transcript signature from LTBI. The signatures were validated out-of-sample using an independent cohort and benchmarked against culture-negative TB patients and Xpert® MTB/RIF, currently used for detection of M. tuberculosis. This thesis provides proof of principle that minimal host blood transcriptional signatures are able to distinguish TB from LTBI and OD regardless of HIV infection. The subsequent transformation of the signatures into a score for every patient may facilitate disease categorisation and potentially development of diagnostic tools.