Introduction

HIV infection increases the risk of progression to active TB in Mtb-infected persons (Parida and Kaufmann 2010). A five-fold increase in TB notification rates has been observed due to HIV infection (Lawn et al. 2009). In 2016, 46% of incident TB cases were HIV-infected (World Health Organization 2017). HIV-infected persons present with paucibacillary disease (low numbers of acid-fast bacilli in sputum) thereby making diagnosis of TB using current diagnostic tools difficult. Sputum culture is a more sensitive technique in diagnosing TB in HIV-infected persons than sputum smear microscopy, however, it takes days to week to confirm the diagnosis (Tornheim and Dooley 2017; Cain et al. 2010; Stephen D Lawn et al. 2006).

Furthermore, a large proportion of HIV-infected persons do not have productive coughs (Cain et al. 2010) and are thus unable to provide sputum specimens. Hence, new tools to improve and shorten time to diagnosis of active TB in HIV-infected persons are essential in the fight against TB.

Transcriptomic signatures of TB are promising tools for the diagnosis of active TB disease. Many diagnostic TB signatures have been shown to yield good diagnostic performance in HIV-uninfected persons, either to differentiate between active TB

cases and healthy infected or uninfected controls, or symptomatic controls with other respiratory diseases (Berry et al. 2010; Blankley et al. 2016; Bloom et al. 2012, 2013;

Kaforou et al. 2013; Maertzdorf et al. 2011; Maertzdorf et al. 2011; Walter et al.

2016; Zak et al. 2016). Only a few studies have developed transcriptomic TB diagnostic signatures in HIV-infected populations (Kaforou et al. 2013; Anderson et al. 2014; Dawany et al. 2014). Colleagues at SATVI and CIDR developed a transcriptomic 16-gene correlate of risk of TB signature capable of predicting TB disease in HIV-uninfected persons more than a year prior to TB diagnosis (Zak et al.

2016). This 16-gene risk signature also performed well in diagnosing active TB disease, when re-parameterised and applied to public, published microarray datasets from the Berry et al 2010, Bloom et al 2012, Kaforou et al 2013 and Maertzdorf et al 2011 (Zak et al. 2016). The performance of this risk signature in diagnosing active TB disease in HIV-infected persons has not been compared to its performance in HIV-uninfected persons from the same population. Hence, we sought to determine the performance of the 11-gene TB risk signature derived from the 16- gene signature (described in Chapter 1) in diagnosing active TB in HIV-infected persons.

An important aim we sought to address in this project was to determine if our transcriptomic signatures could predict recurrent TB in HIV-infected persons.

However, the only available clinical cohort to address this aim was the TRuTH study, in which PBMC samples and no whole blood RNA samples were collected and biobanked. Since the 16-gene signature from which the 11-gene signature was derived was developed on whole blood samples, and PBMC are distinct from whole blood in that they lack neutrophils and other granulocytes. We therefore reasoned

that a careful comparison of signature gene expression between whole blood and PBMC was necessary. This is especially relevant because the Type I IFN response genes found to be associated with active TB are most prominently expressed by neutrophils compared to other cell types (Berry et al. 2010; Singhania et al. 2017;

Scriba et al. 2017).

Previous studies have shown that transcriptomic signatures developed from PBMC samples can diagnose active TB disease in HIV-uninfected persons (Lee et al. 2016;

Lu et al. 2011). Lee and colleagues developed gene signatures to differentiate active TB from healthy Mtb-infected and Mtb-uninfected persons. The study identified 297 genes that were differentially expressed between active TB cases and Mtb- uninfected persons, and 127 genes that were differentially expressed between Mtb- infected and -uninfected persons. There were 169 genes differentially expressed genes active TB and Mtb infection, and 14 of which genes were found to be differentially expressed between Mtb-infected and -uninfected persons. These genes were enriched in immune activation and regulation, cell differentiation, and mRNA transcription and translation modules. Three of the differentially expressed genes in the three comparisons differentiated Mtb-infected from uninfected persons and active TB cases in an external validation cohort from the same setting (Taiwan). Lu and co- workers on the other hand, measured gene expression in active TB cases, Mtb- infected, and uninfected persons using PBMC stimulated with PPD for four hours or unstimulated PBMC as controls. A three-way pair-wise analysis revealed 506 differentially expressed genes amongst the three groups based on fold changes between PPD stimulated and unstimulated samples. Fifty-five genes mainly involved in T cell homeostasis were differentially expressed when active TB and Mtb-infected

individuals were compared to Mtb-uninfected persons. Furthermore, 229 genes related to chemotaxis and responses to external stimulus were differentially expressed for active TB (active TB relative to Mtb-infected and uninfected persons).

Of the 506 genes from the three pair-wise analyses, 30 differentially expressed genes and 22 genes specific for active TB and Mtb infection were translated into qPCR. These genes were applied to a validation cohort of active TB cases, Mtb- infected and uninfected persons using qPCR. Of these, a three-gene signature (CXCL10, ATP10A, TLR6) was found to be the most discriminatory in identifying active TB disease from Mtb infection using decision trees. The three-gene signature had a sensitivity and specificity of 80% and 89% respectively with an accuracy of 85%. These studies suggest that gene signatures developed from PBMC samples can diagnose active TB with a high accuracy.

However, the diagnostic performance of whole blood based transcriptomic signatures in differentiating between active TB disease and Mtb infection using PBMC samples is unknown. We therefore set out to assess the mRNA signal strength of the ACS 11-gene signature in PBMC.

3.2 Aim and hypothesis