Discussion

IMPRESS participants, with some participants having persistently high signature scores at the end of treatment, as was observed in HIV-uninfected persons (Sweeney et al. 2016; Thompson et al. 2017). This suggests underlying inflammation and high type I/II IFN responses in these participants despite clinical and microbiological cure. Indeed, a heterogeneous pattern of pulmonary inflammation, measured by FDG-uptake, was also observed in the CTRC study, with some participants developing new lesions whilst a mixed response was observed in some participants (Malherbe et al. 2016). FDG uptake improved in 60% and resolved in 14% of the CTRC participants. Signature scores decreased in approximately 56% of the IMPRESS cohort at the end of treatment. However, because we did not have a matched control group to define signature scores in HIV-infected persons without TB with a history of previous TB disease, we could not define what a resolved response would be. This precluded distinguishing between resolved and improved response in this cohort. Underlying pulmonary inflammation was observed up to a year after successful cure in the CTRC study, suggestive of residual mycobacterial disease or bacterial activity despite control of disease after successful cure. This was also suggested by the detection of mycobacterial RNA in bronchoalveolar lavage samples (Malherbe et al. 2016). Consistent with this observation in HIV-uninfected patients, we observed underlying type I/II IFN response up to eight months after end of TB treatment in the IMPRESS cohort. The rPSVM.1 signature suggested that type I/II IFN activity increased up to eight months after successful cure, however the reason for this in not clear to us. Residual bacterial products may continue to trigger inflammation after recent TB disease, while circulating HIV loads may also trigger gene expression patterns in the periphery (the effect of HIV plasma viral loads on gene expression is discussed in the next chapter). Finally, this study was performed

in a high Mtb-transmission setting and reinfection may also contribute to elevation of signature scores. We conclude that our TB risk signatures can monitor TB retreatment response in HIV-infected persons, but with a reduced accuracy to that observed in HIV-uninfected persons.

When we applied the ACS 11-gene signature to identify rapid converters at baseline the signature could not correctly identify rapid converters at this time point. Our cohort consisted of HIV co-infected persons, most of who were not on ART (n=34) and had unsuppressed pVL (n=36) (see Chapter 6) at diagnosis. The ACS 16-gene signature could distinguish rapid from late converters at baseline in the CTRC and most importantly, treatment failures were identified at 24 weeks after the start of therapy (Thompson et al. 2017). This suggests that the ACS 16-gene signature could be a useful tool for customising duration of therapy, especially in individuals with treatment failure. The ACS 11-gene and rPSVM.1 signatures measure expression of ISGs. Expression of the transcripts in our signatures were higher around the TB diagnosis time points. However, both HIV infection and active TB disease induce expression of ISGs. Thus, the effect of active TB and HIV co- infection on type I/II IFN signature gene expression resulted in the inaccurate differentiation between early and late converters at TB diagnosis. After two months of treatment, signature expression remained elevated in some persons likely due to the presence of actively replicating bacteria. Thus late converters could be distinguished from early converters by the ACS 11-gene signature at month two. The gene signatures does not improve upon the diagnostic performance of culture conversion, but can be performed in individuals who cannot produce sputum and can be done in a much shorter time than culture, offereing some advantages.

Xpert MTB-RIF Ct values have been used as a measure of bacterial load and have been shown to predict treatment response and treatment failure in HIV-uninfected persons at diagnosis (Shenai et al. 2016). Time to culture positivity at diagnosis has been associated with sputum culture conversion at two months of TB treatment (Hesseling et al. 2010). In addition, sputum smear grading at diagnosis in the same study could predict treatment response. Thus, mycobacterial loads measured at baseline can be used to monitor treatment response in HIV-uninfected persons. In the IMPRESS study, culture conversion at month two was not associated with time to positivity at diagnosis. This raises the question whether mycobacterial load may be poorly predictive of treatment response in HIV-infected persons. In addition to HIV infection, the methodologies used to determine the time to positivity is different between the two studies. Hesseling and colleagues used the BACTEC 12B liquid radiometric culture method to determine time to positivity whilst we used the more sensitive MGIT automated system, thus we might be identifying participants as early converters who could have been identified as late converters using the radiometric system. Time to culture positivity used in conjunction with a biomarker may increase its’ accuracy in monitoring treatment response (Hesseling et al. 2010). The ACS 11- gene and rPSVM.1 signature scores were not associated with time to culture positivity in our cohort, contrary to what was observed in the CTRC study (Thompson et al. 2017). In addition, Thompson and colleagues observed an increase treatment response monitoring when mycobacterial load was combined with signature score into a new model. In our cohort, assessing signature scores and time to culture positivity at baseline did not improve clustering accuracy at baseline, month two, or at the end of treatment in monitoring treatment response. Based on this result, we

believed that it was not warranted to develop a model that combined time to culture positivity with signature scores into a classifier. We observed that HIV infection increased ACS 11-gene signature scores (Chapter 3) leading to a decreased specificity and diagnostic accuracy for active TB disease. We propose that high signature scores in all participants in the IMPRESS cohort resulting from HIV load is the most likely explanation for the poor classification of early and late converters (more discussion in Chapter 6).

The small sample size, particularly in the late converters study arm, was a limitation of our study. In addition, we could not determine the utility of signature scores in identifying treatment failures because no participants were identified with clinically defined treatment failure.

To conclude, the ACS 11-gene and rPSVM.1 signatures could not predict early conversion at TB diagnosis but showed promise as treatment monitoring tools in HIV-infected TB retreatment patients.