Longitudinal Dynamics of a Blood Transcriptomic Signature of Tuberculosis

Rationale Performance of blood transcriptomic tuberculosis (TB) signatures in longitudinal studies and effects of TB-preventive therapy and coinfection with HIV or respiratory organisms on transcriptomic signatures has not been systematically studied. Objectives We evaluated longitudinal kinetics of an 11-gene blood transcriptomic TB signature, RISK11, and effects of TB-preventive therapy (TPT) and respiratory organisms on RISK11 signature score, in HIV-uninfected and HIV-infected individuals. Methods RISK11 was measured in a longitudinal study of RISK11-guided TPT in HIV-uninfected adults, a cross-sectional respiratory organisms cohort, or a longitudinal study in people living with HIV (PLHIV). HIV-uninfected RISK11+ participants were randomized to TPT or no TPT; RISK11− participants received no TPT. PLHIV received standard-of-care antiretroviral therapy and TPT. In the cross-sectional respiratory organisms cohort, viruses and bacteria in nasopharyngeal and oropharyngeal swabs were quantified by real-time quantitative PCR. Measurements and Main Results RISK11+ status was transient in most of the 128 HIV-negative participants with longitudinal samples; more than 70% of RISK11+ participants reverted to RISK11− by 3 months, irrespective of TPT. By comparison, reversion from a RISK11+ state was less common in 645 PLHIV (42.1%). Non-HIV viral and nontuberculous bacterial organisms were detected in 7.2% and 38.9% of the 1,000 respiratory organisms cohort participants, respectively, and among those investigated for TB, 3.8% had prevalent disease. Median RISK11 scores (%) were higher in participants with viral organisms alone (46.7%), viral and bacterial organisms (42.8%), or prevalent TB (85.7%) than those with bacterial organisms other than TB (13.4%) or no organisms (14.2%). RISK11 could not discriminate between prevalent TB and viral organisms. Conclusions Positive RISK11 signature status is often transient, possibly due to intercurrent viral infection, highlighting potentially important challenges for implementation of these biomarkers as new tools for TB control.

Up to 1.7 billion people are estimated to be infected with Mycobacterium tuberculosis (Mtb) (1); however, less than 10% may progress to tuberculosis (TB) disease in their lifetime (2,3). Diagnosis and treatment currently depend on screening for symptoms consistent with TB and collection of sputum for microbiological testing. However, prevalence of subclinical (asymptomatic) TB is high, such that symptom-based casefinding approaches would fail to detect approximately 50% of cases (4).
Biomarkers that allow identification of individuals with both subclinical and clinical disease are urgently needed. Such tests may be useful to guide confirmatory testing, or for screen-and-treat strategies to direct short-course TB-preventive therapy (TPT) to persons at highest risk of progression to TB. Many blood transcriptomic TB signatures with promise as TB triage tests (5), and as tests for predicting progression to TB disease (6), have been reported (7)(8)(9).
We developed a PCR-based 11-gene signature (RISK11) that predicted development of TB disease in individuals with Mtb infection up to 12 months preceding TB diagnosis and had promising diagnostic performance for TB (10)(11)(12). Diagnostic performance of RISK11 for TB in a prospective study of symptomatic HIVuninfected adults in a TB endemic setting was very promising (13). Prognostic performance for short-term prediction of incident TB was also good, but RISK11-guided TPT with 3 months of weekly rifapentine and isoniazid (3HP) did not reduce progression to TB disease over 15 months (13).
Although treatment of TB disease has been shown to reduce RISK11 score (12,14), the effect of TPT on transcriptomic signature scores in individuals without TB disease is not known. Longitudinal dynamics of RISK11 or other transcriptomic signatures in healthy persons without TPT have also not been described. If transcriptomic signatures are to be implemented for serial TB screening, an understanding of biomarker dynamics in the face of changing host, pathogen, and treatment factors is needed.
One factor that might affect transcriptomic signature score is viral infection. Among participants without TB disease, detectable HIV plasma viral load is known to be associated with elevated signature scores compared with undetectable viral load, likely due to induction of type I IFN and elevated expression of IFNstimulated genes (ISGs), which comprise RISK11 (12). Another ISG-inducing viral infection, influenza (15), has also been shown to affect transcriptomic signature scores. However, the effect of other common upper respiratory infections on TB signatures remains unexplored.
We evaluated RISK11 dynamics in adults with and without HIV, assessed the effect of TPT on signature dynamics, and evaluated the effect of upper respiratory tract organisms on RISK11 score. Some of the results of these studies have been previously reported in the form of an abstract (16

Study Design and Study Populations
We enrolled HIV-uninfected volunteers into two observational cohorts nested in the CORTIS (Correlate of Risk Targeted Intervention Study) trial (13) (ClinicalTrials.gov: NCT02735590) and HIV-infected volunteers into a cohort nested in the observational CORTIS-HR (Validation of Correlates of Risk of TB Disease in High Risk Populations) study (17), respectively ( Figure 1; Figure E1 in the online supplement). These three cohorts allowed  ORIGINAL ARTICLE evaluation of 1) transcriptomic signature dynamics in HIV-uninfected and HIVinfected individuals; 2) transcriptomic signature dynamics with and without TPT; and 3) effects of upper respiratory organisms on the transcriptomic signature. Parent studies. Briefly, healthy adult (18-60 yr) community volunteers with and without HIV residing in TB endemic communities in South Africa were recruited. Eligible participants did not have known TB disease, or household exposure to a person with multi-drug-resistant TB, within the last 3 years. Participants were investigated for TB at enrollment, end of study, and upon detection of symptoms consistent with TB during study follow-up of 15 months. Participants were tested with the RISK11 biomarker at screening and stratified as RISK11 1 or RISK11 2 based on an a priori test threshold of 60% (13). HIV-uninfected RISK11 1 participants in the CORTIS trial were randomized to receive TPT (weekly high-dose isoniazid and rifapentine for 12 weeks [3HP]) or no intervention; HIVuninfected RISK11 2 participants did not receive TPT. People living with HIV (PLHIV) were referred for standard-of-care antiretroviral therapy (ART) and TPT. Study protocols were approved by the institutional human research ethics committees of participating sites. The CORTIS trial was approved by the South African Health Products Regulatory Authority. All participants provided written, informed consent for participation.
Longitudinal cohorts. In the HIVuninfected longitudinal cohort, we aimed to determine the dynamics of the RISK11 signature over 12 months of follow-up, and the effect of 3HP on RISK11 signature dynamics, by repeat measurement of RISK11 at 3 and 12 months. RISK11 1 participants in the 3HP 1 treatment arm, and both RISK11 1 and RISK11 2 participants in the 3HP 2 arm, were enrolled at two sites (Worcester and Ravensmead).
In the HIV-infected longitudinal cohort, we enrolled RISK11 1 and RISK11 2 participants at five sites (Worcester, Ravensmead, Durban, Klerksdorp, and Rustenberg) and measured RISK11 at enrollment and Month 3 to determine RISK11 dynamics through 3 months.
Participants with prevalent TB at enrollment and those without RISK11 measurement at Month 3 or Month 12 (HIV-uninfected longitudinal cohort) were excluded from analysis of signature dynamics. TB screening and RISK11 assay were performed as described (13). Details are in the online supplement.
Respiratory organisms cohort. A subset of HIV-uninfected participants screened for CORTIS were contemporaneously and consecutively enrolled ( Figure E1), irrespective of symptoms or signs of upper respiratory tract infections, at one site (Worcester). Paired nasopharyngeal and oropharyngeal flocked swabs (FLOQSwabs; COPAN Diagnostics Inc.) were collected and stored in 1.5 ml of Primestore buffer (Longhorn Vaccines and Diagnostics) at 280 C. Participants coenrolled into the CORTIS trial were additionally investigated for TB disease at baseline; those only enrolled into the respiratory organisms cohort were not investigated for TB ( Figure E1).

Respiratory Pathogen Assay
For detection of respiratory viral and bacterial organisms, nucleic acid was extracted from swab samples using the Qiasymphony Virus/Bacteria Mini Kit (Qiagen) and quantified using a multiplex real-time quantitative PCR assay kit (Fast Track Diagnostics Respiratory Pathogens 33 Kit), according to the manufacturer's instructions, on the CFX96 Touch System lightcycler platform (Bio-Rad). This kit is designed to detect a panel of respiratory tract bacteria, viruses, and fungi (Table E4).

Statistical Analysis
Statistical analyses were performed in RStudio version 1.0.153 (RStudio PBC) or Stata version 16 (StataCorp). Receiver operating characteristic area under the curve (AUC) was computed to estimate diagnostic or prognostic performance of RISK11 measured at enrollment versus Month 3 through 15 months' follow-up and to determine whether RISK11 can differentiate participants with or without organisms. The 95% confidence interval (CI) for AUC and differences in paired AUC were estimated using a percentile bootstrap with 10,000 iterations. A Wilcoxon rank-sum test was used to test for differences in RISK11 scores between participants with and without respiratory organisms. Participants with viral and bacterial upper respiratory tract codetection were included in the viral organism group, as they had similar RISK11 score distributions.
A multivariable linear regression model was used to estimate the effect of detection of upper respiratory organisms on RISK11 score. A logit transformation was performed on RISK11 score before fitting the model, as RISK11 score is a proportion. The multivariable model was built using the likelihood ratio test method. First, an initial model with just RISK11 was fitted. Next, nested models were fitted and compared with the initial model. The variable in the model with the smallest Akaike information criterion value and making the most significant contribution was then added to the initial model. The process was repeated until no variable made a significant contribution to the model. Analyses in the 286 participants from the respiratory organisms cohort who were coenrolled in CORTIS were adjusted with sampling weights to reflect the screened population (13). Enrollment into CORTIS was based on RISK11 status. For reasons of trial efficiency, approximately 79% of all eligible RISK11 1 and only 13% of all eligible RISK11 2 participants were enrolled. Because of this enrichment in RISK11 1 participants in the enrolled population, sampling weights of 1.263 and 7.920 were assigned to RISK11 1 and RISK11 2 individuals, respectively, to obtain estimates of the screened population. An a of less than 0.05 was considered statistically significant in all analyses. All analyses in this substudy were performed post hoc without a prespecified statistical analysis plan.
Among RISK11 1 participants at screening who received 3HP, most reverted to RISK11 2 by Month 3 (74.4%, 32/43) or Month 12 (79.1%, 34/43) ( Figure 2B). Surprisingly, we observed the same phenomenon for RISK11 1 participants who did not receive 3HP, most of whom reverted to RISK11 2 by Month 3 (73.1%, 19/26) or Month 12 (84.6%, 22/26) ( Figure 2C). Only 4 of 71 RISK11 1 participants (5.6%) remained persistently RISK11 1 through the 3-and 12-month visits, none of whom developed incident TB. Together these results suggest that RISK11 positivity is transient in most people without HIV infection and that 3HP, which is thought to clear Mtb infection, did not increase the RISK11 reversion rate above that observed in 3HP-untreated individuals. Prevalence of IFN-g release assay-positive individuals were not different between the groups (Table E1).

RISK11 Reversion Is Less Common in PLHIV, but ART Initiation Leads to RISK11 Reversion
Next, we evaluated the dynamics of RISK11 in the longitudinal HIV-infected cohort. RISK11 data were available at both enrollment and Month 3 for 645 participants without prevalent TB (422 RISK11 2 and 223 RISK11 1 at enrollment) ( Figure 1C and Table E2). Nineteen of 645 participants progressed to incident TB. Among the 626 participants (417 RISK11 2 and 209 RISK11 1 ) without incident TB, 21.1% (88/ 417) of those who were RISK11 2 at enrollment converted to RISK11 1 at 3 months after enrollment, a conversion rate similar to that observed in the HIVuninfected cohort (13.6%, Fisher's exact test, P = 0.2) ( Figure 3A). However, the percentage of PLHIV without incident TB who were RISK11 1 at enrollment and reverted to RISK11 2 by Month 3 (42.1%, 88/ 209) ( Figure 3A) was significantly lower than in the HIV-uninfected cohort (73.1%, P = 0.022) ( Figure 2C), suggesting that HIV infection may contribute to persistence of RISK11 1 status.

Viral Upper Respiratory Tract Organisms Are Associated with Elevated RISK11 Scores
Because IFN-stimulated gene pathways are upregulated in HIV viraemia, we asked if other infections were also associated with elevated RISK11 scores. In total, 1,000 participants (148 RISK11 1 , 851 RISK11 2 , and 1 RISK11 indeterminate) were enrolled into the respiratory organism cohort ( Figure  1B and Table E3). Of these, 23 were coenrolled into the HIV-uninfected longitudinal cohort and 286 were coenrolled in CORTIS, which required prospective TB investigation ( Figures 1B and E1). All 999 participants with RISK11 results were screened for upper respiratory tract organisms regardless of symptoms. Viral       Table E4).
No viruses were detected in participants with prevalent TB. However, the proportion of individuals with a viral organism was significantly higher (P = 0.02) in those who progressed to incident TB (44.4%, 4/9) than in those who remained healthy (12.4%, 33/ 266). Participants with a viral organism were five times more likely to progress to pulmonary TB than those without a viral organism (incident rate ratio, 4.99; 95% CI, 0.99-23.17).

Discussion
We evaluated longitudinal kinetics of the RISK11 host blood transcriptomic signature of TB risk in HIV-uninfected and HIVinfected individuals, with and without TPT. Our results suggest that sustained RISK11 positivity is rare. Rather, positive RISK11 status is transient in people not treated with TPT, reverting to negative within 3 months in the majority of HIV-uninfected individuals and a large proportion of PLHIV. Those receiving TPT displayed the same kinetics, suggesting that TPT was not the primary mediator of RISK11 reversion. Our data suggest that intercurrent respiratory viral infections, such as rhinovirus and coronavirus, drive signature conversion and transient positive TB risk status. It is also clear that HIV infection elevates RISK11 signature scores, as reported previously (12,17). A recent study of healthy TB-exposed individuals living in a TB-endemic setting demonstrated that the majority had detectable Mtb DNA in peripheral blood leukocytes (18). Detectable Mtb DNA was not associated with IFN-g release assay positivity, but significantly decreased after isoniazid preventive therapy. These findings suggest that undiagnosed, subclinical Mtb infection may contribute to transient RISK11 1 results observed in our study. RISK11 positivity was more frequent in PLHIV than those without HIV; RISK11 scores were higher in ART-naive than ARTexperienced PLHIV; and RISK11 scores fell after ART initiation. Despite these findings, we note that RISK11 had good diagnostic performance for symptomatic TB and excellent prognostic performance for incident TB within 6 months of testing in HIV-uninfected (13) and HIV-infected people (17).
Nevertheless, we infer that upper respiratory and other systemic viral infections contributed to the RISK11 1 prevalence rate in the CORTIS (9.3%) and CORTIS-HR (34.8%) study populations (13,16,17). Since TPT had a negligible effect on RISK11 status, it appears that RISK11 positivity due to active Mtb infection is relatively uncommon, consistent with the much lower incidence of TB disease compared with RISK11 1 prevalence. Our results highlight the important effects of upper respiratory and systemic viral infections on ISG-based transcriptomic signatures, as suggested previously (19)(20)(21). This illustrates the need for confirmatory testing and/or combined TB and viral screening, which is being implemented in TB-endemic settings for coronavirus disease . Alternatively, further development and validation of signatures, such as those proposed by Singhania and colleagues and Esmail and colleagues (15,19), which are not markedly influenced by ISG modulation, is required.
A key question is whether upper respiratory viral infections not only affect transcriptomic signatures independent of TB risk, due to induction of type I IFN, but whether they also induce changes in immune control of Mtb infection and precipitate progression to TB disease (22). For example, we have shown that HIV infection increases ISG-based signature score in healthy controls, independent of TB, in addition to the known elevated risk for progression to TB in PLHIV. We present preliminary evidence of a possible association between respiratory viral organisms and risk of incident TB, supporting the hypothesis that respiratory viral infections may trigger TB disease progression, or that immune dysfunction increases susceptibility to both viral and Mtb infection. Participants with respiratory viral organisms had fivefold increased risk of TB progression compared with those in whom no viral organisms were detected (Table E6).
Although based on small numbers, which limits the significance of the result, the intriguing finding that viral coinfection may be associated with increased risk of TB should be tested more rigorously in future studies. Previous studies have reported that viral respiratory coinfections are associated with accelerated progression or more severe TB disease (23,24). Patients with TB with nasopharyngeal viral-bacterial coinfection were also likely to have more severe TB disease (25), while murine influenza infection led to impaired control of Mtb infection via a type I IFNdependent mechanism (26). However, no association was found between upper respiratory organisms and pulmonary TB in a Cape Town pediatric cohort (27).
Our study had several limitations. First, our study cohorts were small and included too few incident TB cases to allow robust assessment of longitudinal signature dynamics in relation to progression to TB disease. Second, only 29% of participants in the respiratory organisms cohort provided sputum for TB investigation, and therefore TB disease could not be excluded in the remainder. We also did not seek alternate diagnoses in symptomatic participants without detected respiratory tract organisms or TB disease, nor did we seek to identify lower respiratory or gastrointestinal organisms, which may also modulate RISK11 scores. We therefore hypothesize that the remainder of the variation in RISK11 score that is not explained by the model may be explained by organisms and host factors that we did not measure. Finally, we acknowledge that a number of the respiratory organisms we tested for are not typically pathogenic and occur as commensals.
Several characteristics of the study design strengthen our results. First, participants were enrolled consecutively as they presented for screening and enrollment into the parent studies, thereby minimizing selection bias. Second, this study used both oropharyngeal and nasopharyngeal samples, which increases the detection rate of organisms. Third, to our knowledge this is the first study to investigate temporal dynamics of a transcriptomic signature of TB risk in HIV-uninfected and HIV-infected cohorts, with and without TPT, and to investigate perturbation of an ISG-based signature by common respiratory viral organisms in individuals at risk for incident TB disease.

Conclusions
This study provides insight into the association between common viral coinfections and one of several published TB signatures (5,6,28), highlighting challenges for implementation of these biomarkers as new tools for TB control. Control for confounding factors associated with elevated host blood transcriptomic signature scores, including viral infection, may be critical for implementation as potential TB biomarker tests. It is not yet known to what degree these results are generalizable to other host blood TB transcriptomic signatures, a question that needs to be addressed.