Multi-country evaluation of RISK6, a 6-gene blood transcriptomic signature, for tuberculosis diagnosis and treatment monitoring

There is a crucial need for non-sputum-based TB tests. Here, we evaluate the performance of RISK6, a human-blood transcriptomic signature, for TB screening, triage and treatment monitoring. RISK6 performance was also compared to that of two IGRAs: one based on RD1 antigens (QuantiFERON-TB Gold Plus, QFT-P, Qiagen) and one on recombinant M. tuberculosis HBHA expressed in Mycobacterium smegmatis (IGRA-rmsHBHA). In this multicenter prospective nested case–control study conducted in Bangladesh, Georgia, Lebanon and Madagascar, adult non-immunocompromised patients with bacteriologically confirmed active pulmonary TB (ATB), latent TB infection (LTBI) and healthy donors (HD) were enrolled. ATB patients were followed-up during and after treatment. Blood RISK6 scores were assessed using quantitative real-time PCR and evaluated by area under the receiver-operating characteristic curve (ROC AUC). RISK6 performance to discriminate ATB from HD reached an AUC of 0.94 (95% CI 0.89–0.99), with 90.9% sensitivity and 87.8% specificity, thus achieving the minimal WHO target product profile for a non-sputum-based TB screening test. Besides, RISK6 yielded an AUC of 0.93 (95% CI 0.85–1) with 90.9% sensitivity and 88.5% specificity for discriminating ATB from LTBI. Moreover, RISK6 showed higher performance (AUC 0.90, 95% CI 0.85–0.94) than IGRA-rmsHBHA (AUC 0.75, 95% CI 0.69–0.82) to differentiate TB infection stages. Finally, RISK6 signature scores significantly decreased after 2 months of TB treatment and continued to decrease gradually until the end of treatment reaching scores obtained in HD. We confirmed the performance of RISK6 signature as a triage TB test and its utility for treatment monitoring.


Results
Sociodemographic and clinical characteristics. A total of 141 patients with bacteriologically confirmed pulmonary ATB were included in the study. Their sociodemographic and clinical characteristics were compared at baseline. The median age was 28 years, 66% were male, and 51.8% were smokers. Among them, 48.2% had a positive sputum smear microscopy with a high grade at baseline (2+ or 3+). 97 of these patients were followed at least until the end of treatment and have been successfully treated for TB. The remaining participants included 26 individuals with LTBI and 71 healthy donors ( Table 1).
Performance of the RISK6 signature as a screening and triage test for pulmonary TB disease. To investigate the use of RISK6 score as a screening and triage test for TB, we compared RISK6 scores between patients with ATB disease (n = 141), treated TB patients who have been successfully treated for TB (TREATED, n = 97, with negative sputum culture at T2 and/or T3), the individuals with LTBI (n = 26), and healthy donors (HD, n = 71). In all cohorts, RISK6 scores were significantly higher in ATB patients at baseline compared to HD (p < 0.001) and TREATED TB patients (p < 0.001) (Fig. 1a). Moreover, RISK6 score levels of TREATED patients became indistinguishable from HD. Remarkably, in the Madagascar cohort that includes the enrolled LTBI individuals, we observed a significant difference for the RISK6 scores between ATB and LTBI group (p < 0.001) but not between the LTBI group and the TREATED TB patients or the HD group. Remarkably, when we compared the RISK6 scores levels between study sites, we found that the RISK6 scores levels in ATB, TREATED TB patients and the HD recruited from Bangladesh were higher than the levels observed in the other study sites (Fig. 1a).
We then generated a receiver operating characteristic curve (ROC) and the respective areas under the curve (AUC) for each cohort to evaluate, by country, the performance of RISK6 signature as screening or triage test (Fig. 1b). First, we assessed the performance of RISK6 as a screening test for the discrimination between ATB patients and HD. Remarkably, the performance of the RISK6 signature was similar in the four different cohorts, with outstanding AUC values ranging from 90.1% (Bangladesh; 95% CI 80.7-99.4) to 96.4% (Georgia; 95% CI 90.5-100) (Fig. 1b). Secondly, ROC analysis was also performed to determine the potential of RISK6 signature as a triage test to discriminate between different stages of TB infection. Results demonstrated a powerful classifying potential to discriminate patients with ATB from LTBI or TREATED TB patients with an AUC of 92.8% (95% CI 85.6-100) and 96.1% (95% CI 91.7-100) respectively (Fig. 1b). Remarkably, we also found that the discrimination between ATB and HD was lowest in the cohort of Bangladesh when compared to other study sites (Fig. 1b). Performance of RISK6 as a confirmatory test for pulmonary TB disease. Our next aim was to evaluate the performance of RISK6 signature in sputum smear-negative and culture-confirmed TB individuals. Based on the TPP criteria set by the WHO as a reference 19,27 , we found that RISK6 achieved the minimal sensitivity of > 60% with 100% specificity for an initial TB diagnostic test for sputum smear-negative TB to replace smear microscopy in the cohort from Georgia (Table 2). Similarly, in the same cohort, RISK6 signature also reached the minimum criteria of 65% sensitivity and 100% specificity for a confirmatory test. However, RISK6 signature detection failed to meet these WHO requirements in the other study sites ( Table 2). As most ATB patients had a positive sputum smear microscopy with a high grade at baseline, we wondered if RISK6 scores and mycobacterial loads were correlated. We therefore performed a sub-analysis on stratified sputum smear microscopy results among ATB patients, defined as follow: negative smears, low-grade positive smears (1+ or scanty) and high-grade positive smears (2+ or 3+). RISK6 scores in the negative smear group showed a significant difference (p < 0.001) compared to HD (Fig. 2). Moreover, RISK6 scores were significantly lower (median = 0.31, IQR 0.22-0.40) in negative smears than in individuals with low-or high-grade positive smears (p < 0.001). While not statistically different (p > 0.05), RISK6 scores in the high-grade smear group were higher (median = 0.5, IQR 0.40-0.56) than in the low-grade mycobacterial load group (median = 0.46, IQR 0.38-0.52).
Performance of RISK6 signature compared to IGRAs. Next, we assessed the performance of RISK6 signature compared to two assays based on IFN-γ release: the commercial QFT-P, and the non-commercial IGRAs-rmsHBHA. Compared to the QFT-P assay, the RISK6 signature achieved better performance in AUC (94.1% vs 57.2%), sensitivity (90.9% vs 50.9%) and specificity (87.8% vs 57.2%) to discriminate ATB patients from an asymptomatic population (LTBI + HD) (Table 3). However, a comparative sub-analysis indicated a lower positive (79.7%) and negative (50%) predictive values of the RISK6 signature when compared to QFT-P assay (100% and 63.9%, respectively) in detection of Mtb-infected individuals (ATB + LTBI) from uninfected ones (HD). Notably, the RISK6 signature showed a higher performance (AUC 90.9%, 95% CI 87.2-94.5), with 90.1% sensitivity and 72.2% specificity than the IGRAs-rmsHBHA (AUC 75.3%, 95% CI 68.6-82) that achieved RISK6 as a biomarker for TB treatment monitoring. Patients with successful treatment (defined as negative sputum culture at T2) were selected to determine whether RISK6 signature was a clinically relevant biomarker for TB treatment monitoring. Overall, in all cohorts combined, we observed a significant drop in RISK6 scores after two months of treatment (T1, p < 0.001) and until treatment completion (T2, p < 0.001). Moreover, RISK6 scores were significantly higher in cured TB patients (T2, p > 0.05) when compared to HD, however, in each of the four cohorts, there were no significant difference between these two groups (p > 0.05) (Fig. 3a). Similarly, analytical performance demonstrated capacity of RISK6 signature to significantly discriminate patients at baseline and two months after treatment initiation (AUC 69.7%, 95% CI 57.1-79.6) ( Fig. 3b and Supplementary  Table 6). Furthermore, we evaluated whether RISK6 allows the discrimination of cured TB patients (n = 104) from those with a treatment failure (defined as positive sputum culture at T2, n = 2). Thereafter, patients were stratified into drug-sensitive (DS) and drug-resistant TB (DR-TB) cases and the RISK6 signature scores were compared within these groups. We found that RISK6 scores decreased throughout treatment among DS-TB patients independently of treatment outcome ( Supplementary Fig. 1). In contrast, the RISK6 score remained stable at baseline and during treatment in a DR-TB patient with a treatment failure. Importantly, RISK6 score levels during TB treatment seem to be higher in patients with treatment failure among both DS and DR-TB cases. However, in a univariate or multivariate analyses, no significant association of the RISK6 score at baseline with treatment failure was found (Supplementary Table 8). www.nature.com/scientificreports/ Table 2. Receiver operating characteristic curve analysis of the performance of the RISK6 signature to distinguish active TB cases (ATB) from healthy donors (HD) and from latent TB infected individuals (LTBI) in cohorts from Georgia, Madagascar, Lebanon, and Bangladesh. The performance of the signature is benchmarked against the WHO TPP for a non-sputum based screening/triage test (at a sensitivity of > 90%, the minimum specificity as set out in this TPP should be ≥ 70%), for an initial TB diagnostic test to replace sputum smear (at minimum 60% sensitivity, the minimum specificity as set out in this TPP should be > 98%) and for a confirmatory test (at minimum 65% sensitivity, the minimum specificity as set out in this TPP should be > 98%) 19 . ATB active TB, LTBI latent TB infection (were only recruited from Madagascar), HD healthy donors, CLT + positive sputum culture, AFB − negative AFB smear microscopy, AUC area under the curve, CI confidence interval, Vs versus.

Discussion
TB remains one of the major infectious causes of death globally. In this study, we aimed to evaluate the relevance of RISK6, a PCR-based six-gene blood transcriptomic signature 26 , in the context of TB diagnosis and treatment monitoring. This was conducted in four independent cohorts enrolling ethnically and geographically diverse participants, including ATB patients, LTBI individuals, and HD, in both high-and low-TB incidence settings. We first evaluated the performance of RISK6 signature as a screening test for TB and showed that it displayed similar performance in the four different cohorts with excellent near-identical ROC AUC values (> 90.1%). Furthermore, RISK6 signature satisfied the minimum criteria set by the WHO TPP for a non-sputum-based screening test 19 . Notably, our findings suggest that, compared to IGRAs, the RISK6 signature showed a better performance as a screening test for discriminating between ATB patients and HD. Importantly, compared to previous RISK6 results reported by Penn-Nicholson et al. 26 , our study found similar data in terms of score range  www.nature.com/scientificreports/ and score changes over time despite the heterogeneity of both cohorts and study designs. In addition, marked technical differences are also apparent between our studies: we performed the RISK6 scores measurements on RNA manually isolated from whole blood collected directly in Tempus Blood RNA tubes and from blood samples first collected in lithium heparin tube and then transferred in Tempus Blood RNA Tubes, while this measurement was done by Penn-Nicholson et al. using RNA extracted manually or by an automated processes from whole blood collected in PAXgene Blood RNA tubes. Collectively, these results highlight the robustness of this PCR-based host-blood transcriptomic signature. Besides, the higher RISK6 score levels detected in the cohort of Bangladesh compared to the other study sites was a remarkable result. We hypothesized that these RISK6 scores observed in Bangladesh may be influenced by the differing epidemiology, geographical locations as well as differences in gene expression levels between ethnic populations that may have contributed to a stronger transcriptomic signal in Bangladesh.
Our AUC data showed that RISK6 scores had a powerful ability to distinguish ATB from HD, with better or equal results to what was found with other transcriptomic signatures [28][29][30][31][32][33][34] . Moreover, while these previous signatures have shown promise as diagnostic tests, it should be noted that results of a three gene signature were not generalizable 28,34 , while other signatures 33 require measurement of a high number of genes, thus limiting their possible application in resource-limited settings. Moreover, while RISK6 signature seems to meet or exceeded the TPP criteria based on each of our four cohorts, only two among the previous signatures (Sweeney3 28 and Sambarey10 32 ) satisfied the sensitivity and specificity TPP criteria set by the WHO for a triage test 35 . However, it would be interesting to validate those signatures in other independent cohorts 28,36 .
An important finding of our study is that RISK6 signature allowed to stratify TB patient's stages. Thus, when applied to the cohort of Madagascar, the only one including LTBI cases, the RISK6 signature demonstrated a significantly higher score in ATB individuals at baseline compared to those with LTBI. This is consistent with a previous study showing that a 3-gene transcriptomic signature was significantly higher in ATB patients versus LTBI 28 individuals, in addition to a 20-gene signature set that also discriminated ATB patients from LTBI and healthy controls 18 . In the same way, some gene-signatures were also evaluated 18 and showed high specificity and sensitivity to distinguish ATB patients from those with LTBI 23,28,31,33 . In our study, at > 90% sensitivity, RISK6 signature discriminated ATB from both LTBI and HD with a specificity > 70% which met the WHO TPP for a triage test for TB. Besides, no significant differences in the classification performance of RISK6 signature were observed between LTBI and HD, in line with recent transcriptomic studies demonstrating failure in discriminating LTBI from HD 18,28 . Moreover, while no previous studies has compared the levels of a transcriptomic signature between LTBI and treated TB patients, our data showed that the RISK6 signature reached the same score levels in treated TB patients when compared to LTBI individuals. Hence, it will be of interest to validate RISK6 signature in cohorts with larger number of latently infected individuals.
An additional finding of our study is that RISK6 signature also achieved the minimal WHO criteria in the Georgia cohort, for (i) an initial TB diagnostic test for sputum smear-negative TB to replace smear microscopy, using culture-confirmed TB as a gold standard (ii) and a non-sputum-based confirmatory test for sputum smearnegative TB. In this context, Turner et al. 37 reported a comparison of 27 signatures in cohorts of 181 patients for discriminating TB and no TB disease. They found that no previously published signatures achieved the minimal WHO sensitivity (65%) and specificity (98%) performance for a non-sputum-based confirmatory test for sputum smear-negative TB. Thus, our results are promising but further validation of RISK6 signature in larger cohorts will allow testing such performance. Furthermore, we found that ATB patients with low-or high-grade positive smears had significantly higher RISK6 scores compared with those with negative smears. Similarly to previous reported results with either Xpert MTB/RIF test or the C-reactive protein (CRP) concentration measurements 38,39 , our findings suggest that RISK6 signature scores directly correlate with sputum smear grade, and may possibly represent a useful tool in the identification of patients with high transmission risk.
In the present study, we also attempted to compare the performance of different TB blood-based tests; RISK6 versus two IGRAs (QFT-P and IGRAs-rmsHBHA). Our results indicate that the performance of RISK6 was greater than that of QFT-P assay for ATB case-finding. Given that QFT-P was not recommended for the diagnosis of ATB but for LTBI diagnosis, we and others have shown that this assay is a better indicator for the detection of Mtb infection 12,40,41 .
Our next aim was to evaluate variations in the RISK6 scores throughout successful treatment. We found that the RISK6 signature scores were significantly higher in ATB at baseline compared to HD, and continued to decrease progressively until the end of treatment reaching scores obtained in HD. Moreover, we also demonstrated that the RISK6 signature enables discrimination with high accuracy between untreated (T0), treated (T1 and T2), and post-treated (T3) TB patients who achieved a clinical cure. Taken together, these results showed the RISK6 genes might be modulated during anti-TB treatment as early as 2 months. Notably, the well-established data by Penn Nicholson et al. 26 also included additional earlier time points (week 1 and week 4) and found that RISK6 signature scores decrease over the course of successful treatment as early as 1 week. Data obtained with RISK6 is consistent with previous studies showing that transcriptomic signatures could be used as a powerful tool to monitor TB treatment response 30,[42][43][44][45][46] . In this context, it has been previously reported that reduced gene expression levels occurred rapidly during the first and the second weeks of TB treatment 47,48 . An additional report showed that ATB gene set decreased after 4 months of anti-TB treatment, however, no tests were performed at earlier time points, or during TB treatment course 49 . To note, we showed that the RISK6 signature scores returned to normal levels (compared to HD) after 6 months of treatment, which confirmed previous data 26 but contrasted with another transcriptomic study showing that normal levels were reached 12 months after the treatment initiation 30 . Subsequently, these results indicate that RISK6 scores significantly stratified end of treatment from pre-treatment baseline. Taken together, our findings suggest that RISK6 signature could be used as a useful tool to monitor the response to anti-TB treatment. It may represent a potential alternative of the current tests used www.nature.com/scientificreports/ to assess TB treatment efficacy and used comparing its result with those obtained by sputum culture that are crucial to evaluate drug resistance occurrence.
Remarkably, RISK6 relies on the use of qRT-PCR that could detect low levels of gene expression 50 and could be integrated into clinical poor settings in contrast to other complex methods. Besides, this signature requires the measurement of a small number of genes with subsequent reduced complexity and costs. Moreover, a key advantage of RISK6 is that it is a blood-based test, which is an easily accessible sample. Blood transcriptomic tests will improve the diagnosis of TB allowing faster treatment and thus reduction of transmission, especially in children, HIV co-infected TB patients and paucibacillary pulmonary TB patients. In such populations, microbiological tests are not always feasible due to the limited ability to produce good quality sputum samples or due to low bacterial loads in their samples. In the future, it will be of interest (i) to evaluate if RISK6 is able to predict the risk of progression to TB as demonstrated by the RISK11 signature 25 and (ii) to assess the diagnostic performance of RISK6 signature as a prototype cartridge assay as it has already been evaluated for the 3-gene signature against a microbiological reference standard 51 .
This study was subject to several limitations. Indeed, the sample size was relatively small and LTBI individuals were recruited from only one country. Hence, validation of our findings in cohorts with larger number of LTBI individuals is required to better estimate specificities and sensitivities for a triage test. Moreover, only two patients had failed treatment. Therefore, further validation is required to better understand how RISK6 signature tracks with response to treatment. Additionally, we excluded diabetic and HIV-positive patients and immunosuppressed individuals in general and our study was restricted to adults. Thus, similar validation studies are needed for children and HIV-positive patients. Moreover, in future studies, it would be relevant to evaluate the specificity of the RISK6 scores in comparison to other respiratory diseases than TB, which is considered as most difficult to distinguish with.
In conclusion, data from this study provide strong proof that RISK6 can be applied as a non-sputum-based screening and triage test that met the WHO TPP benchmarks. This host response-based gene signature may be used for stratifying patients according to their TB infection status, as well as for monitoring patients over the course of treatment. RISK6 signature is applicable using a robust and simple qRT-PCR platform which facilitates its implementation in the clinical laboratories located in resource-poor settings. Our overall findings support the efforts to incorporate RISK6 signature into a point-of-care test ensuring rapid and accurate detection of ATB cases. Indeed, such simple tests are highly needed to reduce TB spread and transmission especially in areas with high TB burden that are usually disturbed with poverty.

Methods
Study design and population. This evaluation of the RISK6 signature was a nested case-control multicenter prospective cohort study evaluating the prognostic value of blood-based immunological biomarkers for monitoring TB treatment outcome. It was conducted within the GABRIEL Network 52 in four different countries including Bangladesh, Georgia, Lebanon and Madagascar.
In total, 238 participants were recruited and followed-up between August 2018 and September 2020. Participants included patients with ATB disease (n = 141), HD (n = 71) and individuals with LTBI (n = 26). Enrolled ATB patients aged ≥ 15 years old, newly diagnosed with pulmonary ATB: scoring positive for TB following bacteriological (culture positive and/or sputum smear microscopy positive) and/or molecular analysis (GeneXpert positive results) were recruited at primary healthcare TB clinics in each country: National Center for Tuberculosis and Lung Diseases (NCTLD) in Tbilisi, Georgia; Tuberculosis screening and treatment center (CHUSSPA) related to National Tuberculosis Programs (NTPs) in Antananarivo, Madagascar; NTP centers in Tripoli and Akkar, Lebanon and International Centre for Diarrhoeal Disease Research, Bangladesh (icddr,b) in Dhaka, Bangladesh. Clinically asymptomatic healthy donors; who do not have a previous TB history and who have no recent TB contacts were also recruited in all sites. In Madagascar, participants with positive QFT-P results (IFN-γ production ≥ 0.35 IU/mL) were defined as latently Mtb infected individuals. Patients with negative cultures at inclusion, ATB patients with Human Immunodeficiency Virus (HIV) or with diabetes mellitus comorbidities and patients under immunocompromising treatment were excluded (Fig. 4).
Enrolled ATB patients were followed-up during the treatment course at four different time points and classified as follow: (i) ATB at baseline T0: patients who didn't start TB treatment; (ii) treated active TB at T1 and T2: patients with ATB followed-up during the treatment and tested after 2 months of the start of the treatment (T1), and at the end of treatment (T2); (iii) treated active TB at T3: treated TB patients tested at 2 months after treatment completion. www.nature.com/scientificreports/ staining method and/or Auramine O staining. In addition to positive culture, active TB status was defined by positive Xpert MTB/RIF (Cepheid). Patients were re-evaluated by sputum smear and culture during the intensive phase of treatment (T1) thereafter at the end of treatment (T2) and 2-months after treatment completion (T3) to confirm that they were successfully treated and cured. Drug susceptibility testing (DST) methods were performed according to standard protocols 53 .
Demographic and clinical data collection. At enrollment and at each follow-up visit, medical history, clinical and demographic data were collected using standardized questionnaires to feed the cloud-based database system CASTOR (CASTOR Electronic Data Capture, Version 1.4, Netherlands). Pre-amplification PCR. Prepared cDNA was pre-amplified using specific sequences of TaqMan primerprobes as previously described by Penn-Nicholson et al. 26    www.nature.com/scientificreports/ 4 min, and hold at 4 °C. The pre-amplified PCR products were diluted 1:25 with nuclease-free water and stored at − 20 °C for long-term conservation.

Quantitative Real-Time PCR (qRT-PCR) assay and gene expression analysis. For every target
to amplify, 4 μL of pre-amplified DNA was subjected to a real time nucleic acid amplification using 10 µL of TaqMan Universal PCR Master Mix (Applied Biosystems by Thermo Fisher Scientific), 1 µL of primers-probe mix (20×) and 4 µL of nuclease-free water using the following conditions: 2 min at 50 °C, 10 min at 95 °C, followed by 95 °C for 15 s and 60 °C for 1 min for 40 cycles. For analytical reasons, all the PCR reactions were performed in duplicate.
RISK6 score generation. Polymerase chain reaction signals were analyzed using CFX Manager Software version 3.1 (BioRad) in regression mode and expressed as cycle threshold (Ct) values. The step-by-step procedure for computing the 6-gene signature (RISK6) scores was performed as described by Penn-Nicholson et al. 26 .
Briefly, the mean of Ct values was calculated for every targeted genes and combined to generate a score. The score was computed with R script available on https:// bitbu cket. org/ satvi/ risk6/ src/ master/.
QuantiFERON-TB Gold Plus and IGRAs-rmsHBHA assays. 1 mL of whole blood was collected directly into each of the QFT-P tubes (Qiagen, Hilden, Germany, 622526) (Nil: Negative Control, TB-Antigens (TB1/TB2) and Mitogen: Positive control) and an extra 1 mL of blood was collected in a heparin tube and stimulated with 10 µg/mL of rmsHBHA (UNICATT, Rome, Italy 10101010 ). After 16-24 h incubation at 37 °C, plasma samples were harvested and stored at − 80 °C prior subjected to QFT-P ELISA (Qiagen, Hilden, Germany, 622120), following the manufacturer instructions. Briefly, 50 µL of plasma samples were tested, optical density results were compared to log-normalized values from freshly reconstituted IFN-γ kit standards. To account for potential immunomodulation phenomena unrelated with TB treatment, baseline IFN-γ level values (Nil tubes) were subtracted from antigen-stimulated IFN-γ values (TB1, TB2, Mitogen and rmsHBHA). According to the kit's sensitivity range, the maximum for IFN-γ level values was set at 10 IU/mL and negative values were rescaled to 0.