This descriptive study was nested within a multicenter prospective cohort study assessing the prognostic value of blood-based immunological markers for TB treatment monitoring. The study was based in five institutions from the Mérieux Foundation GABRIEL network (31), with the approval of national TB programs and the following ethical committees: the international center for diarrheal diseases and research, Bangladesh (icddr,b) in Dhaka, Bangladesh; the National Center for Tuberculosis and Lung Diseases (NTCLD) in Tbilisi, Georgia; the Laboratoire Microbiologie, Santé et Environnement (LMSE, Université Libanaise), in Tripoli, Lebanon; the Institut Pasteur de Madagascar in Antananarivo, Madagascar; and the Instituto de Investigationes en Ciencias de la Salud (Universidad Nacional de Asunción; IICS-UNA) in Asunción, Paraguay. All recruited patients provided written informed consent and standard biosecurity and institutional safety procedures were followed in all study sites.

The sample size was evaluated to detect a difference in rmsHBHA IFN-γ between baseline and end of treatment, with the following assumptions: we aimed for a level of significance of 95% and a power of 80%, assuming a minimum average expected difference of 1.6 IU/ml in rmsHBHA IFN-γ levels throughout treatment based on reported estimates (32), with an expected standard deviation of 3 IU/ml at each repeated measurement. We calculated (33) that a sample size of 117 was required to reach significance. As this study was nested in a cohort study with a sample size of 200, we aimed to enroll more patients to account for missing data. Patients were recruited if diagnosed with microbiologically confirmed pulmonary TB (positive culture and/or sputum smear and/or GeneXpert). Patients with HIV or diabetes mellitus and children under 15 years were excluded. In downstream analyses, patients under immunocompromising treatment, patients with negative cultures at inclusion, and patients who were lost-to-follow-up were excluded. Detailed procedures for microbiological diagnosis, drug sensitivity testing, and therapeutic regimen composition were described previously (29).

Patients were followed up: at inclusion (T0), after two months of treatment (T1), at the end of therapy [T2; 6 months for drug-susceptible (DS-TB) patients, nine to 24 months for drug-resistant (DR-TB) patients]. Data were presented for all patients followed up until T2 at least. Patients were on Directly Observed Treatment (DOT) and received treatment according to standard protocols (2, 3, 34). In this study, culture conversion at T1 was used to define three patient subsets: fast converters (definitive culture conversion between T0 and T1), slow converters (definitive culture conversion between T1 and T2), and patients with poor treatment outcome (positive culture results at T2: treatment failure; or positive culture at T3: relapse).

At every follow-up visit, 10 ml of whole blood were drawn: 4 ml was used for other downstream analyses, 1 ml was collected in EDTA tubes and used to measure whole blood cell counts by standardized automated systems available in the study sites as listed previously (29), and 5 ml was used for in vitro blood stimulation. For the QFT-P assay, 1 ml of whole blood was seeded directly into each of four QuantiFERON-TB Gold Plus (QFT-P, Qiagen) tubes as per the manufacturer’s instructions. The NIL tube contained no antigens and was used as a negative control. The TB1 and TB2 QFT-P tubes are coated with commercial M. tuberculosis-specific antigenic peptide pools. TB1 tubes contain two mycobacterial peptides, ESAT-6 (>15aa) and CFP-10 (8–13aa), which elicit specific immune responses from CD4+ T lymphocytes (35). TB2 tubes contain an additional commercial peptide pool designed to induce CD8+ T lymphocyte stimulation. MIT tubes are coated with commercial phytohemagglutinin-like bacterial antigens and were used as a positive control (35). For the rmsHBHA assay, 1 ml of blood was seeded into a NIL tube which was complemented with rmsHBHA (provided by the Delogu laboratory, UNICATT, Rome, Italy (23)), at a final concentration of 5 µg/ml. Within 2 h of blood collection, samples were placed at 37°C in a 5% CO₂ atmosphere and incubated for 24 h. After incubation, plasmas were separated from the cell fraction by decantation, and stored at −80°C until further use.

IFN-γ secretion was quantified using the QFT-P ELISA kit (Qiagen) according to the manufacturer’s instructions. Briefly, plasma samples were thawed at room temperature, and 50 µl of plasma was 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 (MIT, TB1, TB2, 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.

To optimize comparability, a sample handling and processing protocol common to all study sites was developed, and on-site trainings were performed to standardize experimental processes such as instrument settings and timings. A tracking sheet was developed and used to follow sample shipment and standardize storage conditions in all sites. As the models of measurement instruments used in the different sites could not be homogenized, instrument readings were tested with QuantiFERON Control Panel (Qiagen) prior to launching the project. Finally, internal controls were added to each ELISA run to control for experimental variation and verify storage quality. Briefly, whole blood from a healthy donor was collected and stimulated with QFT-P antigens following the same protocol as described previously. Plasma was then separated and aliquoted and added to each ELISA run.

Standardized clinical report and data collection forms were implemented to ensure dataset homogeneity as described previously (29). Data were entered into the CASTOR database system (Version 1.4, Netherlands) (36), and cleaned and analyzed in R (version 3.6.2). Discrete variables were analyzed using Fisher’s Exact test with Bonferroni’s post-hoc test (37). Normal, continuous variables were analyzed with Student’s t-test. Non-normal, continuous variables were analyzed with the Mann–Whitney test, or the Kruskal–Wallis rank sum test with Dunn’s Kruskal–Wallis Multiple Comparisons post-hoc test (38). Repeated measures of non-independent continuous variables were analyzed using the Friedman rank sum test, with the Wilcoxon–Nemenyi–McDonald-Thompson post-hoc test (39). As the HBHA IGRA was not commercialized and QFT-P was designed to screen latent rather than active TB patients, we used Receiver Operating Curve (ROC) analyses to identify optimal IFN-γ thresholds adapted for this cohort, discriminating culture positive from culture negative patients. The overall QFT-P test was considered positive if either TB1 or TB2 was above their respective thresholds. ROC analyses and logistic regression were performed as described previously (29). In particular, multivariate logistic regression analyses were adjusted with the combination of variables that minimized the Akaike Information Criterion (AIC) for most tested predictors, while including important adjustment variables (age, sex, drug sensitivity, country).

Between December 2017 and September 2020, 199 eligible patients with culture confirmed active pulmonary TB were recruited in Dhaka (Bangladesh), Tbilisi (Georgia), Tripoli and Akkar (Lebanon), Antananarivo (Madagascar), and Asunción (Paraguay). As of September 2020, 132 of them were followed at least until the end of treatment and had available IGRA data ( Figure 1 ). Among these patients, 21.2% (28/132) were diagnosed with DR-TB. The sociodemographic characteristics of DS-TB and DR-TB patients were similar at inclusion ( Table 1 ). Sociodemographic characteristics were compared between study sites, and significant differences were observed concerning age, BMI at inclusion, and BCG vaccination rates ( Supplementary Table 1 ). All enrolled patients were sputum culture positive at inclusion. Most patients were also positive for sputum smear microscopy (sensitivity: 78.0%, 103/132) and/or GeneXpert (98.4%, 125/132). Three (3.9%) cases of treatment failure and one (0.7%) case of relapse were recorded ( Table 1 ).

Plasma IFN-γ levels in response to TB1, TB2, or HBHA stimulations were measured during anti-TB treatment ( Figure 2 ). The median IFN-γ response to TB1 and TB2 remained constant over time, while the median response to rmsHBHA increased significantly ( Figure 2A ). Individual IFN-γ levels were heterogeneous in all three stimulation conditions ( Figure 2B ). To account for individual variations, rmsHBHA/TB1 and rmsHBHA/TB2 IFN-γ ratios were evaluated, and a significant increase in both ratios was still observed overall ( Supplementary Figures 1A–C ). We also measured the TB2-TB1 IFN-γ response, as a proxy for the QFT-P CD8⁺ T-cell response ( Supplementary Figure 1D ). No significant difference was detected over time. The impact of sociodemographic parameters on IFN-γ levels was assessed but no significant association was detected (data not shown).

Overall, QFT-P positivity rates remained constant during treatment (T0 vs. T2: 52 vs. 55%, p = 0.71), whereas rmsHBHA positivity rates increased significantly (T0 vs. T2: 31 vs. 67%, p < 0.001 ( Table 2 ). We also calculated the slopes of rmsHBHA and QFT-P IFN-γ variations during treatment ( Table 2 ). An increased INF-γ response to TB1, TB2, and rmsHBHA was observed in 55.3% (73/132), 56.8% (75/132), and 77.3% (102/132) of patients respectively.

IFN-y levels over time were then stratified per study site ( Figures 2C–E ). Similar trends were observed in all cohorts for TB1 and TB2 IFN-γ levels, except in the Madagascar site in which an increase in TB1 IFN-γ was recorded between T0 and T2. Variation in IFN-γ levels produced by rmsHBHA-stimulated samples was different between study sites: similar in increase and order of magnitude in the Bangladesh and Georgia cohorts on the one hand, as well as in Paraguay and Lebanon on the other hand; however, no increase was observed in the Madagascar cohort, as well as lower IFN-γ values ( Supplementary Table 2 ). Mitogen IFN-γ levels were also significantly lower in the Madagascar cohort than in the Georgia cohort at all timepoints ( Supplementary Table 2 ).

We analyzed the distribution of neutrophil percentages, and stratified IFN-γ results according to three groups: low neutrophils (less than the first quartile), intermediate neutrophils (between first and third quartiles), and high neutrophils ( Figure 3A ; threshold values are available in Supplementary Table 3 ). Similar analyses were performed with lymphocyte percentages ( Figure 3C ). We also evaluated the proportion of QFT-P and rmsHBHA positivity at each timepoint, stratified by neutrophil ( Figure 3B ) and lymphocyte percentages ( Figure 3D ). As HBHA stimulation was not performed using a commercial kit, Receiver Operating Curve (ROC) analyses were performed to identify the optimal rmsHBHA IFN-y threshold value differentiating culture-positive patients from culture-negative patients at any timepoint. The resulting Area Under the Curve (AUC) was maximized for an IFN-y cutoff value of 0.24 IU/ml (AUC 0.725, 95% CI 0.674–0.777). Overall, neutrophil and lymphocyte percentages directly impacted IFN-γ responsiveness to TB-specific antigens: QFT-P and rmsHBHA IFN-y levels and positivity rates were significantly higher in patients with low neutrophil ( Figures 3A, B ) or with high lymphocyte proportions ( Figures 3C, D ). This statistically significant trend was also observed when comparing the subgroup of patients with both low neutrophil and high lymphocyte percentages to the rest of the cohort (data not shown).

Overall, 112 patients had available culture data at T0, T1, and T2. Most patients were fast converters (definitive culture conversion between T0 and T1; 82.1%, 92/112) or slow converters (definitive culture conversion between T1 and T2; 14.2%, 16/112). Poor treatment outcomes were recorded in four patients (treatment failure, 2.7%, 3/112; relapse, 0.9%, 1/112; data not shown). Among successfully cured patients (n = 108), median IFN-γ levels ( Figure 4A ) as well as QFT-P and rmsHBHA IGRA positivity rates ( Figure 4B ) were stratified according to the culture conversion profiles. In slow converters, TB1 and TB2 IFN-γ levels at T0 and T1 and rmsHBHA IFN-γ levels at T2 were significantly lower than in fast converters. Similarly, QFT-P positivity rates at T1 and rmsHBHA positivity rates at T1 and T2 were significantly lower in slow converters.

Then, we calculated the sensitivity, specificity, and accuracy of the QFT-P and rmsHBHA IGRAs for TB treatment monitoring at T1 and T2, using culture as a reference standard ( Supplementary Table 4 ). At T1 and T2 respectively, the accuracy of the QFT-P IGRA was of 44 and 46%, and the accuracy of TB2-TB1 was of 52 and 55%. For the rmsHBHA IGRA, we evaluated the test performances of negative rmsHBHA results (i.e. rmsHBHA IFN-γ ≤ 0.22 IU/ml), since lower rmsHBHA IFN-y values were observed before treatment. The accuracy of the rmsHBHA IGRA was of 64 and 65% at T1 and T2, respectively. Finally, we generated a score which was positive when the QFT-P result was positive and the rmsHBHA result was negative. The sensitivity of this combined score was of 86% at T1 and 82% at T2, and its accuracy reached 77% at T1 and 81% at T2, but its specificity remained inferior to 30% at both timepoints. Similar results were observed with a score combining rmsHBHA and TB2-TB1.

We compared the immune cell counts ( Supplementary Table 5 ) and the baseline sociodemographic characteristics ( Supplementary Table 6 ) of patients according to their culture conversion profiles. No difference was detected between slow and fast converters for immune cell counts, but at T0 and T1, patients with treatment failure or relapse had significantly higher neutrophil percentages (at T0, median 84%, interquartile range (IQR) 81.5–86.5; at T1, 79%, IQR 75–81.75), and lower lymphocyte percentages (at T0, 12.5%, IQR 9.2–15.2; at T1, 15.5%, IQR 11–21.2) than successfully treated patients. The BMI at inclusion was significantly lower in slow than in fast converters, and slower conversion rates were observed in the Madagascar cohort.

Then, logistic regression analyses were performed to identify associations between slow culture conversion and immune cell counts or IGRA results ( Table 3 ). In univariate analyses, significant associations were detected between slow conversion and MIT IFN-γ at T0 (odds ratio (OR) 0.78, p = 0.001) and T1 (OR 0.84, p = 0.021), with QFT-P IGRA positivity at T0 (OR 0.19, p = 0.008), and with rmsHBHA IGRA positivity at T1 (OR 0.24, p = 0.015) and T2 (OR 0.29, p = 0.029). The BMI at inclusion was also associated (OR 0.791, p = 0.025).

In multivariate analyses, significant associations were maintained for MIT IFN-γ at T0 (adjusted OR 0.65, p = 0.009), QFT-P IGRA positivity at T0 (aOR 0.045, p = 0.013), and HBHA IGRA positivity at T1 (aOR 0.076, p = 0.045). No significant association was found otherwise ( Supplementary Table 7 ). Adjusting the models with neutrophil and monocyte proportions at baseline yielded similar results, but with higher AIC values ( Supplementary Table 8 ).

Overall, we observed a slow converter profile including consistent clinical patterns at baseline (low BMI, high neutrophil percentages, low lymphocyte percentages, low TB1 and TB2 IFN-γ responses), as well as a downregulated rmsHBHA response at the end of treatment.