To our knowledge, this is the first cohort study to identify antigen-specific biomarkers with substantial predictive accuracy for adolescents at high risk of progressing to active tuberculosis. The biomarkers identified in this study pave the way for developing more precise diagnostic tools for tuberculosis, making it possible to implement targeted interventions early.

Tuberculosis (TB), caused by Mycobacterium tuberculosis (M.tb), likely reemerged as the world’s leading cause of death from a single infectious agent in 2023. That year, an estimated 10.8 million new TB cases were reported worldwide, with China accounting for approximately 741,000 cases, ranking third among the 30 high-burden TB countries, behind India and Indonesia. This follows a three-year period during which COVID-19 temporarily surpassed it as the most significant infectious cause of mortality (Module 1: tuberculosis surveillance, 2024). The global burden of TB remains substantial, with approximately one-fourth of the world’s population harboring a latent tuberculosis infection (LTBI) and an annual incidence of over 10.8 million new active TB (ATB) cases. Globally, TB accounts for an estimated 1.09 million deaths each year, underscoring its lasting impact on public health (World Health Organization). However, 5–10% of individuals infected with M.tb will progress to ATB disease within months to years after the initial infection (Organization WH, 2018). Early detection and treatment of ATB, along with preventive treatment for individuals with LTBI, are thus regarded as two fundamental pillars of TB control. These strategies are essential for achieving the TB elimination targets outlined by the World Health Organization (WHO) (Organization WH, 2014).

The diagnosis of LTBI currently lacks a definitive gold standard. The two most widely used diagnostic tools, the tuberculin skin test (TST) and interferon-γ release assays (IGRA), have limitations. Neither method can reliably differentiate LTBI from ATB, and they are associated with low positive predictive values (PPV)—approximately 2.7% for TST and 1.5% for IGRA (Diel et al., 2012). Consequently, a significant number of individuals testing positive with TST or IGRAs would need treatment to prevent progression to ATB, highlighting the need for more accurate diagnostic tools. Given the differences in treatment approaches, an immunodiagnostic test capable of distinguishing between LTBI and ATB would substantially advance clinical care. The immune response to mycobacterial infections, particularly the role of IFN-γ, has been extensively studied (Kaufmann, 2001; Ruhwald and Ravn, 2009). Urgently needed are easily measurable biomarkers or biosignatures, such as cytokines associated with the immune response to M.tb, to enhance the prediction of disease progression (Kaufmann, 2001; Ruhwald and Ravn, 2009; Clifford et al., 2015). These developments are crucial for enhancing global TB control efforts. Furthermore, combinations of biomarkers may offer greater sensitivity than individual markers (Nemeth et al., 2010; Frahm et al., 2011). Cytokines often synchronize to regulate inflammation, immune cell activation, and tissue repair, influencing each other’s expression through feedback mechanisms.

Despite this, existing studies mostly focus on adults, lacking dynamic immune marker analysis of LTBI progression to ATB in adolescents. We hypothesized that adolescents are optimal subjects for identifying biomarkers of TB progression due to their simpler immune profiles, which exhibit fewer comorbidities and less immune senescence than adults, thereby minimizing confounding factors in cytokine expression (Eming et al., 2017; Daniel et al., 2023). In this study, we assessed TB-antigen-stimulated cytokine levels in adolescents who advanced to ATB compared to those who did not. We aimed to reveal the distinct immune profiles between these groups and identify predictors for the transition to infectious TB disease.

While IGRAs generally exhibit limited prognostic capability, the IGRA test can effectively identify additional biomarkers. Stimulating T cells with M.tb antigens triggers a surge in a complex network of cytokines, reflecting the immune responses specific to TB. In this study, we conducted an exploratory analysis of soluble biomarkers to predict progression to ATB. We utilized single and combined cytokine biomarkers through LASSO regression and multivariate logistic regression analysis to identify significant factors distinguishing ATB from non-TB cases. Our findings revealed that the predictive ability of single cytokines was relatively low. However, this predictive capability significantly improved when combined with positive QuantiFERON test results and was greatly enhanced using combined cytokine biomarkers through LASSO regression and multivariate Logistic regression analysis.

IGRAs have shown reliable performance in detecting LTBI, making them valuable adjuncts in diagnosing ATB, with superior accuracy over the TST in predicting ATB. However, their limited ability to distinguish ATB from non-TB cases reduces their clinical utility (Auguste et al., 2017; Lu et al., 2021; Hamada et al., 2023). Host biomarker studies have been at the center of interest for TB immunodiagnostics due to their utility in developing a point-of-care assay. In this study, we undertook an exploratory analysis of soluble biomarkers capable of predicting the progression of TB disease. To contextualize our findings, we compared them with studies employing diverse methodologies to identify predictive biomarkers for TB progression (Lacombe et al., 2014; Penn-Nicholson et al., 2019; Daniel et al., 2023). Our results align with and extend previous research using QuantiFERON-based, transcriptomic, proteomic, and multi-omic approaches. For example, Daniel et al. (2023) identified IP-10, CCL19, IFN-γ, IL-1ra, CCL3, and GM-CSF in QuantiFERON supernatants as predictors of ATB in household contacts, consistent with our identification of FGFbasic and MPIF-1/CCL23. Similarly, Penn-Nicholson et al. (2019)) reported a proteomic signature(TRM5 and 3PR)for TB progression, supporting the role of inflammatory markers akin to our cytokine ratios. Transcriptomic analyses by Gupta et al. (Ji et al., 2019) and Roe et al. (Sanchez et al., 2023) identified gene expression profiles linked to immune activation, which align with our cytokine networks. Multi-omic studies by Huang et al. (Scriba and Mendelsohn, 2020) integrate proteomic and transcriptomic data to enhance predictive accuracy, reinforcing the superiority of our combined biomarker models. Rodríguez-Molino et al. (2024)) reported higher rates of nonrespiratory TB and severe disease in immunocompromised children, highlighting the poor sensitivity of immune-based tests like QFT, which supports our focus on novel biomarkers to improve diagnostic accuracy in high-risk groups. These studies collectively suggest that our LASSO-derived biomarkers could be tested in broader populations, including immunocompromised individuals, to confirm their predictive utility. These datasets provide opportunities for future validation of our nomogram across diverse cohorts, potentially broadening its clinical applicability. We found a significant difference in the levels of 3 analytes (FGFbasic, GM-CSF, and MPIF-1/CCL23) of the 67 and 2 combinations (FGFbasic/GM-CSF and FGFbasic/MPIF-1/CCL23) in progressors as compared to non-progressors. Macrophages produce FGF-basic and plays a role in mitogenic and angiogenic activity (Henke et al., 1993). While its association with TB has not been consistently established (Awoniyi et al., 2016; Sariko et al., 2019), increased expression of FGFbasic may contribute to mechanisms that promote TB pathogenesis, although the exact mechanisms remain unclear. FGFbasic has been implicated in other inflammatory diseases, such as rheumatoid arthritis and pulmonary fibrosis, where it modulates inflammation and tissue remodeling, suggesting a potential role in TB progression through pathways like granuloma formation and fibrosis (Gudbjörnsson et al., 2004; Ghanem et al., 2024).

During LTBI, FGFbasic could potentially support the establishment of immune tolerance, inhibiting the immune system’s ability to clear M.tb. By modulating cytokine balance, FGFbasic may help the immune system maintain “tolerance” to the pathogen, allowing M.tb to persist in the host. This could facilitate the transition from LTBI to ATB. Indirect support for this hypothesis is provided by a study showing a decrease in FGFbasic expression following the initiation of TB treatment. GM-CSF is a growth factor that acts as a chemoattractant, recruiting neutrophils, lymphocytes, and macrophages. It plays a crucial role in supporting the inflammatory response during the productive phases (Olsson et al., 2006; Cheknev et al., 2014) and is involved in granuloma formation in TB to contain the infection (Szeliga et al., 2008). Previous studies show that lower GM-CSF expression may serve as a potential biomarker for predicting TB, although the underlying mechanism remains unclear (Balcells et al., 2018; Sariko et al., 2019).

In contrast, reduced levels of MPIF-1/CCL23 in progressors suggest a disruption in granuloma formation, which facilitates disease progression—this finding diverges from previous studies (Li et al., 2023; Yang et al., 2023). Pang et al. and Zhang et al. observed higher expression of MPIF-1/CCL23 in ATB cases compared to non-TB controls, though without statistical significance, proposing it as a predictive biomarker for TB (Li et al., 2023; Yang et al., 2023). This discrepancy may be explained by the possibility that the immune profile of the ATB group is in an early or “standby” state, which could contribute to TB development (Yang et al., 2023). The distinct immune profiles observed at different disease stages suggest that MPIF-1/CCL23 may have a stronger predictive ability for the progression to ATB. Our study identified the FGFbasic/GM-CSF and FGFbasic/MPIF-1/CCL23 ratios as highly promising predictive markers for ATB, outperforming single markers. However, further studies are needed to validate these marker combinations in their respective cohorts.

The study population included a higher proportion of female participants, which may influence immune marker expression. Female hormones, such as estrogen, have been shown to modulate immune responses, potentially affecting cytokine production and TB progression. For instance, estrogen can enhance macrophage activation and cytokine release, which may explain the observed differences in FGFbasic and GM-CSF levels in our cohort (Straub, 2007; Kovats, 2015). Further studies are needed to explore gender-specific immune profiles in TB progression.

The host response to LTBI and ATB involves a complex network of cytokines and chemokines at various levels of the immune system. Other diseases can also trigger a broad array of cytokines and chemokines, complicating the interpretation of infection status. Several studies have suggested using a multiplex model of cytokines and chemokines as biomarkers to differentiate TB infection states, offering improved sensitivity and specificity compared to single cytokine or chemokine tests (Wang et al., 2012; Yang et al., 2023). Through logistic regression, we identified a combination of “QFT, FGFbasic, IL-1ra, and I-309/CCL1” and increased the AUC to 0.932, sensitivity to 0.818, and specificity to 0.923. Compared to traditional logistic regression, LASSO regression offers key advantages such as automatic variable selection, reduced overfitting, better handling of multicollinearity, and model simplification for more straightforward interpretation. Additionally, LASSO improves predictive performance, especially in high-dimensional data (Sukarti et al., 2024). Our study also confirmed that, compared to logistic regression, the sensitivity for predicting ATB was increased to over 0.9 as it achieves the optimal requirement of ≥90% sensitivity for biomarker-based non-sputum tests to predict TB progression per the target product profile (Penn-Nicholson et al., 2019; Daniel et al., 2023).

An intriguing observation is that the multivariate models include variables that do not show statistical significance when considered individually, such as QFT, MPIF-1/CCL23, IL-1ra, and I-309/CCL1. Like IGRAs, which have limited prognostic ability and a low positive predictive value, the IGRA test may still be valuable for identifying additional biomarkers to improve predictive accuracy (Daniel et al., 2023). The decision tree analysis further supported that cytokine predictive capabilities are enhanced in individuals with positive QFT results. Additionally, IL-1ra, an antagonist of IL-1 produced during inflammatory processes, has been shown in previous studies to be associated with increased susceptibility to LTBI and elevated levels in ATB cases (Ji et al., 2019; Sanchez et al., 2023). Notably, IL-1ra was detectable in the serum of LTBI patients and found to be at lower levels in those treated for LTBI compared to untreated individuals, suggesting its potential as a marker for the progression to active disease (Juffermans et al., 1998; Zhang et al., 2020).

Our study has several notable strengths. First, the cohort consisted entirely of adolescents, which helped avoid potential cross-reaction with cytokines from other chronic diseases. Second, all participants were from a school outbreak, indicating that LTBI was likely recent. Third, the nomogram was tested using independent cohorts from different centers, enhancing its reliability and generalizability to a large extent.

However, we acknowledge several limitations in the current study. First, the small sample size (n=50, with 11 active tuberculosis [ATB] cases) limits statistical power, reducing the precision and generalizability of our findings. This restricted cohort may lead to overfitting in the LASSO model, despite 10-fold cross-validation, and could bias biomarker selection, potentially compromising the nomogram’s robustness. The limited number of outcome events further restricts our ability to detect subtle differences in cytokine profiles. Larger, multi-center studies with diverse populations are essential to validate our findings, enhance statistical power, and ensure robust biomarker performance. Second, the LASSO logistic regression model has not been independently validated in other cohorts. Finally, we cannot rule out the possibility that some individuals in the ATB group may have been in the subclinical or early stages of TB. Third, event times for ATB development were not recorded, limiting our ability to perform time-to-event analyses such as Cox regression. This represents a methodological limitation of using logistic regression, which does not account for the timing of disease progression.

To address the need for multi-center cohort validation, we plan to collaborate with regional TB control centers in Jiangsu and neighboring provinces to recruit a larger, diverse cohort for external validation. This will involve prospective enrollment of adolescents and adults with recent LTBI exposure, using standardized IGRA and cytokine assays, and longitudinal follow-up for at least 2 years to confirm the predictive accuracy of our nomogram.

Our study identified soluble TB-specific biomarkers that could be effectively used as potential short-term risk predictors for TB. Specifically, ratios such as FGFbasic/GM-CSF and FGFbasic/MPIF-1/CCL23 may serve as promising markers for predicting the progression from non-TB to ATB, particularly in individuals with positive IGRA results. Additionally, the combination of multiple immune indicators analyzed through LASSO logistic regression demonstrated significant potential in enhancing the diagnosis of ATB.