Evaluation of serological assays for the diagnosis of childhood tuberculosis disease: a study protocol

The aim of this study is to evaluate the diagnostic performance of serological assays in detecting children with TB disease, and in distinguishing those subjects from (1) healthy TB-exposed children, (2) children with non-TB lower respiratory tract infection, and (3) children with TB infection.

This study will utilise plasma samples obtained from three different prospective multicentric observational studies: the Childhood Tuberculosis in Switzerland (CITRUS) study (NCT03044509), the Pediatric TB Research Network in Spain (pTBred), and the Procalcitonin guidance to reduce antibiotic treatment of lower respiratory tract infections in children and adolescents (ProPAED) study (ISRCTN 17,057,980) (Table 1).

CITRUS is a multicentric prospective diagnostic study done at nine centres across Switzerland (Bern, Basel, Zurich, Lausanne, Geneva, Aarau, St. Gallen, Lucerne, Bellinzona). Its primary objective is to evaluate and validate novel immunodiagnostic assays for childhood TB [14, 15]. The study includes children under the age of 18 years, with or without a history of Bacillus Calmette-Guérin (BCG) vaccination, who are undergoing evaluation for TB disease, infection, and exposure. Children who have received any anti-mycobacterial treatment for five days or more before inclusion or who have been previously treated for TB disease or infection are excluded. Recruitment for the CITRUS study began in May 2017 and is currently ongoing.

PTBred is a multidisciplinary collaborative network established in 2014 in Spain, recruiting children < 18 years with TB. Since 2017, different types of samples have been stored in the Biobank of the Gregorio Marañon Hospital or in the individual collection registered as C.0006631 in the National Biobank Collections Registry. For this study, a common protocol for sample processing was implemented in October 2019, including children with children with TB disease, infection, and exposure irrespective of their BCG-vaccination status. The pTBred and CITRUS study follow the same inclusion and exclusion criteria [16].

The ProPAED study collected samples from children and adolescents presenting with fever and cough at two emergency departments in Switzerland (Basel and Aarau), from January 2009 to February 2010. For the ProPAED study, children with severe immunocompromise or known HIV infection, those undergoing immunosuppressive treatment, children with M. tuberculosis infection, neutropenia, cystic fibrosis, viral laryngotracheitis, hospital stay within the preceding 14 days, or other severe infections (e.g., osteomyelitis, endocarditis, or deep tissue abscesses) were excluded [17].

In this study, we will use the published criteria of compound TB case definitions proposed by Graham et al. [18]. Briefly, confirmed TB disease is defined as the presence of bacteriologically confirmed TB disease through culture or nucleic acid amplification tests (NAAT). Unconfirmed TB disease is defined as the absence of bacteriological confirmation in the presence of at least two of the following criteria: symptoms or signs suggestive of TB disease, chest radiograph consistent with TB disease, close TB exposure or immunologic evidence of M. tuberculosis infection, positive response to TB treatment. TB infection is defined as the presence of immunologic evidence of M. tuberculosis infection, including a positive TST of ≥ 5 mm (in accordance with the Swiss and Spanish guidelines [19, 20]) or a positive IGRA without meeting the criteria for confirmed or unconfirmed TB disease. Healthy TB-exposed children are defined as asymptomatic individuals with negative results on IGRA or TST test (single or repeat testing according to age, time since exposure as defined by national guidelines), making them unlikely to have TB. Children with non-TB lower respiratory tract infection will be the sick control group and are defined as presenting with fever (core body temperature ≥ 38.0° C) and at least one symptom (cough, sputum production, pleuritic pain, poor feeding) and at least one sign (tachypnea, dyspnoea, wheezing, late inspiratory crackles, bronchial breathing, pleural rub) lasting for fewer than 14 days.

The study will analyse antibody concentrations and properties in children stratified into distinct age groups: 0 to < 2, 2 to < 5, 5 to < 10, and ≥ 10 years, as proposed by Cuevas et al. [21]. This stratification is crucial due to the differences and dynamics of the nature of TB disease across age. In the youngest age range (infants and children < 2 years old), disseminated diseases and heightened susceptibility to progression from TB infection to TB disease is well-documented [22]. The risks for progression from infection to disease, as well as the subsequent mortality risk following development of disease, consistently declines during childhood, reaching its lowest point between 5 and 10 years of age [23]. Transitioning into adolescence and the onset of puberty, typically beyond the age of 10 years, the phenotype of TB disease becomes more adult-like. Pulmonary TB becomes more prevalent during this phase, contributing to an upsurge in TB-related mortality rates [24, 25].

Some previous studies in children have demonstrated improved specificities achieved by combining both protein and glycolipid antigens within serological assays [26‐29]. Furthermore, several studies have illuminated the potential for heightened sensitivity through the combined analysis of multiple antigen targets, effectively overcoming the interindividual heterogeneity of the human humoral immune response to M. tuberculosis [26‐33].

We will analyse antibodies concentrations and properties against single protein antigens, single glycolipid antigens [12, 34‐40], as well as multiple antigens in combination (Table 2). The types of antigens include cell wall fractions, whole cell lysates, and total lipids of M. tuberculosis. The selection of protein antigens is based on results from large protein microarray studies in adults [41‐46], one large multiplex bead-based study in children [31], and published and unpublished data from an adult study performed in the U.K (MIMIC study; personal communication M. Tebruegge) [47]. In order to enhance specificity, the overlap of the antigen targets for M. tuberculosis with Bacillus Calmette-Guérin (BCG) and other non-tuberculous mycobacteria will be reduced.

Together with targeted M. tuberculosis antigens, this study will evaluate the following distinct properties of the antibodies: isotypes and their subclasses, FcR binding profiles, and antibody glycosylation patterns (refer to Fig. 1). The rational for this is to obtain further information about the immune response to the antigen. TB disease results from a combination of the mycobacteria infecting and the resulting pathologic immune response. Therefore, antibody concentrations may only reflect on exposure, timepoint, and burden of mycobacteria, whereas additional properties such as FcR may reflect on the fact if the immune response producing tissue damage and pathology or not. This is shown in studies in children with TB disease that have demonstrated the potential enhancement of serological assay sensitivity through the integration of diverse antibody isotypes [48‐50]. Recent advancements in adult research have indicated that an evaluation of certain antibody properties, such as FcRs binding profiles and glycosylation patterns, could potentially enable the differentiation between TB disease and infection [12, 13].

As a quality control and potential normalisation variable, we will measure the total antibody concentration of each isotype and the total antibody concentration binding to distinct FcRs.

Upon plasma sample collection, preservation is ensured through storage in a − 80 °C freezer until the initiation of laboratory assays. Customised multiplex antigen-coupled beads will be produced to evaluate antigen-specific antibodies concentrations and properties in plasma samples. The protein antigens will be coupled to carboxylated beads through covalent NHS-ester linkages, using 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride and Sulfo-NHS (Thermo Scientific), following the manufacturer’s recommendations [51, 52]. Glycan antigen LAM, single lipid antigens (e.g., TDM and TMM), and multiple lipid antigen from Mycobacterium tuberculosis total lipids will be modified using 4-(4,6-dimethoxy [1, 3, 5] triazin-2-yl)-4-methyl-morpholinium (DMTMM) dissolved in ethanol and conjugated beads following the COOH-DMTMM method [53].

The antigen-specific antibodies concentrations and properties will be measured using different PE-labelled detection antibodies as follows: for the isotypes and subclasses, PE-coupled detection antibodies (anti-IgG, anti-IgA, anti-IgM, anti-IgG₁, anti-IgG₂, anti-IgG₃ and anti-IgG₄) at a concentration of 1 µg/mL; [52] for the FcR binding profiles, FcRs (FcγRIIIa/CD16a, FcγRIIIb/CD16b, FcγRIIa/CD32a H167, FcγRIIb/CD32b, FcγRI/CD64 from R&D Systems) will be labelled with PE and added to the samples at a concentration of 1 µg/mL; and for the glycosylation profiles, PE-labelled lectins (SNA for sialic acid, ECL for galactose, LCA for fucose and PHA-E for N-acetylglucosamine) will be used at a concentration of 20 µg/mL. After 2 h of incubation at room temperature, the beads will be washed with PBS-0.05% Tween20, and PE signal will be measured using xMAP technology. (refer to Fig. 2)

All data will be securely entered and shared through password-protected and encrypted systems to uphold the confidentiality of health-related personal information. Adhering to Swiss legal requirements for data protection (Ordinance HRO Art. 5), our procedures for storing biological samples and handling health data are meticulously governed. Coding mechanisms and personalised logins are implemented to grant exclusive access to the study database and source documents for authorised personnel, thereby preventing third-party disclosure. Unique identification numbers are assigned to the biological samples and health-related personal data.

To assess the performance of each individual antigen specific antibody feature as a diagnostic assay, sensitivity and specificity will be calculated based on cut-off values determined by the highest Youden’s index. Receiver operating characteristic (ROC) analysis will be performed, and area under the curve (AUC) will be calculated (confidence interval will be determined using the DeLong method).

In subsequent analyses, we aim to evaluate the combined interpretation of antigen-specific antibodies concentrations and properties using different strategies:

Strategy one involves defining cut-off values based on a specificity of ≥ 98%, in accordance with the minimal WHO’s TPP requirement for a biomarker-based detection test. We will calculate the corresponding sensitivity. Similarly, we will determine cut-off values based on a sensitivity of ≥ 66% and calculate the corresponding specificity. To assess the combined interpretation of multiple antigen targets, the test for a specific antibody or antibody property will be scored positive if at least one antibody level against a specific antigen exceeds the cut-off value in an individual’s plasma sample, and negative if all antibody levels against all antigens in a plasma sample are below the cut-off values.

Another strategy for the combined interpretation of multiple antibody concentrations and properties will involve feature selection using the least absolute shrinkage and selection operator (LASSO). This approach will help identifying the most informative features that could be used in diagnostic assays. To validate the predictive power of the selected features (k features), we will train and evaluate an additional model using only those k features. In a further step, we will include the selection of antibody concentrations and properties in the training of the model. By performing feature selection using LASSO, we aim to maximize prediction performance using all features and select the k most informative features after the training stage. This procedure is based on the concept that selecting the most informative features from a well-performing prediction model will also yield a well-performing prediction model when one only has access to the selected subset of features. Recent advances in machine learning research will enable us to incorporate feature subset selection directly into the training step of a model [54, 55]. Therefore, we optimise not only the prediction performance but also the subset selection of k features during training. The choice of subset size, k, should be based on external constraints. The diverse sensitivities and specificities observed in paediatric TB serological tests make a precise sample size determination challenging. To estimate the sample size for our experiments, we used data generated from a cohort of adults with latent infection (n = 20) and active pulmonary disease (n = 22) from South Africa [56]. For the analysis of 75 antibody features, linear regression was conducted to assess the association between diagnosis and antibody feature, while controlling for age and gender. For the thirteen features exceeding a false discovery rate threshold of 10%, the partial correlation coefficient of 0.50 or higher was observed between diagnosis and antibody feature. Using this estimate as the effect size of biologically active antibody features, 68 individuals in an independent cohort (34 LTB, 34 ATB) would provide a statistical power of 80% to observe significant differences in top antibody features between tuberculosis infection and disease at an alpha level of 0.0005. This alpha level represents the threshold for significance required by the Bonferroni-Holm correction method, set at 0.0005 to accommodate the testing of 100 antibody features.

Findings of this study will be disseminated through peer-reviewed journals, scientific conferences, and other relevant platforms. Participants will receive a summary of the results. All scientific data generated from this project will be made available as soon as possible, and no later than the time of publication or the end of the funding period, whichever comes first. The data and related metadata underlying reported findings will be deposited in a public data repository. A data access committee will support third parties who wish to perform further research with the data. Data will be curated in the repository following accepted standards and a persistent identifier, a DOI, is created for each data set published. If intellectual property is developed, dissemination of data will occur after appropriate protections for intellectual property are put in place.