Exploring non-sputum biosignatures for tuberculosis triage and diagnosis in children living with HIV

Abstract

Introduction:

Pediatric tuberculosis (TB), especially among children co-infected with HIV, remains a significant global health challenge. Traditional sputum-based diagnostics are less effective in this population due to difficulties in sample collection and paucibacillary disease, necessitating alternative non-sputum diagnostic approaches.

Areas Covered:

This review examines recent advances in non-sputum biosignatures for pediatric TB diagnosis, including host- and pathogen-based biomarkers detectable in blood, urine, stool and breath.

Expert Opinion:

Despite rapid advances in TB biosignature discovery, translation into usable diagnostics lags behind—especially for children and those with HIV. Studies often exclude these key populations, lack diverse validation, and depend on complex laboratory platforms. Bridging this gap requires early integration of feasibility, usability, and health system factors into product development. Multimodal, point-of-care tools adapted for low-resource settings and inclusive of high-risk children are essential. Implementation science and technology adaptation are critical to ensure real-world impact of these promising innovations.

Summary

Non-sputum biosignatures offer an opportunity for early diagnosis of TB in children with HIV. However, these signatures need to be validated and be translated into affordable point-of-care tools that can be integrated with current diagnostic approaches and implemented in low resource settings.

1. INTRODUCTION

1.1. Tuberculosis in children: brief overview

In 2024 alone, children under 15 years accounted for an estimated 12% of pulmonary tuberculosis (TB) cases worldwide, with those co-infected with HIV facing the highest risk of death due to diagnostic and treatment delays[1]. Childhood TB remains an under-recognised public health challenge, particularly in high-burden, low resource settings.

Current diagnostic algorithms for paediatric TB rely on symptom screening, chest radiography, tuberculin skin tests (TSTs), interferon-gamma release assays (IGRAs), and molecular tools such as Xpert MTB/RIF Ultra[2]. However, none of these tools offer a definitive, timely, and child-friendly solution. Sputum samples are difficult to obtain from young children and, when available, are often negative on smear, culture and molecular testing due to having paucibacillary disease[3]. Alternative approaches such as nasogastric aspirates and induced sputum require skilled personnel and infrastructure that may be unavailable in many low-resource settings. Radiological findings are frequently non-specific and prone to inter-reader variability[4], while TSTs and IGRAs cannot distinguish latent from active TB disease[5]. These limitations contribute to both under- and over-diagnosis[6], undermining global TB elimination goals[7].

HIV co-infection further compounds these diagnostic challenges. HIV progressively depletes CD4+ T lymphocytes, weakens granuloma formation, and disrupts cytokine signaling (including IFN-γ and TNF-α)[8,9], while also impairing innate immune responses such as dendritic and natural killer (NK) cell function[10]. In immunocompetent individuals, TB infection typically triggers granuloma formation that helps contain the bacilli and produces the classic clinical symptoms such as cough or weight loss. However, in children with HIV (CHIV), this immune response is often reduced or absent, leading to atypical disease presentation. As a result, TB in CHIV may lack classic signs and may instead present with non-specific symptoms including fever or poor growth. Radiological findings may also be subtle or non-specific, lacking hallmark features such as cavitation or hilar lymphadenopathy. Furthermore, due to their impaired immune containment of the infection, these children are at risk for disseminated or extrapulmonary disease forms, which are harder to detect using sputum-based tests[2].

These diagnostic blind spots have spurred global scientific interest in alternative diagnostic modalities that do not rely on sputum[11]. Among these, non-sputum biosignatures—sets of host- or pathogen-derived biomarkers detectable in accessible samples such as blood, saliva, urine, stool or breath—are the highest priority [2]. Host-based biosignatures can support triage and screening, while pathogen-based biosignatures may offer diagnostic confirmation[12].

To guide innovation in this area, the World Health Organization (WHO) and the Stop TB Partnership developed Target Product Profiles (TPPs) for new TB diagnostics[13]. These profiles outline minimal and optimal performance benchmarks for tests suitable for children and deployable at the point-of-care (POC). For example, triage tests should achieve at least 90% sensitivity and 70% specificity, while confirmatory diagnostics should attain a minimum of 65% sensitivity and 98% specificity. Importantly, the TPPs emphasize not only test accuracy, but also affordability, ease of use, and ability to function across diverse populations, including those with HIV or malnutrition[14].

In this review, we explore the current landscape of non-sputum biomarkers and biosignatures for TB, and discuss next steps to translate these into tests that can improve screening and/or diagnosis of TB in CHIV.

2. METHODS

A scoping literature review was conducted to identify and analyze studies evaluating non-sputum-based biomarkers for triage and diagnosis of TB in children living with HIV. We evaluated literature published between January 2010 to May 2025, utilizing the following databases and authoritative websites: PubMed, Google Scholar, ScienceDirect, Cochrane Library, World Health Organization (WHO), UNAIDS, Foundation for Innovative New Diagnostics (FIND), and the Stop TB Partnership. Reference lists of key articles and WHO technical documents were also manually reviewed to identify additional relevant studies. Search terms included both MeSH and free-text keywords, combined with Boolean operators: “Tuberculosis” AND “biomarkers” AND “children” AND (“HIV” OR “HIV co-infection”); “Non-sputum diagnostics” AND “pediatric TB” AND “HIV-positive”; “Blood-based biomarkers” OR “urine-based biomarkers” OR “host-response signatures” AND “childhood tuberculosis”; “Point-of-care diagnostics” AND “HIV-infected children”; and “Breath-based diagnostics” OR “volatile organic compounds” AND “childhood TB” AND “HIV”. Truncation and filters were applied to limit results to studies in English and within the defined time frame. We further examined studies that 1) enrolled children (aged 0–14 years) who were HIV-negative, HIV-infected or HIV-exposed; and 2) Investigated non-sputum-based biomarkers for TB diagnosis. We did not consider studies that 1) relied solely on sputum-based diagnostics; 2) did not report accuracy; or 3) were editorials, commentaries, or conference abstracts without sufficient data.

3. HOST-BASED BIOSIGNATURES

These markers, detectable in samples including blood, stool, urine, saliva or breath, measure changes in the expression of host mRNA, cytokines, proteins, or metabolites to distinguish TB disease from infection and other conditions (Table 1).

Table 1.

Categories of host-based biosignatures evaluated in children

Category	Biosignature	Sensitivity1,2	Specificity1	Key Limitations/ Considerations	References
Gene Expression	Sweeney3 (GBP5, DUSP3, KLF2 or TBP)	76–89% (general); ~50% in CHIV	18–66%	Low specificity in overall cohort; paucity of CHIV data; mostly controlled cohorts; requires GeneXpert; cost & supply chain issues.	(15–17)
	Anderson 51-gene score	83%	84%	Large panel that could increase complexity and cost; limited CHIV data; POC translation challenging.	(14)
	RISK6 (6-gene IFN-related)	70–90%	~90%	Limited CHIV data; interferon elevated in other infections; predictive/prognostic utility needs evaluation.	(18,19)
Cytokine	IP-10 (CXCL10)	90–95%	85–90%	Limited CHIV data; non-specific inflammation affects specificity.	(27)
	3-cytokine panels (e.g., IL-1ra, IL-7, IP-10; TNF-α, IL-2, IL-17A)	72–100%	75–100%	Multiplex assays costly; limited CHIV validation; infrastructure-dependent.	(21,22)
Omics Approaches	9-metabolite signature; metabonomic approach	AUC: 0.72 69%	83%	Small sample sizes; exploratory; limited CHIV validation; high-cost instrumentation.	(30,32)
	3–6 protein panels	97%	~70%	Complex analysis; standardization & POC translation required.	(31)
	Integrated multi-omics	AUC: 0.66–0.77		High cost; complex platforms; reproducibility & standardization issues; pediatric data limited.	(38)

^1.Range noted when multiple reference standards used

^2.AUC noted if sensitivity and specificity were not reported

3.1.1. Gene Expression Signatures

Three transcriptional signatures that have been examined in children and CHIV include:

1. The Anderson et al. 51-transcript signature[15]. Pediatric microarray/RNA-sequence data was analyzed from children with presumptive TB enrolled in South Africa and Malawi. In the independent test set in Kenya, the risk score achieved a sensitivity of 83% (95% Confidence Interval (CI): 69–94) and a specificity of 84% (95% CI: 75–93) to diagnose culture-confirmed TB in children[15]. The accuracy was higher among CHIV, with a sensitivity of 90% and specificity of 92%, although it only included 10 children who had TB and HIV. The advantage of a large transcript set is that it may capture a richer, more robust disease signal across heterogenous presentations and age groups. However, the 51-gene panel may be a challenge for routine POC translation, as it raises technical, cost, and analytical complexity for platform development.

2. Sweeney3 Signature. This consists of two upregulated genes, GBP5, DUSP3, and one downregulated gene, KLF2[16]. The TB risk score based on these genes was derived by analysing pooled microarray/RNA-sequence data mostly from adults, but also with inclusion of the paediatric data from Anderson et al. Notably, this signature has been adapted into a prototype polymerase chain reaction (PCR) cartridge (Xpert Host Response (HR), with the same genes or with TBP replacing KLF2) for use in GeneXpert machines with only a finger-prick of blood. In children, including those who were HIV-positive, the three-gene blood signature had moderate accuracy (Area under the Curve [AUC] 0.61 to 0.85) depending on if a culture-based reference standard or a composite reference standard (CRS) was used that incorporates clinically-diagnosed unconfirmed TB[17,18]. In the Rapaed-TB study, Xpert HR was examined as a diagnostic across five countries, and was found to have a moderate sensitivity of 59.8% (95% CI: 50.8 – 68.4) to detect culture-positive TB, at a high specificity of 90.3% (95% CI: 85.5 – 94.0). However, although sensitivity dropped across the reference standards, this further dropped to 41.6% when sputum Xpert Ultra was included in the definition of TB, and further reduced to 29.6% against the CRS[18]. When stratified by HIV status, the sensitivity was lower in CHIV compared to those without HIV (50% vs. 61%). The COMBO study also evaluated Xpert HR in children as a triage test in Uganda and the Gambia, and found that sensitivity ranged from 75.7–88.5% at specificities ranging from 30.3–33.3% depending on the reference standard. Moreover, the positive predictive value (PPV) and negative predictive value (NPV) of 17.6% and 94.0%, respectively, against the microbiological reference standard[17]. When applied to the small number of CHIV, the sensitivity was high at 100% but at 0% specificity. These studies suggest that this signature has moderate performance to detect culture-confirmed TB, but it has lower accuracy in children with paucibacillary, culture-negative disease and CHIV. This may be expected given that the signature was derived from largely adult data[16], and highlights the need for a pediatric-specific host signature. Also, dependence on the GeneXpert infrastructure limits decentralisation to sites without that platform.

3. RISK6 (six-gene interferon-related signature) [19,20]. This is a host transcriptional signature composed of six interferon-stimulated genes (ISGs) related to TB-associated immune activation. This signature is measured with transcriptome-wide approaches and has been validated using targeted PCR assays. Pouzol et al[19] evaluated RISK6 in children with presumptive TB in Bangladesh, and found that the sensitivity ranged from 60–79% and specificity 56–58% for children ≤12 months across reference standards, and it was 56–85% sensitive and 37–56% specific for children older than 12 months. None of the children had HIV. The low specificity is consistent with other studies that showed ISGs can be elevated in other viral or inflammatory conditions[21]. While the specificity was low, the authors found that it was more sensitive than other standard tests including TST and chest x-ray in children age ≤12 months, and combining these tests increased sensitivity. This emphasizes the need to evaluate any new biosignature in comparison to current testing.

3.1.2. Cytokine Biomarkers

Cytokines, chemokines, and acute-phase proteins have been measured to distinguish TB disease using methods that include enzyme-linked immunosorbent assays (ELISAs), multiplex bead-based immunoassays, mass spectrometry, and lateral flow assays[22,23]. Key markers and signatures in children and CHIV include:

• CRP (C-reactive protein). This is an acute-phase inflammatory protein that rises in response to infection or tissue injury, and it is currently endorsed by the WHO for TB screening and triage among adults with HIV[24]. Kagujje et al. performed CRP testing in 280 children from Zambia and found a sensitivity and specificity of 80% and 73%, respectively, against a microbiological reference standard (PCR or culture positive for TB)[25]. Sensitivity further dropped to 36% when the CRS was used. There were only five confirmed TB cases, and so assessment of accuracy among CHIV was limited. Relatedly, among Ugandan children, when Jaganath et al. set the CRP cut off level at 10mg/dL, the sensitivity and specificity of the test was 50% (95% CI: 37–63) and 63% (95% CI: 55 – 71) respectively[26]. Among CHIV, the sensitivity increased to 63% against the MRS, but specificity dropped to 38%. A study among Kenyan children found that CRP levels were unable to differentiate children with and without TB[27]. Taken together, current studies do not support CRP as a triage tool for children with or without HIV.

• IP-10 (interferon-gamma inducible protein also known as CXCL10). The performance of IP-10 has paralleled IFN-γ, but it is more abundant in blood and urine and thus attractive for diagnostic use. However, data has been mixed on its accuracy in children. For example, Strzelak et al. evaluated 225 children, including those with active TB, and measured IP-10 and IFN-γ levels in plasma[28]. The study found that children with active TB exhibited significantly higher IP-10 levels compared to the uninfected children, with an optimal IP-10 threshold that achieved a sensitivity of 90–95% and specificity of 85–90%. Druszczynska et al. demonstrated that children with active TB exhibited significantly higher levels of urinary IP-10 compared to healthy controls (HC) or those with non-mycobacterial pneumonia (NMP) in Bangladesh. This study further observed a moderate to strong correlation between urinary and serum IP-10 level elevation in active TB. In contrast, a study by Petron et al. among Ugandan children aged 1–14 years demonstrated that there was no significant difference in IP-10 levels (urinary or serum) between children with active TB and NMP or HC[29]. The implications of these studies are that the utility of IP-10 may be context-specific and varies by geographic or population characteristics. Also, the accuracy of IP-10 among HIV-positive children remains unclear with limited CHIV representation in these studies.

• Combination cytokine panels. Togun et al. studied Gambian children where the researchers combined and generated a three-marker protein biosignature including IL-1ra, IL-7 and IP-10 measured with a multiplex cytokine assay[22]. This biosignature distinguished children diagnosed with TB disease irrespective of microbiologic confirmation with demonstrated sensitivity and specificity of 72% and 75% respectively. In an Indian study by Kumar et al[23], the combination of TNF-α, IL-2 and IL-17A achieved a sensitivity of 100% and 98% specificity to distinguish children with Confirmed versus Unlikely TB. In both studies, none of the children were living with HIV. The combination of cytokine markers thus has the potential to improve accuracy over single analytes[23], but further validation is needed with inclusion of CHIV.

3.1.3. Omics-based approaches

Omics approaches include the comprehensive profiling of proteins, metabolites, lipids and the microbiome to distinguish TB status. These markers are usually measured using mass spectrometry (MS), nuclear magnetic resonance (NMR) spectroscopy, or high-throughput proteomic platforms[30]. Key markers and signatures in children and CHIV include:

• Proteomic signatures. Fossati et al[31] studied 511 children with and without HIV from four countries and used machine learning to develop four protein-based biosignatures comprising 3 to 6 proteins. These biosignatures achieved an AUC of 0.87–0.88, and surpassed the WHO’s TPP for TB screening tests with ≥70% specificity and ≥90% sensitivity. A limitation of this study was that it was based on mass spectrometry analysis, and further validation is needed including with platforms that can be implemented without the need for significant laboratory infrastructure, and at or near the point-of-care. They also did not perform a subgroup analysis by HIV status.

• Metabolic signatures. Certain amino acids, lipids, and energy metabolites have been identified as elevated or depleted in active TB. However, their performance has been moderate and heterogenous. For instance, Nellis et al. evaluated the metabolome of children with and without TB from the Gambia, India, Peru, South Africa, and Uganda, and demonstrated that children with confirmed TB had reduced levels of creatine, alanine, retinol, citrulline, fumarate and tryptophan while cortisol, nicotinamide and butyrylcarnitine levels were increased[30]. The researchers also demonstrated moderate accuracy in identifying confirmed TB using a nine-metabolite signature with an AUC of 0.72, but with heterogeneity across countries. Notably, the nine-metabolite plasma signature had similar accuracy in CHIV and HIV-negative children.

In addition, Andreas et al. studied 112 symptomatic children with presumptive TB from the Gambia, with data acquired using ¹H Nuclear Magnetic Resonance. They found that their metabonomic approach could achieve a sensitivity, specificity and AUC of 69% (95% CI: 56–73%), 83% (95% CI, 73–93%), and 0.78 respectively, for active TB[32]. However, without a specific metabolite panel, further validation and translation into a point-of-care test could be challenging. None of the children were living with HIV.

In a diagnostic accuracy study in India, targeted metabolomic and transcriptomic profiling was performed on children with confirmed TB versus age- and sex-matched household contacts to evaluate the role of kynurenine, tryptophan, their ratio (K/T), and IDO-1 gene expression[33]. The K/T ratio achieved an AUC of 0.667 with 81.5% sensitivity for diagnosing pediatric TB; however, kynurenine, tryptophan, and IDO-1 individually had lower AUCs: 0.667, 0.602, and 0.463, respectively. The K/T ratio was not evaluated among CHIV. Overall, metabolomic profiles appear to distinguish TB disease in children, but their performance has been heterogenous across setting and currently they have not been able to achieve goal accuracy and with limited assessment in CHIV.

• Breath-based metabolic biomarkers. These aim to identify volatile organic compounds (VOCs) or other metabolites exhaled by patients with TB. The VOCs identified include alkanes, methylated alkanes, and ketones, which are believed to reflect both host inflammatory responses and mycobacterial metabolism. These VOC profiles, however, can be altered by diet, environment, or co-infections, and thus raises concerns about specificity. Modalities for detection include gas chromatography–mass spectrometry (GC-MS), electronic noses (eNoses), and nanosensor arrays. Although most studies to date have focused on adults, some pilot studies in children have shown promise. For instance, a study by Bobak et al. demonstrated that a “breath print” could distinguish children with TB from those without TB, at 80% sensitivity and 100% specificity[34]. However, sample sizes were small, and the performance was lower in children co-infected with HIV. A 2023 pilot test of the TSI-3000(I) Breath Analyzer in India reported 95.7% sensitivity, 91.3% specificity, and an AUC of 0.935 in both adults and children ≥10 years old[35]. Similarly, Bijker et al[36] demonstrated that exhaled breath analysis showed moderate diagnostic accuracy for detecting TB in children with a sensitivity of 86% (95% CI:62–96) but low specificity of 42% (95% CI: 30–55). A study by Mosquera-Restrepo et al. measured exhaled breath condensate (EBC) and found that IL-17 was particularly elevated in paucibacillary children and adults. The study further noted that biomarkers like oleate (fatty acid), IL-17 and MCP-1 in EBC may serve as non-invasive early biomarkers of TB[37]. At present, no breath-based test for TB has been WHO-endorsed, and further validation studies in children, particularly those under 5 years and CHIV, are urgently needed.

• Multi-omics approaches. Integration of transcriptomics, proteomics, and metabolomics can improve disease discrimination and identify risk signatures predictive of progression. For example, Dutta et al. identified several metabolites associated with TB diagnosis and treatment response in children- pyridoxate, quinolinate, and N-acetylneuraminate[38]. The latter marker had an AUC of 0.66 at diagnosis, while quinolinate achieved an AUC of 0.77 after 1 month of treatment. Omics-based approaches may identify predictive biomarkers, but the high cost and sophisticated instrumentation currently limit use in resource-constrained, high-burden settings. Also, most studies are exploratory, with small sample sizes and limited CHIV-specific validation. Other limitations include the challenge of standardization and reproducibility across laboratories and populations, and translating multi-omics signatures into simple POC tests.

3.1.4. Other sample types

• Adenosine Deaminase (ADA). An enzyme involved in purine metabolism, ADA is released during cellular immune activation, and has been assessed as an adjunctive biomarker in children and adolescents for both pleural TB and TB-meningitis (TBM). For example, in a systematic review by Na et al, the pooled sensitivity and specificity of ADA (at a cut-off of 40 U/L in pleural fluid) was approximately 85% and 58%, respectively, for detecting paediatric pleural TB[39]. Daniel et al. found the sensitivity and specificity of ADA in cerebral spinal fluid (CSF) for detecting paediatric TBM was approximately 95% and 91%, respectively, at a cut off of 10 U/L [40].

• Saliva. Saliva provides a non-invasive matrix for immune biomarker detection, with potential for child-friendly diagnosis. Host-based salivary markers include cytokines such as IP-10, IL-6, and transcriptional signatures. Other candidate analytes include chemokines, and acute-phase proteins typically measured by ELISA or multiplex immunoassays[41,42]. For example, Mutavhatsindi et al. identified a 5-protein biosignature that diagnosed TB with sensitivity of 100% (95% CI: 76–100) and specificity of 91% (95% CI: 59–100) in adults[42]. However, currently there are no validated salivary biomarker panels for childhood TB[43].

• Stool. Biomarkers in stool include calprotectin, lactoferrin, myeloperoxidase, and neopterin, and these reflect intestinal or systemic inflammation[44]. These biomarkers are measured using ELISA or lateral flow assays. Larsson et al[45] demonstrated that fecal calprotectin could differentiate intestinal TB from pulmonary TB, highlighting potential for disease localization and follow up. However, there is sparse TB-focused evidence in children, and most data are from enteropathy/diarrheal diseases with no CHIV-focused studies.

4. PATHOGEN-BASED BIOSIGNATURES

Pathogen-based biosignatures aim to detect Mycobacterium tuberculosis (Mtb) or its components directly in non-sputum samples such as urine, oral swab, breath, stool or blood. These biosignatures rely on identifying antigens, nucleic acids, or metabolic products of the bacilli, offering the potential for a direct TB diagnosis, especially in children who struggle to produce quality sputum. These signatures are typically used for diagnostic confirmation rather than triage[15]. Compared to host-based signatures, we found there has been limited work in identifying non-sputum pathogen-based biosignatures for children. Key categories include:

4.1. Lipoarabinomannan (LAM) assays:

Lipoarabinomannan (LAM) is a glycolipid component of the Mtb cell wall, released into the circulation and excreted in urine during active TB disease. Because the samples are easily obtainable, LAM-based assays provide a non-invasive diagnostic alternative, particularly valuable in children and those unable to produce sputum. Currently, the only available lateral flow assay is the Determine TB LAM Ag Test (Abbott, Chicago USA)[46]. Sensitivity of LAM is higher in those with HIV and correlates with lower CD4 count, likely for multiple reasons including greater risk of disseminated disease including renal involvement. In a Tanzanian pediatric cohort (n=132), Determine LAM showed 50% sensitivity in CHIV compared to 0–13% in HIV-negative children, with specificity reaching 97%[47]. Based on such findings, the WHO recommends Determine LAM primarily for CHIV[48]. Despite its ease of use, Determine LAM has low sensitivity in HIV-negative children and those with early or paucibacillary disease. Moreover, there is reduced specificity and greater false positives in children, possibly due to cross-reactivity from the use of polyclonal antibodies coupled with urinary contamination from bagged urine specimens[49]. New LAM assays are being developed to address these issues, such as the Fujifilm SILVAMP TB LAM (FujiLAM) test that uses high-affinity monoclonal antibodies and silver amplification to enhance sensitivity. Meta-analyses suggest FujiLAM achieves 55–60% sensitivity in both HIV-positive and HIV-negative children, with specificity >90%. Sensitivity is again notably higher in those who are severely malnourished or immunosuppressed (low CD4 counts)[3,49]. FujiLAM and similar next generation assays therefore show potential as more reliable tools for the diagnostic rule-in of TB among CHIV and severely ill children, though further validation studies are still ongoing[50].

4.2. Mtb-specific proteins.

Several assays have been developed to detect secreted Mtb proteins. ESAT-6 (Early Secreted Antigenic target 6kDA) and CFP-10 (Culture Filtrate Protein-10), two proteins used in interferon-gamma release assays, have been detected in blood by functionalized nanodiscs with matrix-assisted laser desorption ionization–time of flight mass spectrometry (MALDI-TOF MS)[51]. In a study in the United States of 105 banked samples from children with confirmed or unconfirmed TB or close contacts as controls, nanodisc could measure ESAT-6 or CFP-10 at any detectable level with 86% sensitivity and 100% specificity. However, there was only one child with HIV[52]. Another study used a nanopore assay to measure serum ESAT-6/CFP-10 in children from South Africa, and achieved a sensitivity of 94% to detect microbiologically confirmed TB, with 81% specificity for Unlikely TB[53]. HIV subgroup analysis was not performed. Further validation is needed as these studies were conducted in a single setting and used either healthy controls or with a small sample size, including for CHIV.

4.3. Cell-free DNA (cfDNA).

Small fragments of Mtb DNA released after bacterial breakdown have been measured in a variety of sample types including blood and urine. Metagenomic sequencing of plasma in 30 children with and without TB had low sensitivity (20–40%), with higher sensitivity among those who had smear positive disease[54]. Targeted next generation sequencing (tNGS) of plasma samples of children with presumptive TB in China had a sensitivity of 32% (95% CI: 22.9–42.4) and specificity of 97% (95% CI 82–99.8)[55]. CRISPR-based assays have been developed to detect Mtb cfDNA in plasma, and were assessed in a study in children from Eswatini and Kenya[56]. The sensitivity in Eswatini was 83% with 95% specificity; among CHIV from Kenya, the assay could detect cfDNA in all 13 children with confirmed TB and 85% of unconfirmed cases[56]. CHIV had advanced HIV disease, were not yet started on antiretroviral therapy, and were admitted to the hospital, which may have overestimated performance. A point-of-care CRISPR assay has been developed to further facilitate testing in under an hour[57], but further validation is needed.

4.4. Volatile Organic compounds (s).

Pathogen-based VOCs are released as a result of Mtb metabolism or host-pathogen interaction and can be detected in exhaled breath, urine or stool. VOC classes include (i) Methyl nicotinate from Mtb lipid metabolism, (ii) Methyl-branched fatty acid derivatives from mycolic acid catabolism, (iii) Terpenoids from host-pathogen oxidative pathways, (iv) Alkanes/alkenes (C8—C20 range) from oxidative stress following host-pathogen interaction, and (v) Short-chain aldehydes from lipid peroxidation[35]. However, studies in children with and without HIV are lacking.

5. CONCLUSION: A Vision for the Future

In conclusion, both host-related and pathogen-derived biosignatures have the potential to transform the diagnostic landscape for paediatric TB. Host signatures have the advantage that they can discriminate disease even when bacillary burden is low and can thus be particularly valuable in children with culture-negative disease. However, these markers can be non-specific, especially in the presence of co-morbidities including HIV and malnutrition. Conversely, pathogen-derived biosignatures offer a more direct detection of Mtb and so are inherently more specific. However, there has been limited work to identify new non-sputum, pathogen-specific markers, the low bacillary burden in children reduces their sensitivity, or it requires highly sensitive assays that may be challenging to translate to the point-of-care.

The next phase of research must therefore prioritize paediatric validation, sample-type versatility, and implementation pathways that move pathogen-based biosignatures from proof-of-concept into scalable, community-accessible diagnostic tools.

6. EXPERT OPINION: Reimagining TB Diagnosis in Children Living with HIV

The landscape of non-sputum biosignatures for paediatric TB diagnosis is evolving rapidly, yet its translation into impactful, real-world tools remains unfulfilled. Despite the identification of novel host- and pathogen-based biomarkers, scientific innovation is outpacing implementation. Most candidates remain in the discovery or early-phase validation stage, particularly in paediatric populations. This translational lag is most evident among CHIV, who often exhibit an atypical immune response and non-classical TB presentation, rendering adult-derived diagnostics inadequate. This translational gap is not simply a matter of time or funding, but a reflection of deeper structural and scientific challenges embedded within the development pipeline. Drawing from emerging evidence and recent studies, we offer the following perspective:

A). Study design limitations undermine clinical relevance:

Many promising biosignature studies have been limited by methodological shortcomings that impede generalizability. The use of highly controlled case-control designs, often involving well-characterized cases and healthy controls, creates spectrum bias and overestimates diagnostic accuracy. Moreover, sample sizes are frequently small, and studies often exclude key paediatric subgroups such as children under five, those with HIV, or those with malnutrition—despite these groups accounting for a significant proportion of TB-related morbidity and mortality. As a result, many biosignatures fail to reflect the true diagnostic complexity of childhood TB in routine settings.

B). Limited validation across epidemiologic and geographic contexts:

A major barrier to implementation lies in the lack of cross-context validation. Diagnostic performance of biosignatures can be highly variable depending on TB prevalence, co-morbidities, and health system characteristics. Yet, few studies attempt replication in diverse geographic or programmatic settings. For example, transcriptomic signatures identified in adults in South Africa may not replicate in malnourished children in West Africa or in those with advanced HIV in Asia. Rigorous external validation is critical to understand the utility of these biomarkers and any adjustments in thresholds depending on the setting and population. In addition to sensitivity and specificity, positive and negative predictive value and yield are important metrics to assess clinical utility.

C). Need to rigorously assess and validate biomarkers for extrapulmonary TB (EPTB):

EPTB poses an especially large diagnostic challenge in children and particularly CHIV. These groups are disproportionately affected by EPTB, often presenting with atypical features and paucibacillary disease, while invasive procedures needed to obtain diagnostic samples may not be feasible or safe[58,59]. As a result, many cases go undetected, leading to delayed treatment, preventable complications, and excess mortality. To close this gap, there is an urgent need not only to discover but also to validate and rigorously assess biomarkers specifically for EPTB. The current biomarker landscape has predominantly focused on pulmonary TB (PTB)[60]; while pathogen-derived markers may share common molecular signatures across different body compartments, the analytical performance and limit of detection may differ depending on the sample type (i.e., CSF versus blood versus urine)[61]. In contrast, host-response markers may vary significantly between EPTB and PTB reflecting differences in bacterial burden, tissue environment, immune compartmentalization, and site-specific complications. Thus, candidate biomarkers in blood, urine, CSF, or tissue could transform case detection for EPTB, particularly for syndromes like TB meningitis where rapid diagnosis is critical to survival. However, their promise will remain unrealized without structured validation to ensure reliability, reproducibility, and applicability across diverse patient populations.

D). Technological constraints block translation to point-of-care:

While many biosignatures show promise on high-throughput laboratory platforms, these technologies are ill-suited for low-resource environments. Translation into POC assays—such as lateral flow strips or isothermal amplification—is technically complex and rarely prioritised during early discovery. Most candidate signatures require further miniaturisation, multiplexing, and robustness testing, yet diagnostic development efforts remain underfunded and poorly integrated with scientific discovery. Limited public-private partnerships and dedicated translational pathways has compounded delays in bringing products to scale.

E). The case of children with HIV highlights the urgency:

CHIV represent a particularly vulnerable population in whom current diagnostics underperform. Immunosuppression alters biomarker expression, and extrapulmonary or paucibacillary TB forms are more common. Yet, few biosignatures have been validated specifically in this group. As Kasule et al. highlight, specificity in HIV-infected children remains suboptimal for many platforms, limiting its clinical utility[62]. Also, discovery efforts often have limited representation of CHIV, and thus identified markers may not reflect their unique immune responses to TB or the distinct pathogenesis of childhood TB, which is more often a primary disease rather than reactivation. Therefore, prioritizing HIV-exposed and infected children in future biomarker research is essential—both for equity and diagnostic impact.

F). Multimodal approaches hold promise:

Recognizing that no single test is likely to meet all performance, cost, and operational targets across diverse settings, multimodal and multi-sample strategies are emerging as the most promising path forward. These might combine host-based biosignatures (e.g., transcriptomics or inflammatory proteins), pathogen-derived markers (e.g., urine LAM or blood-based Mtb proteins), with current approaches to increase diagnostic sensitivity and maintain feasibility.

Emerging studies suggest that multimodal approaches may approach or exceed these TPP benchmarks under controlled conditions. For instance, transcriptomic signatures such as RISK11 and RISK6, when used alone or in combination with clinical and radiological features, have demonstrated area under the curve (AUC) values exceeding 0.85 in HIV-negative children; however, specificity was often reduced in HIV-positive populations[20]. Similarly, studies in Africa have shown that combining computer-aided detection (CAD) scores, symptom screening, and blood-based biomarkers enhanced the overall diagnostic yield compared to any single modality[63,64]. State-of-the-art studies that build well-characterized clinical, radiological and biological repositories can support the combined assessment of emerging tools and biomarkers for TB diagnosis in children[65]. However, integration into existing healthcare systems may require investments in infrastructure, training, and digital health tools, which may not be feasible in all settings. Moreover, the cost and operational complexity of using multiple diagnostic layers may be a barrier to scale-up, especially in primary care or rural contexts.

G). The next phase must include greater efforts in implementation science.

Laboratory accuracy alone does not guarantee that a diagnostic will have clinical utility in field settings. Instead, attention must turn to how these tools can be integrated into existing health systems, how they affect clinical decision-making, and how they are perceived by end-users, including healthcare workers, caregivers, and children themselves. Key questions include: Can the test be used in peripheral clinics? Does it reduce diagnostic delay? Is it cost-effective compared to current strategies? Such questions require rigorous evaluation using mixed-methods studies, health economic models, and pragmatic implementation trials in high-burden, resource-constrained contexts. The recent TB-Speed studies have shown that even simple diagnostic innovations can face major bottlenecks if usability and health system integration are not addressed from the outset[66–68].

Table 2:

Categories of non-sputum pathogen-based biosignatures for children

Category	Biosignature Example	Sensitivity	Specificity	Key limitations/ Considerations	References
Antigen-based	Determine TB-LAM	45–55% overall; 65–70% (CHIV)	90–95%	Lower specificity in immunocompetent children. WHO-recommended for CHIV	(44,45)
	Fujifilm SILVAMP TB-LAM	60–70% overall ~60% in CHIV	90–95%	Limited data in younger children. Improved sensitivity and specificity over Determine LAM	(48,56)
	ESAT-6/CFP-10	86–94% 84–100% (CHIV)	81–100% 93 – 100% (CHIV)	Point-of-care assay needs to be developed	(49–51)
cfDNA	Plasma cfDNA	20–83%	95–97%	Lower sensitivity since children are paucibacillary Expensive; sophisticated lab equipment; limited CHIV-specific validation	(52,55,57)

Article Highlights.

• Non-sputum biosignatures are urgently needed to improve the diagnosis of tuberculosis of children with and without HIV, and reduce delays in treatment

• Host-based and pathogen-based biomarkers have been identified that may have promise and utilize less invasive sample types including blood, urine, stool and breath

• However, limitations in study design, lack of validation, and laboratory infrastructure requirements have prevented further translation to a point-of-care test

• Multimodal, multi-sample biosignatures have improved performance

• Implementation science approaches are needed to assess how to best integrate these biosignatures into clinical care

Abbreviations:

cfDNA

cell-free DNA

CHIV

Children with HIV

LAM

lipoarabinomannan

ESAT-6

Early secretory Antigenic target – 6kD

AUC

Area Under the Curve

GBP-5

Guanylate Binding Protein-5

DUSP-3

Dual Specificity Phosphatase-3

KLF-2

Kruppel-like transcription Factor 2