A Mycobacterium tuberculosis fingerprint in human breath allows tuberculosis detection

Tuberculosis remains a leading cause of death from a single infectious agent, with an estimated 10 million cases and 1.5 million deaths in 2020. Approximately 25% of the global population is infected with Mtb and at risk of disease. Current diagnostics largely rely on sputum, which is difficult to obtain and insensitive in children, HIV-infected individuals, and extrapulmonary or paucibacillary disease. The WHO prioritizes development of point-of-care, non-sputum biomarker-based tests. Exhaled breath condensate (EBC) is an easily accessible, non-invasive sample reflecting airway lining fluid and may contain pathogen-derived markers. The research question is whether non-volatile Mtb molecules (e.g., LAM, specific lipids, proteins) can be detected in EBC to diagnose active pulmonary TB and monitor treatment efficacy across patient populations, including smear- and culture-negative cases.

Prior work includes urine-based LAM assays recommended for seriously ill HIV patients, though with suboptimal sensitivity for broader screening. EBC has been proposed as a non-invasive matrix reflecting lung pathophysiology, with earlier studies showing oxidative stress and inflammatory mediators can distinguish TB patients from controls, but nucleic acid detection in EBC has been inconsistent. Mtb produces species-specific lipids (e.g., sulfolipids, mycolic acids, tuberculostearic acid derivatives) and proteins that may serve as biomarkers. Structural features of LAM (e.g., arabinan termini and mannose caps) influence immunoassay performance, and antibody epitope specificity (e.g., to MTX motif or Ara6 termini) affects sensitivity. Combining pathogen-derived biomarkers could improve specificity over host-response markers.

Study design: Proof-of-concept analysis of EBC from adult and pediatric pulmonary TB patients and controls (healthy and community-acquired pneumonia). Clinical categorization followed national and international TB guidelines, with smear and culture status defined by Ziehl-Neelsen staining and mycobacterial culture; pediatric string-test used as applicable. Exclusion criteria included HIV, diabetes, cancer, autoimmune disease, immunosuppression, smoking, and prior TB. Participants: Table 1 cohorts included adults (smear-positive n=8; culture-negative n=7; culture-positive n=14), pediatric TB (string-test smear- or culture-positive n=5; smear- and culture-negative n=12), controls (adult pneumonia n=8; healthy adults n=7; healthy children n=15). ROC analyses used n=46 TB cases and n=45 controls. EBC collection: R-tube device with 0.3-µm filter and saliva trap; participants breathed for 15 min; ≥1 mL collected, frozen at −20 °C. Saliva contamination assessed by amylase; positive samples discarded. Samples lyophilized and normalized to 250 µL LPS-free water. Immunoassays: Dot-blot on nitrocellulose; primary antibodies included anti-LAM (CS-35/CS-51) and anti-GroEL2 (CS-44); HRP-conjugated secondaries; detection by ECL and densitometry (ImageJ). Standards: purified LAM from M. tuberculosis H37Rv and Mtb cell lysate. Monosaccharide analysis: EBC hydrolyzed (2 M TFA, 110 °C, 2 h). APTS labeling and CE-LIF quantification for D-arabinose, mannose, glucose; GC-MS of trimethylsilyl derivatives for corroboration and absolute configuration (D- vs L-arabinose). Arabinose used as surrogate for LAM. LAM structural analysis: Pooled EBCs (Ad S* pool, n=50; Ch S/C pool, n=50) processed to enrich LAM: lipid extraction (chloroform/methanol partition), enzymatic digestion (phospholipase D, RNase, proteases), dialysis (6–8 kDa cutoff). NMR (600 MHz, cryoprobe) 2D HSQC and 1H-13C spectra to assign glycosidic residues; mannose cap analysis by mild hydrolysis and electrophoretic methods; SDS-PAGE/Western blot for MPI anchor assessment. Lipidomics: Lipid extracts analyzed by MALDI-TOF MS (negative ion mode) and SFC-HRMS (ESI-QTOF, positive/negative). Detected classes included PIMs, sulfolipids (SGLs), phthiocerol dimycocerosates (PDIM/DIM), mycolic acids (α- and methoxy-MA; free forms), and tuberculosinyl adenosine (TbAd/1-TbAd). Internal standard: 1,2-diradienoyl-sn-glycero-3-phosphocholine; extracted ion chromatogram areas used for relative quantification. Proteomics: Label-free quantitative proteomics on Orbitrap (nanoLC-MS/MS). Proteins reduced/alkylated, trypsin-digested; database searches against Mtb H37Rv and human UniProtKB. Selected proteins validated by dot-blot immunoassay (e.g., GroEL2). Longitudinal follow-up: Six pediatric smear/culture-negative patients sampled at baseline and months 1 and 3 during antibiotic treatment to monitor biomarker trajectories. Statistics: ROC analysis for LAM immunoassay in EBC (AUC, sensitivity, specificity; LOOCV). Correlations between biomarkers assessed (Pearson r) with P-values.

- LAM detection in EBC: Dot-blot immunoassay detected LAM in all baseline TB patients, but not in controls (healthy or community-acquired pneumonia). Apparent LAM amounts by immunoassay in TB EBCs ranged approximately 40–947 ng per EBC sample; chemical quantification of arabinose (proxy for LAM) yielded 0.7–30 µg per EBC, confirming high abundance. - Diagnostic performance: ROC AUC = 0.997 (95% CI ~0.993–1.002), n=46 cases, n=45 controls; P < 0.0001. At a 64 ng/EBC threshold, sensitivity 93.4% and specificity 100%. Leave-one-out cross-validation: sensitivity 0.93, specificity 0.98, accuracy 0.96. Signal-to-noise ratio >70. Smear- and culture-negative adults and pediatric patients were distinguishable from controls; even pediatric smear-/culture-negative patients under treatment with apparent LAM <480 ng/EBC were separated from controls. - Longitudinal monitoring: In 6 pediatric patients followed for 3 months, apparent LAM declined in 5/6; 4 fell below the diagnostic threshold by month 1 and remained low at month 3. One child showed no decline and experienced treatment failure, later investigated for primary immunodeficiency. - LAM structure in EBC: NMR of LAM-enriched fractions from pooled adult and pediatric EBCs showed overall intact polysaccharide architecture versus broth-grown Mtb LAM, but with notable differences: presence of t-α-Ara3 and t-α-Ara5 (“non-mature” arabinan termini), reduced Ara6 motifs, fewer mannose caps (~1 mono-, 1.5 di-, 0.3 tri-mannosyl units per LAM; 3–4× fewer di-/tri-caps than broth), absence of MTX and certain succinyl-type modifications, and intact MPI lipid anchor; slightly lower apparent MW consistent with truncated arabinan termini. - Lipid biomarkers: EBC contained Mtb lipids (PIMs, SGLs, PDIM/DIM, free α- and methoxy-mycolic acids, TbAd). Lipid molecular species profiles were consistent across patients and differed from broth-grown bacteria, suggesting in vivo biofilm-like metabolism using host lipids. Relative abundance patterns of MA, PIM, and TbAd paralleled LAM patterns. Correlations: LAM vs MA r = 0.54 (P < 0.0001); MA vs TbAd r = 0.92 (P < 0.0001). Detection sensitivity higher for MA and TbAd than PIM under the tested conditions. - Proteins: Up to 1432 Mtb proteins detected across individual TB EBCs; per-patient detections ranged 127–1288 proteins. Few proteins were detected in smear-negative patients. GroEL2 immunoassay tracked with LAM (r = 0.90, P < 0.0001), distinguishing TB patients from controls though less sensitive for smear-negative cases. - Method concordance: LAM immunoassay and chemical analyses correlated modestly (r = 0.45, P = 0.0002), with differing numeric ranges, likely due to epitope-specific antibody recognition and structural modifications of LAM in vivo.

Detecting Mtb-derived molecules in EBC enables non-invasive diagnosis of pulmonary TB, including in cases where sputum-based tests perform poorly (children, smear-/culture-negative). The strong diagnostic accuracy of LAM immunoassay in EBC, supported by lipid and protein biomarker profiles, demonstrates the feasibility of an EBC-based diagnostic approach. Longitudinal declines in EBC LAM during therapy suggest utility for monitoring treatment response and potentially identifying treatment failure early. Structural characterization of LAM in EBC reveals in vivo modifications (reduced Ara6, altered capping, absence of some motifs) that inform rational design of improved diagnostic antibodies (e.g., targeting non-mature Ara termini) and may guide vaccine or therapeutic development. Lipid profiles (mycolic acids, TbAd, SGLs, PDIM) reflect the metabolic state of Mtb in human lungs, consistent with biofilm-associated growth and host lipid utilization, offering additional Mtb-specific targets that can enhance specificity when combined with LAM. Overall, EBC-based biomarker detection aligns with point-of-care needs: it is rapid, affordable, and non-invasive, and could facilitate immediate diagnosis and therapy initiation, including in pediatric and paucibacillary settings.

This proof-of-concept study establishes that exhaled breath condensate contains abundant Mtb-specific molecules—LAM, characteristic lipids, and proteins—that can diagnose active pulmonary TB with high accuracy and monitor treatment response in both adults and children. Structural insights into LAM from patient EBCs highlight epitope differences from broth-grown bacteria, informing development of more sensitive and specific antibodies. Combining LAM with Mtb-specific lipids (e.g., SGLs, TbAd, mycolic acids) and proteins may further improve diagnostic performance and specificity. Future work should standardize EBC collection/normalization, validate performance in larger, diverse cohorts (including HIV-positive and extrapulmonary TB), develop robust, rapid point-of-care assays (e.g., lateral flow) targeting EBC biomarkers, and refine antibody specificity toward in vivo LAM epitopes.

- Proof-of-concept with limited cohort sizes for some subgroups; some adults were already on antibiotics (<2 weeks), potentially lowering biomarker levels and contributing to false negatives. - Variability in EBC collection and processing can affect quantitation; standardization is needed to reduce inter- and intra-subject variability, though binary detection may be less impacted. - Discrepancies between LAM immunoassay and chemical quantification (r = 0.45) suggest epitope-dependent detection and structural heterogeneity of LAM in vivo; assay calibration and antibody selection are critical. - LAM is not Mtb-specific (shared across mycobacteria), potentially limiting species specificity when LAM is present at low levels; combining with Mtb-specific lipids/proteins is advisable. - Some protein assays (e.g., GroEL2) lacked sensitivity for smear-negative cases; broader antibody panels and more sensitive detection methods are needed. - Mass spectrometric lipid detection used a fraction of EBC volume (~6%), suggesting potential underestimation of sensitivity; extraction strategies influenced lipid detection profiles.