Plasma multiomics distinguishes pulmonary tuberculosis from other respiratory infections

This study identifies a highly accurate five-marker plasma signature (IFN-γ, IL-22, IL-10, methionine, and oxoproline) that effectively distinguishes pulmonary tuberculosis from other respiratory infections, meeting WHO criteria for rapid, non-sputum diagnostic and triage tests.

The Gist

The Big Picture: Finding a "Blood Test" for Tuberculosis

Imagine Tuberculosis (TB) is a sneaky thief hiding in a house (your lungs). Right now, the only way to catch this thief is to ask the house to cough up a sample of its "trash" (sputum) and send it to a high-tech lab to be examined. This is slow, requires special equipment, and sometimes the house just can't cough up the trash.

The Goal: The researchers wanted to find a better way. They wanted a simple blood test that could instantly tell the difference between a house with a TB thief and a house with a different kind of trouble (like pneumonia or a cold).

The Detective Work: Two Types of Clues

The researchers didn't just look for one clue; they looked for two different types of evidence in the blood, like a detective using both a magnifying glass and a lie detector.

1. The "Lie Detector" (Cytokines): These are tiny chemical messengers your immune system sends out when it's fighting a battle. Think of them as the immune system's "sirens." When TB is present, specific sirens go off louder than usual.
2. The "Magnifying Glass" (Metabolites): These are the tiny fuel particles and waste products your body burns while fighting. Think of them as the "footprints" or "trash" left behind by the battle. Different diseases leave different footprints.

The Experiment: Mixing the Clues

The team took blood samples from 391 adults who were sick with respiratory problems.

• Group A (The TB Thieves): 187 people who definitely had TB.
• Group B (The Imposters): 204 people who were sick with other things (like pneumonia) or had TB symptoms but tested negative.

They measured 28 different sirens and 118 different footprints in everyone's blood.

The Discovery: The "Magic 5"

By using a computer to analyze all these clues, they found that looking at just five specific items together was the key to solving the mystery. It wasn't just one siren or one footprint; it was a specific combination:

• 3 Sirens: IFN-gamma, IL-22, and IL-10.
• 2 Footprints: Methionine and Oxoproline.

The Result: When they tested this "Magic 5" combination, it was incredibly accurate.

• It correctly identified 98% of the TB cases (Sensitivity).
• It correctly ruled out 98% of the non-TB cases (Specificity).

Why this matters: The World Health Organization (WHO) has a "Gold Standard" for what a good TB test should look like. This new blood test signature met those standards perfectly. It's like finding a key that fits the lock exactly, even when the lock is rusty or the key is slightly bent.

The "Imposter" Challenge

A common mistake in these studies is testing the new key only against people who are perfectly healthy. But in the real world, you need to distinguish TB from other sick people (like those with pneumonia).

The researchers were smart. They included people with other serious lung diseases in their "control group." Even when the "imposters" were very sick, the "Magic 5" signature could still tell the difference between TB and the other diseases. This makes the test much more reliable for real-world hospitals.

What About HIV?

Many people with TB also have HIV, which confuses the immune system. The researchers worried this might mess up their test.

• The Good News: The "Magic 5" signature worked just as well for people with HIV as it did for people without HIV. It didn't get confused by the extra noise.

The "Crystal Ball" (Predicting Treatment)

The researchers also tried to use these clues to predict if a patient would get better or worse after starting treatment.

• The Result: This was harder. The "sirens" (cytokines) weren't very good at predicting the future. However, the "footprints" (metabolites, specifically fatty acids) showed some promise. It's like the footprints tell a story about how the body is recovering, but the story is still a bit blurry.

The "Why" (The Metabolic Story)

The study also looked at why these clues were different.

• The Fatty Acid Mystery: They found that TB patients had very low levels of certain fats (fatty acids) in their blood.
• The Analogy: Imagine the TB bacteria are like a hungry wolf. When it infects your body, it eats up all your fat reserves to grow. The "footprints" (low fat) show that the wolf has been feasting. This explains why the body looks so different when TB is present compared to other infections.

The Bottom Line

This paper is a major step forward. It proves that by combining chemical messengers (sirens) and fuel particles (footprints) from a simple blood draw, we can create a highly accurate test for Tuberculosis.

Why should you care?
If this test can be turned into a quick, cheap, point-of-care device (like a pregnancy test stick), doctors in remote villages could diagnose TB in minutes without needing a lab or a sputum sample. This could save millions of lives by catching the "thief" before it steals too much.

Note: This is a preprint, meaning it's new research that hasn't been fully checked by other scientists yet, but the results are very promising.

Technical Summary

1. Problem Statement

Tuberculosis (TB) remains a leading cause of infectious death globally. While the Xpert MTB/RIF assay is the current gold standard, it requires high-quality sputum samples, laboratory infrastructure, and trained personnel, making it difficult to implement in low-resource settings. Furthermore, approximately 23% of adults evaluated for TB cannot produce sputum. There is an urgent need for accurate, blood-based, point-of-care (POC) diagnostic tools that meet the World Health Organization (WHO) Target Product Profiles (TPP):

• Diagnostic Test: >65% sensitivity at 98% specificity.
• Triage Test: >90% sensitivity at 70% specificity.

Existing blood-based biomarker studies often suffer from limitations: they frequently use asymptomatic controls (overestimating specificity), rely on single-omics layers (transcriptomics or proteomics alone), or lack validation in clinically relevant cohorts with symptomatic non-TB respiratory diseases.

2. Methodology

Study Cohort:

• Total Participants: 391 adults (≥18 years) presenting with respiratory symptoms.
• Cases (PTB): 187 individuals with microbiologically confirmed Pulmonary TB (via Xpert MTB/RIF and/or M. tuberculosis sputum culture).
• Controls (ORD): 204 individuals with PTB excluded. This group was rigorously defined to include:
  • 165 patients hospitalized with non-TB respiratory illnesses (e.g., pneumonia, COPD exacerbation).
  • 39 ambulatory household contacts of TB patients who reported TB symptoms but tested negative.
• Subgroups: The cohort included 80 people living with HIV (PLWH).

Data Acquisition:

• Multiomics Profiling: Plasma samples were analyzed for:
  • Metabolomics: 118 targeted metabolites using high-resolution LC-MS/MS (Orbitrap Q Exactive).
  • Cytokines: 28 cytokines using the U-PLEX electrochemiluminescence assay (Meso Scale Discovery).
• Treatment Outcomes: For 178 of the PTB patients, treatment outcomes were tracked (122 cured, 56 failed/relapsed/deceased).

Statistical & Machine Learning Approach:

• Preprocessing: Data were adjusted for covariates (age, sex, HIV status) using linear models (limma).
• Feature Selection:
  • Diagnosis: Random Forest (via randomForestSRC) was used on a discovery set (70% of data) to select top features.
  • Treatment Prediction: The Boruta algorithm was used to identify features predictive of treatment failure.
• Model Building: Support Vector Machines (SVM) with Radial Basis Function (RBF) kernels were trained to classify PTB vs. ORD and predict treatment outcomes.
• Validation: Models were tested on an independent validation set (30% of data). Performance was evaluated using Area Under the Curve (AUC), sensitivity, and specificity at WHO-recommended thresholds.
• Pathway Analysis: Untargeted metabolomics data (9,516 positive/8,804 negative ion features) were analyzed using Mummichog for pathway enrichment. Correlation networks between cytokines and metabolites were constructed.

3. Key Contributions

1. Rigorous Control Group Design: Unlike many prior studies, this research utilized a control group comprising both hospitalized patients with other respiratory diseases and symptomatic household contacts. This provides a more realistic assessment of biomarker specificity in a clinical setting where differential diagnosis is challenging.
2. Integrated Multiomics Signature: The study successfully integrated cytokine and metabolite data, demonstrating that a combined approach outperforms single-omics models, particularly at high specificity thresholds.
3. WHO TPP Compliance: The identified biomarker signature meets the stringent WHO criteria for both a non-sputum diagnostic test and a triage test.
4. Mechanistic Insight: The study elucidates the interplay between metabolic pathways (specifically fatty acid and amino acid depletion) and immune signaling (cytokine upregulation) in TB pathogenesis.

4. Key Results

A. Diagnostic Performance (PTB vs. ORD)

• The 5-Marker Signature: The optimal signature consists of 3 cytokines (IFN-γ, IL-22, IL-10) and 2 metabolites (Methionine, Oxoproline).
• Performance Metrics:
  • AUC: 0.97 (95% CI: 0.95–1.00) in the validation set.
  • Sensitivity at 98% Specificity: 84% (exceeds the 65% WHO target for diagnostics).
  • Sensitivity at 70% Specificity: 98% (exceeds the 90% WHO target for triage).
• Robustness: The signature maintained high performance across subgroups:
  • HIV Status: AUC 0.99 (HIV+) and 0.98 (HIV-).
  • Control Type: AUC 0.97 (Hospitalized) and 0.98 (Ambulatory).
• Comparison: The combined model significantly outperformed cytokine-only or metabolite-only models, particularly in PLWH and hospitalized controls. It also outperformed previously published signatures (e.g., Kynurenine/Tryptophan ratio, IFN-γ alone).

B. Treatment Outcome Prediction

• Predictive Power: Models predicting treatment failure (cure vs. failure) showed modest performance (AUC ~0.69–0.70).
• Key Markers: The Boruta algorithm identified 5 metabolites (Tyrosine, Arachidonic acid, Malate, EPA, Valine) and 1 cytokine (IL-17A).
• Finding: Metabolites showed better predictive potential than cytokines for treatment outcomes, suggesting that metabolic shifts may be more indicative of treatment response than acute inflammatory markers.

C. Biological Mechanisms

• Pathway Enrichment: "De novo fatty acid biosynthesis" and "fatty acid activation" were the most significantly enriched pathways. Plasma fatty acids were significantly lower in PTB, suggesting increased host lipid utilization by M. tuberculosis.
• Correlation Networks: Significant negative correlations were found between downregulated metabolites (Tryptophan, Retinol, Histidine) and upregulated pro-inflammatory cytokines (IFN-γ, IL-6, TNF-α). This suggests that metabolic depletion drives or is driven by hyper-inflammation.
• IL-10 Anomaly: IL-10 was downregulated in PTB in this study, contrary to some literature. The authors attribute this to the inclusion of hospitalized pneumonia patients in the control group, who exhibited higher IL-10 levels, highlighting the importance of control group selection.

5. Significance and Limitations

Significance:

• Clinical Translation: The 5-marker signature represents a viable candidate for a rapid, minimally invasive, plasma-based POC test that could replace or supplement sputum-based diagnostics in resource-limited settings.
• Scientific Advancement: It validates the utility of multiomics integration, showing that combining metabolic and immune data captures complementary biological signals that single-layer approaches miss.
• Pathophysiology: The findings reinforce the concept of systemic lipid depletion and amino acid catabolism as hallmarks of active TB, offering new targets for therapeutic intervention.

Limitations:

• Geographic Bias: Most participants were from South Africa; external validation in diverse geographic populations is required.
• Symptomatic Bias: Participants were enrolled after presenting with symptoms; performance in community screening (sub-clinical TB) is unknown.
• Sample Size for Outcomes: The subgroup of PLWH with treatment outcome data was small (n=8), limiting the power to predict outcomes in this high-risk group.
• Extrapulmonary TB: The study focused solely on Pulmonary TB; performance in extrapulmonary cases was not assessed.

Conclusion:
This study provides a robust, multi-omics biomarker signature that effectively distinguishes pulmonary TB from other respiratory infections, meeting WHO diagnostic targets. It highlights the critical role of integrating metabolomics with cytokine profiling to improve diagnostic accuracy and offers insights into the metabolic-immune crosstalk driving TB pathology.