Nasal Gene Expression in ART-Naive Adults with HIV and Pulmonary Tuberculosis in Uganda

Technical Summary: Nasal Gene Expression in ART-Naive Adults with HIV and Pulmonary Tuberculosis in Uganda

Problem Statement
Diagnosing pulmonary tuberculosis (TB) in people living with HIV (PLHIV) remains a critical challenge, particularly in ART-naive populations. PLHIV often present with paucibacillary disease and difficulty expectorating sputum, limiting the efficacy of standard sputum-based diagnostics like culture or Xpert MTB/RIF Ultra. While blood-based mRNA biomarkers have shown promise as triage tests, their performance is often suboptimal in PLHIV, potentially due to the dominance of interferon-stimulated genes (ISGs) driven by HIV viremia or opportunistic infections. Furthermore, the initial pathogen-host interaction in TB occurs in the upper airway, yet transcriptomic analysis of nasal specimens has not been explored for TB diagnosis. This study addresses the gap in understanding host gene expression profiles in the nasal mucosa of PLHIV with active TB compared to those without.

Methodology
The study enrolled 40 ART-naive adults in Uganda with newly diagnosed HIV and presumptive TB symptoms (cough, fever, night sweats, weight loss, hemoptysis). Participants were classified into two groups based on sputum Xpert MTB/RIF Ultra results: 20 with confirmed active TB disease and 20 without TB (controls). Exclusion criteria included prior TB treatment, current/former smoking, diabetes, severe malnutrition, or recent antibiotic/corticosteroid use.

• Sample Collection: Nasal cells were collected via cytology brush from the lateral nasal wall, and blood was drawn into PAXgene RNA tubes. Samples were collected prior to the initiation of TB or HIV treatment.
• RNA Sequencing: Total RNA was extracted, ribosomal RNA depleted, and libraries prepared for Illumina NovaSeq sequencing (paired-end 150 bp).
• Bioinformatics: Raw data underwent quality control (FastQC, Trimmomatic) and alignment to the human genome (GRCh38) using STAR. Differential gene expression (DEG) analysis was performed using DESeq2, identifying genes with an adjusted p-value < 0.05 and log2 fold-change ≥ 1.
• Machine Learning: Six supervised machine learning algorithms (Random Forest, SVM with radial/linear kernels, Elastic Net, PLS, KNN) were trained on the DEG sets to classify TB status. Performance was evaluated using 10-fold cross-validation repeated 10 times, reporting Area Under the Curve (AUC), sensitivity, and specificity.
• Pathway and Cell Analysis: Gene Set Enrichment Analysis (GSEA), Over-Representation Analysis (ORA), and Ingenuity Pathway Analysis (IPA) were used to identify enriched biological pathways. Immune cell composition was estimated using CIBERSORTx with the LM22 signature matrix.
• Validation: The study evaluated the performance of 59 published blood-based TB signatures on the nasal dataset and compared them against the blood dataset.

Key Results

• Differential Gene Expression: The study identified 44 differentially expressed genes (DEGs) in nasal samples and 238 DEGs in blood samples between TB and control groups. Only three genes overlapped between the two sample types. Notably, most nasal DEGs were downregulated in TB, whereas most blood DEGs were upregulated.
• Diagnostic Performance: Models trained on the 44 nasal DEGs achieved a cross-validated AUC between 0.87 and 0.90. Models trained on the 238 blood DEGs achieved an AUC between 0.89 and 0.98. A Random Forest model on nasal data achieved the highest AUC (0.90).
• Simplified Signature: A parsimonious four-gene nasal signature (SPIB, SHISA2, TESPA1, and CD1B) was identified based on variable importance across all models. This signature achieved an AUC of 0.87–0.94, with five out of six models meeting the World Health Organization (WHO) criteria for a TB triage test (>90% sensitivity and >70% specificity).
• Pathway Divergence: Pathway analysis revealed opposing immune responses. In nasal samples, pathways related to the innate immune system, inflammatory response, interferon signaling (IFN-γ), neutrophil degranulation, and chemokine signaling were significantly downregulated in TB. Conversely, in blood samples, these same pathways were upregulated.
• Published Signatures: Only 6 out of 59 published blood-based TB signatures performed well (AUC ≥ 0.7) in the nasal dataset, whereas 41–44 performed well in the blood dataset, highlighting the distinct nature of the nasal transcriptome in TB.
• Cell Composition: While no significant differences in nasal immune cell proportions were found between groups, blood samples showed a higher proportion of neutrophils and lower natural killer cells in the TB group.

Significance and Claims
The authors present this work as a proof-of-concept study demonstrating that distinct nasal gene expression patterns associated with TB exist in ART-naive PLHIV and are not captured in blood. The study claims that:

1. Novel Biomarkers: Nasal transcriptomics can identify host signatures for TB that differ fundamentally from systemic (blood) responses, specifically showing a blunted local mucosal immune response (downregulation of innate immunity) concurrent with systemic inflammation.
2. Diagnostic Potential: The identified nasal gene signatures, particularly the four-gene panel, show promise as non-sputum-based diagnostic biomarkers that meet WHO triage criteria.
3. Complementary Value: The finding that nasal models correctly classified TB cases missed by blood models (and vice versa) suggests potential value in combining nasal and blood biomarkers for improved diagnostic accuracy.
4. Feasibility: The study establishes the feasibility of using minimally invasive nasal samples for transcriptomic analysis in a high-burden, difficult-to-diagnose population (ART-naive PLHIV).

The authors explicitly state that further validation in larger, independent, and geographically distinct cohorts is required before clinical application. They also note limitations, including the reliance on sputum Ultra as the reference standard (potentially missing paucibacillary cases), the small sample size preventing external validation splits, and the inability to adjust for all potential confounders like nasopharyngeal colonization.