Dataset Selection Criteria: GSE97298 and GSE83456 were selected based on:

1) Human whole-blood transcriptomes from pulmonary infection patients;

2) Inclusion of both NTM/TB and matched healthy controls;

3) Platform compatibility (Affymetrix/Illumina) for cross-dataset normalization.

NTM Cohort: Gene Expression Omnibus (GEO) dataset GSE97298 ( https://www.ncbi.nlm.nih.gov/geo/ ) [12] , containing whole-blood transcriptomes from 32 pulmonary NTM patients and 9 healthy controls.

TB Cohort: GEO dataset GSE83456 [13] , including 45 active TB patients and 61 controls.

Ethical Compliance: All data was de-identified and publicly available under GEO’s data usage agreements.

1) Raw Data Processing:

Affymetrix CEL files: Normalized using the RMA algorithm in the affy R package (v1.78.0) [14] .

Illumina IDAT files: Processed with limma (v3.56.0) using neqc normalization for background correction and quantile normalization [15] .

2) Batch Correction: Combat algorithm in sva (v3.48.0) [16] adjusted for batch effects (NTM: GPL11532 vs. TB: GPL10558).

Conducted separately for NTM and TB cohorts using limma with linear models.

Thresholds: |log₂FC| ≥ 1.5, adjusted *p* (FDR) < 0.05 (Benjamini-Hochberg correction).

Mapped probes to gene symbols using Bioconductor annotation packages:

hugene11sttranscriptcluster.db (v8.8.0) for NTM

illuminaHumanv4.db (v1.26.0) for TB

1) Tools: WebGestalt (2024 release, https://www.webgestalt.org ) [17] :

Parameters: Over-Representation Analysis (ORA), FDR < 0.05, minimum 5 genes per category.

Reference set: Human genome (GRCh38, Ensembl v104).

2) Metascape ( https://metascape.org ) [18] :

Integrated GO, KEGG, and Reactome databases.

Significance threshold: *p* < 1 × 10⁻³ after Bonferroni correction.

STRING Database (v12.0, https://string-db.org ) [19] : Interaction confidence score > 0.7 (high confidence). Excluded disconnected nodes. Cytoscape (v3.9.1) [20] : Network visualization and topological analysis. Hub genes identified using: cytoHubba (v1.6.1) [21] (Maximal Clique Centrality algorithm). MCODE (v2.0.0) [22] (cluster score > 5).

Spearman correlation between ELANE and DEFA4 calculated via Hmisc R package (v5.1-0). Significance threshold: *p* < 0.001.

KEGG pathway reconstruction (hsa04621: NOD-like receptor signaling) performed using pathview (v1.40.0) [23] .

R Environment: v4.3.1 with critical packages:ggplot2 (v3.4.2) [24] for visualization. ComplexHeatmap (v2.16.0) [25] for hierarchical clustering.

Figure 1 is the work flow of study.

In the NTM cohort (GSE97298), comparative analysis of 32 patients versus 9 controls identified 349 differentially expressed genes (DEGs) under stringent thresholds (|log₂FC| ≥1.5, FDR < 0.05), with 53 upregulated and 296 downregulated genes. Notably, interferon-gamma (IFNG), a critical mediator of anti-mycobacterial immunity anti-mycobacterial immunity ( Figure 2 ), showed marked suppression (log₂FC = -2.1, *p* = 1.2 × 10⁻⁵), while the negative TLR regulator IRAK3 was significantly elevated (log₂FC = +0.62, *p* = 0.003).

The TB cohort (GSE83456) exhibited more pronounced transcriptional alterations, with 756 DEGs detected between 45 patients and 61 controls. Key findings included downregulation of the pattern recognition receptor NOD2 (log₂FC = -1.10, *p* = 4.5 × 10⁻⁶) and metalloproteinase MMP9 (log₂FC = −0.71, *p* = 0.001), contrasting with upregulation of lymphoid enhancer-binding factor 1 (LEF1, log₂FC = +0.79, *p* < 0.001), a transcription factor governing T-cell differentiation. Hierarchical clustering ( Figure 3 ) revealed disease-specific patterns: NTM patients showed prominent upregulation of antimicrobial effectors (CLEC4D, BPI), while TB signatures were dominated by T-cell activation markers (LEF1, CD28).

Venn analysis ( Figure 4 ) identified 48 genes common to both infections, representing 13.8% (48/349) of NTM DEGs and 6.1% (48/784) of TB DEGs. Striking bidirectional regulation patterns emerged upon cross-disease comparison ( Table 1 & Table 2 ):

adj.P.Val thresholds: NTM (FDR < 0.05), TB (FDR < 0.001). Full gene lists are provided in Supplementary Tables.

Table 2. Differential expression analysis of 48 shared genes in TB (GSE83456).

adj.P.Val thresholds: NTM (FDR < 0.05), TB (FDR < 0.001). Full gene lists are provided in Supplementary Tables.

Functional enrichment analysis through WebGestalt (2024 release, https://www.webgestalt.org/ )revealed significant over-representation of immune-related processes (FDR < 0.01):

Biological processes:

Regulation of immune system process (GO:0002682, 19 genes); Response to bacterium (GO:0009617, 12 genes) ( Table 3 ).

Molecular functions:

Signaling receptor activity (27.1% of genes); Cytokine binding capacity (14.6%).

Metascape analysis further identified neutrophil degranulation (Reactome R-HSA-6798695, *p* = 2.45 × 10⁻⁸) and Fc receptor signaling (GO:0038093, *p* = 1.15 × 10⁻⁴) as core pathways. Disease association mapping demonstrated strong links to pulmonary tuberculosis (DisGeNET C0041327, *p* = 1.58 × 10⁻⁶) and bacterial infections (C0004623, *p* = 3.98 × 10⁻⁸) ( Table 4 ).

Protein–protein interaction (PPI) network analysis, performed using the STRING database, elucidated key immune pathways significantly enriched among the host response genes. The top enriched immune pathways are summarized in ( Table 5 ). Notably, “Regulation of immune system process” (GO:0002682) exhibited the most significant enrichment (FDR = 1.72 × 10⁻⁵), involving 18 genes, with IRAK3, NOD2, and CD28 identified as pivotal regulatory nodes. Similarly, “Positive regulation of immune system process” (GO:0002684) was highly represented (FDR = 1.72 × 10⁻⁵), comprising 15 genes, with key contributions from LEF1 and CD40LG. The “Response to bacterium” pathway (GO:0009617) was also significantly enriched (FDR = 0.001), highlighting the involvement of BPI, CLEC4D, and DEFA4. In addition, the “Regulation of innate immune response” pathway (GO:0045088) (FDR = 0.002) and the “Antimicrobial humoral response” pathway (GO:0061844) (FDR = 0.007) were enriched, with IRAK3, FCGR1A, DEFA4, and ELANE serving as central components. Collectively, these results underscore the pivotal role of immune regulatory networks in mediating the host response to pathogen challenge, with several key genes occupying central positions within these pathways.

Consensus analysis integrating cytoHubba (topological analysis), cytoNCA (node centrality assessment), and MCODE (molecular complex detection) robustly identified neutrophil elastase (ELANE) and defensin alpha 4 (DEFA4) as central hub genes within the protein–protein interaction network ( Figure 5 ). Both ELANE and DEFA4 were consistently recognized as core regulators by all three computational approaches, as demonstrated by their intersection in the Venn diagram.

To further substantiate these findings, internal validation was performed using the original dataset. The expression levels of ELANE and DEFA4 were found to be significantly different (P < 0.05), confirming their pivotal roles with robust statistical significance ( Figure 6(a) , Figure 6(b) ). This internal validation provides additional confidence in the reliability of the computational predictions and underscores the central regulatory functions of ELANE and DEFA4 within the immune response network.

A robust and consistent positive correlation was observed between ELANE and

DEFA4 expression across both cohorts, as demonstrated by Spearman correlation coefficients of r = 0.86 and r = 0.94, respectively (P < 0.001 for both comparisons). Notably, this strong co-expression was evident even though DEFA4 did not independently exhibit significant differential expressions. These findings indicate a potential co-regulatory mechanism underlying their involvement in antimicrobial responses. Biologically, both ELANE and DEFA4 are intimately associated with neutrophil-mediated host defense, with ELANE participating in neutrophil degranulation and DEFA4 functioning as a key antimicrobial peptide. This functional synergy suggests that, while DEFA4 alone may not be differentially expressed, its expression is tightly coordinated with ELANE and may contribute cooperatively to pathogen clearance. The scatter plots in Figure 7(a) , Figure 7(b) further illustrate the strong positive relationship between ELANE and DEFA4 in both NTM and TB cohorts, underscoring their collective role in the immune response to mycobacterial infection.

The bidirectional regulation of LDHB (NTM: log₂FC = −1.86 vs. TB: +0.61) highlights divergent host metabolic adaptations. In TB, LDHB upregulation may fuel glycolytic reprogramming to support pro-inflammatory macrophage responses, consistent with prior reports linking enhanced glycolysis to M. tuberculosis control [26] . Conversely, LDHB suppression in NTM could promote a metabolically quiescent state conducive to chronic infection, akin to metabolic adaptations observed in latent viral infections [27] . This dichotomy aligns with clinical observations of TB’s acute inflammation versus NTM’s indolent course, suggesting that targeting metabolic checkpoints (e.g., lactate transporters) might recalibrate host responses.

The paradoxical upregulation of NOD2 in NTM (log₂FC = +0.63, FDR = 0.007) contrasts sharply with its suppression in TB (log₂FC = −1.10, FDR < 0.001). While NOD2 typically activates NF-κB via RIPK2 in TB [28] , its role in NTM may involve non-canonical pathways such as autophagy induction or mitochondrial antiviral signaling (MAVS) crosstalk [29] . This hypothesis is supported by NOD2’s known capacity to trigger LC3-associated phagocytosis, a mechanism exploited by intracellular pathogens to evade lysosomal degradation [30] .

The inverse regulation of IRAK3 (NTM: +0.62 vs. TB: −0.80) underscores its role as a TLR signaling rheostat. In NTM, elevated IRAK3 likely dampens TLR-driven inflammation through SOCS1-mediated ubiquitination [31] , fostering immune tolerance. In TB, IRAK3 downregulation may permit hyperactive TLR responses that, while enhancing bacterial clearance, risk collateral tissue damage—a phenomenon observed in murine TB models with IRAK3 deficiency [32] . Similarly, CD28 upregulation in TB (log₂FC = +0.87) correlates with enhanced T-cell co-stimulation and IFN-γ production [33] , whereas its suppression in NTM may reflect T-cell exhaustion, as seen in chronic viral infections [34] .

The strong *ELANE-DEFA4* correlation (r = 0.86, *p* < 0.001) in NTM suggests coordinated neutrophil degranulation, potentially limiting bacterial dissemination via NETosis [35] . However, ELANE’s downregulation in TB (log₂FC = −0.82) may reflect M. tuberculosis evasion strategies, such as secretion of nuclease EsxA to degrade NETs [36] . These findings echo recent work showing that excessive NETosis exacerbates lung pathology in murine TB [37] , highlighting the need for therapeutic strategies that balance antimicrobial efficacy and immunopathology.

Our study has limitations. First, the use of whole-blood transcriptomes may obscure tissue-specific responses, as pulmonary immune activity often diverges from peripheral signatures [38] . Second, sample size disparities (NTM: *n* = 32 vs. TB: *n* = 45) could affect statistical power, though our FDR-controlled analysis mitigates this risk [39] . Third, Potential confounders including age, sex, comorbidities (e.g., COPD), and prior treatments were not analyzed due to limited metadata in public datasets. Future studies should incorporate covariate-adjusted models to validate target robustness. Also, future work should validate mechanistic links:

1) In vitro NETosis assays measuring ELANE/DEFA4 release in NTM/TB-infected neutrophils;

2) Murine models evaluating NET inhibition (e.g., DNase I) on lung pathology;

3) Spatial transcriptomics to localize ELANE+ neutrophils in patient lung tissues.

The bidirectional regulators identified here—particularly IRAK3 and CD28—represent promising targets for host-directed therapies. IRAK3 inhibitors (e.g., ND-2158) have shown efficacy in dampening pathological inflammation in sepsis models [40] , while CD28 agonists are being explored to reverse T-cell exhaustion in chronic infections [41] . Validating these interventions in mycobacterial models could pave the way for precision immunotherapies tailored to infection context.