A total of 308 peripheral blood samples were collected from patients at Weifang No. 2 People’s Hospital between March 2021 and March 2024. Blood was drawn into EDTA-K₂ anticoagulant tubes and centrifuged at 3000 rpm for 15 min at room temperature. The plasma fraction was then carefully separated and stored for subsequent RNA extraction. In addition to these clinical samples, miRNA expression profiles were obtained from publicly available datasets in the Gene Expression Omnibus (GEO) database, including GSE116542¹, GSE31568, and GSE61741 (18, 19).

Total RNA was isolated from plasma samples using TRIzol™ LS Reagent (Thermo Fisher Scientific, Waltham, MA, United States) in accordance with the manufacturer’s instructions. Complementary DNA (cDNA) was synthesized from the extracted RNA using the PrimeScript™ II 1st Strand cDNA Synthesis Kit (Takara Bio, Shiga, Japan). Quantitative real-time PCR (qRT-PCR) was then conducted under the following thermal cycling conditions: initial denaturation at 95 °C for 30 s, followed by 40 amplification cycles consisting of denaturation at 95 °C for 5 s and annealing/extension at 60 °C for 30 s.

Primers specific for U6 small nuclear RNA (used as the endogenous control) and the three target miRNAs were synthesized by Sangon Biotech (Shanghai, China). Relative expression levels of miRNAs were calculated using the 2^−ΔCt method, where ΔCt = Ct (miRNA) – Ct (U6).

The target miRNA sequences, sourced from the miRDB database, were as follows:

Verification against the GenBank database identified Mus musculus U6 small nuclear RNA-RnU6 (GeneID: 19862, 107 bp):

5′-GTGCTCGCTTCGGCAGCACATATACTAAAATTGGAACGATACAGAGAAGATTAGCATGGCCCCTGCGCAAGGATGACACGCAAATTCGTGAAGCGTTCCATATTTTT-3′.

Differentially expressed miRNAs were identified using the GEO2R online analysis tool.² For the LC group, the selection threshold was set at |log₂ fold change (FC)| ≥ 0, whereas for the PTB group, the threshold was |log₂FC| ≥ 1 with a significance level of p ≤ 0.05. Here, FC represents the ratio of the mean gene expression level in patients compared with HC. To maximize the inclusion of potentially informative biomarkers, no false discovery rate (FDR) adjustment or other multiple-testing corrections were applied during the initial screening, and statistical significance was defined as p ≤ 0.05. Furthermore, candidate miRNAs were required to show opposing expression patterns between the PTB and LC groups to enhance their diagnostic relevance.

A binary logistic regression model was established using SPSS software, version 27.0.1 (IBM Corp., Armonk, NY, USA). The diagnostic performance of the model was assessed by generating receiver operating characteristic (ROC) curves and calculating the AUC with MedCalc Statistical Software, version 20.022 (MedCalc Software bvba, Ostend, Belgium). Similarly, decision curve analysis (DCA) was conducted using R software, version 3.5.0 (R Development Core Team, Vienna, Austria), to evaluate the clinical net benefit of the predictive model.

Putative mRNA targets of the selected miRNAs were predicted using three publicly available databases: TargetScan version 7.2³, miRWalk⁴, and miRDB.⁵ Functional enrichment analyzes of the predicted target genes were then performed using Gene Ontology (GO) annotations and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis via the Sangerbox online platform.⁶

An independent t-test was applied to evaluate differences between two groups, whereas one-way analysis of variance (ANOVA) was used for comparisons involving more than two groups. Sensitivity comparisons among diagnostic methods were performed according to the following criteria: when the sample size was n ≥ 40 and all expected frequencies (T) were ≥ 5, Pearson’s chi-square test was used; when n ≥ 40 with at least one expected frequency of 1 ≤ T < 5, the continuity-corrected chi-square test was applied; and when n < 40 or any expected frequency was T < 1, Fisher’s exact test was employed. All statistical analyzes and graphical visualizations were conducted using GraphPad Prism software, version 8.0.2 (GraphPad Software, San Diego, CA, United States), and R software, version 3.5.0 (R Development Core Team, Vienna, Austria). A p-value of < 0.05 was considered statistically significant.

A total of 308 participants were initially recruited for this study, comprising 128 patients with PTB, 109 patients with LC, and 71 HC. PTB cases were identified based on clinical symptoms and chest imaging findings consistent with tuberculosis. Among these, 86 patients were confirmed by positive results from Mycobacterium tuberculosis (MTB) culture, acid-fast bacilli (AFB) smear microscopy, the TB-LAMP assay, or the Xpert MTB/RIF test. The remaining 42 PTB cases were diagnosed based on clinical presentation, chest radiographic features, and a favorable response to anti-TB therapy, in accordance with the Chinese Clinical Guideline for Diagnosis of PTB (WS 288-2017). LC cases were confirmed primarily through histological and/or cytological examination, supplemented by clinical evaluation when necessary. The HC group included individuals without respiratory symptoms and those with normal chest imaging results obtained within 3 months before enrollment. None of the participants had a history of PTB or other malignancies, and all were free from medical treatment within the 4 weeks preceding recruitment.

To minimize potential confounding, all participants were screened to exclude co-infections such as HIV and hepatitis B virus. Plasma samples that were hemolyzed or failed to meet RNA quality control criteria were excluded, resulting in 272 eligible samples for downstream analysis. The sample selection process is outlined in Figure 1, and the demographic and clinical characteristics of the 272 included participants are summarized in Table 1.

The primary objective of this study was to identify a robust miRNA signature capable of distinguishing PTB from LC. In the current study, GEO datasets were analyzed, and 50 differentially expressed miRNAs were identified in a PTB dataset, along with 75 and 192 in two independent LC datasets. Comparative analysis revealed six overlapping candidates: hsa‑miR‑199a‑3p, hsa‑miR‑199b‑3p, hsa‑miR‑342‑3p, hsa‑miR‑502‑3p, hsa‑miR‑361‑5p, and hsa‑miR‑423‑5p. Because hsa-miR‑502‑3p, hsa‑miR‑361‑5p, and hsa‑miR‑423‑5p were consistently upregulated in both PTB and LC, they lacked discriminatory specificity and were excluded. Therefore, hsa‑miR‑342‑3p, hsa‑miR‑199a‑3p, and hsa‑miR‑199b‑3p were selected as the candidate biomarker panel (Figure 2).

To establish a rapid, non‑invasive diagnostic model, the plasma expression levels of these three miRNAs were quantified by RT‑qPCR. Significant differences were observed in expression levels between PTB patients (n = 77) and LC patients (n = 68), as well as between PTB patients and healthy controls (HC, n = 65) (Figure 3), supporting their potential clinical applicability.

The diagnostic model was trained on the cohort using binary logistic regression analysis [n = 145 (77 PTB and 68 LC)]. The diagnostic score for each case was calculated using the following equation:

The diagnostic model demonstrated excellent performance in plasma samples from patients with PTB, yielding an AUC of 0.911 (95% CI: 0.852–0.952, p < 0.001; Figure 4A), with a sensitivity of 0.974 and a specificity of 0.765. Based on these findings from the training cohort, the model was further evaluated for robustness and accuracy in an independent validation cohort (n = 62; 33 patients with PTB and 29 patients with LC). Validation testing confirmed the strong diagnostic application of the model, with an AUC of 0.886 (95% CI: 0.780–0.953, p < 0.001), a sensitivity of 0.879, and a specificity of 0.759 (Figure 4B).

For comparison, models constructed using only two miRNAs were also assessed; however, their AUC values were lower than those of the three-miRNA model, as summarized in Supplementary Table 1.

The clinical utility of any diagnostic tool depends on its ability to maximize accurate detection while minimizing the risks associated with misdiagnosis. In suspected LC cases, invasive procedures are often necessary, and both false-positive and false-negative results can lead to severe clinical consequences. To assess the practical applicability of the three-miRNA panel, DCA was performed (Figure 4C). In the test cohort, the DCA demonstrated that the miRNA panel consistently provided a higher net clinical benefit across all threshold probabilities compared with the default strategies of treating all patients or treating none. These results suggest that the panel may offer tangible clinical advantages by reducing the need for unnecessary invasive procedures and lowering the risk of diagnostic errors.

To examine the potential influence of age on miRNA expression and its implications for diagnostic accuracy, correlations between age and the expression levels of the three selected miRNAs were analyzed in patients with PTB. A negative correlation was observed, with expression levels decreasing as age increased (Figures 5A–C). The correlation coefficients were R = −0.45 (p = 0.008) for hsa-miR-342-3p, R = −0.45 (p = 0.009) for hsa-miR-199a-3p, and R = −0.48 (p = 0.005) for hsa-miR-199b-3p.

To further assess the effect of age on diagnostic performance, participants were stratified into two groups: younger (≤ 50 years) and older (≥ 51 years). The ROC curve analysis indicated improved diagnostic performance in the younger group, with an AUC of 0.950, compared with an AUC of 0.834 in the older group (Figures 5D,E). These results suggest that the diagnostic application of the miRNA panel may be particularly enhanced in younger individuals.

Early detection of PTB is critical for improving clinical outcomes. In the validation cohort, patients were initially stratified into early or late stages according to the presence or absence of pulmonary cavities. Recognizing that this single parameter has inherent limitations and may not adequately reflect the overall disease status, additional severity-related clinical indicators, including monocyte count, hemoglobin (Hb) levels, C-reactive protein (CRP), and erythrocyte sedimentation rate (ESR), were also evaluated to provide a more comprehensive assessment (20–24).

In PTB patients with cavitary disease, ESR, CRP, and monocyte counts were elevated compared with non-cavitary cases, whereas hemoglobin levels were significantly reduced (Figures 6A–E). These results underscore apparent clinical differences between cavitary and non-cavitary stages of PTB. The diagnostic utility of the three-miRNA panel was then compared with conventional methods. Across PTB subgroups, the panel consistently demonstrated high sensitivity. In cavitary PTB, it significantly outperformed smear microscopy and culture in detection sensitivity (p < 0.05). For non-cavitary PTB, the panel achieved sensitivity comparable to culture, although it did not surpass Xpert MTB/RIF (p < 0.05). When all PTB cases were assessed together, the miRNA panel showed higher sensitivity than Xpert MTB/RIF, smear microscopy, and culture (p < 0.05). These findings suggest that the panel may serve as a viable diagnostic alternative in clinical practice, with performance matching or, in some cases, exceeding that of Xpert (Figure 6F).

To investigate the potential biological roles of the selected miRNAs in TB, GO enrichment analysis was conducted on their predicted target mRNAs. Target genes were identified based on the intersecting results from three prediction databases: TargetScan 7.2, miRWalk, and miRDB.

The GO analysis revealed that target genes of hsa-miR-342-3p were significantly enriched in biological processes, including positive regulation of gene expression, transcriptional regulation by RNA polymerase II, and RNA metabolic regulation (Figure 7A). Similarly, target genes of hsa-miR-199a/b-3p were associated with pathways involved in positive regulation of gene expression, macromolecule metabolic processes, and overall metabolic regulation (Figure 7B).

Then, KEGG pathway enrichment analysis was performed to identify signaling pathways that these miRNAs may regulate. Genes targeted by hsa-miR-342-3p were mainly enriched in the Ras, TGF-β, and Jak–STAT signaling pathways (Figure 7C). However, target genes of hsa-miR-199a/b-3p showed significant enrichment in the PI3K-Akt, thyroid hormone, and sphingolipid signaling pathways (Figure 7D).