Research on tracheal stenosis based on molecular biomarkers

Abstract

Background

Tracheal stenosis is a common respiratory disease characterized by abnormal narrowing of the tracheal lumen, which restricts airflow and can lead to symptoms such as dyspnea and even asphyxiation. Although the widespread use of imaging and bronchoscopy has improved diagnostic efficiency, the molecular mechanisms underlying tracheal stenosis remain unclear, particularly regarding the identification of predictive molecular markers.

Methods

This was a retrospective study involving 43 patient samples. A range of statistical methods—including Spearman correlation analysis, multivariable logistic regression, LASSO regression, and ROC curve validation—were employed to systematically screen for molecular targets associated with tracheal stenosis and assess their potential for risk prediction. The study also investigated the roles of ion channels and cell cycle-related molecules through literature analysis. Additionally, a tracheal stenosis animal model was constructed, and RT-qPCR was used to detect differences in mRNA expression levels of candidate molecules (CACNA1D, MKI67, CCNB2, CDKL5, and VAMP2).

Results

Respiratory failure and scar diathesis were significantly associated with the occurrence of tracheal stenosis (P < 0.05). Patients with respiratory failure had a notably higher risk of developing tracheal stenosis (P < 0.001), while those with scar diathesis also showed increased susceptibility (P = 0.009). Other clinical characteristics, including age, sex, hypertension, and diabetes, were not significantly associated with stenosis risk (P > 0.05). Regarding molecular markers, positively correlated genes—such as CACNA1D, AMPH, and PTPRD—were expressed at significantly higher levels in patients with tracheal stenosis (P < 0.001), suggesting their role in promoting cell proliferation or ion channel dysfunction. In contrast, negatively correlated markers—such as CALML4, CDKL5, and VAMP2—were expressed at lower levels (P < 0.001), potentially exerting protective effects through inhibition of cell proliferation or maintenance of cellular structure. LASSO regression identified CCNB2, PTPRD, and CACNA1D as positively associated with stenosis risk, whereas VAMP2 and CDKL5 were negatively associated. ROC curve analysis demonstrated excellent model performance (AUC = 0.91), indicating strong predictive capacity of the selected biomarkers. Subgroup analysis showed that elevated MKI67 expression was significantly associated with stenosis risk in patients with respiratory failure, but not in those without. Experimental validation using a tracheal stenosis animal model confirmed the differential expression trends observed in clinical samples: upregulation of CACNA1D, MKI67, and CCNB2, and downregulation of CDKL5 and VAMP2.

Conclusions

This study, for the first time, identified several key molecular targets associated with tracheal stenosis through integrated clinical, computational, and experimental analyses. Dysregulation of ion channel-related genes (e.g., CACNA1D) and cell cycle regulators (e.g., MKI67, CCNB1, CCNB2) may contribute to stenosis by promoting airway smooth muscle proliferation and calcium signaling disturbances. Conversely, protective markers such as CALML4 and CDKL5 may inhibit proliferation and preserve cellular homeostasis. Notably, respiratory failure may amplify the pro-stenotic effects of MKI67 via enhanced inflammation or oxidative stress. The identified markers (e.g., CACNA1D, MKI67) show promise as early predictors for tracheal stenosis, facilitating early screening and risk stratification. Furthermore, targeted therapies—such as calcium channel blockers—against high-risk markers may offer novel treatment options. However, further studies with larger cohorts and mechanistic experiments are warranted to confirm these findings and explore clinical translation.

Introduction

Tracheal stenosis is a common respiratory condition characterized by abnormal narrowing of the tracheal lumen, which restricts airflow and leads to symptoms such as dyspnea, wheezing, and even asphyxiation [1]. The etiology of tracheal stenosis is diverse; however, the use of artificial airways—such as prolonged intubation or tracheostomy—remains a major cause of benign stenosis [2, 3]. Although advances in imaging and bronchoscopy have significantly improved diagnostic efficiency, the underlying pathogenesis of tracheal stenosis remains poorly understood [4, 5]. Current clinical management primarily relies on mechanical dilation or surgical intervention, both of which are associated with recurrence and reduced quality of life. Therefore, identifying reliable early molecular markers and elucidating their potential roles in the pathogenesis of tracheal stenosis have become critical priorities in both clinical and research settings.

Among the intracellular signaling pathways, the B cell transforming signaling (BTS) pathway plays a central role in regulating cell proliferation, apoptosis, and metabolic homeostasis via PI3K/AKT and MAPK cascades. However, most studies have focused on the functions of individual genes (e.g., AKT1, ERK1), with limited exploration of the global regulatory network. Moreover, the crosstalk between the BTS pathway and other pathways, such as Wnt/β-catenin and JAK/STAT, remains largely uncharacterized.

With the rapid advancement of bioinformatics, large-scale molecular marker screening has become increasingly feasible. Aberrant expression of molecular markers is now recognized as a major contributor to the onset and progression of various diseases, playing critical roles in risk prediction, disease classification, and targeted therapy [6–9]. However, studies focusing on molecular biomarkers in tracheal stenosis remain scarce, particularly in terms of systematic screening and comprehensive evaluation of associated risk factors.

In this study, we conducted an integrated analysis of clinical characteristics and molecular marker expression, employing statistical techniques including Spearman correlation analysis, multivariable logistic regression, LASSO regression, and ROC curve validation. Our aim was to systematically identify key molecular targets associated with tracheal stenosis and evaluate their potential utility in early risk prediction.

Materials and methods

Clinical data

Patients with benign tracheal stenosis who visited the Department of Respiratory Medicine at the Second Hospital of Hebei Medical University from March 2024 to December 2024 were enrolled in the experimental group. The control group consisted of patients with a history of artificial airway use but without tracheal stenosis, who were also treated at the same hospital during the same period. The presence or absence of tracheal stenosis in all enrolled patients was confirmed by neck and chest computed tomography (CT) imaging combined with bronchoscopy. Comorbidities were controlled for in both groups.

Inclusion criteria

Tracheal Stenosis Group: Patients who developed benign tracheal stenosis following artificial airway use and presented with clinical symptoms. Age between 18 and 100 years, with no gender restriction.

Control Group: Patients who had a history of artificial airway use but did not develop benign tracheal stenosis. Age between 18 and 100 years, with no gender restriction.

All participants in both groups had a history of artificial airway use to ensure comparability and to eliminate the possibility of including latent cases in the control group (i.e., patients who may appear normal but are at risk of developing stenosis following injury).

Exclusion criteria

Patients with tracheal stenosis caused by non-artificial airway factors, such as bronchial tuberculosis, thyroid enlargement, or relapsing polychondritis.

Patients with a history of genetic or immune-mediated dermatologic diseases (e.g., eczema, psoriasis, seborrheic dermatitis, urticaria), or those who had received related treatments.

Patients with other severe conditions that could interfere with study outcomes or pose increased risk, as determined by the investigators (e.g., malignancy, severe cardiovascular disease, diabetes, thyroid dysfunction, hepatic insufficiency).

Patients currently taking, or who had taken within two weeks prior to sampling, anticoagulants (e.g., warfarin) or antiplatelet agents (e.g., clopidogrel), as well as long-term aspirin users. Patients with known coagulation disorders or bleeding tendencies were also excluded.

Patients with known allergies to commonly used local anesthetics (e.g., lidocaine)

Data collection

Clinical and molecular data were systematically collected from all participants. General information, including age, gender, diagnosis, and treatment history, was recorded. Tracheal tissue samples and peripheral venous blood were obtained from both the tracheal stenosis group and the control group. For tissue sampling, bronchoscopic biopsy or surgical resection was performed, and the specimens were immediately preserved in liquid nitrogen or an RNA stabilization reagent to prevent RNA/protein degradation.

Molecular expression levels were measured using quantitative real-time PCR (qRT-PCR). Total RNA or DNA was extracted from the collected tissues, followed by reverse transcription (for RNA) into cDNA. Specific primers were designed, and fluorescence-based qPCR assays were conducted. Gene expression was quantified using the ΔCt method and normalized to housekeeping genes such as GAPDH and β-actin.

To assess the severity of tracheal stenosis, both imaging and clinical evaluations were performed. The degree of stenosis was quantified using CT or bronchoscopy by measuring the diameter and cross-sectional area at the stenotic site compared to a normal segment. The following formulas were used:

Diameter stenosis rate = (normal diameter – stenotic diameter)/normal diameter × 100%.

Area stenosis rate = (normal area – stenotic area)/normal area × 100%.

Based on these measures, stenosis was classified into four categories: mild (< 50%), moderate (50%–70%), severe (> 70%–90%), and extremely severe (> 90%). Clinical symptoms were also used to grade severity: patients with mild stenosis were asymptomatic or experienced exertional dyspnea only, those with moderate stenosis had limited activity tolerance, and those with severe stenosis experienced dyspnea or wheezing at rest, potentially requiring emergency intervention.

Statistical methods

Statistical analyses were performed using SPSS version 25.0. Patients were categorized into two groups according to the presence (coded as 1) or absence (coded as 0) of tracheal stenosis. For continuous variables with normal distribution, data were expressed as mean ± standard deviation (x̄ ± s), and comparisons between groups were conducted using independent sample t-tests. For non-normally distributed variables, data were presented as median and interquartile range [M (P25, P75)], and non-parametric tests were employed for group comparisons. A P value ≤ 0.05 was considered statistically significant.

To assess associations between tracheal stenosis and clinical or molecular variables, Spearman correlation analysis was conducted. Feature selection was further refined using Least Absolute Shrinkage and Selection Operator (LASSO) regression to reduce dimensionality and eliminate weakly associated variables. A Bayesian logistic regression model was applied to quantify the direction and strength of associations between gene expression levels, age, respiratory failure, and other relevant variables with tracheal stenosis.

Additionally, receiver operating characteristic (ROC) curves were constructed to evaluate the performance of the selected variables in distinguishing patients with and without tracheal stenosis, based on area under the curve (AUC) values.

Spearman correlation analysis

Spearman correlation coefficients were computed to evaluate monotonic relationships between variables and to identify highly collinear features that may introduce multicollinearity. Analyses were performed using R software (version 4.2.1). The cor() function was used to calculate the correlation matrix, and heatmaps were generated using the corrplot package (version 0.92). Features with |r| >0.9 were considered highly correlated; in such cases, the variable with greater variance was retained while the redundant variable was removed to reduce dimensional redundancy.

LASSO regression

LASSO regression was used for feature selection and regularization to identify variables most strongly associated with tracheal stenosis in high-dimensional data. The model employed L1 regularization to minimize overfitting. Implementation was conducted using the LassoCV module from sklearn.linear_model in Python (version 3.9). The optimal regularization parameter α was automatically selected using 5-fold cross-validation by default.

RT-qPCR

Animal Model Construction: Specific pathogen-free (SPF) grade male Sprague-Dawley (SD) rats (220–300 g) were used in this study, with 6 rats per group, including a sham-operated control group. After a one-week acclimatization period, rats were fasted for 12 h prior to surgery and anesthetized via intraperitoneal injection of sodium pentobarbital. The neck region was shaved and disinfected. A midline cervical incision was made to separate the muscles and expose the tracheal cartilage rings (2nd to 4th). A 3-mm longitudinal incision was created in the anterior tracheal wall, and a nylon brush was used to circumferentially scrape the tracheal mucosa 10–15 times. The incision was closed with layered sutures, and postoperative anti-infective care was administered for 3 days.

At 7 days postoperatively, neck computed tomography (CT) was performed to assess tracheal narrowing. The stenosis rate was calculated using the formula: Stenosis rate (%) = (normal tracheal diameter – stenotic diameter)/normal tracheal diameter × 100%. A stenosis rate > 50% was considered indicative of successful modeling.

RNA Extraction and Quantitative PCR: Total RNA was extracted from tracheal tissues using standard protocols. Real-time quantitative PCR (RT-qPCR) was performed to assess the mRNA expression levels of five target genes: CACNA1D, MKI67, CCNB2, CDKL5, and VAMP2. GAPDH was used as the internal reference gene. The relative expression levels of target genes were calculated using the 2^−ΔΔCt method.

The primer sequences used for qRT-PCR were as follows (5CR wer.

CACNA1D, forward: GCATTTAAGCCCAAGGGGTA, reverse: TGAGGGAGTCAAACGTGTTC;

MKI67, forward: CAGGGCTTAGGAAACAGTCC, reverse: GGTTCTAACTGGTCTTCCTGG;

CCNB2, forward: CGGTGTCCAGTGATTTGAAGAA, reverse: GGTATTCTGAGGTTTCTTAGCCA;

CDKL5, forward: ATAGAGTGCCATCTCCACGA, reverse: ATCAGTGTTGGGCACTGTTT;

VAMP2, forward: GCACCTCCTCCAAATCTTACC, reverse: AACATGCTTAGCACACCTCAT.

Results

Sample size and power analysis

Under the current data characteristics, when the sample size reaches 35, the model achieves a statistical power of 95.5% for detecting an AUC ≥ 0.8. When the sample size increases to 40, the power reaches 100%. Therefore, the current sample size (n = 43) is considered adequate to support the applicability of this cohort for high-dimensional data analysis (Table 1).

Table 1.

Sample size and power analysis

Sample size	Efficacy(AUC ≥ 0.8)
20	0.885
25	0.915
30	0.885
35	0.955
40	1.000

Patient clinical baseline characteristics

A total of 43 patients were included in this study, divided into two groups: tracheal stenosis group (n = 18) and non-stenosis group (n = 25). Table 2 presents the clinical variables and molecular marker expression levels analyzed, including age, gender, hypertension, diabetes, hyperlipidemia, cerebrovascular disease, cardiovascular disease, respiratory failure, scar diathesis, and the expression levels of AMPH, MKI67, CCNB2, CDKL5, VAMP2, CACNA1D, PTPRD, CCNB1, CALML4, SH3GLB2, and MTSS1.

Table 2.

Baseline characteristics: clinical variables and tracheal stenosis

Clinical Variables	Tracheal Stenosis (Yes 1, No 0)
		0(n = 25)	1(n = 18)	P-value
Age	< 60 years	7(16.3%)	10(23.3%)	0.068
	≥ 60 years	18(41.9%)	8(18.6%)
Gender	Male	16(37.2%)	8(18.6%)	0.244
	Female	8(18.6%)	10(23.3%)
Hypertension (Yes 1, No 0)	0	14(32.6%)	8(18.6%)	0.455
	1	11(25.6%)	10(23.3%)
Diabetes (Yes 1, No 0)	0	19(44.2%)	13(30.2%)	0.779
	1	6(14.0%)	5(11.6%)
Hyperlipidemia (Yes 1, No 0)	0	24(55.8%)	16(37.2%)	0.367
	1	1(2.3%)	2(4.7%)
Cerebrovascular Disease (Yes 1, No 0)	0	13(30.2%)	11(25.6%)	0.553
	1	12(27.9%)	7(16.3%)
Cardiovascular Disease (Yes 1, No 0)	0	17(39.5%)	16(37.2%)	0.110
	1	8(18.6%)	2(4.7%)
Respiratory Failure (Yes 1, No 0)	0	4(9.3%)	12(27.9%)	< 0.001
	1	21(48.8%)	6(14.0%)
Scar Diathesis (Yes 1, No 0)	0	24(55.8%)	12(27.9%)	0.010
	1	1(2.3%)	6(14.0%)
AMPH (ΔCt)		0.64(0.35 ~ 0.97)	2.37(1.69 ~ 2.91)	< 0.001
MKI67(ΔCt)		0.13(0.05 ~ 0.24)	0.81(0.55 ~ 0.99)	< 0.001
CCNB2(ΔCt)		0.32(0.25 ~ 0.78)	2.49(1.69 ~ 3.23)	< 0.001
CDKL5(ΔCt)		2.60(2.31 ~ 2.99)	0.31(0.22 ~ 0.51)	< 0.001
VAMP2(ΔCt)		2.57(2.22 ~ 3.04)	0.33(0.23 ~ 0.51)	< 0.001
CACNA1D(ΔCt)		0.31(0.26 ~ 0.36)	2.61(2.09 ~ 2.84)	< 0.001
PTPRD(ΔCt)		0.09(0.06 ~ 0.12)	1.06(0.85 ~ 1.37)	< 0.001
CCNB1(ΔCt)		0.27(0.23 ~ 0.33)	2.35(2.11 ~ 2.84)	< 0.001
CALML4(ΔCt)		2.40(2.09 ~ 2.69)	0.34(0.26 ~ 0.42)	< 0.001
SH3GLB2(ΔCt)		2.52(2.20 ~ 2.72)	0.35(0.25 ~ 0.42)	< 0.001
MTSS1(ΔCt)		2.38(2.01 ~ 2.59)	0.36(0.27 ~ 0.40)	< 0.001

Significant differences were observed between the two groups in the prevalence of respiratory failure and scar diathesis (P < 0.05), whereas no significant differences were found for other clinical variables such as age, gender, hypertension, and diabetes (P > 0.05). Additionally, several molecular markers showed significantly different expression levels between groups (P < 0.001) (Table 2).

Box plot analysis (Fig. 1) further illustrated these differences in gene expression. The median expression levels of six molecular markers—CACNA1D, AMPH, PTPRD, MKI67, CCNB1, and CCNB2—were significantly higher in the tracheal stenosis group. In contrast, CDKL5, VAMP2, CALML4, SH3GLB2, and MTSS1 exhibited lower expression in the stenosis group. Notably, markers such as CCNB1, CCNB2, CDKL5, SH3GLB2, VAMP2, and MTSS1 displayed substantial inter-individual variability, with many outliers. All group comparisons yielded statistically significant results (**** indicates P < 0.0001).

Fig. 1.

Box plot comparing gene expression levels between patients with and without stenosis. This figure displays the expression levels of selected molecular targets under two conditions: with stenosis (purple) and without stenosis (pink). The x-axis indicates gene symbols, including CACNA1D, AMPH, PTPRD, MKI67, CCNB1, CCNB2, CALML4, CDKL5, SH3GLB2, VAMP2, and MTSS1. The y-axis represents normalized gene expression levels ranging from 0 to 4.5 Overall, most genes—such as CACNA1D, AMPH, and CCNB2—demonstrate significantly higher expression in the stenosis group compared to the non-stenosis group. For genes like MKI67 and CCNB1, the differences between groups are less pronounced but remain statistically significant. The distribution patterns vary across genes, with some exhibiting tightly clustered expression levels and others displaying broader dispersion

Spearman correlation between tracheal stenosis and Clinical/Biological variables

To investigate the associations between clinical characteristics, molecular markers, and tracheal stenosis, we performed Spearman correlation analysis. The results demonstrated that several clinical variables, including age (r = −0.316, P = 0.039), respiratory failure (r = −0.517, P < 0.001), and scar diathesis (r = 0.392, P = 0.009), were significantly correlated with tracheal stenosis. Specifically, age and respiratory failure showed moderate negative correlations, whereas scar diathesis exhibited a positive correlation. In contrast, other clinical factors such as gender, hypertension, diabetes, and cardiovascular comorbidities showed no significant associations (P > 0.05).

In addition to clinical variables, multiple molecular markers were significantly associated with tracheal stenosis (P < 0.001). Among them, CACNA1D, AMPH, PTPRD, MKI67, CCNB1, and CCNB2 displayed strong positive correlations with the occurrence of stenosis. Conversely, CALML4, CDKL5, VAMP2, MTSS1, and SH3GLB2 exhibited negative correlations. The correlation coefficients ranged from moderate to strong (Table 3).

Table 3.

Spearman correlation analysis between tracheal stenosis and clinical Characteristics/Biomarkers

	Correlation	P-value
Gender	−0.249	0.107
Age	−0.316*	0.039
Hypertension	0.114	0.466
Diabetes	0.043	0.786
Hyperlipidemia	0.138	0.379
Cerebrovascular Disease	−0.091	0.564
Cardiovascular Disease	−0.244	0.115
Respiratory Failure	−0.517**	< 0.001
Scar Diathesis	0.392**	0.009
CACNA1D(ΔCt)	0.646**	< 0.001
AMPH(ΔCt)	0.631**	< 0.001
PTPRD(ΔCt)	0.703**	< 0.001
MKI67(ΔCt)	0.665**	< 0.001
CCNB1(ΔCt)	0.672**	< 0.001
CCNB2(ΔCt)	0.657**	< 0.001
CALML4(ΔCt)	−0.646**	< 0.001
CDKL5(ΔCt)	−0.745**	< 0.001
SH3GLB2(ΔCt)	−0.638**	< 0.001
VAMP2(ΔCt)	−0.741**	< 0.001
MTSS1(ΔCt)	−0.669**	< 0.001

*When the confidence level (double test) is 0.05, the correlation is significant

**When the confidence level (double test) is 0.01, the correlation is significant

The correlation heatmap (Fig. 2) provided a visual overview of the relationships among variables. Strong positive correlations were observed between AMPH and PTPRD (r = 0.89), CCNB1 and CCNB2 (r = 0.89), and MKI67 and CCNB1 (r = 0.87). Similarly, CACNA1D and CCNB1 were highly correlated (r = 0.81).

Fig. 2.

Correlation analysis heatmap of clinical and genetic features. This heatmap visualizes the correlation coefficients among clinical parameters and genetic features, titled “Correlation Analysis of Clinical and Genetic Features.” Rows and columns represent individual variables. The color scale denotes the strength and direction of Pearson correlation coefficients: Red indicates positive correlations, with darker shades representing stronger associations Blue indicates negative correlations, with darker shades representing stronger inverse relationships A color bar on the right maps the gradient from − 1.0 (strong negative correlation, dark blue) to + 1.0 (strong positive correlation, dark red) Several notable correlations are observed: CACNA1D shows strong positive correlations with multiple genetic variables, including AMPH, PTPRD, and MKI67 CALML4 demonstrates significant negative correlations with several other genetic markers These correlations suggest possible co-regulatory relationships or shared pathways between clinical and molecular variables

On the other hand, CDKL5 (r = −0.74), CALML4 (r = −0.65), VAMP2 (r = −0.74), and MTSS1 (r = −0.67) showed strong negative correlations with tracheal stenosis, implying a possible inhibitory role. Notably, age was also negatively associated with CALML4 expression (r ≈ −0.65).

Multivariate logistic regression analysis

To assess the independent effects of clinical characteristics and molecular markers on tracheal stenosis, multivariate logistic regression was conducted. The results revealed that AMPH (ΔCt value) showed a positive association with tracheal stenosis (B = 0.047, OR = 1.048, 95% CI: 0.103–10.644), though the association was not statistically significant (P = 0.969), indicating a limited independent effect. Similarly, age (B = 0.59, OR = 1.804, P = 0.702) and gender (B = −0.771, OR = 0.463, P = 0.615) were not significantly correlated with tracheal stenosis. Cardiovascular disease also showed a negative but non-significant association (B = −2.38, OR = 0.093, P = 0.343).

Importantly, CDKL5 (ΔCt value) emerged as a significant protective factor (B = −2.848, OR = 0.058, 95% CI: 0.005–0.649, P = 0.021), suggesting that higher expression of CDKL5 may reduce the risk of tracheal stenosis (Table 4).

Table 4.

Multivariate logistic regression analysis of tracheal stenosis with clinical characteristics and biomarkers

	B	S.E,	Wals	P	OR	95% OR
						Lower	Upper
AMPH(ΔCt)	0.047	1.183	0.002	0.969	1.048	0.103	10.644
Age	0.59	1.543	0.146	0.702	1.804	0.088	37.11
Gender	−0.771	1.533	0.253	0.615	0.463	0.023	9.336
Cardiovascular Disease	−2.38	2.51	0.899	0.343	0.093	0.001	12.678
CDKL5(ΔCt)	−2.848	1.232	5.342	0.021	0.058	0.005	0.649

LASSO regression for risk factor selection

To identify key molecular features contributing to tracheal stenosis, a LASSO regression model was applied. The results indicated that CCNB2, PTPRD, and CACNA1D were positively associated with tracheal stenosis risk, with coefficients of 0.034, 0.037, and 0.057, respectively. In contrast, VAMP2 and CDKL5 exhibited negative coefficients (−0.126 and − 0.141).

Among these variables, CDKL5 and VAMP2 demonstrated relatively stronger effects due to their higher absolute coefficient values, while the other three genes contributed more modestly. To enhance the robustness and generalizability of the model, 5-fold cross-validation was employed. The model yielded strong predictive performance with an average accuracy of 0.93, AUC of 0.87, sensitivity of 0.83, and specificity of 0.98, underscoring the reliability of the selected biomarkers in risk prediction (Fig. 3).

Fig. 3.

Bar chart of risk factors identified by LASSO regression analysis. This figure presents a bar chart of variables selected by the least absolute shrinkage and selection operator (LASSO) regression model, displaying features with non-zero standardized coefficients. The y-axis lists the selected variables: CCNB2, PTPRD, CACNA1D, VAMP2, and CDKL5, while the x-axis represents their standardized coefficients, ranging from − 0.150 to 0.050 Positive coefficients (bars extending to the right) indicate a positive association with the outcome variable. For example, CCNB2 (0.034), PTPRD (0.037), and CACNA1D (0.057) are positively correlated Negative coefficients (bars extending to the left) reflect a negative association. Notably, VAMP2 (–0.126) and CDKL5 (–0.141) are inversely associated with the target variable These features were retained by the LASSO model for their predictive relevance in distinguishing outcome status

ROC curve validation

The ROC curve analysis demonstrated the excellent discriminative performance of the random forest classifier in binary classification. As shown in Fig. 4, the ROC curve (orange line) is clearly positioned above the diagonal reference line (blue dashed line, representing random guessing), indicating that the model significantly outperforms chance level.

Fig. 4.

Receiver operating characteristic (ROC) curve of the Random Forest classifier. This plot illustrates the ROC curve of the Random Forest model used to evaluate classification performance. The x-axis represents the false positive rate (FPR), defined as the proportion of actual negatives incorrectly classified as positives. The y-axis represents the true positive rate (TPR), or sensitivity, defined as the proportion of actual positives correctly identified The orange curve represents the ROC curve of the Random Forest classifier, depicting the trade-off between sensitivity and specificity across different classification thresholds The blue dashed diagonal line from (0, 0) to (1, 1) represents the line of no-discrimination (i.e., random chance). A model with a curve approaching the top-left corner demonstrates better discriminative ability The area under the ROC curve (AUC) can be used to quantify overall model performance, with values closer to 1.0 indicating superior classification accuracy

The curve rises steeply toward the upper-left corner, reflecting a high true positive rate and a relatively low false positive rate across varying threshold settings. The area under the curve (AUC) reached 0.91, suggesting strong overall predictive accuracy and reliability in distinguishing between tracheal stenosis and non-stenosis cases. (Fig. 4).

Bayesian logistic regression analysis

Table 5 presents the convergence diagnostics and parameter estimates from the Bayesian logistic regression model. The standardized expression of CACNA1D showed a significant effect on tracheal stenosis (mean = 1.693; 97% HDI: [0.317, 2.914]). Other parameters yielded non-significant estimates.

Table 5.

Bayesian logistic regression convergence diagnostics

	mean	sd	hdi_ 3%	hdi_ 97%	mcse_mean	mcse_sd	ess_ bulk	ess_ tail	r_ hat
alpha	0.211	0.736	−1.103	1.645	0.015	0.011	2439	2593	1
Standardizd age	−0.007	0.475	−0.902	0.881	0.009	0.008	2663	2438	1
Respiratory failure	−0.585	0.754	−1.981	0.839	0.015	0.011	2482	2295	1
Standardizd MKI67 gene expression	0.913	0.713	−0.439	2.220	0.015	0.011	2135	2430	1
StandardizdCACNA1D gene expression	1.693	0.691	0.317	2.914	0.015	0.011	2129	2206	1

Convergence diagnostics indicated robust model performance. All parameters had effective sample sizes (ess_bulk and ess_tail) above 2000, and all (r_hat) values equaled 1.0, confirming satisfactory convergence of the MCMC chains and reliability of the posterior estimates.

Figure 5 illustrates the posterior distributions and MCMC trace plots for the intercept (alpha) and regression coefficients (betas). All parameters exhibit stable posterior distributions and well-mixed chains.

Fig. 5.

Bayesian logistic regression: posterior distributions and MCMC sampling traces. This figure presents the posterior probability distributions and Markov Chain Monte Carlo (MCMC) sampling trajectories of parameters from the Bayesian logistic regression model Top-left panel: Posterior distribution of the intercept parameter (α), showing a unimodal, bell-shaped curve centered near zero Top-right panel: Trace plot of α across MCMC iterations, indicating stable sampling around a central value without apparent drift Bottom-left panel: Posterior distributions of selected regression coefficients (βs), shown as distinct colored curves. Variations in peak position and spread reflect differences in parameter uncertainty Bottom-right panel: Trace plots of βs across iterations, with each line representing a different coefficient. All chains exhibit convergence and mixing, supporting the reliability of the sampling process All x-axes labeled as “betas” refer either to parameter values or MCMC iterations, depending on panel context. Overall, the figure demonstrates robust posterior estimation and well-behaved sampling chains

Figure 6 displays the 94% high-density intervals (HDIs) for the odds ratios of the four standardized predictors: betas[0] (Age_z): The odds ratio interval includes 1, indicating a non-significant effect of age on tracheal stenosis. betas [1] (Respiratory Failure): Also non-significant, with the odds ratio interval overlapping 1. betas [2] (MKI67_z): The odds ratio interval is entirely greater than 1, indicating a statistically significant positive association. Increased MKI67 expression correlates with higher risk of stenosis. betas [3] (CACNA1D_z): Similarly, its odds ratio interval lies entirely above 1, confirming that higher CACNA1D expression significantly increases the likelihood of tracheal stenosis. These findings suggest that MKI67 and CACNA1D are key risk factors, while age and respiratory failure alone do not significantly influence stenosis risk in this model.

Fig. 6.

Odds ratio plot of Bayesian logistic regression parameters associated with stenosis. This figure illustrates the estimated odds ratios and their 94% highest density intervals (HDIs) derived from a Bayesian logistic regression model. The x-axis displays the odds ratio values, calculated as the exponential transformation of the model coefficients (exp(β)), ranging from 0 to 15. The y-axis lists the regression coefficients, labeled as β[0] to β [3], each corresponding to a specific independent variable Each point on the plot represents the posterior mean of exp(β), while the horizontal line denotes the 94% HDI, indicating the most credible range of values with 94% posterior probability. Odds ratios greater than 1 suggest a positive association with stenosis, while those below 1 indicate a negative association

Figure 7 presents subgroup analyses for MKI67 under different respiratory failure statuses. In patients without respiratory failure, the effect of MKI67_z on stenosis was not statistically significant. However, in patients with respiratory failure, MKI67_z showed a significant positive effect on stenosis risk.

Fig. 7.

Subgroup analysis of MKI67_z: comparison of regression coefficients by respiratory failure status. This figure presents the estimated regression coefficients (β₀ and β₁) for the variable MKI67_z across two subgroups stratified by respiratory failure status. The x-axis displays coefficient values ranging from 0 to 15, and the y-axis lists the corresponding parameter labels Two colored lines indicate subgroup estimates: Orange line: subgroup with respiratory failure (respiratory failure = 1), Blue line: subgroup without respiratory failure (respiratory failure = 0) Each point represents the estimated coefficient, and the horizontal line depicts the 95% confidence interval, with endpoints indicating the lower and upper bounds. This comparison highlights the variation in MKI67_z’s effect between clinical subgroups

RT-qPCR detection results

Compared with the control group, the relative mRNA expression levels of CACNA1D, MKI67, and CCNB2 in tracheal tissues of rats in the tracheal stenosis group were significantly upregulated, being 2.5 ± 0.03 fold, 4.1 ± 0.07 fold, and 3.5 ± 0.07 fold that of the control group, respectively (all P < 0.0001). The relative mRNA expression levels of CDKL5 and VAMP2 were significantly downregulated, being 0.2 ± 0.06 fold and 0.3 ± 0.04 fold that of the control group, respectively (all P < 0.0001) (Fig. 8).

Fig. 8.

RT-PCR detection of relative mRNA expression levels of CACNA1D, MKI67, CCNB2, CDKL5, and VAMP2 in the control and tracheal stenosis model groups (* indicates statistically significant differences)

Discussion

This study systematically investigated the molecular markers and predictive risk factors of benign tracheal stenosis through a multi-level analytical approach. A total of 43 patient samples were analyzed using Spearman correlation, LASSO regression, ROC curve validation, and Bayesian inference to identify clinically relevant molecular markers and evaluate their predictive performance. The results revealed that respiratory failure and scar diathesis were significantly associated with tracheal stenosis, while several biomarkers—particularly CACNA1D, AMPH, and PTPRD—exhibited abnormal expression patterns. These markers are potentially involved in disease pathogenesis through mechanisms such as enhanced cell proliferation and ion channel dysfunction.

Regarding clinical characteristics, the study confirmed that respiratory failure and scar diathesis are major risk factors. Patients with respiratory failure had a significantly increased risk of tracheal stenosis (P < 0.001), likely due to chronic hypoxia, airway inflammation, and oxidative stress that promote smooth muscle cell proliferation and extracellular matrix remodeling in the airway [10–14]. Moreover, these patients often require prolonged artificial airway support, further exacerbating mucosal injury and scarring. Similarly, scar diathesis was significantly associated with stenosis risk (P = 0.009), possibly due to aberrant tissue repair and fibrotic responses following mucosal injury [15–17]. In contrast, demographic and systemic factors such as age, gender, hypertension, and diabetes were not significantly correlated with stenosis occurrence (P >0.05), suggesting that local pathological processes following airway trauma—rather than systemic comorbidities—play a more dominant role in the development of benign tracheal stenosis.

This study revealed that multiple molecular markers are significantly associated with the presence of tracheal stenosis, based on comparative expression analyses between stenosis and non-stenosis groups. Positively correlated biomarkers—including CACNA1D, AMPH, PTPRD, MKI67, CCNB1, and CCNB2—were expressed at significantly higher levels in patients with tracheal stenosis (P < 0.001). These molecules are likely involved in promoting the onset and progression of tracheal stenosis.

For instance, CACNA1D encodes a subunit of the L-type voltage-dependent calcium channel, and its overexpression may enhance calcium influx, activate downstream signaling cascades, and stimulate the proliferation and contraction of airway smooth muscle cells [18, 19]. Similarly, overexpression of AMPH and PTPRD may contribute to the pathogenesis by modulating endocytosis, cell adhesion, and signal transduction, thereby affecting cellular communication and tissue remodeling.

Conversely, negatively correlated markers—such as CALML4, CDKL5, SH3GLB2, VAMP2, and MTSS1—showed significantly lower expression levels in the tracheal stenosis group (P < 0.001), suggesting a potential protective role against pathological airway remodeling. CALML4, which encodes a calmodulin-like protein, may be involved in maintaining intracellular calcium homeostasis; its downregulation could disrupt calcium-dependent cellular functions and impair airway integrity [20–23]. CDKL5, known for its role in cell cycle inhibition, may suppress abnormal cell proliferation; reduced expression could diminish this control and increase fibrotic risk [23–25]. Similarly, VAMP2 and MTSS1 are associated with vesicle trafficking and cytoskeletal regulation; their downregulation may compromise membrane stability and cell-cell junctions, thereby facilitating tissue remodeling and stenosis formation.

This study established a risk prediction model for tracheal stenosis based on LASSO regression analysis, identifying CCNB2, PTPRD, and CACNA1D as positively associated with stenosis risk, and VAMP2 and CDKL5 as negatively associated. The expression patterns of these markers were strongly correlated with the likelihood of developing tracheal stenosis. ROC curve analysis confirmed the high predictive performance of this model, with an AUC of 0.91, indicating excellent accuracy and discriminative power.

To further investigate underlying mechanisms, we conducted a preliminary literature-based analysis focusing on ion channel- and cell cycle-related genes. Overexpression of ion channel-related genes such as CACNA1D may enhance calcium influx and activate downstream signaling pathways, promoting airway smooth muscle proliferation and contraction. Similarly, dysregulation of cell cycle-related genes, including MKI67, CCNB1, and CCNB2, may contribute to pathological airway remodeling through enhanced cellular proliferation. In contrast, negatively correlated markers such as CALML4 and CDKL5 may exert protective effects by preserving calcium homeostasis or inhibiting aberrant cell growth.

Subgroup analysis further demonstrated that respiratory failure modulates the effect of molecular markers on stenosis risk. Specifically, elevated MKI67 expression was significantly associated with stenosis in patients with respiratory failure, but not in those without. This suggests that respiratory failure may amplify the pathogenic impact of MKI67 by exacerbating inflammatory responses or oxidative stress in the airway microenvironment.

Limitations

Despite the meaningful results obtained in this study, several limitations should be acknowledged. First, the sample size is relatively small, which may limit the statistical power and generalizability of the findings. Although we expanded the dataset to 43 patients and conducted power analysis to support model adequacy, future studies with larger, multicenter cohorts are needed for further validation. Second, while we performed RT-qPCR validation in an animal model to confirm key gene expression trends, this study lacks deeper functional experiments to elucidate the molecular mechanisms underlying tracheal stenosis. Further in vitro and in vivo studies are warranted. Third, although the prediction model demonstrated high performance (AUC = 0.91), the absence of an external validation cohort and reliance on internal cross-validation may carry a risk of overfitting. Future studies should incorporate independent validation datasets to improve model robustness and clinical applicability.

Conclusion

This study identified key molecular markers associated with tracheal stenosis, including upregulated CACNA1D, MKI67, and CCNB1/2, which may promote airway remodeling via calcium signaling and cell proliferation. In contrast, CALML4 and CDKL5 showed negative associations, suggesting protective roles. These findings provide a preliminary molecular framework for risk prediction and pathogenesis of tracheal stenosis. Further experimental and clinical validation is warranted to explore their therapeutic potential.

Authors’ contributions

Shuxian Ma and Haitao Li have given substantial contributions to the conception or the design of the manuscript. Pei Zhang and Yanjie Yang to acquisition, Shuangna He, Xuze Li, Zhangnan Sun, Zhigang Cai and Lining Huang analysis and interpretation of the data. All authors have participated to drafting the manuscript, Shuxian Ma revised it critically. All authors read and approved the final version of the manuscript. All authors contributed equally to the manuscript and read and approved the final version of the manuscript.

Funding

Medical Science Research Program of Hebei Health Commission: Screening, identification and analysis of key genes for tracheal stenosis and scar constitution after artificial airway (20230051).

Data availability

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the ethics committee of The Second Hospital of Hebei Medical University (2022-R485). The consent that was obtained from all of the participants was informed. The research was conducted in accordance with the Declaration of Helsinki.

This study has been approved by the Ethics Committee of the Second Hospital of Hebei Medical University (2022-R485). Each participant has given their consent.

Consent to participate

Every human participant provided their consent.

Conflict of interest

The authors certify that there is no conflict of interest with any financial organization regarding the material discussed in the manuscript.