Skip to main content
NCBI home page
ACS Omega logoLink to ACS Omega
. 2026 Sep 28;11(40):61052–61064. doi: 10.1021/acsomega.6c04401

Integration of Network Toxicology and Machine Learning to Prioritize Candidate Molecular Features Associated with NNK-Related Lung Cancer

Ying Li †, Hongfeng Wu ‡,§, Ping Liu ∥, Jiaqi Ji ⊥, Quan Zhang ‡,§, Yongqiang Zhang #,*, Yongwen Li ‡,§,*
PMCID: PMC13670790  PMID: 42890033

Abstract

4-(Methylnitrosamino)-1-(3-pyridyl)-1-butanone (NNK) is a tobacco-specific nitrosamine implicated in tobacco-related lung carcinogenesis. Its biological effects in vivo depend substantially on metabolic conversion to NNAL and cytochrome P450-mediated formation of reactive intermediates. In this study, predicted NNK-associated proteins were integrated with lung cancer transcriptomic data sets, followed by network analysis and machine-learning-based feature prioritization. Five genes, SNAI2, PLAU, DNMT3A, FAP, and CCNB1, were prioritized based on their contributions to the multigene classification framework. Exploratory molecular docking generated computational ligand-placement models with modest docking scores; these results were used only as structural visualizations and do not demonstrate direct biochemical binding. In the metabolically limited A549 cell model, micromolar exposure to parent NNK produced higher CCK-8 signals, altered cell-cycle distributions, increased Transwell migration and invasion counts, and changes in candidate-gene mRNA expression. These parallel observations do not establish that the candidate genes mediate the cellular phenotypes. Moreover, because intact NNK is rapidly metabolized in vivo and is not an established biomarker of systemic exposure in human blood, urine, or saliva, the parent-compound exposure used here does not represent an established human internal-exposure scenario. This study provides a hypothesis-generating framework for prioritizing candidate molecular features for future investigation rather than direct evidence of NNK target engagement or causal gene function.

graphic file with name ao6c04401_0010.webp

graphic file with name ao6c04401_0008.webp

1. Introduction

Lung cancer is a leading cause of cancer incidence and mortality worldwide, posing a significant public health challenge. , Tobacco smoking is the primary etiological factor, with robust correlations to histological subtypes such as squamous cell carcinoma and small cell lung cancer. , Ambient air pollution exacerbates pulmonary oncogenic risk. Epidemiological evidence indicates that lung cancer risk declines after smoking cessation, although excess risk may persist for years. Moreover, tobacco smoking has been associated with increased all-cause mortality, cancer recurrence, and second primary cancers. Tobacco smoke is a complex mixture comprising over 7000 chemicals, including over 90 recognized carcinogens; 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone (NNK) is one of the most extensively studied carcinogenic nitrosamines.

NNK is a potent carcinogen classified as a tobacco-specific nitrosamine, prevalent in tobacco smoke. Its primary oncogenic role is promoting lung carcinogenesis. Additionally, NNK and related tobacco-specific nitrosamines have also been implicated in oral, esophageal, and pancreatic malignancies. − Notably, NNK is not an endogenous phytochemical of tobacco; rather, it is synthesized during fermentation and curing and produced during combustion through chemical reactions between precursors such as nicotine and nitrites. Extensive human and animal model research led the International Agency for Research on Cancer to classify NNK as a Group 1 carcinogen, confirming its carcinogenicity in humans.

The oncogenic potential of NNK has been extensively characterized through mechanistic and biomarker research. NNK is a prominent etiological agent in tobacco-associated lung carcinogenesis, with multiple elucidated metabolic activation pathways. In hepatic and pulmonary tissues, NNK undergoes carbonyl reduction to NNAL and cytochrome P450-mediated α-hydroxylation to unstable intermediates that generate DNA-reactive electrophiles, whereas NNAL can undergo further metabolic activation through related pathways. − These metabolites can form DNA adducts that, if unrepaired, contribute to mutational events and genomic instability. Beyond direct genotoxicity, NNK has also been reported to influence signaling and inflammatory pathways associated with lung tumorigenesis. ,, Despite these insights, the integrated molecular framework governing these events remains incomplete, emphasizing the need for investigations to identify therapeutic targets in lung adenocarcinoma.

To generate testable molecular hypotheses, we integrated predicted NNK-associated proteins with lung cancer transcriptomic data sets and applied network analysis and machine-learning-based feature prioritization. Exploratory molecular docking was included only to visualize possible computational ligand placement, not to establish biochemical binding. Cellular assays in A549 cells were used to record phenotypic and transcriptional responses to parent NNK under controlled in vitro conditions. The purpose of this workflow was to prioritize candidate genes for future validation rather than to identify direct NNK targets or establish causal mechanisms of NNK-induced lung carcinogenesis.

2. Materials and Methods

2.1. Compilation of Predicted NNK-Associated Proteins

The chemical structure and canonical SMILES of NNK were retrieved from PubChem (https://pubchem.ncbi.nlm.nih.gov/). The predicted toxicity profile of NNK was evaluated using ProTox 3.0. Potential human protein targets of NNK were predicted using three complementary resources: SEA (https://sea.bkslab.org/), CTD (https://ctdbase.org/), and SwissTargetPrediction (http://www.swisstargetprediction.ch/). Only human proteins were retained. Proteins obtained from the three resources were merged and duplicate entries were removed to generate a computationally predicted NNK-associated protein list. This list was used for hypothesis generation and should not be interpreted as a set of experimentally validated NNK-binding targets.

2.2. Identification of Lung Cancer-Related Data Sets

Five lung cancer transcriptomic data sets were obtained from the GEO database. GSE10072 and GSE75037 were merged as the training cohort, whereas GSE116959, GSE18842, and GSE27262 were used as independent validation cohorts. A sequential normalization protocol was applied to minimize batch-related variation. Initially, surrogate variable analysis was employed to model and adjust for potential confounders within the discovery cohort. Subsequently, residual batch effects were corrected utilizing ComBat within an empirical Bayes framework. Postcorrection principal component analysis (PCA) demonstrated enhanced data set integration and reduced sample heterogeneity.

2.3. Differential Expression Analysis

Differentially expressed genes (DEGs) were identified using the “limma” R package. Candidates with a false discovery rate <0.05 and an absolute log 2 fold change >0.585 (approximately a 1.5-fold change) were considered significant DEGs.

2.4. Weighted Gene Coexpression Network Analysis

The merged data set of GSE10072 and GSE75037 served as the target matrix for constructing a weighted gene coexpression network using the weighted gene coexpression network analysis (WGCNA) package. The analytical workflow included sample quality assessment, soft-thresholding power determination to attain scale-free topology (R 2 > 0.85), outlier detection through hierarchical clustering, module detection through topological overlap matrix (TOM) hierarchical clustering (minimum module size = 30; merge cut height = 0.25), and module–trait relationship analysis involving correlation coefficients (|R| > 0.5, P < 0.05). Additionally, hub genes within modules were identified based on high module membership values (kME >0.8).

2.5. Functional Enrichment

Gene Ontology (GO) and KEGG pathway enrichment analyses were performed using the clusterProfiler package; terms with a significance threshold of P < 0.05 were considered enriched. Visualizations were produced using the R packages “ggplot2,” “circlize,” and “ComplexHeatmap.”

2.6. Machine Learning-Based Selection of Core Genes

To prioritize candidate genes within the integrative NNK–lung cancer framework, a multialgorithm machine-learning strategy was applied to the 16 overlapping genes. The training cohort consisted of GSE10072 and GSE75037, whereas GSE116959, GSE18842, and GSE27262 were used as validation cohorts. The evaluated algorithms included Lasso, support vector machine, random forest, Generalized Linear Model Boosting, Elastic Net, Ridge, Steglm, XGBoost, Linear Discriminant Analysis, Partial Least Squares Regression on Generalized Linear Models, Naive Bayes, and Generalized Boosted Regression Modeling. Hyperparameters were optimized using 5-fold cross-validation in the training cohort. Model performance was assessed using the area under the receiver operating characteristic curve (AUC), accuracy, and F1 score. Candidate genes were prioritized according to their contribution to high-performing models and their recurrence across feature-selection procedures. Gene expression patterns were visualized using the pheatmap package.

2.7. SHAP Model Analysis

To assess feature importance within the predictive model, the Shapley additive explanation (SHAP) method was employed to quantify the contribution of each individual feature to the model output. SHAP values provided insight into the directional influence and magnitude of each gene’s contribution to prediction, facilitating interpretability of feature effects.

2.8. Exploratory Molecular Docking

Exploratory molecular docking was performed to visualize the computational placement of NNK within predicted structural regions of proteins encoded by the prioritized genes. The SDF structure of NNK was obtained from PubChem. Three-dimensional protein structures corresponding to SNAI2, PLAU, DNMT3A, FAP, and CCNB1 were obtained from UniProt or related structural resources. Protein structures were prepared by removing water molecules and adding hydrogen atoms, and docking was conducted using CB-Dock2. Because NNK is a small procarcinogen whose biological activity depends substantially on metabolic activation, and because no biochemical binding assay was performed, docking scores and poses were treated only as exploratory structural outputs. They were not used to infer direct binding, covalent interaction, or cellular target engagement.

2.9. Cell Culture

Human lung adenocarcinoma A549 cells were obtained from the Cell Bank of the Chinese Academy of Sciences. Cells were authenticated through short tandem repeat genotyping and confirmed free of mycoplasma contamination prior to experiments. Cells were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum (FBS; Gibco, USA) and 1% penicillin–streptomycin (Gibco, USA) at 37 °C in a humidified incubator with 5% CO2. Cells were subcultured at a 1:3 ratio, with exponentially growing cells selected for experiments.

2.10. NNK Treatment

NNK was purchased from Taoshu Biotechnology (Shanghai, China). A concentrated stock solution was prepared according to the manufacturer’s instructions and stored at −20 °C protected from light. Before treatment, NNK was diluted to the indicated working concentrations. Vehicle-control cells received the same final concentration of solvent, which did not exceed 0.1%. When A549 cells reached 60–70% confluence, they were treated with 0, 10, or 25 μM NNK for the indicated durations.

Toxicokinetic evidence indicates that NNK is rapidly absorbed and extensively converted to NNAL in vivo. In a direct human urinary LC–MS/MS study, intact urinary NNK was not detected in smokers, whereas free and total NNAL were measurable. Thus, NNAL, rather than intact NNK, is the more established biomarker of systemic NNK exposure. − Because A549 cells show very low CYP2A6 and CYP2A13 activity, direct treatment with 10 or 25 μM parent NNK was used only as a controlled in vitro perturbation and should not be interpreted as a physiologically representative human internal-exposure scenario.

2.11. RNA Extraction and qRT-PCR

Following NNK exposure, total RNA was extracted using RNAiso Plus reagent (TAKARA, Dalian, China). RNA concentration and purity were assessed using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, MA, USA). Complementary DNA was synthesized using a reverse transcription kit according to the manufacturer’s instructions. qRT-PCR was performed using ChamQ SYBR Green qPCR Master Mix (Vazyme, Nanjing, China) to quantify the mRNA expression levels of FAP, CCNB1, SNAI2 (Slug), DNMT3A, and PLAU, with GAPDH used as the internal reference. Amplification specificity was confirmed by melt curve analysis. Relative mRNA expression was calculated using the 2–ΔΔCt method, with the control group as the reference. All reactions were performed in triplicate. Primer sequences are provided in Table S1.

2.12. Cell Proliferation Assay (CCK-8)

A549 cells in exponential growth phase were seeded in 96-well plates at a density of 3–5 × 103 cells per well, with 3–5 replicates per group. After cellular attachment, NNK was administered in varying concentrations, while control wells received vehicle alone. At designated time points (0, 24, 48, and 72 h), 10 μL of CCK-8 solution (APExBIO, TX, USA) was added to each well and incubated at 37 °C for 2 h. Absorbance at 450 nm was assessed using a Synergy H1 microplate reader (BioTek Instruments, VT, USA) to determine cell viability and proliferation responses.

2.13. Cell-Cycle Analysis

Harvested cells were washed twice with phosphate-buffered saline and resuspended at approximately 1 × 106 cells/mL to examine the effect of NNK on cell-cycle distribution. Cells were fixed in cold 70% ethanol at 4 °C overnight. After washing, cells were incubated with RNase A at 37 °C for 30 min, followed by staining with propidium iodide for 30 min in the dark at room temperature. DNA content was analyzed using the NovoCyte 2000R flow cytometry (Agilent, Singapore), and the proportions of cells in G0/G1, S, and G2/M phases were determined using standard flow cytometric analysis software.

2.14. Migration and Invasion Assays

Transwell migration and invasion assays evaluated motility and invasive capacity. Uncoated inserts (8 μm pore size; Corning) were employed for migration assays. For invasion, inserts were precoated with Matrigel and incubated to facilitate polymerization at 37 °C. After NNK treatment, cells were resuspended in serum-free medium, and 1–2 × 105 cells were seeded into the upper chamber. The lower chamber contained medium supplemented with 10% FBS as a chemoattractant. Following 24 h of incubation at 37 °C in a 5% CO2 atmosphere, nonmigrated/noninvaded cells on the upper membrane were removed. Cells adhering to the lower surface were fixed with 4% paraformaldehyde and stained with 0.1% crystal violet. Cells were counted in multiple randomly selected high-power fields, and the average cell number per insert was calculated for statistical analysis.

2.15. Statistical Analysis

All experiments were replicated at least three times. Data are presented as mean ± SD from at least three independent experiments. Comparisons between two groups were performed using an unpaired two-tailed Student’s t-test. For experiments involving multiple NNK concentrations, one-way ANOVA followed by Dunnett’s multiple-comparison test was used to compare each NNK-treated group with the vehicle control. For time-course experiments, two-way ANOVA followed by Dunnett’s multiple-comparison test was used to compare each NNK concentration with the vehicle control at the corresponding time point. A two-sided P value < 0.05 was considered statistically significant (Figure ).

1.

1

Open in a new tab

Overview of the data set analysis workflow.

3. Results

3.1. Toxicity Prediction

The potential toxicity profile and molecular targets of NNK were first assessed based on its chemical structure (Figure A). Using the ProTox 3.0 database, the toxicity profile of NNK was predicted, and its significant potential for organ-specific toxicities was identified. Database- and in silico model-based evaluations indicated a high probability of positive signals across multiple toxicity domains, including immunotoxicity, mutagenicity, cytotoxicity, cardiovascular toxicity, and receptor-related signaling pathways, showing a compound with broad, multifaceted risk (Figure B).

2.

2

Open in a new tab

Identification of putative NNK-associated targets. (A) Chemical structure of NNK. (B) Predicted toxicity profile of NNK based on ProTox 3.0, showing predicted toxicity signals across multiple biological domains, including immunotoxicity, mutagenicity, cytotoxicity, cardiovascular toxicity, and receptor-related end points. (C) Workflow for target prediction using SEA, CTD, and SwissTargetPrediction, followed by integration of the three data sets to generate the final set of predicted NNK-associated targets.

3.2. Compilation of Predicted NNK-Associated Proteins

The potential protein targets of NNK were identified through systematic prediction using SwissTargetPrediction, SEA, and CTD. After data integration and deduplication, 443 candidate targets were retained (Figure C). After integration and removal of duplicates, 443 proteins were retained as a computationally predicted NNK-associated set. These predictions provided an initial hypothesis pool for subsequent overlap analysis and do not demonstrate direct effects of NNK on these proteins.

3.3. Identification of Lung Cancer-Related Genes

PCA was conducted on GSE10072 and GSE75037 to evaluate the global structure of the gene expression data. Prior to batch effect correction, samples clustered strongly by data set, revealing significant technical variability (Figure A). Following data set merging and comprehensive normalization, PCA demonstrated enhanced data congruity and more discernible biological cluster formations (Figure B). Differential expression analysis identified 529 significantly upregulated and 132 downregulated genes in lung cancer, visualized in a volcano plot (Figure C).

3.

3

Open in a new tab

Identification of lung cancer-related genes. (A) PCA plot showing clear separation between GSE10072 and GSE75037 before batch correction, indicating batch effects. (B) PCA plot after batch correction showing improved integration of GSE10072 and GSE75037, demonstrating decreased batch effects. (C) Volcano plot of differentially expressed genes (DEGs) based on logFC and statistical significance. Red dots indicate upregulated genes, green dots downregulated genes, and gray dots nonsignificant genes. (D) WGCNA dendrogram showing hierarchical clustering of genes based on coexpression. Module colors below the dendrogram represent distinct gene modules. (E) Module–trait correlation heatmap showing associations between WGCNA modules and sample traits (control vs tumor). The values in each cell indicate the correlation coefficient and P-value. (F) Scatter plot showing a strong positive correlation between module membership and gene significance. (G) Venn diagram illustrating overlap between DEGs (red) and WGCNA module genes (blue); overlapping genes are shown in purple.

For WGCNA, a soft-thresholding power of β = 5 was selected to satisfy the scale-free network criterion (R 2 ≥ 0.8). Using this parameter, a TOM was constructed, resulting in the delineation of 10 coexpression modules (Figure D). Module–trait correlation analysis revealed associations between specific modules and lung cancer (P < 0.05) (Figure E). The high correlation between module eigengenes and gene significance (correlation = 0.93; P = 1.3 × 10–90) indicated that genes within the green module are relevant to disease-associated transcriptional alterations (Figure F). Integrating DEGs with WGCNA module genes and removing overlaps led to the identification of 735 lung cancer-related genes (Figure G).

3.4. Overlap between Predicted NNK-Associated Proteins and Lung Cancer-Related Genes

Intersection of the computationally predicted NNK-associated protein list with lung cancer-related genes identified 16 overlapping genes (Figure A). These genes were then submitted to the STRING database to construct a protein–protein interaction network, which was visualized using Cytoscape 3.9.0 (Figure B). GO and KEGG enrichment analyses were performed to characterize the biological processes and pathways associated with these overlapping genes (Figure C,D).

4.

4

Open in a new tab

Overlap between predicted NNK-associated proteins and lung cancer-related genes. (A) Venn diagram comparing predicted NNK-associated genes exposure (red) and lung cancer (blue), emphasizing 16 overlapping genes (1.4%). (B) Overlapping genes shared by NNK exposure- and lung cancer-related gene sets. (C) GO enrichment of the overlapping genes across biological process, cellular component, and molecular function. The x-axis indicates gene counts, and the color gradient represents adjusted P-values (darker purple indicates higher significance). (D) KEGG pathway enrichment analysis of the overlapping genes.

GO enrichment analysis showed associations with epithelial cell migration, fibrinolysis regulation, extracellular matrix degradation, chromosome-related cellular components, and protease- and DNA-related molecular functions. KEGG pathway analysis highlighted enrichment in p53 signaling, transcriptional misregulation in cancer, prostate cancer, microRNAs in cancer, proteoglycans in cancer, estrogen signaling, glycerolipid metabolism, oocyte meiosis, nonhomologous end-joining, cellular senescence, cell cycle, folate biosynthesis, and galactose metabolism. These results suggest that the overlapping genes are related to oncogenic signaling, cell-cycle and DNA-damage-associated processes, and metabolic regulation in lung cancer.

3.5. Machine-Learning and SHAP Analyses Prioritize Candidate Genes within the Analytical Framework

To further prioritize candidate genes, 32 classification models were developed based on the 16 overlapping genes. Among the tested models, the Ridge model showed the strongest overall performance across the analyzed cohorts, with the highest AUC value of 0.986 (Figure A). Based on model performance and feature contribution, five genes, SNAI2, PLAU, DNMT3A, FAP, and CCNB1, were selected as candidate core genes. Individual-gene ROC analysis yielded AUC values ranging from 0.657 to 0.851 (Figure B), indicating variable discriminatory performance among the five prioritized genes (Figure B). Their differential expression patterns in lung cancer tissues are shown in the volcano plot (Figure C).

5.

5

Open in a new tab

Identification of candidate core genes in the integrative NNK–lung cancer analysis. (A) Model performance comparison based on AUC values across the training and validation cohorts. (B) ROC curves of the five candidate genes: SNAI2, PLAU, DNMT3A, FAP, and CCNB1. (C) Volcano plot showing differentially expressed genes, with the five candidate genes highlighted. (D) SHAP feature importance ranking based on mean absolute SHAP values. (E) SHAP summary plot showing the contribution of each gene to model prediction. (F) SHAP dependence plots showing the relationship between gene expression and SHAP values. (G) Representative SHAP force plot showing the contribution of selected features to an individual model prediction.

SHAP analysis identified SNAI2 and PLAU as the two most influential predictors in the model, with mean SHAP values of 0.151 and 0.146, respectively (Figure D). DNMT3A also contributed to model prediction, although its effect was directionally different from those of SNAI2 and PLAU in the SHAP summary plot (Figure E). SHAP dependence plots further illustrated the relationship between gene expression levels and model output for selected genes, including SNAI2, DNMT3A, and PLAU (Figure F). In the representative SHAP force plot, PLAU and SNAI2 contributed negatively to the model output for the selected sample, resulting in a predicted value lower than the baseline expectation (Figure G). These analyses prioritized SNAI2, PLAU, DNMT3A, FAP, and CCNB1 as candidate features within the analyzed lung cancer data sets. Their model contributions should not be interpreted as evidence that they are direct NNK-responsive genes.

3.6. Expression and Prognostic Associations of the Prioritized Genes in Lung Adenocarcinoma

In this study, the differential expression of the five key genes between tumor and normal tissues in lung adenocarcinoma (LUAD) was evaluated using the GEPIA bioinformatics platform. PLAU and CCNB1 were significantly upregulated in tumor samples (Figure A), demonstrating potential roles in tumorigenesis and disease progression. Kaplan–Meier analyses revealed prognostic stratification based on gene expression levels (Figure B). Notably, increased CCNB1 expression correlated with worse overall survival (log-rank P = 1.2 × 10–4). Moreover, CCNB1 levels were significantly associated with decreased disease-free survival (log-rank P = 0.0079) (Figure C), supporting its potential as a prognostic marker in LUAD.Exploratory docking generated computational ligand-placement models with scores ranging from −5.2 to −6.9 kcal/mol (Figure S1). Given the modest scores, the metabolic activation required for NNK carcinogenicity, and the absence of biochemical validation, these outputs were not interpreted as evidence of direct binding or target engagement.

6.

6

Open in a new tab

Expression and prognostic associations of the prioritized candidate genes in lung adenocarcinoma. (A) Box plots showing expression levels of the NNK core genes in tumor and normal tissues. The y-axis represents gene expression, and the x-axis lists the genes. Red indicates tumor samples, and black indicates normal tissues. (B) Kaplan–Meier overall survival curves stratified by core gene expression. The x-axis denotes overall survival time, and the y-axis denotes survival probability. Blue lines indicate low-expression groups, and red lines indicate high-expression groups. (C) Kaplan–Meier disease-free survival curves stratified by core gene expression. The x-axis denotes disease-free survival time, and the y-axis denotes disease-free survival probability. Blue lines indicate low-expression groups, and red lines indicate high-expression groups.

3.7. NNK Exposure is Associated with Nonspecific Malignant-like Phenotypes in A549 Cells

We next examined the cellular effects of NNK exposure in A549 cells. qRT-PCR analysis showed that NNK treatment increased the mRNA levels of FAP, CCNB1, SNAI2, DNMT3A, and PLAU, suggesting transcriptional changes related to proliferation, cell-cycle regulation, and phenotypic plasticity (Figure A). CCK-8 assays showed a dose- and time-related increase in A549 cell viability after NNK treatment, with higher viability observed at 24, 48, and 72 h compared with control cells (Figure B). Flow cytometry showed a decrease in the G1-phase population and an increase in the S and G2/M fractions, consistent with accelerated cell-cycle progression under the tested conditions (Figure C). Transwell assays further showed increased migration and invasion after NNK exposure (Figure D). Together, these data show that intact NNK exposure under the tested A549 conditions was associated with proliferative, migratory, and invasive phenotypes, accompanied by altered mRNA expression of the prioritized genes. These observations do not establish that the candidate genes mediate the phenotypes.

7.

7

Open in a new tab

Phenotypic and transcriptional responses of A549 cells to parent NNK under the tested in vitro conditions. (A) Relative mRNA expression levels of FAP, CCNB1, SNAI2, DNMT3A, and PLAU in A549 cells treated with vehicle control (0 μM NNK) or 25 μM NNK for 24 h. (B) CCK-8 assay showing absorbance at 450 nm in A549 cells treated with vehicle control (0 μM NNK), 10 μM NNK, or 25 μM NNK for 0, 24, 48, and 72 h. (C) Flow-cytometric analysis of cell-cycle distribution in A549 cells treated with vehicle control (0 μM NNK) or 25 μM NNK for 24 h. Representative histograms are shown on the left, and the percentages of cells in the G0/G1, S, and G2/M phases are shown on the right. (D) Transwell migration and invasion assays following treatment with vehicle control (0 μM NNK), 10 μM NNK, or 25 μM NNK. Representative stained images are shown on the left, and the corresponding quantitative results, expressed relative to the vehicle control, are shown on the right. Data are presented as mean ± SD from three independent experiments (n = 3). For panels A and C, comparisons between two groups were performed using an unpaired two-tailed Student’s t-test. For panel B, two-way ANOVA followed by Dunnett’s multiple-comparison test was used to compare each NNK concentration with the vehicle control at the corresponding time point. For panel D, one-way ANOVA followed by Dunnett’s multiple-comparison test was used to compare each NNK-treated group with the vehicle control. All statistical comparisons were made against the vehicle control (0 μM NNK). *P < 0.05; **P < 0.01; ***P < 0.001.

4. Discussion

NNK is a tobacco-specific nitrosamine classified as carcinogenic to humans by the International Agency for Research on Cancer. Although intact NNK is present in tobacco smoke and has been measured in airborne particulate matter, its biological effects in vivo depend substantially on metabolic transformation. NNK undergoes carbonyl reduction to NNAL and cytochrome P450-mediated α-hydroxylation, particularly through CYP2A13 and CYP2A6, generating unstable intermediates that produce DNA-reactive electrophiles and DNA adducts. ,,− Toxicokinetic studies in rats have shown that NNK is rapidly absorbed and extensively converted to NNAL after inhalation, intraperitoneal administration, or oral gavage. In humans, intact NNK has not been established as a detectable internal-exposure biomarker in blood, urine, or saliva, whereas NNAL and its glucuronides are commonly used as biomarkers of NNK exposure. − A direct LC–MS/MS study found no detectable urinary NNK in smokers, whereas free and total NNAL were detected. Smoking-related human cancers also exhibit characteristic mutational signatures, although these signatures are not specific to NNK exposure. Therefore, direct exposure of A549 cells to 10 or 25 μM parent NNK does not reproduce an established human internal-exposure scenario or the complete metabolic process responsible for NNK carcinogenicity. The A549 experiments in this study should be interpreted only as controlled parent-compound perturbation experiments.

The computational component of this study integrated protein lists obtained from chemical-associated prediction resources with transcriptomic differences between lung cancer and nontumor tissues. This workflow identified 16 overlapping genes and subsequently prioritized SNAI2, PLAU, DNMT3A, FAP, and CCNB1 using machine-learning analysis. These genes have established relevance to cancer biology in other experimental contexts. PLAU has been implicated in nonsmall cell lung cancer growth and treatment response, whereas CCNB1 is involved in cell-cycle regulation and has been associated with lung adenocarcinoma progression. , DNMT3A contributes to de novo DNA methylation and cancer-associated epigenetic regulation, , while FAP has been investigated as a component and therapeutic target of the tumor stroma. , These studies support the broader cancer relevance of the prioritized genes, but they do not demonstrate that the genes are specific targets of NNK.

The discriminatory performance observed in the machine-learning and GEPIA analyses should be interpreted within the source data sets. The GEO data sets compared lung cancer with nontumor tissues and were not derived from individuals or experimental systems with documented NNK exposure. Accordingly, the model performance indicates that the prioritized genes distinguished tumor from nontumor samples within the analyzed data sets; it does not indicate that these genes can identify NNK exposure or distinguish NNK-associated lung cancer from lung cancer caused by other factors. Similarly, the association between elevated CCNB1 expression and poorer survival supports its prognostic relevance in lung adenocarcinoma but does not establish NNK specificity.

Exploratory molecular docking generated computational ligand-placement models with scores ranging from −5.2 to −6.9 kcal/mol. These scores were modest, and their biological interpretation is particularly limited for a small nitrosamine whose carcinogenicity depends substantially on metabolic activation. ,− No biochemical binding, competition, thermal-shift, or structural assay was performed. Accordingly, the docking results were retained only as supplementary structural visualizations and were not used to support direct NNK–protein binding, covalent interaction, or cellular target engagement.

Parent NNK exposure in A549 cells was accompanied by higher CCK-8 signals, altered cell-cycle distributions, increased Transwell migration and invasion counts, and changes in several candidate-gene transcripts. These findings represent parallel observations obtained in the same experimental setting. They do not establish that CCNB1, PLAU, SNAI2, DNMT3A, or FAP lies upstream of, mediates, or is required for any of the observed cellular phenotypes. CCK-8 primarily reflects cellular metabolic activity and cannot independently distinguish increased proliferation from other changes in cell state. Likewise, increased Transwell counts may be influenced by cell number, survival, metabolic activity, or multiple signaling pathways.

Previous studies have reported NNK-related epigenetic effects involving DNA methyltransferases in mice, human lung cancer samples, and bronchial epithelial models. , NNK has also been reported to promote lung cancer cell migration and invasion through a c-Src/PKCι/FAK signaling loop and to induce pulmonary tumorigenesis through nAChR/Src/STAT3-dependent signaling. , These findings illustrate that NNK-related cellular responses may involve multiple pathways, but the present experiments did not examine these mechanisms. Although changes in CCNB1, PLAU, SNAI2, and DNMT3A mRNA were observed, protein-level validation and gene knockdown, overexpression, or rescue experiments would be required before assigning a functional relationship between any candidate gene and the measured cellular phenotypes.

Several limitations define the scope of the present study. First, the discovery data sets consisted of lung cancer and nontumor tissues rather than samples from individuals with documented NNK exposure. The prioritized genes are therefore lung-cancer-associated candidates that overlap with computationally predicted NNK-associated proteins, not direct NNK-induced transcriptional targets. Second, A549 cells do not adequately reproduce the cytochrome P450-dependent metabolic competence required for the complete bioactivation of NNK. The cellular effects of NNK are strongly influenced by CYP2A13-mediated metabolism, and toxicokinetic evidence shows rapid absorption and extensive conversion of NNK to NNAL in vivo. , Direct micromolar exposure to parent NNK therefore cannot reproduce the formation, distribution, and clearance of NNK metabolites in humans.

Third, experimental validation was limited to a single cell line and mRNA measurements. Protein-level changes and direct NNK–protein interactions were not examined. Fourth, the cellular phenotypes are nonspecific, and no knockdown, overexpression, or rescue experiment was performed to establish a causal role for any candidate gene. The Transwell results may also have been influenced by exposure-related differences in cell number or metabolic state because proliferation was not independently controlled during the assay. Finally, the acute exposure design does not reproduce chronic, low-dose tobacco-smoke exposure. Differentiated air–liquid interface cultures and other metabolically competent airway epithelial systems may provide more physiologically relevant models for evaluating inhaled toxicants that require metabolic activation.

Within these limitations, the contribution of this study is the construction of a hypothesis-generating prioritization framework linking computationally predicted chemical-associated proteins with lung cancer transcriptomic features. The five prioritized genes provide experimentally testable candidates for future studies. However, the present findings do not establish direct NNK targets, physiologically representative human exposure, or a causal molecular mechanism.

5. Conclusion

This study identifies SNAI2, PLAU, DNMT3A, FAP, and CCNB1 as candidate genes within an NNK-related lung cancer analytical framework. The findings are based on public transcriptomic data sets, machine-learning prioritization, exploratory docking, and a simplified A549 parent-compound exposure model. They should be interpreted as hypothesis-generating rather than as evidence of direct NNK target engagement or causal gene function. Future studies using metabolically competent airway models, NNAL or reactive metabolite-focused systems, protein-level validation, and gene perturbation experiments are required to clarify the functional relevance of these candidates.

Supplementary Material

ao6c04401_si_001.pdf (599.6KB, pdf)

Acknowledgments

An GTP 4.5 AI-assisted language-editing tool was used only to improve grammar and readability. The authors reviewed and verified all scientific content, data analysis, and conclusions.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.6c04401.

  • Exploratory molecular docking models and docking scores for NNK with proteins encoded by SNAI2, PLAU, DNMT3A, FAP, and CCNB1 (Figure S1); qRT-PCR primer sequences (Table S1); and ProTox-3.0 toxicity predictions for NNK (Table S2) (PDF) (PDF)

¶.

Y.L., H.W., and P.L. contributed equally to this work.

This work was supported by the National Natural Science Foundation of China (82172569, 81601996), the Natural Science Foundation of Tianjin (23JCYBJC01010), the Tianjin Health Research Project (TJWJ2025MS002), and the Tianjin Key Medical Discipline Construction Project (TJYXZDXK-3-002B, TJYXZDXK-3-006A).

All authors give their full consent to publish the present article.

The authors declare no competing financial interest.

References

  1. Shong L. Y., Lam D. C.. Emerging Trends in Global Lung Cancer Burden. Semin. Respir. Crit. Care Med. 2025;46(5):409–418. doi: 10.1055/a-2651-0612. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Kim S. Y., Park H. S., Chiang A. C.. Small Cell Lung Cancer: A Review. JAMA. 2025;333(21):1906–1917. doi: 10.1001/jama.2025.0560. [DOI] [PubMed] [Google Scholar]
  3. Cao W., You Z., Wang Z., Liang Z., Li H., Chang Z.. et al. Risk factors behind the global lung cancer burden: a pan-database exploration. Transl. Lung Cancer Res. 2025;14(7):2452–2469. doi: 10.21037/tlcr-2025-131. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Minervini F., Lampridis S., Kestenholz P., Pardo E., Crommelinck J., Putora P. M.. et al. The impact of smoking on lung cancer patients. Eur. Respir. Rev. 2025;34(176):240175. doi: 10.1183/16000617.0175-2024. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Ma Z., Lv J., Zhu M., Yu C., Ma H., Jin G.. et al. Lung cancer risk score for ever and never smokers in China. Cancer Commun. 2023;43(8):877–895. doi: 10.1002/cac2.12463. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Warren, G. W. ; Cummings, K. M. . Tobacco and Lung Cancer: Risks, Trends, and Outcomes in Patients with Cancer; American Society of Clinical Oncology educational book American Society of Clinical Oncology Annual Meeting, 2013; pp 359–364. [DOI] [PubMed] [Google Scholar]
  7. Hecht S. S.. Tobacco carcinogens, their biomarkers and tobacco-induced cancer. Nat. Rev. Cancer. 2003;3(10):733–744. doi: 10.1038/nrc1190. [DOI] [PubMed] [Google Scholar]
  8. Bodhicharla R., Ryde I. T., Prasad G. L., Meyer J. N.. The tobacco-specific nitrosamine 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone (NNK) induces mitochondrial and nuclear DNA damage in Caenorhabditis elegans. Environ. Mol. Mutagen. 2014;55(1):43–50. doi: 10.1002/em.21815. [DOI] [PubMed] [Google Scholar]
  9. Verza F. A., Santos-Sousa A. L., Oliveira S. H. P., Bernabé D. G.. Oral cancer cells secrete stress neurotransmitter and proliferate in response to tobacco carcinogen NNK. Endocr. Oncol. 2025;5(1):e240076. doi: 10.1530/eo-24-0076. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Yuan J. M., Knezevich A. D., Wang R., Gao Y. T., Hecht S. S., Stepanov I.. Urinary levels of the tobacco-specific carcinogen N’-nitrosonornicotine and its glucuronide are strongly associated with esophageal cancer risk in smokers. Carcinogenesis. 2011;32(9):1366–1371. doi: 10.1093/carcin/bgr125. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Zhang D., Lei J., Ma J., Chen X., Sheng L., Jiang Z.. et al. β2-adrenogenic signaling regulates NNK-induced pancreatic cancer progression via upregulation of HIF-1α. Oncotarget. 2016;7(14):17760–17772. doi: 10.18632/oncotarget.5677. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Cheng G., Li J., Zheng M., Zhao Y., Zhou J., Li W.. NNK, a tobacco-specific carcinogen, inhibits the expression of lysyl oxidase, a tumor suppressor. Int. J. Environ. Res. Public Health. 2015;12(1):64–82. doi: 10.3390/ijerph120100064. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Smokeless Tobacco and Some Tobacco-Specific N-Nitrosamines. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans; IARC Monographs, 2007; Vol. 89, pp 1–592. [PMC free article] [PubMed] [Google Scholar]
  14. Ge G. Z., Xu T. R., Chen C.. Tobacco carcinogen NNK-induced lung cancer animal models and associated carcinogenic mechanisms. Acta Biochim. Biophys. Sin. 2015;47(7):477–487. doi: 10.1093/abbs/gmv041. [DOI] [PubMed] [Google Scholar]
  15. Ma G., Yu H., Xu X., Geng L., Wei X., Wen J.. et al. Molecular Basis for Metabolic Regioselectivity and Mechanism of Cytochrome P450s toward Carcinogenic 4-(Methylnitrosamino)-(3-pyridyl)-1-butanone. Chem. Res. Toxicol. 2020;33(2):436–447. doi: 10.1021/acs.chemrestox.9b00353. [DOI] [PubMed] [Google Scholar]
  16. Sarkar M., Wang J., Liang Q.. Metabolism of nicotine and 4-(methylnitrosamino)-l-(3-pyridyl)-lbutanone (NNK) in menthol and non-menthol cigarette smokers. Drug Metab. Lett. 2013;6(3):198–206. doi: 10.2174/1872312811206030007. [DOI] [PubMed] [Google Scholar]
  17. Ma B., Zarth A. T., Carlson E. S., Villalta P. W., Upadhyaya P., Stepanov I.. et al. Methyl DNA Phosphate Adduct Formation in Rats Treated Chronically with 4-(Methylnitrosamino)-1-(3-pyridyl)-1-butanone and Enantiomers of Its Metabolite 4-(Methylnitrosamino)-1-(3-pyridyl)-1-butanol. Chem. Res. Toxicol. 2018;31(1):48–57. doi: 10.1021/acs.chemrestox.7b00281. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Aggarwal R. K., Sidoli S., Wang J., Sahu S., Sanawar R., Gupta V.. et al. Smoking-Associated Carcinogen-Induced Inflammation Promotes Lung Carcinogenesis via IRAK4 Activation. Clin. Cancer Res. 2025;31(4):746–755. doi: 10.1158/1078-0432.CCR-24-2182. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Tian J., Zhao W., Wu Y., Shi Y., Yu J., Zhang W.. et al. Diallyl Disulfide Blocks Cigarette Carcinogen 4-(Methylnitrosamino)-1-(3-pyridyl)-1-butanone-Induced Lung Tumorigenesis via Activation of the Nrf2 Antioxidant System and Suppression of NF-κB Inflammatory Response. J. Agric. Food Chem. 2023;71(46):17763–17774. doi: 10.1021/acs.jafc.3c02007. [DOI] [PubMed] [Google Scholar]
  20. Aquilina N. J., Havel C. M., Harrison R. M., Ho K. F., Benowitz N. L., Jacob P. III. Determination of 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone arising from tobacco smoke in airborne particulate matter. Environ. Int. 2022;158:106992. doi: 10.1016/j.envint.2021.106992. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Carlson E. S., Upadhyaya P., Hecht S. S.. Evaluation of nitrosamide formation in the cytochrome P450-mediated metabolism of tobacco-specific nitrosamines. Chem. Res. Toxicol. 2016;29(12):2194–2205. doi: 10.1021/acs.chemrestox.6b00384. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Li Y., Hecht S. S.. Metabolism and DNA adduct formation of tobacco-specific N-nitrosamines. Int. J. Mol. Sci. 2022;23(9):5109. doi: 10.3390/ijms23095109. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Sun Y., Wang H., Chen H., Zhang S., Li J., Zhang J.. et al. Nicotine inhibits the cytotoxicity and genotoxicity of NNK mediated by CYP2A13 in BEAS-2B cells. Molecules. 2022;27(15):4851. doi: 10.3390/molecules27154851. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Hu S. C., Bryant M. S., Sepehr E.. et al. Toxicokinetic and genotoxicity study of NNK in male Sprague Dawley rats following nose-only inhalation exposure, intraperitoneal injection, and oral gavage. Toxicol. Sci. 2021;182(1):10–28. doi: 10.1093/toxsci/kfab049. [DOI] [PubMed] [Google Scholar]
  25. Hu C. W., Hsu Y. W., Chen J. L., Tam L. M., Chao M. R.. Direct analysis of tobacco-specific nitrosamine NNK and its metabolite NNAL in human urine by LC–MS/MS: evidence of linkage to methylated DNA lesions. Arch. Toxicol. 2014;88(2):291–299. doi: 10.1007/s00204-013-1137-y. [DOI] [PubMed] [Google Scholar]
  26. Alexandrov L. B., Ju Y. S., Haase K., Van Loo P., Martincorena I., Nik-Zainal S.. et al. Mutational signatures associated with tobacco smoking in human cancer. Science. 2016;354(6312):618–622. doi: 10.1126/science.aag0299. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Zheng Y., Zhang L., Zhang K., Wu S., Wang C., Huang R., Liao H.. PLAU promotes growth and attenuates cisplatin chemosensitivity in ARID1A-depleted non-small cell lung cancer through interaction with TM4SF1. Biol. Direct. 2024;19(1):7. doi: 10.1186/s13062-024-00452-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Bao B., Yu X., Zheng W.. MiR-139–5p targeting CCNB1 modulates proliferation, migration, invasion and cell cycle in lung adenocarcinoma. Mol. Biotechnol. 2022;64(8):852–860. doi: 10.1007/s12033-022-00465-5. [DOI] [PubMed] [Google Scholar]
  29. Chen B. F., Chan W. Y.. The de novo DNA methyltransferase DNMT3A in development and cancer. Epigenetics. 2014;9(5):669–677. doi: 10.4161/epi.28324. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Holz-Schietinger C., Matje D. M., Harrison M. F., Reich N. O.. Oligomerization of DNMT3A controls the mechanism of de novo DNA methylation. J. Biol. Chem. 2011;286(48):41479–41488. doi: 10.1074/jbc.M111.284687. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Santos A. M., Jung J., Aziz N., Kissil J. L., Puré E.. Targeting fibroblast activation protein inhibits tumor stromagenesis and growth in mice. J. Clin. Invest. 2009;119(12):3613–3625. doi: 10.1172/JCI38988. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Xie Y., Ma J., Tang W., Zhang Y., Zhang C., Chen Y.. Efficacy and safety evaluation of 177Lu-FAP-2286 in the treatment of advanced lung cancer. Clin. Nucl. Med. 2024;49(9):830–837. doi: 10.1097/RLU.0000000000005297. [DOI] [PubMed] [Google Scholar]
  33. Lin R. K., Hsieh Y. S., Lin P.. et al. The tobacco-specific carcinogen NNK induces DNA methyltransferase 1 accumulation and tumor suppressor gene hypermethylation in mice and lung cancer patients. J. Clin. Invest. 2010;120(2):521–532. doi: 10.1172/JCI40706. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Liu X., Wang J., Yang Z., Xie Q., Diao X., Yao X.. et al. Upregulated DNMT3a coupling with inhibiting p62-dependent autophagy contributes to NNK tumorigenicity in human bronchial epithelial cells. Ecotoxicol. Environ. Saf. 2024;286:117157. doi: 10.1016/j.ecoenv.2024.117157. [DOI] [PubMed] [Google Scholar]
  35. Shen J., Xu L., Owonikoko T. K., Sun S. Y., Khuri F. R., Curran W. J.. et al. NNK promotes migration and invasion of lung cancer cells through activation of the c-Src/PKCι/FAK loop. Cancer Lett. 2012;318(1):106–113. doi: 10.1016/j.canlet.2011.12.008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Boo H. J., Min H. Y., Hwang S. J., Lee H. J., Lee J. W., Oh S. R.. et al. The tobacco-specific carcinogen NNK induces pulmonary tumorigenesis via nAChR/Src/STAT3-mediated activation of the renin–angiotensin system and IGF-1R signaling. Exp. Mol. Med. 2023;55(6):1131–1144. doi: 10.1038/s12276-023-00994-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Cao X., Coyle J. P., Xiong R.. et al. Invited review: human air–liquid-interface organotypic airway tissue models derived from primary tracheobronchial epithelial cellsoverview and perspectives. In Vitro Cell. Dev. Biol.: Anim. 2021;57(2):104–132. doi: 10.1007/s11626-020-00517-7. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

ao6c04401_si_001.pdf (599.6KB, pdf)

Articles from ACS Omega are provided here courtesy of American Chemical Society

RESOURCES