Integrated Systems Toxicology Analysis and Molecular Dynamics Study Reveal the Potential Mechanism by which Benzo[a]pyrene Contributes to Depression in Lung Adenocarcinoma via RNASE1 Upregulation

Abstract

Benzo­[a]­pyrene (Bap), a prominent carcinogen found in tobacco smoke and air pollution, is closely associated with lung adenocarcinoma (LUAD) and subsequent depression, though the mechanism underlying this comorbidity remains unclear. LUAD patients were categorized based on depression-related gene expression. Prognostic genes identified through machine learning were cross-referenced with predicted Bap targets. Mendelian randomization (MR) evaluated the causal relationship between RNASE1 and depression. Single-cell RNA sequencing (scRNA-seq), molecular docking, and 100 ns molecular dynamics (MD) simulations (using the CHARMM36 force field and TIP3P water model) were conducted, followed by analyses of RMSD, RMSF, Rg, SASA, and Gibbs free energy landscape. A Bap-exposed mouse model was used to validate key gene expression and depression-like behaviors in vivo. Consensus clustering based on 52 depression-related genes stratified the TCGA-LUAD cohort into two gene signature-based molecular subtypes with distinct expression patterns and prognostic outcomes. Cluster 1 represented a favorable-prognosis subtype with better disease-free survival, whereas Cluster 2 represented a poor-prognosis subtype with stronger depression-related transcriptomic features; however, these clusters should not be interpreted as clinically confirmed depression subgroups because TCGA lacks psychiatric assessment data. Among 57 depression-related prognostic genes intersected with 375 Bap targets, five genes overlapped: CLEC7A, NEK2, RNASE1, ITGAL, and CPA3. MR confirmed that elevated lung RNASE1 increased depression risk (IVW: OR = 1.008, 95% CI: 1.002–1.014, P = 0.005), while CLEC7A and other genes showed no causal link. scRNA-seq localized RNASE1 upregulation to tumor epithelial cells, monocytes, and macrophages. Docking and MD analyses supported stable Bap–RNASE1 binding, involving hydrophobic interactions with key residues and a dominant low-energy conformational state. In vivo, Bap-exposed mice showed prolonged immobility time in the tail suspension test, fewer open-arm entries, and reduced open-arm time in the elevated plus-maze test, accompanied by elevated WBC, MPV, PDW, and increased lung RNASE1 protein expression. Bap may contribute to LUAD-related depression through RNASE1 upregulation. This provides molecular-level mechanistic insight into the cancer–depression comorbidity driven by environmental carcinogens.

1. Introduction

In 2022, global data indicated around 2.5 million new lung cancer cases, accounting for one-eighth of all new malignancies, confirming lung cancer as the most frequently diagnosed cancer. Lung adenocarcinoma (LUAD), the leading histological subtype, comprises roughly half of these cases. The pathogenesis of LUAD is multifactorial, involving genetic predisposition, lifestyle-related exposures, and environmental carcinogens.

Environmental carcinogens represent a major etiological factor in the pathogenesis of lung cancer. Benzo­[a]­pyrene (Bap), a prototypical polycyclic aromatic hydrocarbon found in tobacco smoke and environmental pollutants, is known to induce DNA damage and promote tumorigenesis through metabolic activation and oxidative stress pathways. Exposure to Bap occurs primarily through inhalation of tobacco smoke and ambient air pollution, leading to its bioactivation by cytochrome P450 enzymes into reactive diol epoxide metabolites that form DNA adducts and generate oxidative stress, ultimately driving oncogenic mutations and lung tumor initiation.

Recently, Advances in targeted therapies and immunotherapies have significantly improved survival rates. By January 2022, the United States reported over 18 million cancer survivors, with nearly half surviving more than a decade. Nevertheless, survivors often grapple with a spectrum of physical and mental health issues. Debilitating fatigue, anemia, sleep disturbances such as insomnia, and neuropsychiatric disorders like depression and anxiety frequently coexist in cancer patients; these comorbidities intertwine to further impair functional status, quality of life, and treatment adherence. − Notably, depression is one of the most common neuropsychiatric comorbidities among lung cancer patients. Research indicates that depression affects up to 57.1% of Chinese lung cancer patients. This high prevalence underscores the urgent need for routine psychological screening and integrated mental health interventions in oncology care settings.

The etiology of cancer-related depression is multifactorial. In recent years, increasing attention has been drawn to the potential link between environmental carcinogen exposure and depression. Bap, as a typical air pollutant and component of tobacco smoke, has been shown to exert neurobehavioral effects in multiple studies. Beyond its well-established role in carcinogenesis, emerging evidence suggests that exposure to environmental toxicants like Bap may also disrupt neuroendocrine and immune homeostasis, potentially contributing to neuropsychiatric comorbidities such as depression. Epidemiological investigations have revealed a positive association between polycyclic aromatic hydrocarbon (PAH) exposure and the risk of depression in general populations, with smoking significantly enhancing the pro-depressive effect of Bap. Animal studies have further elucidated the underlying mechanisms: Bap and its metabolites can cross the blood–brain barrier, activate microglial inflammatory responses, upregulate pro-inflammatory cytokines, and IL-18, and simultaneously reduce noradrenergic and serotonergic axonal density in the hippocampal CA1 and CA3 subregions. Moreover, Bap exposure has been shown to dose-dependently inhibit brain acetylcholinesterase activity, reduce serotonin bioavailability, and disrupt hypothalamic–pituitary–adrenal (HPA) axis functionall pathways that play important roles in the pathogenesis of depressive-like behaviors. − Collectively, these lines of evidence suggest that Bap may directly promote depressive-like phenotypes through neuroinflammation, neurotransmitter depletion, and neuroendocrine dysregulation. Inflammatory factors such as IL-6 and TNF-α may penetrate the blood–brain barrier during tumor progression and treatment, disrupting neurotransmitter metabolism, including serotonin and dopamine, thereby inducing depression. Additionally, depressive symptoms are influenced by factors like education level, economic status, constipation, and chemotherapy. Researchers have recently investigated targeted interventions such as dignity therapy, exercise therapy, and music therapy to mitigate depression in these patients. Despite these efforts, the molecular basis linking LUAD, environmental toxicant exposure, and depression remains underexplored.

The Self-rating Depression Scale, such as PHQ-9, is widely used in clinical settings to evaluate depressive symptoms in cancer patients. While user-friendly, it falls short in elucidating the molecular mechanisms linking the tumor microenvironment to neuropsychiatric symptoms. Publicly available transcriptomic data sets have provided useful resources for identifying molecular subtypes and disease-related biomarkers. Consensus Clustering, based on transcriptomic features, emerges as a promising research approach. This advanced unsupervised machine learning technique identifies disease subgroups with shared molecular traits without relying on prior clinical diagnostic labels. The Cancer Genome Atlas (TCGA) and the Gene Expression Omnibus (GEO), as large-scale public resources, provide multi-omics data for multiple cancer types, among which the LUAD cohort contains rich transcriptomic and clinical information. However, the lung cancer cohorts from TCGA and GEO do not contain direct psychiatric assessment records, clinical depression diagnoses, or self-rating depression scale data such as PHQ-9 or HAMD. Therefore, depression status could not be directly determined in individual TCGA patients. In this study, Consensus Clustering was applied to investigate depression-related molecular features in LUAD rather than to assign clinical depression diagnoses. By integrating depression-related gene expression profiles with clinical phenotype data, we aimed to identify LUAD molecular subtypes characterized by depression-related transcriptomic features, offering fresh insights into the molecular interplay between cancer and depression. To address this gap, we established an integrative pipeline combining transcriptomic subtyping, systems toxicology analysis, and structural validation. The study integrated transcriptomic and clinical data sets from TCGA and GEO databases. Consensus clustering identified depression-related molecular subtypes of LUAD, with key features delineated via differential expression analysis and WGCNA. A panel of 117 machine learning algorithms was employed to construct a prognostic model based on depression-related genes. Systems toxicology analysis, together with molecular docking and MD simulations, identified RNASE1 as a Bap-targeted protein common to LUAD and depression, and its involvement was further confirmed by in vivo experiments. These findings provide mechanistic insight into cancer–depression comorbidity driven by environmental carcinogens and establish a basis for targeted interventions.

2. Materials and Methods

2.1. Collection of the Chemical Composition and Targets of Bap

Search for the keyword “Bap” in the PubChem database (https://pubchem.ncbi.nlm.nih.gov/) to obtain its chemical structure and SMILES formula: C1 = CC = C2C3 = C4C­(=CC2 = C1)C = CC5 = C4C­(=CC5 = C5)C = C3. Based on this molecular information, conduct potential target analysis using the ChEMBL (https://www.ebi.ac.uk/chembl/), SwissTargetPrediction (http://www.swisstargetprediction.ch/), and SEA (https://sea.bkslab.org/) databases. For ChEMBL, retain all human target genes with reported biological activities. For SwissTargetPrediction, retain targets with a probability score >0.5; this threshold strikes a balance between sensitivity and specificity, retaining high-confidence predictions while excluding low-probability results. For SEA, retain targets with a maximum Tanimoto coefficient (max tc) > 0.05, which is consistent with the database’s default significance threshold. Subsequently, use the UniProt database (https://www.uniprot.org; https://www.ebi.ac.uk/chembl/niprotkb/) and its ID mapping tool to convert all target identifiers into official symbols of the HUGO Gene Nomenclature Committee (HGNC), thus eliminating gene symbol ambiguities and ensuring naming standardization. Merge multiple gene entries resulting from different database identifiers or synonymous symbols, and finally obtain an integrated target set containing 375 unique Bap target genes.

The gene expression profiles, including FPKM data and clinical information for LUAD patients, were obtained from TCGA database (https://portal.gdc.cancer.gov/). Additional data sets for LUADGSE31210, GSE68465, GSE50081, and GSE72094and major depressive disorder (MDD)GSE98793 and GSE32280were sourced from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/). Detailed information is provided in Table .

1. Collection of Dataset Information.

Source	ID	Trait	Tissue	Counts
TCGA	TCGA-LUAD	LUAD	lung tumor	513
GEO	GSE31210	LUAD	lung tumor	226
GEO	GSE68465	LUAD	lung tumor	443
GEO	GSE50081	LUAD	lung tumor	181
GEO	GSE72094	LUAD	lung tumor	442
				2Tumor, 2Normal
GEO	GSE98793	Major Depressive Disorder	Peripheral blood	128MDD, 64Healthy
GEO	GSE32280	Major Depressive Disorder	Peripheral blood	16MDD, 8Healthy

2.2. Collection of Depression-Related Genes

The keywords related to “Depression” and “Major Depressive Disorder” were input into multiple databases, including GeneCards (https://www.genecards.org) and Online Mendelian Inheritance in Man (OMIM, https://www.omim.org/). Meanwhile, two MDD data sets collected from the GEO database were converted to TPM data type, then merged and batch effects were removed. Differential analysis was performed based on |log₂FoldChange| ≥ 0.585 and P < 0.05. This threshold corresponds to a fold change ≥ 1.5, a widely adopted criterion in transcriptomic studies for defining differentially expressed genes with biological relevance, and was applied in conjunction with a significance threshold of P < 0.05 to ensure statistical robustness. The intersection of the disease-related genes identified in the two databases and the analyzed differential genes was taken to obtain the final list of depression-related merge differentially expressed genes (merge DEGs).

2.3. Consensus Clustering Analysis of Depression-Related Genes in Patients with LUAD

The expression profiles of depression-related genes in LUAD samples from the TCGA database were analyzed using the ConsensusClusterPlus R package. To determine the optimal number of clusters (k), we comprehensively evaluated the silhouette coefficient and the Calinski–Harabasz (CH) index across different k-values, − and the k-value with the highest combined performance was selected. The optimal number of clusters (k) was determined by evaluating the relative change in the area under the cumulative distribution function (CDF) curve, with the clustering parameters set as follows: maxK = 9, reps = 100, pItem = 0.8, pFeature = 1, clusterAlg = “km”, and distance = “euclidean”. The clustering results were validated using a UMAP plot with a neighborhood size of 15 and an embedding looseness of 0.1. Patients with a survival time of less than 30 days were excluded, and Kaplan–Meier survival analysis was performed to examine the prognostic differences among the identified clusters.

2.4. Screening of Differentially Expressed Genes (DEGs) between Depression-Related Molecular Subtypes in LUAD

Based on |log₂Fold change| ≥ 1 and P < 0.05, Depression-Related Molecular Subtypes in LUAD differentially expressed genes (Clusters DEGs) between the two depression-related molecular subtypes (Cluster 1 vs Cluster 2) in the TCGA-LUAD cohort were identified using the limma R package (version 3.40.6), and the results were visualized using the Sangerbox platform (http://vip.sangerbox.com/).

2.5. Weighted Gene Coexpression Network Analysis (WGCNA) of Depression-Related Molecular Subtypes in LUAD

WGCNA can identify gene modules with similar expression patterns based on the variation of gene expression profiles. Previously, samples were divided into Cluster1 and Cluster2 groups according to depression-related genes, and then WGCNA analysis was performed on these two groups. We used the “WGCNA” R package (version 1.73). Genes with the top 6000 gene variability (MAD) and an average expression level of no less than 1 were selected for analysis. To assess the robustness of the module identification results, we repeated the WGCNA analysis using alternative gene selection thresholds of 5,000 and 8,000 MAD-ranked genes. The “pickSoftThreshold” function was used to calculate the Pearson correlation coefficients between all genes in LUAD patients and weight them to determine the soft threshold β. The correlation strength was represented by a power function to construct an adjacency matrix defined as Axy = |cor (genex, geney)| ^β. The adjacency matrix was then converted into a topological overlap matrix (TOM). TOM quantitatively describes the similarity between nodes by comparing the weighted correlations of a node with other nodes. After that, hierarchical clustering was performed based on the topological overlap similarity between gene pairs to identify modules, with each module set to contain at least 75 genes. Genes within the same module usually exhibit a high level of coexpression relationships.

2.6. Identification of Key Genes with Depression-Related Molecular Subtypes in LUAD Patients

The intersection of DEGs between the two depression-related molecular subtypes and WGCNA-derived key module genes was defined as the core genes associated with depression-related molecular subtypes in LUAD. These key genes were visualized using the Microbiome Bioinformatics Platform.

2.7. Construction of a Prognostic Model for Depression-Related Molecular Subtypes in LUAD Using Univariate Cox Regression and 117 Machine Learning Algorithms

Given the significant prognostic heterogeneity among LUAD patients and the unclear molecular mechanisms underlying their prevalent depression-like behaviors, we aim to develop a robust prognostic model. We selected the key genes between the two depression-related molecular subtypes in the aforementioned LUAD cohort as relevant genes for identifying the prognosis of LUAD patients and integrated ten machine learning algorithms: Stepwise Cox, CoxBoost, Lasso, Ridge, Elastic Net (Enet, α = 0.1–0.9), Survival SVM, SuperPC, Generalized Boosting Regression Model (GBM), Partial Least Squares Cox (plsRcox), and Random Survival Forest (RSF). A total of 117 algorithm combinations were tested using the mime R package within the framework of Leave-One-Out Cross-Validation (LOOCV). The model performance was evaluated by the concordance index (C-index) of the validation data set. The TCGA-LUAD data set was used as the training set, and other LUAD cohorts from the GEO database were used for independent validation. The machine learning algorithm with the highest average C-index across all cohorts was selected as the final model and named the Depression-Related Gene Signature-Based Prognostic Model for LUAD (DRPM). Patients were divided into high-risk and low-risk depression groups for LUAD based on the median risk score derived from the prognostic model. The survfit function in the Survival (v3.5–5) and Survminer (v0.4.9) packages was used to generate Kaplan–Meier survival curves to assess the survival differences between groups. Finally, we evaluated the 1-year, 3-year, and 5-year ROC curves of the model in five large LUAD cohorts and compared the model’s C-index and the results of the 1-, 3-, and 5-year AUC values with those of published LUAD models to explore the prognostic performance of the model we developed.

2.8. Functional Enrichment Analysis of Gene Sets Incorporated with Machine Learning

The present study employed GO and KEGG enrichment analyses to elucidate the biological processes and underlying mechanisms associated with pathogenic genes linked to two depression-related molecular subtypes of LUAD. The results were visualized using the “ggplot2” R package, with statistical significance set at P < 0.05. This comprehensive analytical framework offers valuable insights into the functional relationships and molecular pathways governing the two depression-related LUAD subtypes, thereby illuminating the regulatory mechanisms driving subtype differentiation.

2.9. Analysis of Immune Cell Infiltration Characteristics in the Tumor Microenvironment of LUAD

The “GSVA” package was employed to analyze gene expression data, calculating enrichment scores for various immune cells in each sample. Violin plots were used to compare the ssGSEA scores between Cluster1 and Cluster2 groups, effectively illustrating data distribution and highlighting statistically significant differences.

2.10. Mendelian Randomization Analysis

To explore the potential causal association between Bap-related core target genes and depression, the Mendelian randomization method was employed. The common core genes were obtained by taking the intersection of Bap target genes and depression feature genes screened by machine learning. The expression quantitative trait loci (eQTLs) of these genes in lung tissue were extracted from the GTEx v8 database as instrumental variables (P < 5 × 10^–8). The exposure data were the gene expression levels corresponding to the above eQTLs, and the outcome data were derived from the genome-wide association study (GWAS) summary data of depression in the IEU OpenGWAS database (ukb-d-20126_3, 86,895 European populations). Allele harmonization was performed before integration, and palindromic sequences and inconsistent-strand single nucleotide polymorphisms (SNPs) were removed. The inverse variance weighting method was used as the main estimation method, supplemented by MR-Egger, weighted median method, simple median method, and weighted mode method for sensitivity analysis. The analysis was completed using the “TwoSampleMR” package in R software to systematically evaluate the potential causal effect of the expression levels of Bap-related core target genes on the risk of depression.

2.11. Single-Cell Transcriptomic Analysis of Lung Cancer

To reveal the expression patterns of core target genes of Bap at the single-cell level, this study utilized the single-cell data set GSE117570 from the GEO database to conduct single-cell RNA sequencing analysis on tumor and paired adjacent normal tissues (GSM3304007, GSM3304008, GSM3304011, and GSM3304012) from 2 lung cancer patients with a smoking history. Meanwhile, scRNA-seq data from nine resected samples collected from nine untreated patients with lung adenocarcinoma in the GSE189357 data set were applied as an external validation data set. The Seurat package was used to read the expression matrix of each sample and merge them into a single Seurat object. Genes expressed in only a single sample were filtered out, and cells with less than 100 detected genes or a mitochondrial gene proportion exceeding 10% were excluded. After LogNormalize normalization, the top 2000 highly variable genes were identified. Principal component analysis was employed for dimensionality reduction. Based on the elbow plot, the top 8 principal components were selected. The Harmony algorithm was used to remove batch effects according to tissue types. A resolution of 0.4 was selected as the optimal one through clustree, and unsupervised clustering was performed based on the Louvain algorithm, followed by visualization using UMAP and t-SNE. Cell type annotation was carried out manually, PTPRC for immune cells, EPCAM for epithelial cells, and PECAM1 for endothelial cells. By analyzing the expression distribution of core target genes of BaP in each cell type, their cell-specific expression characteristics in the lung cancer microenvironment were revealed.

2.12. Cell Communication Analysis

CellChat package was used for cell communication analysis. A CellChat object was created with the annotated cell types. The human “secreted signaling” ligand–receptor database was selected to identify overexpressed genes and ligand–receptor pairs. After projecting them onto the protein interaction network, the intercellular communication network was inferred, and communication events with less than 5 cells were filtered. The communication results of each signaling pathway were calculated and integrated into the overall communication network. netVisual_circle was used to display the communication frequency and intensity, and netVisual_heatmap was used to present the communication network. Hierarchical diagrams, circle diagrams, chord diagrams, and heatmaps were used for visualizing signaling pathways such as MIF. netVisual_bubble was utilized to show the ligand–receptor pairs between specified cell types, and plotGeneExpression was used to display the expression of pathway-related genes.

2.13. Core Gene Perturbation Analysis

Perform core gene perturbation analysis using the scTenifoldKnk package. Extract the counts expression matrix and filter out genes expressed in less than 5% of cells, ensuring that the target gene (taking RNASE1 as an example) is retained. Use the VST method to select the top 5000 highly variable genes and merge them with the target gene. Randomly select no more than 500 cells, set the mitochondrial gene threshold <10% and the minimum library size ≥1000 to simulate the effect of target gene knockout. Compare the differences in gene regulatory networks before and after knockout, calculate the regulatory distance and P-value. Screen the significantly different genes with P < 0.05 and sort them by regulatory distance. Display the top 5 genes in a bar chart to identify the key molecules that coregulate with the core target gene of Bap.

2.14. Single-Cell Pseudotemporal Analysis

To explore the expression dynamics of RNASE1 during the differentiation of monocytes into macrophages, this study employed the Monocle2 package for pseudotime analysis. First, a CellDataSet object was constructed. After preprocessing the data, highly variable genes with an average expression level greater than 0.1 and a dispersion higher than the model-fitted value were selected as ordering genes. The DDRTree algorithm was used for dimensionality reduction, and the pseudotime values of cells were calculated using the “orderCells” function. In terms of visualization, cell trajectory plots were drawn and colored by cell type and pseudotime, respectively, to show the distribution of cells along the differentiation path and the expression changes of RNASE1 in different cell types and pseudotime.

2.15. Single-Cell RNA Velocity Analysis

To uncover the dynamic change directions of gene expression during the differentiation process of cells associated with lung adenocarcinoma, this study employed scVelo to conduct RNA velocity analysis on single-cell transcriptome data. First, the unspliced and spliced transcript count matrices were extracted from the .loom file. After gene and cell quality control, the data of the normal group and the tumor group were, respectively, merged, and the common genes were retained. Subsequently, scVelo was used for data preprocessing, including filtering and normalizing gene expression and calculating the first- and second-order moments based on the neighborhood graph. On this basis, the gene-specific RNA velocity was estimated using the scv.tl.velocity function based on the splicing kinetics model, and the scv.tl.velocity_graph was used to calculate the velocity similarity between cells to construct a velocity graph. In terms of dimensionality reduction and visualization, the RNA velocity was projected onto the UMAP embedding space, and the velocity streamlines and velocity arrow plots were drawn to display the state transitions and differentiation directions of cells along the differentiation trajectories. Meanwhile, combined with cell annotation information, the epithelial cells, monocytes, M1 macrophages, and M2 macrophages in the normal group and the tumor group were, respectively, analyzed to evaluate the differences in cell differentiation dynamics under different pathological conditions.

2.16. Molecular Docking and Molecular Dynamics Simulation

The 3D structure of Bap was obtained from the PubChem database, and the crystal structure of RNASE1 was downloaded from the RCSB PDB database under accession code 1DZA, determined by X-ray diffraction at 1.65 Å resolution. Molecular docking was performed using AutoDock Vina. Before docking, the protein was preprocessed by removing water molecules, adding polar hydrogens, and assigning Gasteiger charges. The grid box was set to cover the active pocket of the protein. Binding affinity was evaluated by binding energy (kcal/mol), and ligand–receptor interactions were analyzed using PyMOL, with hydrophobic interactions displayed as black dashed lines. The best docking conformation was used for subsequent all-atom molecular dynamics simulations performed with GROMACS. The system was described by the CHARMM36 force field and TIP3P water model. After energy minimization, 2 ns NVT equilibration, and 2 ns NPT equilibration, a 100 ns production simulation was carried out with a 2 fs time step. Periodic boundary conditions were applied in three dimensions, and long-range electrostatic interactions were calculated using the PME method. The system was kept at 300 K and 1 bar. Postsimulation analyses included RMSD, RMSF, radius of gyration, SASA, and Gibbs free energy landscape to evaluate conformational stability.

2.17. Construction of a Mouse Model of Depression Associated with Bap-Induced Lung Cancer

Six-week-old female SPF-grade A/J mice, six in each group, were adaptively housed for 1 week and then randomly divided into a control group (Con) and a model group (Mod). Mice in the model group were intraperitoneally injected with Bap (prepared in olive oil at a concentration of 10 mg/mL, 0.1 mL per mouse each time), while the Con group was injected with an equal volume of olive oil once a week for 8 consecutive weeks. Previous studies have reported that chronic intraperitoneal administration of Bap can induce lung adenocarcinoma formation During the experiment, body weight was measured weekly, and the general conditions of the mice were observed daily. At the 15th week from the start of modeling, the mice were deeply anesthetized with isoflurane, and the cerebral cortex, lung, liver, and kidney tissues were isolated. This experiment was approved by the Experimental Animal Ethics Committee of Heilongjiang University of Chinese Medicine (Approval No.: 2024090601).

2.18. Behavioral Tests in Mice

In the 14th week from the start of modeling, to evaluate the effects of Bap chemically induced lung cancer model on depressive-like and anxiety-like behaviors in mice, the tail suspension test and elevated plus maze test were conducted under the conditions of quiet environment and uniform light. In the tail suspension test, the tail of the mouse was fixed on the crossbar of the tail suspension device, with the head 20 cm away from the bottom of the cylinder, and the immobility time within 240 s was recorded. The elevated plus maze device consists of two open arms (50 cm × 10 cm) and two closed arms (50 cm × 10 cm × 40 cm), with a central area of 10 cm × 10 cm and 50 cm above the ground. At the beginning of the experiment, the mouse was placed in the central area facing the open arm, and the number of entries into the open arms and the time spent in the open arms within 5 min were recorded. After each mouse was tested, the device was wiped with 75% ethanol.

2.19. Complete Blood Count Test

Twenty-four hours after the last injection, the tail vein blood of mice in each group was collected and placed in EDTA anticoagulant tubes. An automatic blood cell analyzer was used to detect indicators such as white blood cell count (WBC), red blood cell count (RBC), hemoglobin (HGB), platelet count (PLT), platelet distribution width (PDW), mean platelet volume (MPV), and plateletcrit (PCT).

2.20. Immunohistochemical Staining

Lung, liver, and kidney tissues from mice in the Con and Mod groups were fixed in 4% paraformaldehyde for 24 h, routinely dehydrated, and embedded in paraffin. Serial sections were cut at 4 μm thickness. After deparaffinization in xylene and rehydration through graded ethanol, antigen retrieval was performed by microwave heating in citrate buffer (pH 6.0) at 95 °C for 10 min. Endogenous peroxidase activity was blocked by incubating sections in 3% H₂O₂-methanol solution at room temperature for 25 min. Sections were blocked with 5% goat serum for 30 min at room temperature. Lung tissue sections were then incubated overnight at 4 °C with primary antibodies against TTF-1, Napsin A, Ki-67, and Caspase-3, while liver and kidney tissue sections were incubated with primary antibodies against Ki-67 and Caspase-3. After washing three times with PBS, sections were incubated with HRP-conjugated goat anti-rabbit universal secondary antibody for 30 min at room temperature. Immunoreactivity was visualized using DAB substrate, and sections were counterstained with hematoxylin, dehydrated, cleared, and mounted with neutral resin.

2.21. Histopathological Examination

The cerebral cortex and lung tissues from the Con and Mod groups were rapidly dissected and fixed in 4% paraformaldehyde for 24 h. After the tissues were dehydrated with gradient ethanol, cleared with xylene, and embedded in paraffin, continuous sections with a thickness of 5 μm were prepared. After dewaxing and rehydration, the sections were subjected to hematoxylin–eosin (H&E) staining, including hematoxylin staining for 10 min, differentiation with hydrochloric acid ethanol for 30 s, and eosin counterstaining for 4 min. After dehydration and clearing, the sections were mounted with neutral gum. The morphology of cortical neurons, cell arrangement, and changes in Nissl bodies in the cerebral cortex, as well as the integrity of the alveolar wall, inflammatory cell infiltration, and interstitial hyperplasia in the lung tissues were observed under an optical microscope.

2.22. Western Blot

A portion of lung tissue from the Con and Mod groups was placed in precooled RIPA lysis buffer (containing 1% PMSF and phosphatase inhibitors) and fully homogenized on ice using a tissue homogenizer. After centrifugation at 12,000 rpm for 15 min at 4 °C, the supernatant was collected, and the protein concentration was determined using a BCA protein quantification kit. The protein samples were separated by SDS-PAGE gel electrophoresis and then transferred to a PVDF membrane by the wet transfer method. The membrane was blocked with 5% nonfat milk powder at room temperature for 2 h. Then, the primary antibody against RNASE1 was added and incubated overnight at 4 °C. After washing four times with TBST, the HRP-labeled secondary antibody was added and incubated at room temperature for 2 h. After another three washes with TBST, enhanced chemiluminescence substrate was used for development, and images were captured using a gel imaging system. β-actin was used as an internal reference, and semiquantitative analysis of the band gray values was performed using ImageJ software.

2.23. Statistical Analysis

All statistical analyses and visualizations were performed using R software (version 4.4.1). A comprehensive description of the statistical tests employed can be found in the corresponding bioinformatics methods section and Figure legends. The specific research roadmap is shown in Figure .

1.

Research roadmap.

3. Results

3.1. A Total of 374 Potential Target Genes of Benzopyrene Were Identified

The molecular structure of benzopyrene was determined (Figure A). We employed an integrated strategy that combined three complementary databasesChEMBL, SEA, and SwissTargetPredictionto systematically predict its potential targets. After merging the outputs and removing duplicates, we identified 374 potential target genes for benzopyrene (Figure B, Table S1).

2.

Comprehensive information diagram of Bap. (A) Chemical structure of Bap; (B) Target information of Bap in ChEMBL, SEA, and SwissTargetPrediction.

3.2. 52 High-Confidence Depression-Related Characteristic Genes Were Identified

The present study employed a comprehensive bioinformatics approach to identify depression-related characteristic genes. Leveraging the GeneCards database, 15,337 genes with a relevance score greater than 1 for the keywords “Depression” and “Major Depressive Disorder” were initially identified. Further, 1,803 key genes were collected from the OMIM database after merging and deduplicating. Two publicly available MDD expression data sets, GSE98793 and GSE32280, were merged and batch effects were removed, as evidenced by the improved data distribution in the PCA plot (Figure A). Differential expression analysis of the processed data revealed 1,766 merge DEGs, including 812 upregulated and 954 downregulated genes (Figure B). The intersection of the GeneCards, OMIM, and differentially expressed gene sets yielded 52 depression-related characteristic genes, as depicted in the Venn diagram (Figure C, Table S2).

3.

Collection of related targets for depression. (A) PCA plots of genes before and after normalization in two GEO data sets. (B) Volcano plot of differentially expressed genes between MDD patients and healthy individuals. (C) Venn diagram of the intersection genes between depression-related genes retrieved from GeneCards and OMIM and the differentially expressed genes in MDD.

3.3. Two Depression-Related Molecular Subtypes of LUAD with Significant Prognostic Differences Were Identified through Consensus Clustering Based on Depression-Related Genes

To determine the optimal number of clusters, we comprehensively evaluated the silhouette coefficient and the CH index within the range of k = 2 to k = 9. Based on the principle that both the silhouette coefficient and the CH index were highest and the difference in Kaplan–Meier survival curves was significant, we selected k = 2 as the optimal number of clusters, consensus clustering subtype output is shown in Table S3. Under this classification scheme, Cluster 1 included 370 patients, while Cluster 2 included 143 patients (Figure A-D, Supplementary Figure S1A-C). The UMAP visualization revealed clear separation between the two clusters (Figure E). The two subtypes exhibited markedly different molecular characteristics: Cluster 1 was characterized by favorable prognosis, whereas Cluster 2 showed stronger depression-related transcriptomic features and poor outcomes. Importantly, Kaplan–Meier analysis confirmed a significant difference in disease-free survival (DFS) between the clusters (Figure F). These clusters represent depression-related molecular subtypes inferred from gene expression patterns rather than clinically diagnosed depression subgroups. These findings underscore the critical role of the 52 depression-related genes in the progression of LUAD and suggest that depression-related transcriptomic features may be associated with LUAD prognosis.

4.

Identifying clusters associated with depression-related LUAD patients. (A) Consensus CDF description of the PAM algorithm; evaluating the number of clusters from 2 to 9 to determine the optimal k = 2. (B) Delta area plot shows the relative change in area under the CDF curve for k and k-1. (C) Tracking plot demonstrates the change in the class to which the sample belongs for different k. (D) Consensus matrix for clustering number k = 2. (E) UMAP plot of the two subtypes obtained by consensus clustering. (F) Kaplan–Meier survival analysis of two clusters.

3.4. Two Depression-Related Molecular Subtypes of LUAD Reveal 206 Core Regulatory Genes

The limma test identified 2,040 differentially expressed genes across two depression-related clusters, comprising 954 upregulated and 1,086 downregulated genes (Figure A, Table S4). Subsequently, WGCNA was applied to identify coexpression gene modules within these clusters, excluding the outlier sample TCGA–44–2668–01. A soft threshold of β = 5 was chosen to achieve a scale-free R² value of 0.85 (Figure B). The dynamic tree-cutting algorithm revealed 12 distinct gene modules (Figure C). WGCNA module gene membership is shown in Table S5. Among these, the ME blue and ME brown modules demonstrated significant correlations within the clusters, with correlation coefficients exceeding |R| > 0.3 and P-values < 0.01 (Figure D-E). These modules encompassed 1741 genes. The intersection of these genes with the differentially expressed genes identified by the limma test yielded 206 genes associated with depression in LUAD, which were selected for further analysis (Figure F). The key modules and hub genes identified across the three different cutoffs (6000, 5000, 8000) showed high consistency, confirming that the WGCNA results were not substantially influenced by the specific 6000-gene threshold (Figure S2A-B).

5.

Identification of key genes in two LUAD subgroups distinguished by depressive features. (A) A volcano plot showing the differentially expressed genes between the two subgroups, Cluster1 and Cluster2. (B) Scale independence and mean connectivity in the WGCNA analysis. (C) A clustering dendrogram of genes. (D) A heatmap depicting the associations between gene modules and two subgroups. (E) A scatter plot presenting the relationship between ME blue, ME brown, and gene significance. (F) A Venn diagram indicating the overlap between module genes and differentially expressed genes.

3.5. Depression-Related Gene Signature-Based Prognostic Model for Constructed Based on Multiqueue Machine Learning Significantly Improves the Prognostic Prediction Efficacy of LUAD

The present study employed the 206 depression-related feature genes to develop a prognostic prediction model. Univariate Cox regression analysis of the TCGA-LUAD cohort identified 69 core genes (Table S6) significantly associated with hazard ratio (HR) and gene expression.

To develop a robust prognostic model, we utilized four large cohort data sets as validation sets within a machine learning-based ensemble framework. Employing “leave-one-out cross-validation” (LOOCV), we fitted 117 prediction models and computed the C-index for each in both training and validation sets (Figure A). Machine learning-based risk scores are shown in Table S7. The optimal model, integrating StepCox­[forward] and Enet­[α=0.3], achieved the highest average C-index of 0.65, termed the DRPM. We classified samples into high- and low-risk groups using the median risk score from the DRPM model and performed a log-rank survival analysis (Figure B). Across all five cohorts, the high-risk group correlated significantly with poor prognosis. In the TCGA-LUAD cohort, the area under the curve (AUC) for 1-year, 3-year, and 5-year survival were 0.68, 0.66, and 0.60, respectively, demonstrating the DRPM model’s predictive strength and robustness (Figure S3A-C). We also assessed the 1-year, 3-year, and 5-year ROC curves for the five LUAD cohort models (Figure C).

6.

Development and evaluation of the DRPM prediction model. (A) Construction of 117 prediction models from five large-scale cohorts and the corresponding heatmap of C-index. (B) Kaplan–Meier curves between the high-risk and low-risk groups defined by the prediction model. (C) ROC curves of the prediction performance at the 1-year, 2-year, and 3-year time points in five large-scale cohorts. (D) Univariate regression analysis results of the prediction model in five independent cohorts.

We performed a meta-analysis of univariate Cox regression across five large cohorts using the optimal model, depicted in Figure D. The meta-analysis revealed HR values of 1.81, 5.41, 2.48, 1.35, and 2.02 for random-effects and fixed-effects models, all exceeding 1, with p-values under 0.05, demonstrating the model’s predictive efficacy. To assess the DRPM model’s superiority, we compared it with existing lung cancer prediction models, including 24 LUAD subtype cases, by calculating risk scores for five cohorts, resulting in a heatmap of HR values (Figure S4A). The DRPM model consistently correlated with poor prognosis across all cohorts, with HR > 1 and P < 0.01. Moreover, the DRPM model’s C-index ranked high among all five cohorts (Figure S4B), and its 1-year, 3-year, and 5-year AUC values consistently ranked in the upper-middle tier (Figure S4C-E). These results indicate that the DRPM model’s predictive accuracy is comparable to existing models.

3.6. Functional Enrichment Analysis Reveals the Molecular Association between Two Depression-Related Molecular Subtypes and Systemic Immune Dysregulation

The core genes identified by the optimal model in our study appear to play regulatory roles in the depressive characteristics of patients with LUAD. To investigate these potential functions, we performed GO and KEGG enrichment analyses on the 69 screened core genes. Functional analysis of the differentially expressed genes in the depressive subtypes of LUAD patients revealed significant enrichment in pathways related to immune regulation and antigen presentation. GO analysis indicated that these genes were concentrated in molecular functions such as MHC class II protein complex assembly, antigen processing and presentation of exogenous peptide, and MHC protein complex. KEGG analysis further showed that the differentially expressed genes were mainly involved in immune system diseases, including Asthma, Allograft rejection, and Autoimmune thyroid disease, as well as immune-related pathways such as Antigen processing and presentation, Th1 and Th2 cell differentiation, and Th17 cell differentiation. These findings suggest that the depressive subtypes of LUAD may be closely associated with systemic immune dysregulation (Figure ).

7.

GO and KEGG analyses of core genes. (A) GO enrichment results of the 69 core genes screened by machine learning. (B) KEGG enrichment results of the genes among the 69 core genes screened by machine learning.

3.7. Two Depression-Related Gene Signature-Based Molecular Subtypes of LUAD Exhibit Significantly Different Patterns of Immune Cell Infiltration

This study utilized the ssGSEA algorithm to assess immune infiltration of 28 immune cell types in the TCGA-LUAD cohort postconsensus clustering, examining their roles in LUAD pathogenesis and progression across two depression-related gene signature-based molecular subtypes of LUAD. Findings indicated significant enrichment of immune cells, including Activated B cells, Activated CD4 T cells, Central memory CD4 T cells, Effector memory CD8 T cells, Type 2 T helper cells, CD56bright natural killer cells, Macrophages, Monocytes, and Natural killer cells, in Cluster 2, with differential enrichment in Cluster 1 (Figure ). These results suggest future research should explore the interactions between immune microenvironment dynamics and depressive complications, potentially offering new strategies to target depressive-like behaviors in LUAD patients.

8.

Violin plots showing the differences in 28 types of immune infiltration between two clusters of LUAD.

3.8. Elevated Expression of RNASE1 in Lung Tissue May Increase the Risk of Depression

The intersection of the 69 depression-related prognostic genes in LUAD patients screened by machine learning and the Bap-related targets was taken, and a total of 5 overlapping genes were obtained (Figure S5), namely CLEC7A, NEK2, RNASE1, ITGAL, and CPA3. The eQTL data of the above genes in lung tissues were extracted for Mendelian randomization analysis. The results showed that 17 SNPs were included for the CLEC7A gene, 8 SNPs for the RNASE1 gene, and there were no eligible instrumental variables for the other three genes (Table , Table S8). MR analysis showed that there was no significant causal association between the expression level of the CLEC7A gene in lung tissues and the risk of depression (IVW: OR = 1.000, 95% CI: 0.996–1.004, P = 0.949). There was a significant positive correlation between the expression level of the RNASE1 gene in lung tissues and the risk of depression (IVW: OR = 1.008, 95% CI: 1.002–1.014, P = 0.005), suggesting that increased expression of RNASE1 in lung tissues may increase the risk of depression.

2. Mendelian Randomization Estimates for the Association of RNASE1 and CLEC7A Expression with Depression Risk.

exposure	method	outcome	nsnp	b	se	pval	OR_CI	P_pleiotropy	Q_pval
RNASE1	IVW	depression	8	0.0079	0.0028	0.0047	1.008 (1.002–1.014)	0.56	0.99
CLEC7A	IVW	depression	17	0.0001	0.0020	0.9490	1.000 (0.996–1.004)	0.92	1

3.9. Single-Cell Sequencing Reveals the Clustering of Six Cell Subpopulations in Lung Cancer Tissues and the Abnormal Expression Pattern of RNASE1

A total of 5,270 cells and 6,638 genes were included in the single-cell transcriptomic analysis. The quality control, dimensionality reduction, and clustering analysis of the data are shown in Figure S6. The t-SNE dimensionality reduction clustering results showed that both types of tissues were successfully clustered into a total of six immune and parenchymal cell subpopulations, including T cells, NK cells, monocytes, macrophages, epithelial cells, and endothelial cells (Figure A). Moreover, each subpopulation was clearly identified for its cell type through classic markers (such as CD3D/CD3E for T cells, S100A8/S100A9 for monocytes, APOE/MSR1 for macrophages, NKG7/GNLY for NK cells, PECAM1/ENG for endothelial cells, and EPCAM/KRT18 for epithelial cells, etc.) (Figure B), which confirmed the reliability of the clustering. Further t-SNE visualization analysis of RNASE1 expression revealed that the overall expression level of RNASE1 in tumor tissues was significantly higher than that in normal tissues. Additionally, the cells with high RNASE1 expression showed a more extensive distribution pattern in tumor tissues. In normal tissues, the expression was mainly in epithelial cells, while in tumor tissues, it was also observed in monocytes and macrophages (Figure C). The external data set contained a total of 118,737 cells and 26,727 genes, and yielded results similar to those described above, as shown in Figure S7. These findings confirm that our marker gene selection was appropriate and that the RNASE1 expression distribution across cell types was faithfully reproduced.

9.

Single-cell transcriptomic t-SNE clustering plots of lung cancer and adjacent normal lung tissues. (A) t-SNE dimensionality reduction and clustering plots; (B) Bubble plots of the expression of classical markers in each cell subset; (C) t-SNE visualization plots of RNASE1 expression in tumor and normal lung tissues.

3.10. Characteristics of the Communication Network among Six Cell Subsets in Lung Cancer Tissues and the Regulatory Role of RNASE1 in Cell–Cell Interactions

The cell communication analysis system dissected the ligand–receptor interaction network of six types of cell subsets in lung cancer tissues. The results showed that there was extensive and active signal communication among cell subsets. Macrophages were the core signal-sending cells, and monocytes and epithelial cells were also key signal output sources, forming a complex intercellular regulatory network in the tumor microenvironment (Figure A,B). Global pathway analysis identified a total of 17 significantly enriched signaling pathways. Among them, MIF, SPP1, and GALECTIN were the core pathways regulating intercellular interactions, and macrophages had the highest communication intensity with T cells, NK cells, epithelial cells, and endothelial cells (Figure C,D). Further detailed analysis of the MIF pathway indicated that key ligand–receptor pairs such as MIF–CD74–CXCR4 and MIF–CD74–CD44 showed a high communication probability among multiple cell types, confirming that the MIF pathway is the core molecular axis mediating immune regulation in the lung cancer tumor microenvironment (Figure E). Differential regulation analysis after virtual knockout of RNASE1 showed that the absence of RNASE1 could significantly reshape the intercellular communication network. The expression regulation of immune-related molecules such as KLRB1, CCL5, and CD7 changed significantly, suggesting that RNASE1 may participate in the maintenance of homeostasis and malignant progression of the lung cancer tumor microenvironment by regulating core pathways such as MIF and the interactions between immune cells (Figure F).

10.

Characteristics of the intercellular communication network in lung cancer tissues and the regulatory effect of virtual knockout of RNASE1 on cell interactions. (A) Visualization of the global intercellular communication network. The left panel shows the number of interactions, and the right panel shows the interaction intensity. (B) Heatmap of the number and intensity of intercellular interactions. (C) Subdiagrams of specific communication networks among cell subsets. (D) The cell interaction network of the MIF signaling pathway (left) and the circular diagram of pathway contribution (right). (E) Scatter plot of the interaction probability of key ligand–receptor pairs in the MIF pathway among different cell pairs. (F) Volcano plot of differential regulation after virtual knockout of RNASE1.

3.11. Single-Cell Trajectory and RNA Velocity Analysis Reveal the Dynamic Upregulation of RNASE1 during the Differentiation of M2 Macrophages in Lung Adenocarcinoma

A total of 2,283 cells were included in the pseudotime analysis (1,312 monocytes, 616 M2 macrophages, and 355 M1 macrophages). The cell trajectory was divided into 5 states, with a pseudotime range of 0–19.78. Monocytes were mainly distributed in the early stage of pseudotime (mean 9.08), M1 macrophages in the middle and late stages (mean 11.20), and M2 macrophages in the late stage (mean 17.30), suggesting a differentiation direction of monocytes → M1/M2 macrophages. The cell differentiation trajectory maps colored by pseudotime gradient and by cell type are shown in Figure A and B.

11.

Analysis of the pseudotemporal differentiation trajectory and RNA velocity of macrophages in lung adenocarcinoma tissues. (A) Cell differentiation trajectory diagram colored by pseudotime gradient, where colors represent the cell differentiation process; (B) Differentiation trajectory colored by cell type; (C) Scatter plot showing the relative expression levels of RNASE1 in different cell states along the pseudotime gradient; (D) Heatmap presenting the expression characteristics of macrophage polarization marker genes (TNF, CD86, CD163, IL1B, TLR2, etc.) in each cell subgroup; (E) RNA velocity differentiation trajectory of epithelial and monocyte/macrophage subgroups in tumor tissues; (F) RNA velocity differentiation trajectory of various cell types in the paracancerous tissues of lung adenocarcinoma.

RNASE1 had the highest expression percentage in M2 macrophages, with an average expression level of 1.60 and a maximum expression level of 67; the average expression level was 0.418 in monocytes and 0.144 in M1 macrophages. The expression of RNASE1, M2 marker genes (MSR1, CD163), and inflammatory factors (TNF, IL1B) changed significantly along the pseudotime, while the change with M1 marker genes (CD86, TLR2) was relatively weak, suggesting that the upregulation of RNASE1 expression was consistent with the differentiation direction of monocytes into M2 macrophages. The scatter plot of the relative expression levels of RNASE1 in different cell states along the pseudotime and the heat map of the expression characteristics of macrophage polarization marker genes are shown in Figure C and D.

The results of RNA velocity analysis showed that the proportion of M2 macrophages in the tumor group was higher than that in the normal group, while the proportion of monocytes was the opposite. The expression level of RNASE1 in the tumor group was approximately 4.4 times that in the normal group. The comparison of velocities between cell types showed that in the tumor group, M1 macrophages had the fastest velocity, followed by monocytes and M2 macrophages; in the normal group, epithelial cells had the fastest velocity, followed by monocytes. In the tumor microenvironment, the proportion of M2 macrophages increased, the expression of RNASE1 was significantly upregulated, and the transcriptional dynamics of macrophages were enhanced, suggesting that the tumor microenvironment may promote the upregulation of RNASE1 by reprogramming the transcriptional state of macrophages, thereby driving the occurrence of lung adenocarcinoma-related depression. The RNA velocity differentiation trajectories of various cell types are shown in Figure E and F.

3.12. Bap Exhibits High Binding Affinity for RNASE1

Molecular docking results (Figure A) showed that Bap had strong binding affinity toward the RNASE1 protein, with a minimum binding energy of −8.5 kcal/mol. Bap formed stable hydrophobic interactions with key residues TYR-215, VAL-143, PHE-159, and PHE-220. 100 ns molecular dynamics simulations confirmed that the Bap–RNASE1 complex remained structurally stable. The protein backbone RMSD plateaued after equilibration (Figure B). RMSF values of key binding residues were significantly lower than those of other regions (Figure C). The radius of gyration and SASA remained steady during the simulation (Figure D,E). The Gibbs free energy landscape showed a single dominant low-energy basin (Figure F). These computational predictions provide structural evidence for a stable Bap–RNASE1 interaction, primarily through hydrophobic contacts, but do not directly demonstrate changes in RNASE1 enzymatic activity or biological function. Further experimental validation is required to determine the functional consequences of this binding.

12.

Molecular docking of Bap with RNASE1. (A) The molecular docking conformation of Bap bound to RNASE1. (B) Root-mean-square deviation (RMSD) of the protein backbone during the 100 ns MD simulation. (C) Root-mean-square fluctuation (RMSF) of RNASE1 residues. Red boxes highlight the key binding residues (VAL-143, PHE-159, TYR-215, PHE-220) with reduced flexibility. (D) Radius of gyration (Rg) of the protein over the simulation time. (E) Solvent-accessible surface area (SASA) of the protein during the simulation. (F) Gibbs free energy landscape of the Bap–RNASE1 complex, constructed using the first two principal components (PC1 and PC2) from the simulation trajectory. The color bar represents relative free energy (kJ/mol), with blue indicating low-energy, stable conformational states.

3.13. Bap Exposure Induces Depression-Like Behaviors in Mice Accompanied by Elevations in White Blood Cell Count, Mean Platelet Volume, and Platelet Distribution Width

By the 12th week of model establishment, mice in the Con group displayed shiny, smooth fur and normal neurological states. In contrast, mice in the Mod group exhibited symptoms such as dry fur, dyspnea, and reduced activity (Figure A). Behavioral tests at week 14 revealed that, compared to the Con group, the Mod group showed a significantly prolonged immobility time during the tail suspension test (148.67 ± 31.68 s vs 81.17 ± 15.74 s, P < 0.01) (Figure D, Table S9). In the elevated plus-maze test, the Mod group had fewer entries into the open arms (2.0 ± 0.89 times vs 4.5 ± 1.05 times, P < 0.01) (Figure B, Table S10) and spent less time in the open arms (24.67 ± 4.63 s vs 54.83 ± 6.18 s, P < 0.001) (Figure C). These findings suggest that mice in the Mod group exhibited significant depression-like and anxiety-like behaviors. Hematological tests indicated that, compared to the Con group, the Mod group had significantly increased white blood cell count (WBC) and mean platelet volume (MPV) (P < 0.01) (Figure E,J, Table S11), as well as an increased platelet distribution width (PDW) (P < 0.05) (Figure I). There were no significant changes in red blood cell count (RBC), hemoglobin (HGB), platelets (PLT), and plateletcrit (PCT) (P > 0.05) (Figure F-H,K).

13.

Changes in behavioral and blood routine indexes of mice in the Mod and Con at 12 weeks. (A) Representative appearance of mice in the two groups at 12 weeks; (B-C) Results of elevated plus maze test, showing the number of open arm entries and open arm time, respectively; (D) Results of tail suspension test, showing immobility time; (E-K) Results of blood routine examination, including white blood cell count (WBC), red blood cell count (RBC), hemoglobin (HGB), platelet count (PLT), platelet distribution width (PDW), mean platelet volume (MPV), and plateletcrit (PCT). Data are presented as mean ± standard deviation, n = 6. Compared with the Con group, *P < 0.05, **P < 0.01, ns indicates no statistical difference.

3.14. Immunohistochemistry Validation of Bap-Induced Formation of Lung Adenocarcinoma in Mice

During tissue sampling at week 15, obvious white protrusions were found in the lungs of mice in the Mod group, while no abnormalities were observed in the Con group (Figure S8). To verify whether the intraperitoneal injection of Bap successfully induced the formation of lung adenocarcinoma and evaluate its potential effects on liver and kidney tissues, immunohistochemical staining was performed on lung, liver, and kidney tissues in this study (Figure ). The results showed that in the lung tumor tissues of the Bap Mod group, TTF-1 was strongly positively expressed in the nucleus and Napsin A was positively expressed in the cytoplasm, while the positive signals in the lung tissues of the Con group were weaker. TTF-1 and Napsin A are highly specific markers of lung adenocarcinoma, and their double-positive expression confirmed that the lung lesions induced by Bap were lung adenocarcinoma. Evaluation of the Ki-67 proliferation index showed that the positive rate of Ki-67 in the lung tumor area of the Bap Mod group was higher than that in the normal lung tissues of the Con group. Although the positive rate of Ki-67 in the liver and kidney tissues also increased compared with the Con group, the increase was lower than that in the lung tissues. Caspase-3 apoptosis detection showed that the positive rate of Caspase-3 in the lung tumor area of the Bap Mod group was lower than that in the normal lung tissues of the Con group. The positive rate of Caspase-3 in the liver and kidney tissues also decreased, but the decrease was lower than that in the lung tissues. The above results indicate that the intraperitoneal injection of Bap successfully induced the formation of lung adenocarcinoma in mice, and the lungs are the main target organs of Bap. Although certain degrees of changes in proliferation and apoptosis were observed in the liver and kidney tissues, the amplitude of these changes was weaker than that in the lung tissues, suggesting that the potential effects of intraperitoneal injection of Bap on the liver and kidneys cannot be ignored, but the lungs are still the main effector organs of pathological changes induced by Bap.

14.

Immunohistochemistry validation of Bap-induced lung adenocarcinoma formation and organ-specific changes in proliferation and apoptosis in lung, liver, and kidney tissues.

3.15. Inflammation, Lung and Brain Damage, and Upregulation of RNASE1 in Mice with Depression Model Related to Lung Cancer Induced by Benzopyrene

Pulmonary histopathology showed that the lung tissue structure of mice in the Con group was normal, with intact alveolar septa and well-expanded alveolar spaces, while in the Mod group, alveolar damage, enlargement, and fusion were observed (Figure A). Cerebral cortex histopathology showed that the cerebral cortex structure of mice in the Con group was clear, with regular arrangement and intact morphology of neurons, while in the Mod group, the cerebral cortex structure was loose, neurons were arranged disorderly, cell nuclei were pyknotic or even fragmented, and interstitial edema and vacuolation were present (Figure B). Western blot results showed that compared with the Con group, the protein expression level of RNASE1 in the lung tissue of mice in the Mod group was significantly increased (P < 0.01) (Figure C), Gray values of RNASE1 and β-actin protein bands in lung tissue of mice in the Con and Mod groups are shown in Table S12, and the original Western blot images are provided in Supporting Information 1.

15.

Changes in pulmonary histopathological morphology and RNASE1 expression levels in the Mod group and Con group. (A) Hematoxylin–eosin (H&E) staining of lung tissue (×100 magnification); (B) Hematoxylin–eosin (H&E) staining of brain tissue (×100 magnification); (C) Protein expression level of RNASE1 in lung tissue detected by Western blot and its grayscale analysis.

4. Discussion

As the number of lung cancer survivors rises, associated mental health issues, − notably depression and anxiety affecting over 40% of survivors, demand attention. These emotional disorders significantly diminish quality of life and hinder long-term recovery. Exposure to environmental carcinogens, particularly Bap, a polycyclic aromatic hydrocarbon commonly present in tobacco smoke and air pollution, has been increasingly recognized not only as a potent inducer of lung cancer development but also as a potential contributing factor to neuropsychiatric sequelae in cancer patients. ,, Epidemiological evidence suggests that individuals with long-term exposure to polycyclic aromatic hydrocarbons have a higher incidence of depressive symptoms. However, the molecular mechanisms among Bap exposure, LUAD progression, and the development of depression remain poorly understood. We employed bioinformatics and machine learning to identify molecular markers of lung cancer-induced depression, facilitating early and precise diagnosis. This approach aims to enhance personalized treatment and improve long-term quality of life for patients.

LUAD was classified into two distinct molecular subtypes based on the expression profile of depression-related genes. Patients in Cluster 2 exhibited a poorer prognosis and increased susceptibility to depression. We developed a prognostic model for DRPM model that integrated key differentially expressed genes identified through differential analysis and WGCNA. This model demonstrated robust predictive performance and risk stratification ability when validated across multiple external cohorts. Further analysis suggested that the 69 genes included in the DRPM model may contribute to the pathogenesis of depression through mechanisms involving neuroinflammation, aberrant antigen presentation, autoimmune responses, and Th1/Th2 immune imbalance. These results align with numerous studies in the field. Research indicates that patients with advanced nonsmall cell lung cancer who exhibit both depression (PHQ-9 ≥ 8) and inflammation (ALI < 24) at baseline experience worsening depressive symptoms over an 8-month follow-up. Inflammatory cytokines may contribute to anxiety and depression in these patients. Elevated serum levels of TNF-α, IL-1β, and IL-17 are strongly linked to heightened anxiety and depression risk among survivors. Moreover, increased C-reactive protein (CRP) and decreased albumin levels significantly correlate with depression in lung cancer patients, with more severe symptoms when both markers are abnormal. Another study found that patients with EGFR mutations had significantly lower serum CRP levels than those with the wild-type. Nearly a quarter of the impact of EGFR mutations on depression is mediated by CRP regulation, suggesting that EGFR mutations might reduce depression risk by downregulating inflammatory pathways like IL-6/CRP and mitigating neuroinflammatory responses. These findings underscore the concept of “inflammatory depression” in lung cancer patients, highlighting the pivotal role of inflammation in cancer-related depression.

By intersecting the depression-related prognostic genes identified by machine learning with the Bap targets predicted by network toxicology, we obtained five overlapping genes: CLEC7A, NEK2, RNASE1, ITGAL, and CPA3. Among them, RNASE1 emerged as the most promising candidate molecule linking environmental exposure and comorbid depression. A two-sample Mendelian randomization analysis using lung tissue eQTL data provided causal evidence, indicating that increased RNASE1 expression significantly elevated the risk of depression (IVW: OR = 1.008, 95% CI: 1.002–1.014, P = 0.005), while CLEC7A did not show a significant causal association. This finding establishes RNASE1 as a genetically validated key mediator in the exposure-outcome pathway.

RNASE1, also known as human pancreatic ribonuclease, is a secreted endoribonuclease that has traditionally been considered to be mainly responsible for the degradation of dietary RNA in the digestive tract. RNASE1 is mainly expressed by vascular endothelial cells, participates in extracellular RNA degradation, and plays a significant role in maintaining vascular homeostasis, regulating immune responses, and intercellular signal communication. Recent evidence indicates that RNASE1 also has an oncogenic function in lung cancer. It binds to and activates wild-type ALK, triggering downstream signaling pathways that enhance proliferation, migration, and tumorigenesis. This RNASE1-driven ALK activation (RDAA) is observed in about 8.5–10.4% of nonsmall cell lung cancer patients, who exhibit significant sensitivity to FDA-approved ALK inhibitors. RNASE1 has also been identified as a hub gene associated with gefitinib sensitivity in lung adenocarcinoma, indicating its potential use as a predictor of response to targeted therapy. In extracellular RNA signaling, RNASE1 acts as a key immunomodulator. Under pro-inflammatory or pro-thrombotic conditions, endothelial cells release RNASE1 to degrade extracellular RNA, which otherwise functions as a potent endogenous danger signal and activates pattern-recognition receptors such as Toll-like receptors, thereby amplifying chronic inflammation. Moreover, RNASE1 secreted by tumor cells can be internalized by CD8+ T cells via endocytosis, where it interacts with and inactivates STAT1. This interaction results in the reduced expression of effector cytokines, including IFN-γ and IL-2, while increasing the expression of immune checkpoint proteins such as PD-1, TIM-3, and LAG-3. Consequently, T cell dysfunction is induced, aiding tumors in evading immune destruction. This immunomodulatory role places RNASE1 at the interface between inflammation and disease pathogenesis. Notably, blood transcriptomic analysis identified RNASE1 as one of four candidate biomarkers with state-dependent expression in patients with late-onset major depression, demonstrating high discriminant validity with a sensitivity of 91.3% and a specificity of 87.5%, and this result was further validated in a depression animal model. Research has shown that extracellular RNA can act as damage-associated molecular patterns (DAMPs) to activate pattern recognition receptors such as Toll-like receptors, inducing the release of inflammatory factors. The degradation of extracellular RNA mediated by RNASE1 may be involved in limiting abnormal inflammatory activation. The occurrence and development of depression are closely related to chronic low-grade inflammation. Elevated levels of pro-inflammatory cytokines such as IL-6 and TNF-α can promote neuroinflammation, oxidative stress, and damage to neural plasticity. Additionally, inflammatory activation can enhance the activity of indoleamine 2,3-dioxygenase (IDO), promoting the conversion of tryptophan to the kynurenine pathway, resulting in reduced synthesis of 5-hydroxytryptamine (5-HT) and the production of neurotoxic metabolites, thus participating in the imbalance of neurotransmitters associated with depression. Therefore, abnormal expression of RNASE1 may further regulate the IDO-kynurenine pathway and the neuroimmune environment by affecting extracellular RNA homeostasis, inflammatory signal transduction, and oxidative stress levels, and thereby participate in the occurrence and development of depression. Additionally, RNASE1 mRNA levels in peripheral blood mononuclear cells from heart failure patients are significantly higher than those in healthy controls, underscoring its broader relevance to systemic inflammatory states.

Single-cell transcriptomic analysis revealed that RNASE1 was not only upregulated in tumor epithelial cells within the tumor microenvironment of lung adenocarcinoma but also highly expressed in monocytes and macrophages. This cell-type-specific expression pattern suggests that RNASE1 may be involved in both tumor cell-autonomous processes and the reshaping of the immune landscape. Molecular docking and molecular dynamics simulations predicted a strong binding affinity between Bap and the RNASE1 protein, involving hydrophobic interactions with key residues TYR-215, VAL-143, PHE-159, and PHE-220. These computational findings provide atomic-level structural evidence for the Bap–RNASE1 interaction and offer a testable hypothesis for how exposure to environmental carcinogens may perturb the function or expression of RNASE1. However, as these results are derived from computational predictions, further experimental validation is required to confirm the functional consequences of this binding. In our in vivo experiments, mice exposed to Bap exhibited significant depression-like behaviors, characterized by an extended immobility time in the tail suspension test and a reduced number of entries and time spent in the open arms of the elevated plus maze. These behavioral changes were accompanied by an increase in peripheral blood leukocyte count and mean platelet volume, indicating systemic inflammation activation, and a significant upregulation of RNASE1 protein levels in lung tissue.

These findings, supported by computational evidence from systems toxicology analysis, molecular docking, and molecular dynamics simulations, together with association-based evidence from prognostic analysis, Mendelian randomization, and single-cell RNA-sequencing, and further validated by experimental evidence from the Bap-exposed mouse model for behavioral changes and RNASE1 upregulation, establish a coherent exposure-response pathway in which Bap exposure leads to the upregulation of RNASE1 in lung tissue and immune cells, which in turn drives systemic inflammation and ultimately manifests as a depression-like phenotype. The discovery that RNASE1 serves as a key mediator linking Bap exposure to depression associated with lung adenocarcinoma holds several important implications. First, it expands our understanding of the lung-brain axis in cancer, demonstrating that exposure to lung carcinogens can have remote neuropsychiatric consequences through specific molecular mediators. Second, RNASE1 can serve as a clinically actionable biomarker for identifying lung adenocarcinoma patients at increased risk of depression, especially those with a significant history of environmental exposure. Third, the RNASE1-extracellular RNA-inflammation axis represents a potential therapeutic target for preventing or ameliorating cancer-associated depression. Future research should explore whether pharmacological regulation of RNASE1 activity or its downstream pathways can alleviate depressive symptoms without compromising antitumor immunity.

However, this study still has several limitations. First, the transcriptomic data were from public databases, and although batch effects were corrected, technical biases and missing clinical information may persist. In particular, the TCGA-LUAD cohort lacks psychiatric assessment records or clinically confirmed depression diagnoses; thus, the identified subtypes represent depression-related molecular phenotypes rather than clinically confirmed depression subgroups. Second, the mouse model was not subjected to multiomics profiling, limiting mechanistic interpretation. Third, the consensus clustering algorithm itself may not fully capture the complexity of clinical depression, and the resulting subtypes therefore require validation in independent cohorts with standardized depression assessments. Fourth, the scRNA-seq analysis was based on a limited number of samples, and the cell-type-specific expression of RNASE1 needs further validation in larger single-cell cohorts or experimental models. Fifth, intraperitoneal Bap administration may cause systemic exposure to extrapulmonary organs. Although we confirmed lung adenocarcinoma by histopathology and TTF-1/Napsin A immunostaining, we did not systematically examine other abdominal organs, and thus cannot exclude their potential contribution to the observed depressive-like behaviors. Future studies using intratracheal instillation are warranted to more specifically validate the lung-specific contribution of Bap-induced lesions to depressive phenotypes. Finally, the direct physical interaction between Bap and RNASE1 has not yet been experimentally verified by molecular interaction assays, such as surface plasmon resonance (SPR) or cellular thermal shift assay (CETSA), and the downstream mechanisms by which RNASE1 contributes to depression-related phenotypes remain to be fully elucidated. Future studies should combine molecular interaction assays, functional experiments, and conditional knockout models to further validate the proposed mechanism.

Taken together, our study provides preliminary evidence that RNASE1 may be involved in the potential link between Bap exposure and depression-related phenotypes in LUAD. By combining public database–based transcriptomic analysis, systems toxicology analysis, molecular dynamics study, and experimental validation, we identified RNASE1 as a candidate molecule associated with Bap exposure, LUAD prognosis, inflammatory changes, and depression-like behaviors. These findings should be interpreted cautiously because of the absence of clinical depression assessments in TCGA and the need for further biochemical and mechanistic validation. Nevertheless, this work offers a useful hypothesis-generating framework for future studies on environmental carcinogen-associated cancer–depression comorbidity.

5. Conclusions

This study elucidates the molecular link between Bap exposure and lung adenocarcinoma-associated depression through an integrated bioinformatics and network toxicology framework. Consensus clustering based on depression-related genes stratified LUAD patients into two distinct subtypes, with Cluster 2 identified as a high-risk group for depression. Machine learning and Mendelian randomization analyses identified RNASE1 as a key mediator, with elevated RNASE1 expression in lung tissue causally associated with increased depression risk. Single-cell transcriptomics revealed RNASE1 upregulation in tumor epithelial cells, monocytes, and macrophages. Molecular docking and molecular dynamics simulations confirmed high-affinity binding between Bap and RNASE1. In vivo experiments further validated that Bap exposure induces depression-like behaviors and elevates RNASE1 protein levels in lung tissue. These findings establish RNASE1 as a pivotal biomarker linking environmental carcinogen exposure to lung cancer progression and depression pathogenesis, and provide a reusable framework for investigating environmental exposure-associated tumor comorbidities.

The data sets generated and analyzed during this study are derived from publicly available sources, with the original data accessible from The Cancer Genome Atlas Lung Adenocarcinoma (TCGA-LUAD) project (https://portal.gdc.cancer.gov/) and the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/) under accession numbers GSE31210, GSE68465, GSE50081, GSE72094, GSE98793, and GSE32280.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.6c04956.

• Figures S1–S9: Supplementary figures from the manuscript (PDF)

• Raw data and supplementary analyses supporting the findings of this study, including Tables S1–S12 (XLSX)

#.

J.G. and F.H. contributed equally to this work as cofirst authors. J.G. was responsible for conceptualization, methodology, bioinformatics analysis, data curation, and writing the original draft. F.H. participated in conceptualization and was responsible for animal experiments, histopathology, and behavioral tests. K.S. was responsible for hematological analysis, Western blot experiments, and data curation. S.Z. was responsible for behavioral tests, visualization, and formal analysis. Y.X. was responsible for visualization and software support. H.L. and J.G. provided methodology guidance, research resources, and funding support. H.S. and G.Y. were responsible for conceptualization, methodology, data analysis guidance, resources, funding acquisition, supervision, and validation. N.Z. contributed to supervision, project administration, manuscript review and editing, and served as the cocorresponding author. X.W. provided overall supervision, funding acquisition, project administration, manuscript review and editing, and served as the lead corresponding author. All authors have read and approved the final version of the manuscript.

This study was supported by the Lateral Project of Dong’e Ejiao Co., Ltd. (22042240002); the Heilongjiang Provincial Key Research and Development Plan (Grant No. 2022ZX02C04); and the National Multidisciplinary Innovation Team Project in Traditional Chinese Medicine (ZYYCXTD-D-202407).

The authors declare the following competing financial interest(s): Authors Haibin Liu and Jianjun Gu were employed by the company Dong'e Ejiao Co., Ltd. All authors declare that this work was conducted in the absence of any other commercial or financial relationships that could be construed as a potential conflict of interest.