The TCGA_LUAD dataset was obtained from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov). Additionally, the datasets GSE72094, GSE11969, GSE26939, GSE41271, GSE31210, GSE10072, GSE46539, and GSE32863 were downloaded from the Gene Expression Omnibus database (https://ncbi.nlm.nih.gov/gds). All patients included in this study were diagnosed with lung adenocarcinoma, and the datasets used contained detailed clinical information. All datasets were processed using the methodologies previously outlined in our studies (23, 24). Ethical approval was not required since the datasets were publicly available.

Consensus clustering (CC) was performed using the ‘ConsensusClusterPlus’ package in R 4.1.0 with the following parameters: 80% item resampling, 80% gene resampling, 1000 repetitions, and Pearson correlation as the distance metric. Nonnegative matrix factorization (NMF) clustering was conducted using the ‘NMF’ package with the’brunet’algorithm and 100 iterations (25).

The TCGA_LUAD cohort was used as the training set for CRRS development, whereas the datasets GSE72094, GSE41271, and GSE31210 served as validation sets.

The 42 genes co-expressed with CYP4B1 (Supplementary Table 1; correlation coefficient 0.5; P ≤ 0.001) was identified from the TCGA_LUAD dataset. Based on the expression levels of CYP4B1 and its co-expressed genes, and integrating CC and NMF, we stratified the patients in the TCGA_LUAD dataset into two subgroups and identified 38 differentially expressed genes (DEGs) that were significantly associated with prognosis (Supplementary Table 2; |logFC| ≥ 2; Adjusted P value ≤ 0.01). The Least Absolute Shrinkage and Selection Operator (LASSO) Cox regression model was applied to these genes, yielding seven key genes and their coefficients. A risk score for each patient was calculated using the following formula: score = ∑_i Coefficient(Gene i) × Expression(Gene i) (26).

The drug sensitivity analyses were conducted using the “limma” and “pRRophetic” packages in R 4.1.0, considering P < 0.001 and setting the seed to 12345.

A nomogram was developed to quantify patient prognosis based on CRRS, gender, age, tumor stage, T stage, and N stage. This was constructed using the “survival,” “survminer,” “timeROC,” “rms,” and “regplot” packages in R 4.1.0.

Human LUAD cell lines H1299 and PC-9 were obtained from the Cell Bank of the Shanghai Institute of Biological Sciences, Chinese Academy of Sciences. The PC-9 and H1299 human LUAD cell lines were selected to represent distinct genetic backgrounds relevant to LUAD pathogenesis. PC-9 cells harbor an EGFR exon 19 deletion mutation, which is a common driver in LUAD, while H1299 cells are p53-null, representing another frequent genetic alteration. Using these two models strengthens the generalizability of our findings across different LUAD genomic contexts. CYP4B1 knockdown/overexpression lentiviruses were purchased from Shanghai Jikai Biotechnology Co., Ltd. For infection, cells were seeded in 6-well plates and at approximately 60% confluence, were incubated with the lentiviral particles and 8 μg/mL Polybrene for 24 hours. Subsequently, the medium was replaced with fresh complete medium. Stable cell pools were selected using 2 μg/mL puromycin for at least one week.

LUAD cells were seeded into 6-well plates at a density of 500 cells per well and cultured for 2 weeks. The cells were subsequently fixed with cold methanol and stained with crystal violet for colony visualization.

The LUAD samples were obtained from the Pathology Department of Shanxi Provincial People’s Hospital (Taiyuan, China). The use of clinical specimens was authorized by the ethics committee of Shanxi Provincial People’s Hospital (2021-196). All procedures were carried out in accordance with relevant guidelines and regulations. All experimental protocols were approved by the ethics committee of Shanxi Provincial People’s Hospital. The informed consent was obtained from all participants and/or their legal guardian(s).

LUAD and adjacent noncancer tissues were processed into tissue microarrays. The slides were stained with immunohistochemistry (IHC), scanned using a Panoramic MIDI (3DHISTECH, Ltd., Budapest, Hungary), and analyzed with the Panoramic Viewer v. 1.15.3 and Nuclear Quant application for PV v.2.0.0.46136 (3DHISTECH). The H-score was calculated to quantify the protein levels of target genes in the tissues, using the formula: H-score = ∑ (pi × i) =(percentage of weak intensity × 1) + (percentage of moderate intensity × 2) + (percentage of strong intensity × 3), where (i) denotes the intensity of staining, and (pi) is the percentage of stained tumor cells (27). For immunofluorescence (IF), cells were seeded on confocal dishes. Upon reaching ~60% confluence, cells were fixed with 4% paraformaldehyde for 15 min, permeabilized with 0.3% Triton X-100 for 10 min, and blocked with 5% bovine serum albumin (BSA) for 1 hour at room temperature. Cells were then incubated with primary antibodies overnight at 4 °C, followed by incubation with fluorochrome-conjugated secondary antibodies for 1 hour at room temperature in the dark. Nuclei were counterstained with DAPI. Images were captured using a Zeiss LSM 900 confocal laser scanning microscope. Information on the antibodies used for IHC and IF is provided in Supplementary Table 3. For IHC/IF, negative controls were performed by omitting the primary antibody, resulting in no specific staining. Staining of adjacent normal lung tissue served as the positive internal control for CYP4B1.

The cells were plated in six-well plates and subjected to a starvation treatment for 12 h in a medium supplemented with 1% FBS. Following the starvation, the cells were cultured for an additional 24 h in a complete culture medium. Subsequently, the cell cycle distribution was assessed using a cell cycle assay kit (AC12L553, Life-iLab, China) and analyzed by flow cytometry. The cell cycle populations were gated based on DNA content (PI fluorescence) after excluding doublets and debris. Representative gating strategies is provided in Supplementary Figure 3.

The cells were seeded in laser confocal dishes and subjected to a starvation treatment for 12 h using a medium supplemented with 1% FBS. Following this, the cells were cultured in a complete medium for an additional 24 h. The cells were treated with BeyoClick EdU-555 (C0075L, Beyotime, China) to assess cell proliferation and subsequently labeled with DAPI for nuclear visualization. The cells were detected using a confocal laser scanning microscope.

Cells were seeded in 96-well plates at a density of 2,000 cells per well. At the indicated time points (0, 24, 48, 72 hours), 10 μL of CCK-8 reagent (AC11L054, Life-iLab) was added to each well and incubated for 1 hours at 37 °C. The absorbance at 450 nm was measured using a microplate reader (BioTek, USA). Each experiment was performed with at least four replicate wells per group.

Cells in the logarithmic growth phase (PC-9 and H1299) were digested with 0.25% trypsin. After centrifugation at 1000 rpm for 5 minutes, the cells were resuspended in complete medium, counted, and adjusted to a concentration of 3×10⁵ cells/mL. 2 mL of the cell suspension was added to each well of a 6-well plate, gently shaken to ensure uniform distribution of the cells, and then placed in an incubator for 24–48 hours until the cell confluency reached 80%-90%. 1straight line were scratched in the center of the cell monolayer in each well using 200μL sterile pipette tips, with the tips held perpendicular to the wall of the 6-well plate. Subsequently, the wells were rinsed twice with sterile PBS. 2 mL of serum-free medium was added to each well. Under an inverted phase-contrast microscope, 3 fixed fields of view in the scratched area were selected, and images were captured and recorded as 0 h data. After continuous culture in the incubator for 48 hours, images were captured again. The scratch width was measured, and the cell migration rate was calculated using the formula: Migration rate (%) = (Scratch width at 0 h - Scratch width at 48 h)/Scratch width at 0 h × 100%. Each experimental group was set up with 3 replicate wells. To assess collective cell migration, assays were performed in serum-free medium without the use of proliferation inhibitors, acknowledging that the measured “migration rate” may incorporate a component of proliferation at the wound edge.

Matrigel Dilution: Dilute Matrigel with pre-chilled serum-free medium at a ratio of 1:8 (operation on ice), and the diluted Matrigel must be used within 30 minutes. Matrigel Coating: Add 80 μL of diluted Matrigel to the upper chamber of the Transwell insert, ensuring it evenly covers the entire surface of the membrane. Gelation: Place the Transwell insert with coated Matrigel into a 37 °C incubator with 5% CO₂, and incubate for 2–4 hours to allow the Matrigel to polymerize into a solid state. Hydration: Remove the insert from the incubator and carefully aspirate the excess liquid (uncoagulated) from the upper chamber. Add 300 μL of serum-free medium to the upper chamber and 500 μL of complete medium to the lower chamber. Place the insert back into the incubator for hydration for 1–2 hours to balance the environment inside and outside the membrane. After hydration, aspirate the medium from both the upper and lower chambers. Cell Preparation: Digest cells (PC-9 and H1299) in the logarithmic growth phase with trypsin. Resuspend the cells in serum-free medium, then centrifuge and wash them 1–2 times to remove serum. Resuspend the cells again in serum-free medium, count the cells, and adjust the cell density to 2×10⁵ cells/mL. Lower Chamber Liquid Addition: Add 600 μL of complete medium containing 10% FBS to the lower chamber. Upper Chamber Seeding: Carefully add 100 μL of the cell suspension to the upper chamber of the Transwell insert. Culture: Gently place the culture plate into a 37 °C incubator with 5% CO₂ and incubate for 48 hours. Experiment Termination: Carefully remove the Transwell insert and gently rinse it twice with PBS. Fixation: Place the insert into a well containing 4% paraformaldehyde and fix at room temperature for 20–30 minutes. Washing: Take out the insert and gently rinse it 2–3 times with PBS. Staining: Place the insert into a well containing 0.1% crystal violet staining solution and stain at room temperature for 15–30 minutes. Washing Again: Gently rinse the insert with PBS several times until the eluate becomes colorless. Use a wet cotton swab to carefully wipe off the non-migrated cells on the surface of the membrane in the upper chamber. Air-Drying: Invert the insert and allow it to air-dry naturally. Photography and Counting: Photograph the insert and count the number of cells that have migrated through the membrane. Each group was set up with 3 duplicate wells.

Total RNA was extracted from cells using Trizol reagent (Life-iLab, AN51L758) according to the manufacturer’s protocol. Reverse transcription was performed with SuperRT III One-Step Reverse Transcription Kit (Biosharp, BL1020B), followed by cDNA amplification and detection using 2 x qPCR Mix(SYBR Green) (Life-iLab, AN19L918). GAPDH served as the internal loading control.

We cloned the wild-type sequence of 2000 bp upstream of the transcription start site of the human CYP4B1 gene, as well as sequences with mutations in single or multiple potential binding sites of NFIA, into the pGL3-Basic vector individually. The resulting plasmids were co-transfected into 293T cells alongside a Renilla luciferase expression construct. Forty-eight hours after transfection, the cells were harvested and cell lysates were prepared. Luciferase activity was subsequently determined using the Dual-Luciferase Reporter Assay Kit (YEASEN, 1142ES60).

The ChIP assay was performed using the Beyotime ChIP Assay Kit (P2078, China) with PC-9 cells. The ChIP assay was performed strictly in accordance with the manufacturer’s protocol. After immunoprecipitation with NFIA antibody (CST, 69375) or normal rabbit IgG (CST, 2729), the enriched DNA was analyzed by qPCR using primers flanking the putative NFIA binding site on the CYP4B1 promoter and primers for a negative control region. Primer sequences were provided in Supplementary Table 3.

Total proteins were extracted from the cells using RIPA lysis buffer and quantified using a BCA protein quantification kit (AP12L025, Life-iLab). Equal amounts of protein (30 μg) were separated by SDS-PAGE and transferred onto NC membranes. The membranes were blocked with 5% non-fat milk in TBST for 1 hour and then incubated with primary antibodies overnight at 4 °C. After washing, membranes were incubated with HRP-conjugated secondary antibodies for 1 hour at room temperature. Protein bands were visualized using an Enhanced Chemiluminescence (ECL) substrate (AP34L024, Life-iLab) and imaged with a ChemiDoc Imaging System (5200, Tanon). GAPDH was used as a loading control. Information on the antibodies used for Western blotting is provided in Supplementary Table 3.

Six-week-old female nude BALB/c mice were obtained from Beijing Vital River Laboratory Animal Technology Co., Ltd. The cells were mixed with a matrix adhesive in a 1:1 ratio and injected subcutaneously into the dorsal region of the mice, with a total of 5 × 10⁶ cells per mouse (five replicates per group). The tumor volumes were measured starting from day 8 post-injection, with measurements taken every 3 days until the mice were euthanized. Euthanasia was performed using the Animal Multi-functional Anesthesia Machine CO₂ Euthanasia Device (HOPE-MED8160, HePu Industry and Trade Co., Ltd., Tianjin, China). Place the mice in the sealed chamber, open the CO₂ flow valve (using 100% pure CO₂ with the flow rate set to 20% of the container volume per minute), and continue introducing CO₂ until the breathing stops, then maintain this for another 2 minutes to ensure thorough death. The animal research protocol was approved by the ethics committee of Shanxi Provincial People’s Hospital (2021-191). All methods were conducted in accordance with relevant guidelines and regulations, and all procedures adhered strictly to The Animal Research: Reporting of In Vivo Experiments (ARRIVE) guidelines.

The bioinformatics section utilized R software version 4.1.0 for statistical analysis, data evaluation, and visualization. The Wilcoxon rank-sum test is used to assess differences in the expression of the target gene between the two groups. The log-rank test is used to analyze differences in survival between the two groups of patients. Results of in vitro (n = 3) and in vivo (n = 5) experiments were plotted using GraphPad Prism 9, and Student’s t-test were applied to analyze differences between the two groups. The sample sizes for in vitro experiments were determined based on preliminary data and established practices in the field to ensure detectable effect sizes. For in vivo experiments, a sample size of n = 5 per group was initially chosen based on common practice in xenograft studies. A P value <0.05 indicated a statistically significant difference (^*P < 0.05; ^**P < 0.01; ^***P < 0.001).

As an extrahepatic cytochrome P450 isoform, CYP4B1 is predominantly expressed in the lung, with lower expression levels in other organs (28). However, its expression patterns and functional mechanisms in LUAD remain incompletely understood, necessitating further investigation. Prior studies have implicated CYP4B1 as a tumor suppressor in LUAD (29). To characterize CYP4B1 expression in LUAD, we performed multi-dataset analyses across TCGA_LUAD, GSE32863, GSE46539, GSE43458, GSE10072, and GSE31210. These analyses revealed a significant downregulation of CYP4B1 mRNA in LUAD tissues compared to normal lung tissues (Figures 1A–C). Consistently, in clinical tissue samples, we detected a marked reduction in CYP4B1 protein expression in LUAD tissues relative to normal lung tissues (Figures 1D, E; Supplementary Figure 1), and Western blot (WB) assays further confirmed this downregulation in tumor tissues (Figure 1F). We next evaluated the diagnostic potential of CYP4B1. ROC curve analyses across the above six datasets showed area under the curve (AUC) values > 0.78, indicating robust discriminatory power for distinguishing LUAD from normal tissues (Figures 1G–I; Supplementary Figure 2). Survival analyses in three independent datasets (TCGA_LUAD, GSE72094, GSE31210) further demonstrated that higher CYP4B1 expression correlated with better patient prognosis (Figures 1J–L). Collectively, our integrative multi-omics analyses validate that CYP4B1 downregulation serves as a molecular hallmark of LUAD pathogenesis and a clinically actionable prognostic biomarker.

To systematically validate the tumor-suppressive role of CYP4B1 in LUAD, we first established stable cell line models: CYP4B1-overexpressing (OE-CYP4B1) and CYP4B1-knockdown (sh-CYP4B1) PC-9/H1299 cell lines were constructed via lentiviral infection. WB analysis confirmed the successful establishment of these stable cell lines (Figures 2A, B), and immunofluorescence staining further validated the effective modulation of CYP4B1 protein levels (Figures 2C, D). To evaluate the impact of altered CYP4B1 expression on cellular functions, we performed a series of in vitro assays. CCK-8 assays showed that CYP4B1 overexpression significantly inhibited the proliferation of both cell lines (Figures 3A, C), whereas CYP4B1 knockdown (using two distinct targeting sequences) exerted the opposite effect (Figures 3B, D). EdU incorporation assays confirmed these findings, with CYP4B1 overexpression reducing the proportion of EdU-positive cells from 22.1 ± 2.4% to 8.4 ± 1.1% in PC-9 cells and from 26.5 ± 2.1% to 15.8 ± 1.5% in H1299 cells, while CYP4B1 silencing increased the EdU-positive rate to 42.1 ± 1.2% (vs. 23.9 ± 1.5% in controls) and 46.9 ± 1.6% (vs. 25.7 ± 1.8% in controls), respectively (Figures 3E, F). Mechanistic exploration via cell cycle analysis revealed that CYP4B1-overexpressing cells exhibited increased G1 phase arrest (PC-9: 70.2 ± 0.2% vs. 54.5 ± 1.0% in controls; H1299: 61.4 ± 1.1% vs. 46.1 ± 1.6% in controls) and reduced S phase proportion (PC-9: 19.5 ± 1.1% vs. 31.1 ± 1.2% in controls; H1299: 29.2 ± 0.9% vs. 42.7 ± 1.0% in controls), whereas CYP4B1 knockdown resulted in the opposite trends (Figures 3G–I). The results of the colony formation assay showed that the proliferative capacity of PC-9 and H1299 cells decreased by approximately 52.3% and 46.9% respectively after CYP4B1 overexpression, while it increased by approximately 103.5% and 128.2% respectively after CYP4B1 knockdown (Figures 3J, K). The results of the wound healing assay showed that the migratory capacity of PC-9 and H1299 cells decreased by approximately 56.7% and 60% respectively after CYP4B1 overexpression, while it increased by approximately 71% and 59% respectively after CYP4B1 knockdown (Figures 3L, M). Transwell assay indicated that the invasive capacity of PC-9 and H1299 cells decreased by approximately 46.7% and 55.6% respectively after CYP4B1 overexpression, while it increased by approximately 78.6% and 97.8% respectively after CYP4B1 knockdown (Figures 3N, O).

To translate these in vitro findings to an in vivo context, we established PC-9 and H1299 xenograft tumor models in nude mice, monitored tumor growth longitudinally, and after euthanizing the animals, subjected the excised tumors to photographic documentation and gravimetric assessment (Figure 4A). By day 25 post-implantation, tumors in the OE-CYP4B1 groups were significantly smaller in both volume and weight compared to the OE-Control groups (Figures 4B–E): in the PC-9 model, tumor volumes were 0.689 ± 0.103 cm³ vs. 0.408 ± 0.065 cm³ for OE-control vs. OE-CYP4B1 subgroups and 0.595 ± 0.098 cm³ vs. 1.048 ± 0.148 cm³ for sh-control vs. sh-CYP4B1 subgroups (Figure 4F), while tumor weights were 0.2869 ± 0.054 g vs. 0.169 ± 0.033 g for OE-control vs. OE-CYP4B1 subgroups and 0.265 ± 0.036 g vs. 0.455 ± 0.101 g for sh-control vs. sh-CYP4B1 subgroups (Figure 4H); in the H1299 model, tumor volumes were 0.747 ± 0.103 cm³ vs. 0.425 ± 0.062 cm³ for OE-control vs. OE-CYP4B1 subgroups and 0.748 ± 0.08 cm³ vs. 1.083 ± 0.143 cm³ for sh-control vs. sh-CYP4B1 subgroups (Figure 4G), while tumor weights were 0.406 ± 0.065 g vs. 0.192 ± 0.046 g for OE-control vs. OE-CYP4B1 subgroups and 0.341 ± 0.035 g vs. 0.532 ± 0.092 g for sh-control vs. sh-CYP4B1 subgroups (Figure 4I).

To reveal the molecular mechanism by which CYP4B1 inhibits the proliferation of LUAD cells, we conducted a series of experiments. Firstly, the results of WB experiments confirmed that overexpression of CYP4B1 could significantly inhibit the phosphorylation of PI3K, AKT, and mTOR in PC-9 and H1299 cells, while the PI3K/AKT/mTOR signaling pathway was abnormally activated after CYP4B1 silencing (Figures 2A, B). This result was also confirmed by cellular immunofluorescence in both cell lines (Figures 2C, D). In addition, the results of IHC detection showed that CYP4B1 could inhibit the PI3K/AKT/mTOR signaling pathway in clinical LUAD samples (Figure 2E). To further confirm that CYP4B1 exerts its antitumor effect by inhibiting the PI3K signaling pathway, we treated CYP4B1-overexpressing H1299 and PC9 cells with 740 Y-P (10 μM), a PI3K agonist. The results showed that 740 Y-P effectively reversed the decreased cell proliferation induced by CYP4B1 overexpression (Supplementary Figure 4). Based on the above experimental results, we propose that CYP4B1 inhibits the proliferation of LUAD cells, potentially through suppressing the PI3K/AKT/mTOR signaling pathway.

To investigate the regulatory mechanisms governing CYP4B1 expression in LUAD, we performed co-expression analyses using public datasets (GSE31210, GSE72094, GSE26939, and GSE11969). These analyses revealed a significant positive correlation between CYP4B1 and NFIA at the mRNA level (Figures 5A–D). Additionally, using JASPAR database (http://jaspar.genereg.net/) we identified two putative NFIA binding sites within the CYP4B1 transcriptional regulatory region: a distal site (-1764 to -1755 bp) and a proximal site (-759 to -750 bp) (Figure 5E). Collectively, these findings implicate NFIA as a potential upstream regulator of CYP4B1 in LUAD. Dual-luciferase reporter assays demonstrated that co-transfection with an NFIA expression vector enhanced wild-type CYP4B1 promoter activity by 4.07-fold (Figure 5F; P < 0.001). Mutagenesis of both predicted binding sites abolished this activation, resulting in a 79.7% reduction in luciferase activity compared to wild-type controls (Figure 5F; P < 0.001). Discrete mutation of the distal and proximal sites reduced activity by 35.5% and 40.2% respectively (Figure 5F; P < 0.001), indicating both sites contribute to NFIA-mediated transactivation with the proximal site playing a dominant role. ChIP assays in PC-9 cells confirmed direct physical interaction between NFIA and the CYP4B1 promoter, showing a 3.77-fold enrichment of NFIA binding at the -759 to -750 bp region upon NFIA overexpression (Figure 5G; P < 0.001). Consistent with these results, siRNA-mediated knockdown of NFIA in PC-9 and H1299 cells significantly reduced CYP4B1 protein levels as detected by WB (Figure 5H). In conclusion, our integrated bioinformatics and functional analyses establish that NFIA directly binds the CYP4B1 promoter to enhance its transcriptional activity. This NFIA/CYP4B1 axis represents a potential therapeutic target in LUAD and highlights the critical role of transcriptional networks in LUAD pathogenesis.

In this study, the TCGA_LUAD cohort served as the training set for developing the CYP4B1-related risk score (CRRS), whose effectiveness and universality were subsequently validated using three external datasets (GSE72094, GSE41271, and GSE31210). From the TCGA_LUAD dataset, we identified 42 genes co-expressed with CYP4B1 (Supplementary Table 1; correlation coefficient 0.5; P ≤ 0.001). Leveraging the expression profiles of CYP4B1 and these co-expressed genes, we applied consensus clustering and nonnegative matrix factorization (NMF) to stratify patients in the TCGA_LUAD dataset. Through analyzing the consensus matrix for K values from 2 to 10 and the consensus clustering cumulative distribution function (CDF) plot, we determined that K = 2 was the optimal parameter for the consensus clustering method, which classified 271 and 190 patients into C1 and C2 clusters, respectively, with patients in cluster C2 showing a longer overall survival (OS) as indicated by Kaplan–Meier (KM) analysis (Figures 6A, B; Supplementary Figure 5A). Similarly, based on the cophenetic coefficient and the consensus matrix for ranks ranging from 2 to 10, we found that a rank of 2 was the most suitable number of subgroups for the NMF method, and this approach divided 211 patients into group C2, which also demonstrated a longer OS (Figures 6C, D; Supplementary Figure 5B). By focusing on the patients in the intersection of the two grouping methods for further analysis, we assigned 248 patients to cluster A and 189 patients to cluster B (Figure 6E). Principal component analysis (PCA) results revealed that patients in cluster A and cluster B could be effectively differentiated (Figure 6F), and KM analysis further indicated that patients in cluster B had a longer OS (Figure 6G). We identified 38 differentially expressed genes (DEGs) significantly associated with patient prognosis from patients in cluster A and B (Supplementary Table 2; |logFC| ≥ 2; Adjusted P value ≤ 0.01). Incorporating these 38 DEGs into the LASSO-Cox regression model, 7 core genes and their coefficients were selected to develop the CRRS (Figures 7A, B), calculated as Score = (0.149 × ANLN) + (-0.002 × C4BPA) + (-0.004 × CRTAC1) + (-0.018 × CYP4B1) + (-0.018 × IRX2) + (-0.003 × NAPSA) + (-0.008 × SFTPB). Patients in both the training and validation sets were divided into high- and low-risk groups based on the median CRRS, with KM analysis showing significantly longer OS for low-risk patients in both sets (Figures 7C–F). Additionally, we note that higher CRRS (indicative of a worse prognosis) correlates with predicted resistance to etoposide and paclitaxel—two chemotherapeutic agents commonly used in lung cancer treatment regimens (Figure 7G; Supplementary Figure 6; |R| ≥ 0.5; P ≤ 0.001), validating CRRS as an effective tool for evaluating LUAD prognosis and guiding personalized cancer therapy. The negative correlation suggests that patients in the high-risk group, characterized by lower CYP4B1 expression and a more aggressive tumor phenotype (Supplementary Figure 7), may be less sensitive to these agents, possibly due to enhanced survival signaling or drug efflux mechanisms. It is crucial to emphasize that these are computational predictions based on genomic data, and their clinical utility requires prospective validation in pharmacokinetic and clinical response studies. We performed Kaplan-Meier survival analysis stratified by clinical stage (Stage I-II vs. Stage III-IV) within the TCGA cohort. CRRS effectively stratifies patients into high- and low-risk groups with significant survival differences within both early-stage and late-stage subgroups, underscoring its potential clinical applicability across different disease severities (Supplementary Figure 8). A nomogram integrating CRRS, age, gender, tumor grade, T stage, and N stage was developed (Figure 7H). ROC analysis demonstrated AUC values > 0.72 for 1-, 3-, and 5-year OS predictions (Figure 7I), while calibration curves showed strong agreement between predicted and observed outcomes (Figure 7J). With superior performance indicated by the C-index and ROC curves compared to other indicators (Figures 7K–N), the nomogram emerged as an excellent quantitative tool for assessing LUAD prognosis.