Clinical significance and molecular mechanism of CDX2-CBX3 regulatory axis in lung adenocarcinoma progression

Highlights

• •Four CBX-based molecular subtypes of LUAD were identified, with significant differences in patient survival outcomes.​.
• •These subtypes showed distinct patterns in clinical stage distribution, DNA repair pathway activation, and immune cell infiltration characteristics.​.
• •The CDX2-CBX3 regulatory axis was functionally validated, demonstrating that CDX2 directly activates CBX3 transcription by binding to conserved promoter regions.
• •CDX2 overexpression promoted malignant phenotypes such as enhanced cell migration, invasion, and xenograft tumor growth, while CBX3 knockdown significantly attenuated these effects.

Abstract

Lung adenocarcinoma (LUAD) remains a leading cause of cancer-related mortality, with chemotherapy resistance and tumor heterogeneity posing significant challenges. The Chromobox (CBX) protein family, crucial epigenetic regulators in tumor progression, has not been systematically characterized in LUAD. This study aimed to develop a CBX-based molecular classification system for LUAD and explore the mechanistic role of the CDX2-CBX3 regulatory axis in tumor progression. Through multiomics analysis of TCGA-LUAD data, four distinct CBX subtypes were identified, each associated with variations in survival, clinical stage, DNA repair pathway activation, and immune cell infiltration. Mechanistic investigations (ChIP-qPCR, luciferase assays, and gain/loss-of-function experiments) confirmed that CDX2 directly upregulates CBX3 transcription via conserved promoter binding. CDX2 overexpression enhanced migration, invasion, and xenograft growth, whereas CBX3 knockdown suppressed these phenotypic changes. In conclusion, this study defines clinically relevant CBX molecular subtypes in LUAD and reveals the CDX2-CBX3 transcriptional cascade as a novel driver of tumor progression, offering potential targets for precision therapy.

Introduction

Lung cancer is the leading cause of cancer-related mortality globally, with lung adenocarcinoma (LUAD) being the most prevalent histological subtype, accounting for over 40 % of all lung cancer cases [1,2]. Global cancer statistics for 2022 indicate that, due to the absence of early specific symptoms, >60 % of patients with LUAD are diagnosed at an advanced stage [3], often with lymph node or distant metastasis, resulting in a five-year survival rate of <15 % [4]. Although targeted therapies for driver mutations such as epidermal growth factor receptor (EGFR) and anaplastic lymphoma kinase (ALK) [[5], [6], [7]], as well as immune checkpoint inhibitors like PD-1/PD-L1, have improved the prognosis for some patients [8], clinical treatment continues to face significant challenges. In particular, patients without driver mutations or those who develop acquired resistance still rely on chemotherapy as the primary treatment [9,10]. However, chemotherapy resistance and severe side effects significantly reduce its clinical efficacy [[11], [12], [13]]. Therefore, investigating the molecular mechanisms underlying LUAD progression and identifying novel prognostic markers and therapeutic targets are critical for improving patient outcomes.

The Chromobox (CBX) protein family, comprising eight members, plays a key role as epigenetic regulators. Recent studies have shown that CBX family members exhibit distinct expression patterns and functional characteristics across various tumor types [14]. In multiple solid tumors, CBX2, CBX3, CBX4, and CBX8 are typically overexpressed and closely associated with malignant tumor progression [[15], [16], [17], [18]]. In ovarian cancer, elevated CBX3 expression correlates with chemotherapy resistance [19]. Additionally, several studies have highlighted CBX3′s role in regulating the tumor immune microenvironment. CBX3-deficient CD8⁺ T cells can sustain effector functions through the LEF-1/IL-21 receptor signaling pathway, thereby enhancing the anti-tumor immune response [20]. Pan-cancer analyses have shown that CBX3 expression is inversely correlated with the infiltration of most immune cell types [21,22]. Notably, high CBX3 expression in LUAD is significantly associated with larger tumor size, lymph node metastasis, and poor patient prognosis [23]. These findings establish a strong theoretical foundation for considering CBX family members as diagnostic biomarkers and therapeutic targets in cancer.

As a key developmental regulator, the caudal-related homobox (CDX)2 protein exhibits significant functional heterogeneity across various tumors. While CDX2 is known to act as a tumor suppressor in colorectal cancer [24], recent studies suggest it may have a pro-cancer role in certain malignancies. For instance, in gastric cancer, CDX2 phosphorylation promotes tumor growth and angiogenesis [25], and in glioma, CDX2 overexpression drives tumor cell invasion [26]. Notably, bioinformatics analyses have identified CDX2 as significantly elevated in LUAD tissue, with its expression level strongly correlating with CBX3. Based on this observation, and considering CDX2′s function as a transcription factor [27] and the presence of conserved binding sites in the CBX3 promoter region, it is hypothesized that CDX2 may promote LUAD progression by directly regulating CBX3 transcription.

This study introduces a novel LUAD molecular classification system by integrating multi-omics analysis of CBX family expression patterns, enabling patient stratification into subclusters with distinct tumor microenvironment profiles and treatment susceptibility. Despite evidence of transcriptional regulators, such as CDX2, playing roles in tumor progression, the functional mechanisms underlying CDX2 in LUAD and its upstream and downstream regulatory networks remain unclear. In addition to establishing a CBX-based LUAD molecular classification system for prognostic and therapeutic predictions, this study seeks to investigate the functional role and mechanistic pathway of the CDX2-CBX3 regulatory axis in LUAD progression using cell-based and nude mouse tumor models. These findings aim to uncover clinically actionable biomarkers for LUAD prognosis and treatment.

Methods and materials

Data collection and processing

Gene expression profiles, clinical data, and mutation data from 585 LUAD samples (526 tumor tissues and 59 adjacent tissues) were obtained from the TCGA database. The raw count data were log₂(count+1) transformed, and gene annotation was performed using the GENCODE database (ftp://ftp.sanger.ac.uk/pub/gencode/Gencode_human/). Genes with zero expression in over 50 % of the samples were excluded. Batch effect correction and normalization were performed using the normalizeBetweenArrays function from the limma package (v3.9.19). Clinical data with missing rates exceeding 50 % were removed to ensure data quality.

CBX family expression and prognostic analysis

Differential expression analysis was conducted using the GEPIA platform (http://gepia.cancer-pku.cn), and the t-test was applied to compare CBX family gene expression between tumor and normal tissues. The selection criteria included |logFC| > 0.585 and FDR < 0.05. Survival analysis was performed by categorizing patients into high and low expression groups based on the median gene expression. The Kaplan-Meier method was used to assess overall survival differences, and the Log-rank test determined statistical significance.

Subtype clustering analysis of the CBX family

After min-max standardization of the TCGA-LUAD dataset (n = 483) using the scale function in R, unsupervised clustering was performed using the non-negative matrix factorization (NMF) algorithm (R-NMF package). The final subtype classification was determined by evaluating the cophenetic coefficient curve across varying cluster numbers (K = 2–6) and selecting the optimal K corresponding to the inflection point. The cophenetic correlation coefficient (cophenetic value) is a crucial measure for assessing clustering performance and is primarily used to determine the optimal number of clusters. Following standard criteria, the optimal number of clusters is identified at the point just before the maximum change in the cophenetic value as the number of clusters (K) increases. As shown in Fig. 2A, the cophenetic value experiences the most significant drop (i.e., the largest change) when K transitions from 4 to 5. Therefore, K = 4 was selected as the optimal number of clusters. To obtain an indication of the expression profile of CBX genes in each cluster, an expression heatmap was generated for all 8 members of the CBX family (CBX1-CBX8) across the four subtypes. Additionally, the expression differences of individual CBX genes were further evaluated between subtypes.

Fig. 2.

Subtype clustering of the CBX family (A) Consensus parameter selection plot for NMF unsupervised clustering. (B) Heatmap of LUAD samples clustered based on the consensus parameter. (C) Kaplan-Meier survival curves illustrating overall survival differences among the CBX-based clusters.

Multi-omics analysis of subtype characteristics

A multidimensional analysis was conducted to systematically assess the characteristic differences among subtypes: (1) Clinical prognosis analysis: The Kaplan-Meier method was used to generate survival curves, with the log-rank test applied to evaluate the statistical significance of group differences (P < 0.05 as the significance threshold); (2) Clinical characteristic association analysis: The chisq.test() function from the R package was used to perform chi-square tests for clinical parameters such as age, gender, and TNM stage; (3) Signal pathway analysis: Pathway enrichment was calculated using the GSVA (v1.36.3) and GSEABase (v1.50.1) packages based on the Hallmark gene set (h.all.v7.4.symbols) from the MSigDB database; (4) Drug sensitivity prediction: The pRRophetic package (v0.5) was employed to predict the sensitivity (predicted IC50 value) of samples to anticancer drugs listed in possibleDrugs2016; (5) Immune microenvironment analysis: The CIBERSORT package (LM22 feature matrix) was used to assess the infiltration abundance of 22 immune cell types. The Kruskal-Wallis test was applied to analyze differences in immune cells across subtypes This study investigated the correlation between CBX gene expression and prognosis in all LUAD tumor samples (n = 483) using the non-negative matrix factorization (NMF) clustering method (hierarchical clustering). The correlation between CBX family genes and these differentially expressed immune cells was also analyzed. Data standardization (normalization) was performed using the min-max normalization algorithm of the scale function in R to eliminate dimensionality effects between different genes. The NMF function from the R package NMF (version 0.28) was then applied for clustering, with the "brunet" method (default setting) selected and the rank set to range from 2 to 6. The ESTIMATE package (v1.0.13) was used to calculate the immune score (ImmuneScore), stromal score (StromalScore), tumor purity and overall score (ESTIMATEScore). All inter-group comparisons were performed using one-way analysis of variance (ANOVA), with Tukey's HSD post hoc test conducted for significant results (P < 0.05). All statistical analyses were performed in the R 4.0.2 environment.

To perform subtype-specific pathway analysis, a comparison of pathway GSVA score differences was performed among the 4 groups. Targeting the 4 CBX-based subtypes, pathway differences between each specific subtype and all other subtypes were separately compared. The top 10 pathways with the most significant differences were selected in each subtype, merged, and a bubble plot of functional enrichment results was generated for each subtype. Details of the subtype-specific pathway results are shown in Supplementary Table 1.

Western blot

Approximately 20–30 mg of frozen lung cancer tissues and paired adjacent normal tissues were placed into pre-chilled homogenization tubes. Following the NE-PER Kit instructions, CER I reagent (Thermo Fisher Scientific, Cat. No.: 78,833) was added, and high-speed homogenization was performed (6000 rpm, 2–3 cycles, 20 s/cycle, with 30 s ice cooling between cycles). After a 10-min ice incubation, CER II was added, followed by vortexing and centrifugation (∼16,000 × g, 5 min, 4 °C), discarding the cytoplasmic protein supernatant. Pre-chilled NER was added to the pellet, and the sample was vigorously vortexed, incubated on ice with shaking for 40 min, vortexing every 10 min. Centrifugation (∼16,000 × g, 10 min, 4 °C) was performed to obtain nuclear protein extracts, which were aliquoted and stored at −80 °C. Protein concentration was measured using the BCA Kit (Thermo Fisher Scientific, Cat. No.: 23,225). All samples were diluted to 4 μg/μL with RIPA buffer. A 40 μg total protein sample was mixed with 5X SDS-PAGE loading buffer, heated at 95 °C for 5 min (in a metal/boiling water bath) for denaturation, and then immediately cooled on ice. A precast 4–20 % TGX gel was prepared with 1X electrophoresis buffer. Protein marker and denatured samples were loaded, and electrophoresis was performed at 80 V initially, then 120 V after the bromophenol blue entered the separating gel, for 60–90 min until the indicator reached the bottom of the gel. A PVDF membrane (Merck Millipore, Cat. No.: IPVH00010) was cut to the size of the gel and activated in methanol for 1 min. The "sponge pad-filter paper-gel-PVDF membrane-filter paper-sponge pad" sandwich (without bubbles) was assembled and placed in a transfer tank with pre-chilled 1X transfer buffer (20 % methanol). Membrane transfer was conducted at 300 mA (ice bath) for 60 min for CBX3 and 90 min for CDX2/Lamin B1. After transfer, the PVDF membrane was blocked with QuickBlock™ Buffer for 1 h at room temperature on a shaker, followed by three 5-min washes with TBST. Primary antibodies (diluted with Beyotime P0256 buffer) were incubated with the membrane overnight at 4 °C: Anti-CDX2 (Abcam, ab76541, 1:1000), Anti-CBX3 (Abcam, ab109028, 1:2000), and Anti-Lamin B1 (Abcam, ab133741, 1:1000). The next day, the membrane was washed three times with TBST (10 min each). The HRP-conjugated goat anti-rabbit secondary antibody (Abbkine A21020) was diluted to 1:5000 with TBST and incubated with the membrane for 1 h at room temperature on a shaker, followed by three 10-min TBST washes. SuperSignal™ West Pico PLUS Substrate A/B (Thermo Fisher Scientific, Cat. No.: 34,580) was mixed in equal volumes, and the excess liquid was blotted from the membrane. The membrane was placed in an imaging system, and substrate was added for 2 min. Signals were captured, adjusting exposure to ensure clear, non-saturated bands. Band gray value analysis was performed using ImageJ/Tanon software. CDX2/CBX3 gray values were normalized to Lamin B1 (internal reference) for each sample, and protein expression differences between lung cancer and adjacent normal tissues were compared.

Cell culture and transfection

Normal lung epithelial cells HSAEC1-KT and human LUAD cell lines H2106, H2023, H1623, and H2347 (all from ATCC) were cultured in complete 1640 medium (Gibco) supplemented with 10 % fetal bovine serum (FBS) and 1 % penicillin-streptomycin (Beyotime) at 37 °C and 5 % CO₂. When cells reached 80 %−90 % confluence, they were passaged at a 1:2 to 1:3 ratio.

Specific small interfering RNAs (siRNAs) targeting CDX2 and CBX3 genes (si-CDX2 and si-CBX3) were introduced into H2023 and H2347 cells to achieve gene knockdown. The sequences of the siRNAs were as follows: si-CDX2–1: 5′−3′: UCGAUAUUUGUCUUUCGUCCU; 3′−5′: GACGAAAGACAAAUAUCGAGU. si-CDX2–2: 5′−3′: AAGAUUGUGAAAAUGACAGGA; 3′−5′: CUGUCAUUUUCACAAUCUUGG. si-CBX3: 5′−3′: UUAUUGCAGACUUGAAGAGCU; 3′−5′: CUCUUCAAGUCUGCAAUAAAA. Non-targeting siRNA (si-NC) was used as a negative control. Cells, with a confluence of 70 %−80 %, were washed twice with PBS, and Opti-MEM serum-free medium was added before incubation at 37 °C with 5 % CO₂ for equilibration. The transfection complex, consisting of siRNA, Lipofectamine™3000 (Invitrogen), and Opti-MEM (Gibco), was then added to the cells and incubated for 6 h. After a 48-h incubation period, the cells were collected for subsequent analysis.

To achieve stable knockdown of the CBX3 gene in H2347 cells, a lentivirus-mediated shRNA approach was employed using the pLKO.1-puro vector (SHC001, Sigma-Aldrich, USA), which carries specific shRNA sequences targeting human CBX3 (5′-CTGGCGAAAGAGGCAAATATG-3′). Recombinant lentiviral particles were produced in 293T cells by co-transfecting the transfer plasmid with the packaging plasmid psPAX2 (Addgene, #12,260) and the envelope plasmid pMD2.G (Addgene, #12,259) using Lipofectamine 3000 (Thermo Fisher, L3000015). Viral supernatants were collected at 48 and 72 h post-transfection, concentrated, and titered. H2347 cells were infected in the presence of polybrene (8 µg/mL, Sigma-Aldrich, H9268) and selected with puromycin (2 µg/mL, Sigma-Aldrich, P8833) to establish stable knockdown pools. Knockdown efficiency was validated via qPCR using SYBR Green (Applied Biosystems, 4367,659) with CBX3-specific primers and Western blotting with anti-CBX3 (Abcam, ab192579) and anti-GAPDH (CST, 2118S) antibodies, with an expected reduction of at least 70 % in CBX3 expression.

To enhance the expression levels of CDX2 and CBX3 in H2023 and H2347 cells, overexpression plasmids (OE-CBX3 and OE-CDX2) containing full-length CDX2 and CBX3 coding sequences were used. The CDX2/CBX3 expression vector was constructed using the pCMV-HA plasmid (purchased from the Shanghai Cancer Institute, China). Briefly, both the CDX2/CBX3 cDNA PCR product and the pCMV-HA plasmid were digested with EcoRI. The digested PCR product was separated by agarose gel electrophoresis, recovered, and purified. The purified fragment was then ligated into the EcoRI-digested pCMV-HA vector to generate the recombinant plasmid pCMV-CDX2/CBX3-HA. Post-ligation, the plasmid was transformed into Escherichia coli TOP10 competent cells, which were then plated on solid LB medium. Recombinant plasmids were extracted, and their sequences were verified by restriction digestion followed by electrophoresis. H2023 and H2347 cells (1 × 10⁵ cells/well) were seeded into 6-well plates. When cells reached 90 % confluence, they were transfected with either pCMV-CDX2/CBX3-HA or the empty pCMV-HA vector. At 48 h post-transfection, cells were passaged at a 1:10 dilution. The efficiency of CDX2/CBX3 overexpression or siRNA-mediated knockdown was validated by quantitative real-time polymerase chain reaction (qRT-PCR) analysis (Supplementary Fig. 1).

Procedure of qRT-PCR

Cells were washed twice with ice-cold PBS buffer at 4 °C and lysed on ice using RNAisoPlus reagent (Takara, Dalian). Chloroform and isopropanol were sequentially added to the lysate, and RNA was purified by centrifugation. RNA purity was assessed using a NanoDrop 2000, and cDNA synthesis was performed according to the PrimeScript II 1st Strand cDNA Synthesis Kit (Takara, Dalian) instructions. SYBR Green Master Mix (Solarbio, Beijing) was used for qPCR, with the following program: pre-denaturation at 95 °C for 5 min; 40 cycles of denaturation at 95 °C for 15 s and annealing at 60 °C for 30 s. GAPDH was used as the internal reference gene for normalization, and the relative expression of the target gene was calculated using the 2^-ΔΔCt method. The primers for CBX3 RT-qPCR were: Forward: CCTAGCGGGCCATTCCTTAG; Reverse: CAGCAGGTCCTAAACTGCCA. The primers for CDX2 RT-qPCR were: Forward: CCTCGGCAGCCAAGTGAA; Reverse: AAACCAGATTTTAACCTGCCTCTCA.

Cell counting kit-8

The cell concentration was adjusted to 5 × 10³ cells/100 μL, and 100 μL of the cell suspension was added to each well of a 96-well plate, followed by incubation for 24 h. At each time point (0, 1, 2, 3, 4, and 5 days), 10 μL of CCK-8 reagent was added to each well. After gentle shaking for 30 s, the cells were incubated at 37 °C in the dark for 4 h. Absorbance was recorded at 450 nm using a microplate reader (BMG LabTech).

Colony formation assay

For colony formation assays, the cell concentration was adjusted to 500 cells/mL, and 1 mL was added to each well of a 6-well plate. The medium was replaced every 72 h. After 14 days, colonies containing ≥ 50 cells were visualized under an inverted microscope. The cells were fixed with 4 % paraformaldehyde for 40 min at room temperature, followed by staining with 0.1 % crystal violet for 15 min in the dark at room temperature. The cells were then rinsed three times with PBS, and images were captured. Colonies were counted using ImageJ software.

Scratch assay

Cells were adjusted to a density of 5 × 10⁵ cells/mL, and 2 mL of the suspension was added to each well of a 6-well plate. After gentle shaking to ensure even distribution, cells were cultured in a 37 °C, 5 % CO₂ incubator until the monolayer reached 90 % confluence. A straight scratch was made perpendicular to the bottom of the well using a 200 μL sterile pipette tip. The wells were rinsed three times with PBS to remove detached cell debris, and the medium was replaced with serum-free medium to minimize proliferation-related interference with migration. Fixed fields of view were photographed at the initial time point and after 24 h under an inverted microscope.

Transwell assay

Twenty-four hours before the experiment, Matrigel (Corning) was thawed at 4 °C and diluted with pre-cooled serum-free medium. The matrix gel was evenly applied to the upper chamber of the Transwell (Corning) and allowed to solidify for 30 min. A cell suspension (1 × 10⁵ cells) was then added to the upper chamber, and 500 μL of complete culture medium containing 10 % FBS was injected into the lower chamber. After 24 h of incubation, cells in the upper chamber were gently rinsed with PBS. The chamber was subsequently immersed in 4 % paraformaldehyde for 30 min at room temperature and stained with 0.1 % crystal violet at room temperature for 15 min. After acquiring 3–5 fields of view, the average cell number was counted for quantitative analysis. This procedure assessed the cell invasion ability. For migration evaluation, Matrigel was omitted, and the remaining steps were performed as described.

Chromatin immunoprecipitation (ChIP)

The CBX3 promoter region (−2000 to +500 bp) was analyzed using the JASPAR database (https://jaspar.elixir.no/), with a core similarity threshold set to > 0.85 to identify CDX2 binding sites. The relative score threshold (> 80 %) and evolutionary conservation (UCSC multi-species alignment) were used to select three candidate sites: P1 (−1082∼−1087 bp), P2 (−1547∼−1554 bp), and P3 (−1735∼−1742 bp). When cell density reached 80 %, formaldehyde (1 %) was used for cross-linking at room temperature for 10 min, followed by quenching with 0.125 M glycine for 5 min. After cell lysis, chromatin was fragmented by sonication (4 × 15-s pulses, 30 % power, fragments of 200–500 bp). A total of 100 μg of chromatin was incubated overnight at 4 °C with anti-CDX2 antibody (ab76541, Abcam, USA) or IgG control (ab172730, Abcam, USA). The chromatin complexes were captured using Protein A/G magnetic beads (Thermo Fisher Scientific) and washed in a gradient manner. After cross-linking reversal, DNA was purified, and the enrichment of the target region was analyzed by qPCR. The 2^-ΔΔCt method was used, with Input DNA as the internal reference. The primers for CBX3 in ChIP-PCR were as follows: Left Primer 1: GAGGTGGTCCCTGCAGTTAC; Right Primer 1: CAGGGGTCAGTTCCACAGAC; Left Primer 2: TTGGTCCGATTTCCTGCCTC; Right Primer 2: CAGTTCTCACTACAGCGCCA; Left Primer 3: TGAAATTAACGCCGACGGGA; Right Primer 3: TTCAGCAGCGAACTCTCCTG.

Immunofluorescence staining

Immunofluorescence staining was conducted on tissue sections of lung cancer. Tumor specimens were first fixed in 10 % formalin, followed by paraffin embedding and serial sectioning into slices with a thickness of 3 µm. The primary antibodies employed included the following: CCR7 (a marker for memory T cells, catalog number SAB4500329, Sigma-Aldrich), CD27 (a marker for memory B cells, catalog number HPA069501, Sigma-Aldrich), and CBX3 (catalog number HPA004902, Sigma-Aldrich). Double-immunofluorescence staining assays were carried out to co-label CBX3 with either CCR7 or CD27. Cell nuclei were visualized with 4′,6-diamidino-2-phenylindole (DAPI) (D9542, Sigma-Aldrich). Image processing and analysis were implemented using the ImageJ software. All samples presented in a single graph underwent staining and imaging in parallel, with the application of standardized threshold intensity settings. The quantification of immunofluorescence intensity in stained tissues was independently performed by two trained pathologists, who were blinded to the corresponding clinical information and experimental data.

Luciferase reporter assay

To verify the binding activity of CDX2 and CBX3, a luciferase reporter vector containing the P2 site (−1547 to −1554 bp) of the CBX3 promoter region was constructed. Cells were seeded into 12-well plates at a density of 1 × 10⁶/mL. After 24 h of culture, 500 ng of the reporter vector and either OE-CDX2 or si-CDX2 were co-transfected using the Lipofectamine™ 3000 transfection system. After 24 h, cells were lysed, centrifuged to obtain the supernatant, and luciferase activity was measured using the dual luciferase reporter system (Promega). Firefly luciferase activity was detected using the luciferase substrate, and Renilla luciferase was used for normalization.

Nude mouse tumor-bearing assay

H2347 cells with different transfection treatments (control, OE-NC, OE-CDX2, OE-CDX2+sh-NC, OE-CDX2+sh-CBX3) were prepared into a cell suspension (10⁷/mL). After 1 week of adaptive feeding, 6-week-old SPF-grade BALB/c nude mice were subcutaneously injected with 200 μL of the cell suspension (5–10 mm depth) under isoflurane anesthesia (induction 3–4 %, maintenance 1.5–2 %). Tumor long diameter (L) and short diameter (W) were measured weekly, and tumor volume was calculated using the formula V = 0.5 × L × W². After 4 weeks, the mice were euthanized, and the tumor tissue was excised for weighing and imaging. The entire experiment was conducted in An spf environment with a constant temperature (22 ± 1 °C), constant humidity (45 ± 2 %), and a 12-h light/dark cycle. A heating pad was used to maintain body temperature during the procedure. All animal experiments adhered to the ARRIVE guidelines and were approved by the Medical Research Ethics Committee of Hebei Children's Hospital (No. 202,206–10).

Statistical analysis

Data were analyzed using SPSS 27.0 (IBM Corp). Continuous variables with normal distribution were expressed as mean ± standard deviation and compared using an independent samples t-test, while non-normally distributed data were analyzed using nonparametric tests. Bivariate correlations were initially assessed with Spearman's rank analysis, and significantly associated variables were incorporated into multiple linear regression models to establish predictive equations. Intergroup comparisons were performed using one-way ANOVA followed by Bonferroni or SNK post-hoc tests for multiple comparisons. A two-way repeated-measures ANOVA was used to analyze the proliferation and tumor growth experiments. A P-value of < 0.05 was considered statistically significant for all analyses.

Results

Expression and prognostic value of the CBX family in LUAD

RNA-seq expression data for tumor tissues (n = 483) and adjacent tissues (n = 58) were extracted from the TCGA-LUAD project. The analysis revealed significant differential expression patterns for all CBX family genes in LUAD. CBX1, CBX2, CBX3, CBX4, CBX5, and CBX8 were upregulated in tumor samples, while CBX6 and CBX7 were downregulated (Fig. 1A). The prognostic value of these eight CBX genes was assessed using Kaplan-Meier survival analysis and the log-rank test. High expression levels of CBX1, CBX3, and CBX5 were associated with significantly reduced overall survival (Logrank p < 0.05; p(HR) < 0.05), indicating that these three genes are linked to poor prognosis (Fig. 1B). A thorough elucidation of the expression patterns of CBX genes across various subtypes is crucial for understanding the biological basis of these molecular subtypes. To this end, we generated an expression heatmap (Fig. 1C) for all 8 members of the CBX family (CBX1-CBX8) across the four subtypes, which clearly illustrates the unique CBX gene expression signatures of each subtype. Additionally, we further evaluated the expression differences of individual CBX genes between subtypes (Fig. 1D). It was found that the expression levels of CBX1 and CBX3 in Cluster 4-where prognosis is favorable-were lower than those in other subtypes, which is consistent with their role as poor prognostic markers. CBX7 exhibited high expression in Cluster 4, suggesting that this gene may exert a protective effect.

Fig. 1.

Expression and prognostic value of the CBX family in LUAD (A) Box plot showing the expression differences of 8 CBX genes in LUAD and controls based on TCGA data, *P < 0.05. (B) Kaplan-Meier survival curve demonstrating the relationship between the expression levels of the 8 CBX genes and overall survival in patients with LUAD. (C) Expression heatmap for all 8 members of the CBX family (CBX1-CBX8) across the four subtypes. (D) The expression differences of individual CBX genes between subtypes.

Subtype clustering of the CBX family

NMF cluster analysis was performed on LUAD samples using the CBX family genes. The optimal cluster number was determined by examining the cophenetic value's trend relative to the K value, selecting the inflection point with the greatest change. Thus, LUAD samples were classified into four molecular subtypes (Fig. 2A). The sample distribution across subclusters was as follows: cluster 1 (50 cases, 10.4 %), cluster 2 (103 cases, 21.3 %), cluster 3 (67 cases, 13.9 %), and cluster 4 (263 cases, 54.5 %) (Fig. 2B). Survival analysis showed significant differences in overall survival across the subclusters, with cluster 3 exhibiting the poorest prognosis and overall survival, while cluster 4 demonstrated the best survival advantage (Fig. 2C). This distribution suggests that most LUAD cases have a relatively favorable prognosis. A comparison of clinical characteristics across subclusters revealed significant heterogeneity in gender composition, pathological TNM stage, and tumor stage. Notably, cluster 4 contained a higher proportion of early-stage samples compared to cluster 3, which had poorer prognostic outcomes (Table 1).

Table 1.

Analysis of clinical characteristics in different clustering subtypes.

Clinical Characteristics	Cluster 1 (n = 50)	Cluster2 (n = 103)	Cluster3 (n = 67)	Cluster4 (n = 263)	P value	X2
Age (≥60;<60)	11/39	70/33	46/21	198/65	0.347	3.3
M stage (M0; M1/2/X)	29/21	81/22	44/23	172/91	0.036*	8.51
N stage (N0; N1/2/3)	32/18	58/45	41/26	192/71	0.0128**	10.8
T stage (T1; T2/3)	21/29	20/83	11/56	119/153	4.13E-06***	27.73
Gender (male/female)	32/18	58/45	34/33	105/158	0.0018**	15
Tumor stage (stage 1/2/3/4)	27/9/8/6	41/33/22/7	32/14/18/3	167/61/30/5	8.47E-05***	34.13

Notes: * indicates p < 0.05, ** indicates p < 0.01, *** indicates p < 0.001.

Differences in signaling pathways and drug sensitivity between clusters

To further characterize the molecular differences between clusters, tumor-related signaling pathways were investigated using GSVA. A total of 35 significantly different pathways were identified among the four clusters (Table 2). These pathways are involved in key biological processes related to tumor initiation and progression, including DNA repair, the p53 signaling pathway, the PI3K/AKT/MTOR signaling pathway, and the hypoxia signaling pathway. These distinct signaling pathways indicate that the subclusters exhibit unique molecular signatures. The subtype-specific pathway analysis are shown in Fig. 3A and B. Fig. 3A presents a comparison of pathway GSVA score differences among the 4 groups. The results identified 35 pathways that showed significant differences across the 4 subtypes. A deeper red color indicates that the pathway tends to be upregulated in the subtype, i.e., in an activated state. A deeper blue color indicates that the pathway tends to be downregulated in the subtype, i.e., in an inhibited state. In Fig. 3B, targeting the 4 CBX-based subtypes, we separately compared pathway differences between each specific subtype and all other subtypes. Subsequently, we selected the top 10 pathways with the most significant differences in each subtype, merged them, and generated a bubble plot of functional enrichment results for each subtype. Details of the subtype-specific pathway results are available in the table cluster_pathway_enrichment_results.csv. The results in the figure reveal the unique biological characteristics of each subtype: the MYC pathway is significantly downregulated in Cluster 4. Aberrant activation of this pathway is typically associated with poor prognosis, which explains why Cluster 4 has better clinical outcomes compared to other subtypes. To evaluate the potential clinical relevance of CBX family-based classification for personalized treatment, the sensitivity of the subclusters to six commonly used chemotherapy drugs (cisplatin, vinblastine, docetaxel, cyclopamine, doxorubicin, and gemcitabine) was analyzed based on predicted IC50 values (Fig. 3C). The subclusters are predicted to respond differently to these chemotherapy drugs. Specifically, cluster 3, associated with poor prognosis, showed notable sensitivity to docetaxel and gemcitabine, while cluster 2 exhibited significantly reduced sensitivity to doxorubicin. Interestingly, cluster 4, which had a better prognosis, maintained high sensitivity to doxorubicin, docetaxel, and gemcitabine, but showed poor treatment response to cyclopamine (Fig. 3C).

Table 2.

Tumor-related signaling pathways with significant differences.

Signaling Pathways	P value	F value
HALLMARK_MYC_TARGETS_V1	1.79E-21	99.763536
HALLMARK_E2F_TARGETS	1.31E-19	89.596817
HALLMARK_MTORC1_SIGNALING	2.11E-18	83.104852
HALLMARK_MYC_TARGETS_V2	5.24E-18	81.004209
HALLMARK_G2M_CHECKPOINT	6.65E-18	80.455185
HALLMARK_REACTIVE_OXYGEN_SPECIES_PATHWAY	8.38E-16	69.420788
HALLMARK_UNFOLDED_PROTEIN_RESPONSE	3.28E-15	66.346753
HALLMARK_OXIDATIVE_PHOSPHORYLATION	9.58E-15	63.950809
HALLMARK_DNA_REPAIR	4.95E-14	60.299929
HALLMARK_GLYCOLYSIS	1.66E-10	42.662238
HALLMARK_UV_RESPONSE_UP	2.99E-09	36.551589
HALLMARK_SPERMATOGENESIS	3.17E-09	36.423965
HALLMARK_CHOLESTEROL_HOMEOSTASIS	2.03E-08	32.547387
HALLMARK_FATTY_ACID_METABOLISM	4.76E-07	26.065242
HALLMARK_MITOTIC_SPINDLE	1.46E-06	23.791429
HALLMARK_PI3K_AKT_MTOR_SIGNALING	2.42E-06	22.778803
HALLMARK_HEDGEHOG_SIGNALING	3.76E-06	21.886471
HALLMARK_ESTROGEN_RESPONSE_LATE	1.06E-05	19.821209
HALLMARK_HYPOXIA	1.41E-05	19.247677
HALLMARK_PROTEIN_SECRETION	2.77E-05	17.912051
HALLMARK_ADIPOGENESIS	5.73E-05	16.482357
HALLMARK_ALLOGRAFT_REJECTION	8.04E-05	15.819982
HALLMARK_MYOGENESIS	0.000108125	15.241303
HALLMARK_KRAS_SIGNALING_DN	0.000132237	14.849386
HALLMARK_PEROXISOME	0.000665308	11.736673
HALLMARK_IL6_JAK_STAT3_SIGNALING	0.000745466	11.519913
HALLMARK_INFLAMMATORY_RESPONSE	0.000839005	11.295052
HALLMARK_XENOBIOTIC_METABOLISM	0.000926918	11.105796
HALLMARK_KRAS_SIGNALING_UP	0.001067101	10.838793
HALLMARK_ANDROGEN_RESPONSE	0.005134025	7.9039854
HALLMARK_ESTROGEN_RESPONSE_EARLY	0.015542437	5.8958385
HALLMARK_APICAL_JUNCTION	0.020564313	5.3989919
HALLMARK_COAGULATION	0.025006386	5.0550421
HALLMARK_IL2_STAT5_SIGNALING	0.027092379	4.9149316
HALLMARK_P53_PATHWAY	0.031839421	4.6340908

Fig. 3.

Differences in drug sensitivity between clusters (A) Comparison of pathway GSVA score differences. (B) Bubble plot of functional enrichment results for each subtype. (C) Differences in sensitivity to common therapeutic drugs among different CBX subtype clusters, *P < 0.05, **P < 0.01, ***P < 0.001.

Differences in immune infiltration between clusters

Given the critical role of immunotherapy in LUAD, the infiltration of immune cells across different cluster subtypes was analyzed. Among the 22 immune cell types, 7 showed significantly different infiltration levels across clusters, including memory B cells, plasma cells, resting memory CD4 T cells, follicular helper T cells, monocytes, M0 macrophages, and resting mast cells (Table 3). Further analysis of the correlation between the infiltration levels of these 7 immune cell types and the expression levels of the CBX family genes was conducted (Fig. 4A). The results indicated that, except for CBX5, the expression levels of other CBX genes were associated with the infiltration levels of various immune cells. Notably, CBX2, CBX3, and CBX7 showed significant correlations with the infiltration abundances of multiple immune cell types, suggesting that changes in the expression of these CBX genes may influence immune cell infiltration in LUAD. Additionally, the ESTIMATE algorithm was used to calculate stromal, immune, and ESTIMATE scores for the different clusters. Cluster 4 displayed higher stromal and immune scores, resulting in lower tumor purity (Fig. 4B). This finding aligns with the favorable prognostic characteristics of cluster 4, which may be attributed to its microenvironment with heightened immune activity. Immunofluorescence staining for T. cells. CD4 memory. resting and B cells. memory in LUAD samples with varying CBX3 expression levels supported our bioinformatics findings, and these results are shown in Supplementary Fig. 2. Furthermore, we used CIBERSORT to quantify the abundance of 22 immune cell subtypes in all samples. The Kruskal-Wallis test was applied to analyze differences in immune cells across subtypes (Fig. 4D), and the results showed that 7 immune cell subtypes had significant differences among subtypes (p < 0.05). Meanwhile, we analyzed the correlation between CBX family genes and these differentially expressed immune cells. As shown in Fig. 4E, multiple immune cell subtypes were significantly correlated with CBX family genes (p < 0.05).

Table 3.

Immune cell types with differential infiltration abundance among different CBX subtype clusters.

Immune Cells	P value	F value
Resting Memory CD4+ T Cells	6.10E-07	25.56232347
Resting Mast Cells	0.001363796	10.37507772
Memory B Cells	0.001878213	9.772898881
Plasma Cells	0.00222703	9.453738341
M0 Macrophages	0.012375525	6.303707926
Monocytes	0.018722979	5.564894309
Follicular Helper T Cells	0.030848272	1.302280141

Fig. 4.

Differences in immune infiltration between clusters (A) Correlation analysis between CBX family expression levels and immune cell infiltration. The x-axis represents the seven immune cell types, and the y-axis shows different CBX genes. (B) Differences in stromal score, immune score, and ESTIMATE score among CBX-based clusters, (C) Box plots showing differences in immune score, stromal score, and ESTIMATE score among various subtypes. (D) Box plots showing differences in immune cells among various subtypes. (E) Heatmap of the correlation between immune cells and CBX genes, where orange indicates a positive correlation trend between genes and immune cells, and green indicates a negative correlation trend.*P < 0.05, **P < 0.01, ***P < 0.001.

CBX3 and CDX2 promote LUAD cell proliferation and colony formation

Analysis of TCGA data revealed that CDX2 expression was significantly upregulated in primary LUAD tissues (n = 483) compared to normal lung tissues (n = 59) (Fig. 5A, p < 0.001). Survival analysis indicated that high CDX2 expression was significantly associated with poor survival outcomes in patients with LUAD (Fig. 5B). To investigate the roles of CBX3 and CDX2 in LUAD progression, their mRNA expression levels in normal lung epithelial cells (HSAEC1-KT) and four LUAD cell lines (H2106, H2023, H1623, H2347) were examined using qPCR. Both CBX3 and CDX2 were significantly overexpressed in all LUAD cell lines compared to HSAEC1-KT cells (Fig. 5C-D), with H2023 exhibiting the highest expression and H2347 the lowest. Based on these findings, H2023 and H2347 were selected for further functional experiments. Genetic manipulation experiments demonstrated that knockdown of CBX3 or CDX2 in H2023 cells significantly inhibited cell proliferation (Fig. 5E) and reduced colony formation capacity (Fig. 5G-H). Conversely, overexpression of CBX3 or CDX2 in H2347 cells markedly enhanced proliferation rates (Fig. 5F) and promoted colony formation (Fig. 5G-I). Western blotting further validated that both CDX2 and CBX3 protein levels were significantly increased in primary LUAD tissues compared to normal lung tissues (Fig. 5J-L, p < 0.01).

Fig. 5.

CBX3 and CDX2 promote LUAD cell proliferation and colony formation (A) Box plot illustrating differential expression of CDX2 between LUAD tumors (n = 483) and normal lung tissues (n = 59) from the TCGA dataset. (B) Kaplan-Meier survival analysis demonstrating the association between CDX2 expression levels and survival probabilities in patients with LUAD. (C) qPCR analysis of CBX3 and (D) CDX2 mRNA expression in normal lung epithelial cells (HSAEC1-KT) and four LUAD cell lines (H2106, H2023, H1623, H2347), **P < 0.01 versus HSAEC1-KT (n = 6). (E) CCK-8 proliferation assay in H2023 cells after transfection with si-CBX3 or si-CDX2 (n = 6). (F) CCK-8 proliferation assay in H2347 cells following transfection with OE-CBX3 or OE-CDX2 (n = 6). (G) Representative images of colony formation assays. (H) Quantification of colony formation in H2023 cells and (I) H2347 cells (n = 6), ^##P < 0.01 versus si-NC or OE-NC. (J) Representative images of western blotting of CBX3 and CDX2 expression in normal samples or samples from patients with primary lung cancer. (K-L) Relative expression level of CBX3 and CDX2 expression in normal samples or samples from patients with primary lung cancer (n = 8), ^⁎⁎P < 0.01 versus Normal.

CBX3 and CDX2 promote the migration and invasion of LUAD cells

The results of the scratch assay demonstrated that knockdown of CBX3 or CDX2 in H2023 cells significantly inhibited their horizontal migration ability (Fig. 6A-B), while overexpression of CBX3 or CDX2 in H2347 cells significantly enhanced this ability (Fig. 6A and C). Transwell migration assays further confirmed that knockdown of CBX3 or CDX2 reduced the vertical migration ability of H2023 cells (Fig. 6D and F), whereas overexpression of these genes promoted vertical migration in H2347 cells (Fig. 6D and G). Additionally, Transwell invasion assays showed that CBX3 or CDX2 knockdown significantly weakened the invasion ability of H2023 cells (Fig. 6E and H), while their overexpression enhanced the invasion ability of H2347 cells (Fig. 6E and I). These results collectively suggest that CBX3 and CDX2 play pivotal roles in regulating the migration and invasion of LUAD cells.

Fig. 6.

CBX3 and CDX2 promote the migration and invasion of LUAD cells(A) Representative images of wound healing assays in H2023 and H2347 cells with CBX3 or CDX2 modulation. (B) Relative migration rate of H2023 cells and (C) H2347 cells (n = 6). (D) Representative images of Transwell migration assays in H2023 and H2347 cells with genetic modulation of CBX3 or CDX2. (E) Representative images of Matrigel invasion assays in H2023 and H2347 cells with CBX3 or CDX2 modulation. (F) Migrated cell count in H2023 cells and (G) H2347 cells (n = 6). (H) Invaded cell count in H2023 cells and (I) H2347 cells (n = 6), ^##P < 0.01 versus si-NC or OE-NC.

CDX2 binds to the CBX3 promoter and regulates its expression

To explore the regulatory effect of CDX2 on CBX3, overexpression and knockdown experiments of CDX2 were performed in H2347 and H2023 cells, respectively. qPCR analysis revealed that CDX2 overexpression significantly increased the mRNA expression of CBX3 in H2347 cells (Fig. 7A), whereas CDX2 knockdown significantly reduced CBX3 mRNA expression in H2023 cells (Fig. 7B). Reversion experiments further confirmed that the inhibition of CBX3 expression induced by CDX2 knockdown could be reversed by CDX2 overexpression in both H2347 and H2023 cells (Fig. 7C-D). Bioinformatics analysis of the JASPAR database identified three potential binding sites for CDX2 in the CBX3 promoter region (P1: −1082 to −1087 bp; P2: −1547 to −1554 bp; P3: −1735 to −1742 bp). ChIP assay results demonstrated that CDX2 significantly bound to the P2 region of the CBX3 promoter DNA in both H2347 and H2023 cells, suggesting a functional correlation between CDX2 and the P2 region (Fig. 7E). To further validate this, a luciferase reporter plasmid containing the wild-type 5′-untranslated region (5′-UTR) of CBX3 was constructed, along with three mutant reporter constructs in which the P1, P2, or P3 binding sites were individually disrupted. Given that transcription factors typically bind to specific regulatory regions to activate target gene expression, disruption of these binding sites should theoretically reduce transcriptional activity. Our data showed that mutation of the predicted P2 binding site within the CBX3 5′-UTR resulted in a significant reduction in luciferase activity (Fig. 7F-G), indicating that the P2 binding site plays a key role in regulating CBX3 expression. Additionally, luciferase reporter assays demonstrated that CDX2 overexpression significantly enhanced CBX3 promoter activity, while CDX2 knockdown significantly inhibited it (Fig. 7H-I). These results suggest that CDX2 binds to the P2 site in the CBX3 promoter region and positively regulates its transcription.

Fig. 7.

CDX2 binds to the CBX3 promoter and regulates its expression (A) qPCR analysis of CBX3 mRNA levels in H2347 cells following CDX2 overexpression, **P < 0.01 versus OE-NC (n = 6). (B) qPCR analysis of CBX3 mRNA levels in H2023 cells transfected with two independent si-CDX2 constructs (si-CDX2–1 and si-CDX2–2), ^##P < 0.01 versus si-NC (n = 6). (C-D) Rescue experiments showing CBX3 mRNA levels in H2347 and H2023 cells co-transfected with si-CDX2 and OE-CDX2 (n = 6). (E) ChIP assay demonstrating the binding of CDX2 to three predicted binding sites (P1-P3) within the CBX3 promoter in H2347 and H2023 cells (n = 6). (F-G) Luciferase reporter assays in H2347 and H2023 cells with mutant P1-P3 sites in the CBX3 promoter (n = 6). ^##P < 0.01 compared to WT; ^n.s.P > 0.05 compared to WT. (H-I) Luciferase reporter assays in H2347 and H2023 cells validating CDX2-mediated transcriptional activation of the CBX3 promoter (n = 6). ^&&P < 0.01.

Knockdown of CBX3 reverses the effect of CDX2 overexpression on LUAD progression

In H2347 cells, CDX2 overexpression significantly enhanced the horizontal migration ability of cells, an effect that was reversed by CBX3 knockdown (Fig. 8A and D). Transwell migration assays further confirmed that the enhanced vertical migration ability induced by CDX2 was dependent on CBX3 expression levels (Fig. 8B and E). Additionally, the Matrigel invasion assay showed that CDX2 overexpression significantly increased cell invasion ability, which was partially attenuated by CBX3 knockdown (Fig. 8C and F). To evaluate the CDX2-CBX3 regulatory axis in vivo, a nude mouse xenograft tumor model was used. Tumor growth curves revealed that the tumor volume growth rate was significantly accelerated in the CDX2 overexpression group, while this effect was partially alleviated in the CBX3 knockdown group (Fig. 8G). CDX2 overexpression led to a substantial increase in the final tumor volume and mass, with CBX3 knockdown partially reversing this tumor growth-promoting effect (Fig. 8H-I). The qPCR data from xenograft tumors confirmed the expression levels of CDX2 and CBX3 (Supplementary Fig. 3). These results collectively demonstrate that CDX2 promotes LUAD cell migration and invasion, as well as in vivo tumor growth, by upregulating CBX3 expression.

Fig. 8.

Knockdown of CBX3 reverses the effect of CDX2 overexpression on LUAD progression (A and D) Wound healing assays evaluating horizontal migration capacity of H2347 cells transfected with OE-CDX2 or OE-CDX2 + si-CBX3 (n = 6). (B and E) Transwell migration assays assessing vertical migration ability following indicated transfections (n = 6). (C and F) Matrigel invasion assays measuring invasive potential following indicated transfections (n = 6). (G) Nude mouse xenograft growth curves following injection of H2347 cells with different genetic modifications (n = 5). (H) Representative tumor images (n = 5). (I) Tumor weights, **P < 0.01.

Discussion

Epigenetic regulation has been implicated in the development of lung cancer, and its aberrations can serve as potential diagnostic markers and therapeutic targets [28]. As a critical epigenetic regulator, the CBX family plays a pivotal role in the progression of LUAD [29]. This study systematically analyzed the expression characteristics of eight CBX family members in LUAD, establishing, for the first time, a molecular typing system based on CBX gene expression profiles. Patients were classified into four subclusters, each with distinct prognostic implications. Consistent with previous studies, high expression of CBX1 and CBX3 correlated with poor prognosis in LUAD [23], while CBX6 and CBX7 exhibited specific downregulation. Notably, the proportion of early-stage cases in cluster 4, the subgroup with the best prognosis, was significantly higher than in other subclusters, suggesting that the CBX expression pattern may be closely linked to tumor progression stage. Further mechanistic studies revealed significant differences among subclusters in key pathways such as DNA repair, the p53 signaling pathway, and hypoxic stress. These pathway abnormalities not only account for the prognostic differences among subclusters but also highlight the dual role of the DNA repair system in LUAD: while it serves as a guardian of genomic stability in the early stages of tumorigenesis, it may contribute to treatment resistance in later stages [[30], [31], [32], [33]].

This study focused on the expression of CBX genes across different molecular subtypes, with key findings summarized as follows. First, to clarify the biological basis of each molecular subtype, an expression heatmap of all 8 CBX family members (CBX1-CBX8) across the four subtypes was constructed, which clearly revealed the unique CBX gene expression signatures of each subtype. Second, a further analysis of the expression differences of individual CBX genes between subtypes showed that the expression levels of CBX1 and CBX3 in Cluster 4 (where the prognosis is favorable) were lower than those in other subtypes. Third, regarding the association between gene expression and prognosis, the low expression of CBX1 and CBX3 in Cluster 4 was consistent with their role as poor prognostic markers, while the high expression of CBX7 in Cluster 4 suggested that this gene may have a protective effect. Moreover, this study highlighted the relationship between CBX molecular subtyping and chemotherapy sensitivity. Unlike previous studies focusing on individual genes, our analysis showed that different CBX subtypes exhibit specific response patterns to chemotherapy drugs. For instance, cluster 3 displayed more sensitive to docetaxel and gemcitabine, whereas cluster 4 was sensitive to doxorubicin but resistant to cyclopamine. These findings provide valuable insights for clinical decision-making in personalized treatment. In terms of tumor microenvironment regulation, this study enhanced our understanding of the immune regulatory functions of the CBX family. CIBERSORT was used to quantify 22 immune cell subtypes in all samples. The Kruskal-Wallis test identified 7 immune cell subtypes with significant differences across subtypes. Additionally, multiple immune cell subtypes showed significant correlations with CBX family genes. While earlier research indicated that CBX3 is associated with immunosuppression [34], different CBX family members have distinct effects on immune cell infiltration. CBX2, CBX3, and CBX7 were significantly correlated with the infiltration of various immune cells, particularly influencing the differentiation of CD4⁺ T cell subsets and B cell function. This complex regulatory network suggests that the CBX family may contribute to immune escape through multiple mechanisms, offering novel approaches for optimizing immunotherapy strategies. Thirty-five pathways with significant differences across the 4 subtypes were identified. For the 4 CBX-based subtypes, pathway differences between each subtype and others were compared separately. The top 10 most distinct pathways per subtype were selected, merged, and used to generate functional enrichment bubble plots. Notably, MYC pathway is significantly downregulated in Cluster 4; its abnormal activation links to poor prognosis, explaining Cluster 4′s better clinical outcomes.

The most significant breakthrough of this study is the identification of a novel regulatory axis, CDX2-CBX3. By integrating bioinformatics analysis and experimental validation, this study has demonstrated for the first time that the transcription factor CDX2 directly binds to the promoter region of CBX3 and activates its transcription. This discovery not only elucidates the carcinogenic role of CDX2 in LUAD but also offers crucial insights into the hierarchical relationships within epigenetic regulatory networks. Notably, CDX2 is generally regarded as a tumor suppressor in cancers such as colorectal cancer [35,36]. This discrepancy may arise from the tissue-specific nature of CDX2′s downstream target genes [37]. In the colon, CDX2 primarily regulates genes involved in intestinal differentiation [38], whereas in LUAD, it appears to activate oncogenic pathways, including CBX3. Furthermore, LUAD-specific epigenetic modifications may alter CDX2′s DNA-binding properties or co-regulator recruitment patterns [39,40]. Similar tissue-specific functional differences have been observed in other developmental transcription factors, such as those in the SOX and HOX families [[41], [42], [43], [44]], suggesting that the tissue context and microenvironment characteristics must be considered when evaluating the oncogenic functions of transcription factors.

In conclusion, this study provides a comprehensive analysis of the molecular characteristics and clinical relevance of the CBX family in LUAD, incorporating both multi-omics data and experimental validation. Not only has a CBX molecular typing system with clinical applicability been established, but a new therapeutic target has been identified through the discovery of the CDX2-CBX3 regulatory axis. Our findings experimentally validate a previously unreported regulatory mechanism in any cancer type, where CDX2 mediates the transcriptional activation of CBX3. Through extensive in vitro and in vivo functional validation, this study highlights the pivotal roles of CDX2 and CBX3 in promoting malignant phenotypes in LUAD. This novel insight into the CBX gene family’s involvement in lung cancer pathogenesis represents a substantial advancement in understanding LUAD and refining treatment strategies. However, several aspects of this study require further investigation. First, the effector molecule network downstream of the CDX2-CBX3 axis remains incompletely understood. Second, the interaction between this regulatory axis and tumor metabolic reprogramming warrants further analysis. Additionally, targeted intervention strategies for this pathway still need optimization. Future studies should focus on addressing these questions to advance precision treatment strategies for LUAD.

Ethical approval

The study was approved by the Medical Research Ethics Committee of Hebei Children's Hospital (No. 202,206–10).

Data availability

The datasets generated and/or analyzed during the course of this study are accessible from the corresponding author upon reasonable request.

CRediT authorship contribution statement

Shicheng Liu: Writing – original draft, Investigation, Formal analysis, Data curation, Conceptualization. Qingtao Zhao: Visualization, Validation, Resources, Investigation, Data curation. Dahu Ren: Validation, Software, Methodology, Formal analysis. Lingxin Kong: Visualization, Validation, Software, Formal analysis. Hongzhen Zhao: Resources, Methodology, Formal analysis. Guochen Duan: Writing – review & editing, Validation, Supervision, Project administration, Investigation, Funding acquisition, Conceptualization.

Declaration of competing interest

The authors declare that they have no conflict of interest.

Appendix. Supplementary materials

Supplementary Fig. 1: Validation of the efficiency of CDX2 overexpression or siRNA by qPCR analysis

(A) qPCR analysis of CDX2 mRNA levels in H2347 cells following CDX2 overexpression, **P < 0.01 versus OE-NC (n = 6). (B) qPCR analysis of CDX2 mRNA levels in H2023 cells transfected with two independent si-CDX2 constructs (si-CDX2–1 and si-CDX2–2), ^##P < 0.01 versus si-NC (n = 6). (B) qPCR analysis of CBX3 mRNA levels in H2347 cells following CBX3 overexpression, **P < 0.01 versus OE-NC (n = 6). (C) qPCR analysis of CBX3 mRNA levels in H2023 cells transfected with si-CBX3 constructs (si-CBX3), ^##P < 0.01 versus si-NC (n = 6).

Supplementary Fig. 2: Immunofluorescence staining for B cells. memory and T. cells. CD4 memory. resting in LUAD samples with different CBX3 expression levels

(A) Representative immunofluorescence staining for B cells. memory in LUAD samples with different CBX3 expression levels. (B) Representative immunofluorescence staining for T. cells. CD4 memory. resting in LUAD samples with different CBX3 expression levels. (C) The integrated density for CBX3 expression and B cells. memory in LUAD samples, ^⁎⁎P < 0.01 versus Group I (n = 6). (D) The integrated density for CBX3 expression and T. cells. CD4 memory. resting in LUAD samples, ^##P < 0.01 versus Group III (n = 6).

Supplementary Fig. 3: qPCR data from xenograft tumors confirming CDX2 and CBX3 expression levels

(A) qPCR analysis of CDX2 mRNA levels in xenograft tumors, **P < 0.01 versus OE-NC (n = 5). (B) qPCR analysis of CBX3 mRNA levels in xenograft tumors, **P < 0.01 versus OE-NC, ^##P < 0.01 versus OE-CDX2 (n = 5).

Supplementary Table 1. The cluster pathway enrichment results.