CRC patient tumor and paracancerous tissues were collected from the First Department of General Surgery, Shijiazhuang People’s Hospital (Hebei Province, China), and Hebei Medical University Fourth Hospital (China). Pathologists confirmed the diagnosis and staging of all patients, and the Ethical Committee of Shijiazhuang People’s Hospital approved this study. All patients provided written informed consent to participate.

Human HT29 and HCT116 cell lines were grown in McCoy’s 5A media (Gibco, CA, USA) containing 10% FBS and penicillin/streptomycin at 37°C in a 5% CO₂ incubator (Yang et al., 2021).

The Oncomine database (http://www.oncomine.com) (Rhodes et al., 2004) was utilized to evaluate the expression of 21 different CLDN family members in 20 cancer types, using the following criteria for differential expression: p = 0.01, Fold-change > 1.5, gene rank ≤10%.

The GEPIA database (http://gepia.cancer-pku.cn/) (Tang et al., 2017) was used to analyze colon adenocarcinoma (COAD) and rectal adenocarcinoma (READ) tissue samples in order to explore the relationships between different CLDN family members and key clinical outcomes including staging, overall survival (OS), and disease-free survival (DFS). Default GEPIA parameters were utilized for these analyses.

IHC staining data pertaining to the protein level expression of different CLDN family members were assessed with the Human Protein Atlas database (https://www.proteinatlas.org/) in CRC patient tumors and normal tissue samples (Pontén et al., 2008).

The association between different CLDN family members and the infiltration of immune cells into CRC tumors was assessed with the TIMER database (https://cistrome.shinyapps.io/timer/), with scatter plots for different CLDN genes being generated to demonstrate purity-corrected partial Spearman’s r values and corresponding significance metrics (Li et al., 2021).

Staging and nodal metastasis status for COAD (n = 324) and READ (n = 172) patient clinicopathologic parameters were evaluated using the UALCAN database (Lyu et al., 2020). According to multivariate Kaplan-Meier survival analysis, with p-values calculated using the log-rank test (log-rank test). The Ualcan database uses TCGA RNA-seq and clinical data for 31 cancer types.

The mutational status of different CLDN family genes in CRC was assessed using the cBioPortal database (http://cbioportal.org) (Cerami et al., 2012). Assessed mutations included deep deletions, missense mutations, copy number amplifications, and mRNA upregulation.

An interaction matrix for CLDN family genes was generated using the “matrix” command in the GeneNetwork database (Mulligan et al., 2017), enabling a correlation analysis between different CLDN family members.

The STRING database (http://string-db.org; version 11.0) was used to construct a CLDN family gene PPI network incorporating 24 co-expressed genes with a score of >0.4 (Franceschini et al., 2013). The network was visualized using Cytoscape (v 3.8.2).

GO analyses were used to assess the enrichment of particular genes in specific functional categories including biological processes (BPs), cellular components (CCs), and molecular functions (MFs), enabling efficient analyses of transcriptomic datasets (Thomas et al., 2019). The Kyoto Encyclopedia of Genes and Genomes (KEGG) database compiles functional data pertaining to a diverse array of regulatory pathways (Kanehisa et al., 2016). The online DAVID database (https://david.ncifcrf.gov/; version 6.8) was used for GO annotation and KEGG pathway analyses in this study (Huang et al., 2007). R was used to visualize the enrichment data.

Full-length wild-type protein sequences were obtained from UniProt (https://www.uniprot.org/) (UniProt Consortium, 2021), with Uniprot ID. O14493 and Uniprot ID. P01137 being used for CLDN4 and TGFβ1, respectively. Three-dimensional structures for these proteins were obtained from the RCSB PDB database (https://www.rcsb.org/) (Berman et al., 2000), using PDB IDs of 7KP4 and 5VQP, respectively, for docking analyses. The CLDN4 and TGFβ1 proteins were docked with the Zdock server 3.0.2 (https://zdock.umassmed.edu/) (Pierce et al., 2014). Prior to docking, PyMOL was utilized to remove water molecules, heteroatoms, and repeated subunits, with the best docked complex being assessed for hydrogen bonding.

The ISOGEN reagent (Nippon Gene Co. Ltd., Kokyo, Japan) was utilized to isolate RNA from tissue and cell line samples, after which qPCR was conducted by initially reverse transcribing 1 µg of total RNA per sample using a Revert Aid first strand cDNA synthesis kit (ThermoFisher Scientific). Primers used for qPCR are compiled in Supplementary Table S1. A Real-Time PCR system (BIOER Co. Ltd., Kokyo, Japan) was used for all qPCR analyses, and the relative expression of CLDN11, CLDN4, and TGFβ1 was assessed via the ΔCT method as in prior reports (Yang et al., 2021). For appropriate experiments, the HT29 and HCT116 cells were treated for 48 h with the TGFβ1 inhibitor LY364947 (5.00 or 10.0uM).

HT29 and HCT116 cells were treated for 48 h with LY364947 at concentrations of 5, and 10 μM. After being washed three times with PBS, the cells were lysed for 30 min on ice before being centrifuged at 10000 g for 5 min at 4°C for 5 min. The protein concentrations in cell lysates were determined using an ND-1000 Spectrophotometer (National Instruments) (NanoDrop, ThermoFisher Scientific). SDS-PAGE was used to separate out equal amounts of total protein, which was then transferred to PVDF membranes (Millipore, Billerica, MA, USA). After blocking with 5% milk for 2 h, we did overnight immunoblotting with primary antibodies at 4°C. The primary antibodies were raised against TGFβ1 (1:1,000; Abcam, Cambridge, MA, USA), CLDN4 (1:500; abways technology) and β-actin (AC026, at 1 : 10000 dilution). A secondary antibody of goat anti-rabbit IG (KPL074-1,506, at 1: 5,000 dilution) was then used to incubate the membranes for 1 h at 37°C. Fusion FX5 Spectra was used to determine the intensity of protein bands (Fusion, France). The membranes were then incubated for 1 h at 37°C with secondary antibodies of goat anti-rabbit IG (KPL074-1,506, at 1:5,000 dilution). The intensity of protein bands was estimated via executing Fusion FX5 Spectra (Fusion, France).

We began by using the Oncomine database to evaluate CLDN family member expression in CRC. In total, 19 CLDNs were found to be differentially expressed between normal tissues and COAD and READ patient samples. At the mRNA level, CLDN1, CLDN2, CLDN12, and CLDN14 were upregulated whereas CLDN3, CLDN5, CLDN7, CLDN8, CLDN11, CLDN15, and CLDN23 were downregulated (FC > 1.5) in patients with CRC (Figure 1). Specifically, fold-change values for CLDN1, CLDN2, and CLDN12 in COAD tissues were 6.755, 3.006, and 2.096, respectively (Supplementary Table S1), while in READ tissues these respective fold-change values were 4.301, 1.849, and 1.583 in the dataset analysis of Kaiser et al. (2007). CLDN19 expression was altered by 1.117 and 1.158-fold in COAD and READ tissues from this same dataset, while CLDN20 expression was changed by 1.075- and 1.091-fold, respectively. Gaedcke et al. (2010) found fold-change values for the expression of CLDN1, CLDN10, CLDN11, CLDN15, and CLDN22 of 19.563, 1.583, 1.718, 1.417, and 1.05, respectively, in READ tissues. Skrzypczak et al. (2010) further reported CLDN1 and CLDN2 expression levels in CRC tissues that were altered by 6.351- and 7.754-fold, respectively. In the TCGA dataset, CLDN14 expression in READ and COAD was altered by 4.876-fold and 4.368-fold, respectively, while it was altered by 3.471-fold in the dataset generated by Gaedke et al. CLDN18 expression was altered by 1.298-fold and 1.184-fold in COAD data from TCGA and Kaiser et al., and by 1.216-fold and 1.25-fold in READ samples from these two respective data sources.

The GEPIA database was next utilized to compare the expression of different CLDN family members in CRC and in normal colon tissue samples, revealing CLDN1, CLDN2, CLDN3, CLDN4, CLDN7, and CLDN12 to have been upregulated and CLDN5 and CLDN11 to have been downregulated in COAD and READ tissues relative to control samples (Figure 2). Significant differences in CLDN3, CLDN4, CLDN5, CLDN6, CLDN9, CLDN11, and CLDN12 expression were evident as a function of tumor stage (Figure 3). CLDN3, CLDN4 and CLDN6 were related with clinicopathological stage in CRC. The more CLDN5, CLDN9, and CLDN11 are expressed in CRC, the worse the stage. CLDN12 expression in CRC associated with pathological stage. We additionally examined the link between CLDN family gene mRNA levels and CRC patient lymph node metastasis, revealing a significant relationship between the expression of CLDN1, CLDN2, CLDN3, CLDN7, CLDN8, CLDN9, CLDN11, CLDN14, CLDN16 and CLDN23 nodal metastasis in COAD and READ (Figures 4A, B, C, G, H, I, K, M, N, R) (Figures 5A, B, C, E, F, G, I, K, M, O).

Kaplan-Meier curves and log-rank tests were next performed, revealing a relationship between the upregulation of CLDN11, CLDN14, and CLDN23 at the mRNA level and COAD and READ patient OS (Figure 6). Specifically, high levels of CLDN11 and CLDN14 expression were correlated with worse patient OS, whereas CLDN23 overexpression was linked to better OS outcomes.

The Human Protein Atlas yielded findings that were somewhat consistent with the mRNA level data above, with higher CLDN1 and CLDN12 protein levels and lower CLDN11 levels in CRC tumors relative to samples of normal colon tissue (Figure 7).

Using the TIMER database, we detected a negative correlation between the mRNA level expression of CLDN5, CLDN8, CLDN11, and CLDN18 and CRC tumor purity (Figures 8E, H, J, O). Moreover, CLDN12 expression was correlated with B cell infiltration and positively correlated with CD8⁺ T cells (Figures 8K). Additionally, CLDN5, CLDN9, CLDN11, CLDN15, and CLDN20 were significantly positively correlated with CD4⁺ T cell infiltration into CRC tumors, while CLDN7 expression was negatively correlated with such infiltration (Figures 8E, H, J, M, Q). Additionally, a clear positive correlation was observed between macrophage infiltration and the expression of CLDN1, CLDN9, CLDN11, and CLDN16 (Figures 8A,H,J,N). Neutrophil infiltration was negatively correlated with CLDN15 expression and positively correlated with CLDN1 and CLDN16 expression (Figures 8A, M, N), while CLDN11 expression was positively correlated with dendritic cell infiltration (Figures 8J).

Next, we assessed CLDN family gene mutations in 640 COAD samples using the online cBioPortal tool, revealing mutations in 283 samples (44%) (Figures 9A). Mutations in CLDN10 (6%) were attributable to missense mutations and gene amplification.

The Genenetwork database was also used to explore relationships among CLDN gene expression levels in COAD samples, revealing the following correlative relationships: CLDN1 with CLDN2; CLDN3 with CLDN4, CLDN7, CLDN8, CLDN14, and CLDN23; CLDN4 with CLDN7, CLDN8, and CLDN14; CLDN5 with CLDN10, CLDN11, CLDN18, and CLDN24; CLDN7 with CLDN8 and CLDN14; CLDN8 with CLDN14; CLDN9 with CLDN3, CLDN4, CLDN7, CLDN8, CLDN14, CLDN15, and CLDN23; CLDN10 with CLDN18 and CLDN24; CLDN11 with CLDN10, CLDN18, and CLDN24; CLDN15 with CLDN3, CLDN4, CLDN7, CLDN8, CLDN14, and CLDN23; CLDN18 with CLDN24; and CLDN23 with CLDN4, CLDN7, CLDN8, CLDN14, and CLDN23. A co-expression network for these CLDNs is shown in Figures 9B.

A PPI network for these CLDN family genes was generated composed of 24 nodes and 254 edges (Figures 9C).

The DAVID database was utilized to predict the functional roles of different CLDNs in key CRC-related biological processes, molecular functions, and cellular components (Figures 10A). KEGG pathway analyses further revealed a close relationship between these CLDN family members and tight junctions, cell adhesion molecules, leukocyte transendothelial migration, and Hepatitis C (Supplementary Table S2).

A prior study of inflammatory bowel disease demonstrated the ability of TGFβ1 signaling inhibition to suppress CLDN4 expression (Marincola Smith et al., 2021). A molecular docking analysis was therefore conducted to explore putative interactions between CLDN4 and TGFβ1. For this approach, ZDOCK scores were used to estimate protein-protein binding affinity, with higher scores being indicating of greater binding affinity (Pierce et al., 2011). This approach revealed that nine hydrogen bonds were predicted to form between CLDN4 and TGFβ1, with interactions between Glu109 of CLDN4 and Gln177 of TGFβ1, Cys107 of CLDN4 and Gly173 and Arg181 of TGFβ1, Val95 of CLDN4 rand Ser244 of TGFβ1, Ser98 of CLDN4 and His247 and Arg249 of TGFβ1, Trp18 of CLDN4 and Leu242 of TGFβ1, Ala20 of CLDN4 and Glu261 of TGFβ1, and Leu173 of CLDN4 and Thr260 of TGFβ1 (Figures 11A).

At the mRNA level, we found that CLDN11 and CLDN4 expression were respectively reduced and increased in CRC tumor samples relative to paracancerous controls as measured via qPCR (n = 10, p < 0.05). At the protein level, CLDN4 expression was increased by WB (n = 5, p < 0.05) (Figures 11B,C,D) (Supplementary Table S3). Treatment with LY364947 (TGFβ1 inhibitor) at a dose of five or 10 uM was sufficient to suppress CLDN4 mRNA and protein expression at 48 h post-treatment in both HT29 and HCT116 cells (Figure 11E, F, G, H, I, J).