GEPIA (http://gepia.cancer-pku.cn/) is a newly developed interactive website that integrates TCGA and GTEX gene expression information, including extensive RNA raw sequencing expression data (13). Users can quickly find the expression of the target gene in the corresponding tumor by entering the gene symbol or Ensembl ID. In addition, GEPIA can also perform prognostic survival analysis, including two types of overall survival (OS) and disease-free survival (DFS). GEPIA can also analyze target genes and potential transcription factors by analyzing the correlation between two associated genes.

The HPA database (https://www.proteinatlas.org/), dedicated to providing tissue and cellular distribution information for all 24,000 human proteins, is freely available for public inquiry (14). Users can query gene expression (RNA and protein expression) in various tissues by entering the target gene. According to the HPA database, we determined the expression of PRDXs in various solid tumors, immunohistochemical images of normal and cancer tissues from CRC patients, and the localization images of PRDXs in cells.

UALCAN (http://ualcan.path.uab.edu/) is a comprehensive and easily accessible web-based database built on TCGA (15). The database provides gene expression in different cancer tissues and corresponding normal tissues and allows comprehensive analysis of the association between target genes and multiple clinical characteristics. Our study utilized UALCAN to analyze the association between specific PRDXs and CRC clinic characteristics.

The CancerSEA database covers 41,900 single-cell functional annotation information of 25 types of tumors involving 14 cell functional states such as proliferation, invasion, and apoptosis (16). It enables users to investigate the relationship between the functional phenotypes of genes of interest and various tumors. For example, the CancerSEA was utilized to analyze the association between the PRDX family and CRC cell function.

The cBioPortal database (https://www.cbioportal.org) combines 126 tumor genome studies, including the cancer cell line encyclopedia and TCGA (17). cBioPortal integrates genomic data such as somatic mutations, DNA copy-number alterations (CNAs), gene expression (mRNA, microRNA, protein), DNA methylation, and phosphorylated protein enrichment.

The String (https://www.string-db.org) is a database for searching the interactions between known proteins and predicted proteins. The database includes over 5,000 species, 24 million protein species, and over 20 million protein interaction links (18). In addition, we collected and integrated proteins and potential interactions related to PRDX family members by constructing a protein-protein interaction (PPI) network.

The LinkedOmics database (http://www.linkedomics.org) includes multi-omics data and clinical data for 32 cancer types and data from 11,158 patients in the TCGA project (19). We can perform correlation analysis between target genes and other genes by inputting PRDX family genes, selecting cancer types and data sets, GO, KEGG, GSEA, and other analyses. Through this database, we can also obtain a series of visualization results, such as volcano plots of related genes and heat maps of positive and negative related genes.

The CCLE is a database of cancer cell lines maintained by the Broad Institute of MIT, which currently has more than 1,000 cell lines (20). The PRDX4 expression data in CRC cells were obtained directly from the CCLE website (https://www.betastasis.com/). We evaluated the expression of PRDX4 in 57 CRC cell lines by CCLE.

HCT116 cells were purchased from the American type culture collection (ATCC) and cultured in RPMI-1640 medium (Hyclone, UT, USA) containing 10% fetal bovine serum (PAN-Biotech, Adenbach, Bavaria). The short hairpin RNA (shRNA) was synthesized by Sigma-Aldrich (NM_006406). The target sequence was as follows: PRDX4 5′-CCACACTCTTAGAGGTCTCTT-3′ (TRCN000064818, Sigma). In addition, the shRNA vector targeting GFP (SHC005, Sigma) was used as a knockdown control. We observed and photographed the cells using immunofluorescence microscopy, and the number of transfected cells more significant than 95% was considered successful. Moreover, RT-qPCR was applied to detect knockdown efficiency.

We used RNA (Promega, Madison, USA) kit to extract the total RNA. The extracted RNA was then reverse-transcribed into cDNA using the PrimeScript™ cDNA synthesis kit (Takara, Japan). Finally, the results were calculated using the 2^-ΔΔCt method.

An extraction buffer containing protein lysis (Solarbio, Beijing, China) was used to obtain the total protein. Then the extracted protein was quantitatively analyzed using a BCA kit (Solarbio, Beijing, China). First, an equal amount of 30 μg proteins was added to each lane for electrophoresis and membrane transfer. Next, the protein bands were incubated with the primary antibody overnight at 4°C and the secondary antibody for one hour at room temperature. Finally, the western blot results were detected using the ECL chemical reagent method.

A key method for identifying cancer stem cells (CSCs) in vitro is their capacity to form spheres. Therefore, we used the sphere formation assay to assess the stemness of HCT116 cells in this work. First, cells were trypsinized to make a single-cell suspension, then seeded into six-well ultra-low cluster plates (4×10³ cells/well) and cultured in a serum-free medium (DMEM/F-12, Sigma, USA) containing 1% B27 (Gibco, USA), 0.02 μg/ml EGF (Peprotech, USA) and 0.02 μg/ml bFGF (Peprotech, USA) for 15 days. The sphere volume and number were observed under a microscope (Olympus, Japan) every four days, and a new medium was added.

The specific methods are consistent with our previous work (21). Briefly, for colony formation, PRDX4 knockdown and control cells were seeded into 6-well plates at a density of 300 cells/well and incubated in a cell incubator for two weeks. Then, the cells were fixed with 4% formaldehyde for 30 min and stained with crystal violet for 20 min. Colonies were counted under a microscope and images were collected. For apoptosis assay, the PRDX4 knockdown and control cells were disaggregated with trypsin and washed once with PBS. Then, 1 × 10⁶ cells were resuspended in PBS and stained using allophycocyanin and propidium iodide according to the manufacturer’s instructions. For transwell assay, the upper chamber of the 24-well plate was coated with diluted Matrigel and serum-free medium was added, and the lower chamber was filled with medium containing 20% FBS. The transfected cells (5 × 10⁵ cells/well) were seeded into the serum-free medium and incubated in a cell incubator for 48 h. Then, the cells were fixed and stained with 4% formaldehyde and 0.2% crystal violet, respectively. The invaded cells were photographed and counted under an inverted microscope.

The nude mice (BALB/c, four-week-old) were acquired from the Animal Center of North Sichuan Medical College. All experimental procedures were under the ARRIVE guidelines and approved by the Animal Ethics Committee of North Sichuan Medical College (No. SCXK2022-0018). All mice were randomly divided into two groups (sh-Control, n=5; sh-PRDX4, n=5). A total of 5×10⁶ cells were subcutaneously inoculated into each nude mouse. Tumor volume and weight were recorded every four days for four weeks after injection. After treatment, all mice were euthanized to isolate tumors, and tumor weights were measured. The immunohistochemistry (IHC) method is consistent with that described in our previous work (21).

The web database generated all statistical analysis results, which were then calculated using the GraphPad Prism 9.0 software. Student’s t-test analyzed differences between the control and transfection groups. All experiments were repeated more than three times. P < 0.05 represents a statistical difference.

A total of 31 solid tumors were analyzed using the GEPIA database for the RNA expression of each member of the PRDX family ( Figure 1 ). Specifically, PRDX1 was highly expressed in BRCA, CESC, DLBC, GBM, PAAD, TGCT, THYM, and UCEC, and lowly expressed in KICH and LAML ( Figure 1A ). PRDX2 was highly expressed in DLBC, LUSC, PAAD, TGCT, THYM and UCEC, while it was lowly expressed in LAML ( Figure 1B ). PRDX3 was highly expressed in DLBC, PAAD, SKCM, STAD, TGCT and THYM ( Figure 1C ). PRDX4 was highly expressed in BLCA, COAD, DLBC, GBM, KIRC, LGG, LUAD, LUSC, PRAD, READ, SKCM, THYM and UCEC, and was lowly expressed in LAML ( Figure 1D ). PRDX5 was highly expressed in COAD, DLBC, LIHC, PAAD, READ, THYM, UCEC and UCS, and low in ESCA and LAML ( Figure 1E ). PRDX6 was highly expressed in DLBC, GBM, OV and THYM and low in LAML ( Figure 1F ). Furthermore, we focused on the expression of PRDXs in COAD. Although PRDX1, PRDX2, and PRDX3 were highly expressed in COAD, their expression was not statistically significant ( Figures 2A-C ). PRDX6 was lowly expressed in COAD, but also not statistically significant ( Figure 2F ). Among the PRDX family, only PRDX4 and PRDX5 were highly expressed in COAD with statistical significance ( Figures 2D, E ).

IHC results from the HPA database were used to analyze protein expression to confirm the PRDX family’s gene expression trends in CRC. Based on the degree of staining, the protein expression of PRDX2 and PRDX4 was significantly higher in colon cancer tissues than in normal tissues ( Figures 3B, D ). There is no significant difference in the protein expression of PRDX1 and PRDX3 in normal tissues and cancer tissues ( Figures 3A, C ). Conversely, the protein expression of PRDX5 and PRDX6 was lower in colon cancer tissues ( Figures 3E, F ).

It is well known that tumor node metastasis (TNM) staging remains the crucial criterion for determining the prognosis of CRC patients. Therefore, we further evaluated the correlation between PRDX family members and CRC pathological stage. Interestingly, only the expression of PRDX2, PRDX3, and PRDX4 indicated significant variability (P < 0.05) ( Figures 4B-D ). In contrast, there was no correlation between PRDX1, PRDX5, and PRDX6 and the pathological stage of CRC (P > 0.05) ( Figures 4A, E, F ).

We investigated the relationship between mRNA expression and OS and DFS in CRC patients using the GEPIA database. As shown in Figure 5 , PRDX family members showed no significant difference with DFS. Surprisingly, only PRDX4 expression was statistically significant in terms of OS, and low PRDX4 expression was associated with a poor prognosis ( Figure 6D ). Therefore, we further analyzed the relationship between PRDX4 and the clinical characteristics of CRC patients in the UALCAN database. The results revealed that the expression of PRDX4 was significantly correlated with individual cancer stages, race, gender, weight, age, histological subtypes, metastasis status, and TP53 mutation status (P < 0.001) ( Figures 7A-H ). Given these findings, we speculated that PRDX4 might be a potential prognostic marker for CRC.

PRDXs genetic alterations in CRC were analyzed using the cBioPortal database. The results revealed a minimum of four alteration types (PRDX3/5/6) and a maximum of six alteration types (PRDX1/2) ( Figure 8A ). Overall, PRDXs were altered in 61 of the 107 CRC patient samples ( Figure 8B ). Furthermore, PRDX1, PRDX2, PRDX3, PRDX4, PRDX5, and PRDX6 were altered in 15%, 11%, 10%, 7%, 6%, and 8% of CRC samples, respectively ( Figure 8B ). Moreover, the mRNA expression heatmap of PRDXs revealed that PRDX1 and PRDX3 had higher mutation rates ( Figure 8C ).

Furthermore, we performed PPI network analysis for each PRDX using the STRING to explore potential interactions between proteins. The results showed that there were 61 protein nodes related to the SFRP family in the constructed PPI network ( Figure 8D ).

RNA-seq technology’s emergence allows for exploring tumor cells’ functional heterogeneity. The association of each PRDX family member with 14 functional states in CRC cells is shown in Figure 9 . As a potential clinical indicator of CRC progression, PRDX4’s role in cancer biology attracted our attention. According to cell function analysis, PRDX4 positively correlated with proliferation, apoptosis, and invasion of CRC cells but negatively correlated with stemness, which was closely associated with prognosis ( Figure 9 ). Based on these results, it deserves to verify whether PRDX4 can have a similar effect on CRC cells by further in vitro analysis.

We further analyzed RNA-seq sequencing data of PRDX4 from 379 CRC patients using the LinkedOmics database. The results of the volcano plot displayed that 11628 genes (red dots) were positively correlated with PRDX4, while 8200 genes were negatively correlated (green dots) ( Figure 10A ). Moreover, Figures 10C, D show that 50 significant genes were positively and negatively correlated with PRDX4. Among them, the strongest positive correlation was found between PRDX4 and UXT (Pearson correlation = 0.76, P = 1.34e-71), whereas the strongest negative correlation was found with SPTAN1 (Pearson correlation = 0.57, P = 8.02e-34). Furthermore, KEGG pathway analysis showed that PRDX4 positively correlated with several major cellular functional processes, including DNA replication ( Figure 10B ). Moreover, through cellular localization analysis, we found that PRDX4 is mainly localized to the endoplasmic reticulum ( Figure 10E ). In addition, it is partially localized to the cytosol ( Figure 10F ).

The CCLE database was applied to analyze the expression of PRDX4 in various CRC cell lines. The results suggested that among 57 CRC cell lines, the expression of PRDX4 in COLO320, HCT116, HCT15, KM12, LS411N, NCIH716, and RKO was significantly higher than in other cell lines ( Figure 11A ). Therefore, we selected one of the most common HCT116 cells for further in vitro and in vivo analysis. Furthermore, single-cell sequencing data (GSE81861) from the cancerSEA database revealed 14 functional cell states, such as apoptosis, proliferation, invasion, and stemness, in HCT116 cells without prior treatment ( Figure 11B ).

Further analyses of PRDX4 cellular functions were performed both in vitro and in vivo to validate the cancer SEA database findings. Firstly, HCT116 cells with stable knockdown expression PRDX4 were established by lentiviral transfection ( Figure 12A ). RT-qPCR and western blot assays were then performed to validate the knockdown efficiency of the lentivirus. Both mRNA and protein levels were significantly reduced by targeting shRNA to PRDX4 ( Figures 12A, B ).

Numerous studies have found a strong link between the stemness of cancer cells and the prognosis of cancer patients (22). Consequently, a sphere formation assay was utilized to determine the effect of PRDX4 expression level on the stemness of cancer cells. The results showed that knocking down PRDX4 dramatically promoted the sphere-forming capability of cancer cells ( Figure 12C ). As shown in Figures 12D–F , knocking out PRDX4 inhibited HCT116 cell proliferation and invasion while promoting apoptosis. Notably, the experimental results of PRDX4 apoptosis were not consistent with the bioinformatics analysis in Figure 9 . Furthermore, the effect of PRDX4 on cancer cell proliferation in vivo was assessed by injecting sh-Ctrl or sh-PRDX4 cells into nude mice. The results suggested that the tumor volumes were significantly reduced in the sh-PRDX4 group compared to the control group ( Figure 12G ). Proliferating cell nuclear antigen (PCNA) was used to indicate tumor proliferation. IHC staining revealed that the expression of PCNA in the sh-PRDX4 group was significantly lower than in the control group ( Figure 12H ). Taken together, these results demonstrate the involvement of PRDX4 in regulating CRC cell progression.