We collected paraffin specimens from individuals with CC after radical resection at the Department of Gastrointestinal Surgery, Guangxi Medical University Cancer Hospital, between November 2014 and February 2021. In this study, we included 46 adjacent normal tissues and 113 patients with primary CC. The patients underwent radical resection before radiotherapy and chemotherapy, and the diagnosis was established by two pathologists after evaluating the paraffin specimens. The 2019 World Health Organisation Classification of Tumours of the Digestive System was used to guide all pathological diagnoses. Data on survival were collected through telephonic follow-ups. In addition, we collected detailed clinical information, including age, sex, carcinoembryonic antigen (CEA), tumour size, differentiation, peripheral nerve invasion, lymphovascular invasion, deoxyribonucleic acid mismatch repair (MMR), and the American Joint Committee on Cancer tumour node metastasis (TNM) staging system (seventh edition). All participants provided written consent for this research, and this study was approved by the ethics committee of Guangxi Medical University Cancer Hospital (approval number: LW2022123).

The sections that had been embedded in paraffin underwent IHC staining. In summary, sections (4 mm thick) were formalin-fixed, paraffin-embedded, regularly deparaffinised, rehydrated, and heated in antigen retrieval citrate buffer (pH 6.0). After that, non-specific antigens were blocked for 20 min with 10% normal goat serum at 37°C and endogenous peroxidase for 20 min with 3% hydrogen peroxide. Subsequently, the rabbit polyclonal antibody anti-KREMEN2 (orb185896; 1:100 dilution; Biorbyt, Cambridge, UK) was incubated at room temperature for 1 h. This was followed by conjugation to the secondary antibody (KIT-5010; Maixin Inc., Fujian, China) and 3, 3′-diaminobenzidine (DAB) (DAB-1031; Maixin Inc., Fujian, China) staining. A primary antibody that had been replaced with phosphate-buffered saline buffer was employed as a negative control.

Independent evaluation and scoring of the IHC data were performed by two researchers who were unaware of the clinicopathological characteristics of the individuals. A semi-quantitative scoring method based on intensity scores (InSs) and proportion scores (PSs) was used to assess protein expression. The InSs were stated as follows: 0, no staining; 1, light brown; 2, brown; 3, dark brown. The PSs of stained epithelial cells were as stated: 0% (0), 1%–25% (1), 26%–50% (2), 51%–75% (3), and 76%–100% (4). The formula IS = PS × InS was used to determine the final IHC scores (ISs). Lastly, according to the Kremen2 protein expression, the samples were sorted into high and low expression groups (IS ≥ 8 and IS < 8, respectively).

First, we collected gene expression information of different types of tumours in the Tumour IMmune Estimation Resource (TIMER) database (http://timer.cistrome.org/). Further, we downloaded messenger ribonucleic acid sequencing (mRNA-seq) (level 3, HTSeq-Fragments Per Kilobase of transcript per Million mapped reads data) of the colon adenocarcinoma (COAD) cohort and relevant clinical data from TCGA (https://portal.gdc.cancer.gov/). These data were obtained from 30 normal colon tissue samples, 330 colon tumour tissue samples, and 316 patients with CC (basic information, including age, sex, and the TNM stage, while cases without mRNA-seq data and cases with incomplete clinical data and overall survival (OS) of <30 days were excluded). The characteristics of CC patients are shown in Table 1. Perl was used for data processing. Data processing comprised Integration and normalisation of the transcriptome profiling data, the extraction of clinical information, and the merging of the mRNA data.

R packages “ggpubr” and “limma” were used to compare Kremen2 mRNA expression between the colon tumour and normal tissue samples (13). On the basis of the levels of Kremen2 mRNA expression in tumour tissues, 316 patients were subsequently enrolled and separated into two groups (the high-expression and low-expression groups). R packages “ggpubr,” “limma,” “survival,” and “survminer” were employed to assess the significance of the Kremen2 mRNA expression level with different pieces of clinical information and the role of Kremen2 in predicting OS (14). Whether the Kremen2 mRNA expression level could independently predict survival in patients with CC was analysed using a multivariate Cox regression model. Our research complied with the TCGA publication guidelines.

The differences between 22 immune cell subtypes in patients with CC belonging to the high- and low-expression groups were analysed with the CIBERSORT method combined with the LM22 feature matrix to estimate the infiltration status of immune cells (15, 16). The association of the Kremen2 mRNA expression level with immune cell infiltration was analysed using the Spearman correlation coefficient.

R software v4.1, Statistical Package for the Social Sciences v21.0, and GraphPad Prism 8 were utilized to perform the statistical analysis. A non-parametric test was employed to determine the variation in Kremen2 protein expression between CC and adjacent normal tissues. The association of the Kremen2 protein expression level with different pieces of clinical information was analysed with the help of Chi-square or Fisher’s exact tests. The Kaplan–Meier (K–M) analysis was employed to assess the role of Kremen2 in predicting the OS. Receiver operating characteristic (ROC) curves were utilised for measuring the prognostic capacity of Kremen2 protein expression. In order to determine if Kremen2 might be employed as an independent prognostic indicator in individuals with CC, univariate and multivariate Cox regression analyses were carried out. p < 0.05 was considered a significant value.

Based on the TIMER data, we observed Kremen2 mRNA overexpression in various tumours, including CC (p < 0.001) (Figure 1A). TCGA data analysis revealed that Kremen2 mRNA levels were significantly elevated in CC tissues (Figures 1B, C, p < 0.001). We analysed the ISs in 113 tumour tissues and 46 paired adjacent normal tissues to verify Kremen2 protein expression in CC. Using the non-parametric test, we observed that Kremen2 protein expression in CC was remarkably higher in comparison to the adjacent normal tissues (p < 0.001) (Figure 2).

Subgroup analysis was performed based on the age, sex, tumour site, TNM stage, CEA, tumour size, differentiation, peripheral nerve invasion, lymphovascular invasion, and MMR for determining the association of Kremen2 protein expression and clinicopathological parameters. In contrast with individuals with CC having no lymph node metastasis, individuals with CC having lymph node metastasis had substantially greater levels of Kremen2 protein expression (p = 0.036). In addition, we observed that Kremen2 protein correlated with sex and age in patients with CC. Female patients (p = 0.025) and patients aged <55 years (p = 0.013) tended to have a high Kremen2 protein expression. However, there was no considerable association between Kremen2 protein expression in CC and other clinicopathological parameters (p > 0.05) (Table 2).

When we examined the relationship between clinicopathological parameters and Kremen2 mRNA expression in patients with CC based on TCGA, we observed that high Kremen2 mRNA expression was linked to tumour N stage (from N0 to N2 and from N1 to N2) (p = 0.005, 0.012) (Figure 3A). In terms of the T stage, progression from T2 to T3 was correlated with elevated Kremen2 mRNA expression levels (p = 0.032) (Figure 3B). However, Kremen2 mRNA expression was not related to sex, age, or the M stage (Figures 3C–E).

The K–M survival curve was implemented to determine the association of Kremen2 expression and survival in CC. As per the K–M chart, CC cases with elevated Kremen2 mRNA expression had a shorter OS (TCGA, p = 0.043) (Figure 4A) and progression-free survival (TCGA, p = 0.041) (Figure 4B). The K–M survival curve also revealed that high Kremen2 protein expression was linked to a poor prognosis (IHC, p = 0.013) (Figure 4C). Subsequently, the ROC curve analysis was employed to analyse the predictive value of Kremen2 protein expression, promising a prognosis value of the signature for CC over survival. The area under the curve scores over 1, 3, and 5 years (0.769, 0.636, and 0.651, respectively), were observed on the ROC curves (Figure 4D).

Using clinical data from TCGA, we observed that age (p = 0.014), the T stage (p = 0.007), the N stage (p = 0.012), and the M stage (p < 0.001) were independent risk factors for the prognosis of individuals with CC (Figure 5). As per the univariate analysis results, high Kremen2 mRNA expression was strongly related to poor prognosis (p = 0.001), while it was not correlated in the multivariate analysis (p = 0.107). In addition, the univariate Cox regression analysis revealed a substantial correlation between high Kremen2 protein expression and a poor prognosis (p = 0.017). Other clinical parameters, including the N stage (p = 0.009), the M stage (p = 0.000), and peripheral nerve invasion (p = 0.018), were linked to a poor prognosis (Table 3). Furthermore, the multivariate analysis revealed that high Kremen2 protein expression (p = 0.036) and the M stage (p = 0.001) were independently linked with a poor prognosis (Table 3).

Tumour-infiltrating lymphocytes can independently reflect cancer survival (17). Herein, we assessed the link between Kremen2 mRNA expression and the infiltration of 22 immune cell subtypes in TCGA-COAD. Regulatory T cells (Tregs) (p = 0.034) and activated natural killer (NK) cells (p = 0.020) were the immune cells with a substantially higher infiltration level in the high-expression group. CD4⁺ memory resting T cell infiltration level was substantially higher in the low-expression group in comparison to that in the high-expression group (p = 0.031) (Figure 6A). According to the correlation analyses, CD8⁺ T cell and activated NK cell infiltration were positively linked to Kremen2 mRNA expression (Figures 6B, C, r = 0.2, p = 0.022; r = 0.2, p = 0.022).