MUC1 promoter methylation pattern diversity and its association with TET3 expression and prognosis in cholangiocarcinoma

Abstract

Cholangiocarcinoma (CC) is a highly lethal malignancy that urgently requires reliable prognostic biomarkers. Although MUC1 expression and promoter methylation have been implicated in CC, the clinical significance of promoter methylation pattern composition, beyond average methylation levels, remains unclear. Here, we investigated the relationship between MUC1 promoter methylation heterogeneity, MUC1 mRNA expression, and prognosis in CC. We analyzed bisulfite amplicon sequencing data and mRNA expression of MUC1, DNA methylation-related enzymes (TET1, TET2, TET3, Dnmt1, and Dnmt3a), and tumor microenvironment stress markers in 131 CC tissues. In the neoplastic region, high MUC1 mRNA expression was associated with poor overall survival (HR = 0.131, 95% CI: 0.02 to 0.95, p = 0.042) and correlated with the abundance of completely unmethylated promoter patterns (r = 0.386, p < 0.001). Among the enzymes analyzed, only TET3 expression significantly correlated with the abundance of completely unmethylated patterns in the neoplastic region (Cohen's f² = 0.108, p = 0.009), suggesting a potential region-specific regulatory association. We visualized beta-diversity in methylation pattern composition using t-SNE and classified samples into two groups based on a linear decision boundary in the t-SNE space. This classification stratified prognosis independently of clinical factors (HR = 0.291, 95% CI: 0.06 to 0.94, p = 0.037; multivariate p = 0.021). These findings propose a novel, composition-based epigenetic stratification framework in CC, revealing that MUC1 promoter methylation pattern structure-rather than average methylation level-has prognostic relevance. Our results highlight the potential of pattern-resolved methylation profiling in the development of clinically applicable epigenetic biomarkers.

Keywords: Cholangiocarcinoma; Diversity; Mucin; Prognosis; Promoter methylation.

Fig. 1

Association between MUC1 mRNA expression and CC prognosis. (A) Comparison of MUC1 mRNA expression levels between non-neoplastic and neoplastic region in all cases, early-stage cases without lymph node metastasis, and advanced-stage cases with lymph node metastasis. (B) Kruskal–Wallis rank-sum test comparing MUC1 mRNA expression levels among different CC subtypes (d, d + p, i, and p). The effect size was calculated using η². (C) Prognostic comparison between groups with high- and low-MUC1 mRNA expression groups. The threshold was set to -0.57 in the non-neoplastic region and -0.33 in the neoplastic region, determined based on the value that maximized the effect size for prognosis in each respective region. CC, cholangiocarcinoma; i, intrahepatic CC; p, perihilar CC; d, distal CC.

Fig. 2

Evaluation of the alpha and beta diversity of the methylated CpG site sequential pattern in the MUC1 promoter region. (A) Bar chart showing composition of methylation pattern in non-neoplastic and neoplastic region. The colored bars indicate the 10 major and other methylation patterns. The eight methylation sites of the promoter region are noted as letters of U or M to indicate unmethylated and methylated CpG sites, respectively. The eight letters of U or M are arranged from upstream (5-prime end) to downstream (3-prime end) around the transcription start site (TSS). (B) Alpha diversity was monitored using the Simpson and Shannon indices between non-neoplastic and neoplastic region. (C) Beta diversity analysis of methylation patterns using t-SNE and k-means clustering. t-SNE analysis was performed to evaluate the beta diversity of methylation pattern composition in non-neoplastic and neoplastic region. Each point represents an individual sample, with blue dots indicating non-neoplastic region and red dots indicating neoplastic region. K-means clustering (k = 2) identified two distinct clusters, A and B, which are outlined by convex hulls. A contour plot illustrates the spatial distribution of cluster centers and local density variations within the t-SNE plot. (D) Comparison of the abundance of completely unmethylated and completely methylated methylation patterns between clusters identified by k-means clustering on the t-SNE plot. TSS, transcriptional start site; t-SNE, t-distributed stochastic neighbor embedding; DIM, dimension.

Fig. 3

Multivariate Cox regression analysis of MUC1 mRNA expression. Multivariate Cox regression analysis of MUC1 mRNA expression was performed based on the expression of microenvironment stress marker genes and abundance of completely unmethylated methylation patterns. (A) Multivariate Cox regression analysis of GRP78 mRNA expression. The index was calculated as follows: non-neoplastic region, index = 0.08 + 0.37 × UMP + -0.07 × GRP78; neoplastic region, index = -0.08 + 0.38 × UMP + 0.06 × GRP78. (B) Multivariate Cox regression analysis of CA9 mRNA expression. The index was calculated as follows: non-neoplastic region, index = 0.14 + 0.38 × UMP + 0.56 × CA9; neoplastic region, index = -0.07 + 0.21 × UMP + 0.53 × CA9. (C) Multivariate Cox regression analysis based on ACSS2 mRNA expression. The index was calculated as follows: non-neoplastic region, index = 0.06 + 0.35 × UMP + 0.14 × ACSS2; neoplastic region, index = -0.05 + 0.32 × UMP + 0.22 × ACSS2. (D) Multivariate Cox regression analysis of XBP1 mRNA expression. The index was calculated as follows: non-neoplastic region, index = 0.02 + 0.28 × UMP + 0.56 × XBP1; neoplastic region, index = 0.00 + 0.21 × UMP + 0.48 × XBP1. ER, endoplasmic reticulum; UMP, abundance of completely unmethylated CpG site sequence pattern.

Fig. 4

Association between beta diversity of methylated CpG site sequential patterns and CC prognosis. Two groups (A and B) were classified using a linear boundary (Dim2 = -21.5 × Dim1 + 30.0) on the t-SNE plot, representing the beta diversity of methylation patterns. (B) Comparison of MUC1 mRNA expression levels between groups identified using the t-SNE plot. (C) Cox proportional hazards regression analysis comparing OS between Groups A and B in non-neoplastic and neoplastic region. t-SNE, t-distributed stochastic neighbor embedding; DIM, dimension; OS, overall survival.