Pancreatic cancer is a highly malignant tumor with a poor prognosis, exhibiting nearly equivalent morbidity and mortality rates [
19]. The poor prognosis of PC is attributed to the absence of biomarkers for early detection, resulting in 90% of PC patients being diagnosed at a late stage at diagnosis [
20,
21]. In addition, PC is not sensitive to chemotherapy and immunotherapy, and even combination therapy fails to ameliorate unfavorable clinical outcomes [
22]. Therefore, identifying suitable biomarkers for PC diagnosis and developing novel tools to guide precision therapy are crucial in PC treatment.
Glycosylation is the enzymatic process of linking sugars to proteins, lipids, and other glycans, and this major post-translational modification (PTM) occurs in the endoplasmic reticulum/Golgi lumen of all cells and is mediated by the coordinated action of different glycosyltransferases and glycosidases [
23]. Glycosylation is the basis of various biological processes and small changes in glycan structure severely affect cell biology, causing pathophysiology and development [
24], such as cancer [
8]. Malignant transformation is highly associated with aberrant glycosylation [
25]. Previous studies have found that glycosylation-related genes are associated with the prognosis of patients with breast [
26], ovarian [
27], liver [
28] and cervical cancer [
29]. However, current studies on glycation in PC mainly focus on the influence of individual glycation type or single glycation enzyme on tumor progression [
30,
31], and there are no relevant reports on the expression status of numerous glycation related genes in PC and their relationship with PC microenvironment [
32]. Gupta et al. summarized the expression of 207 glycation genes in TCGA database and identified 6 glycation genes (B3GNT3, B4GALNT3, FUT3, FUT6, GCNT3 and MGAT3) that played a unique role in the pathogenesis of PC [
33]. Their functions were investigated using the CRISPR/ cas9 based KD system. The key role of O- and N- linked glycosylation in PC progression was finally demonstrated and the mechanism of GCNT3 action in PC was described. However, this study included only 149 tumor patients and 3 paracancer tissues for analysis, and mainly focused on the function of glycosylation genes on tumor cells. In addition, Yousra et al. performed a bioinformatic analysis including 169 glycosyltransferase RNA sequencing data from resected and unresectable 74 patient-derived xenografts (PDX) and constructed a glyco-signature consisting of 19 genes [
34]. Their research also has limitations including insufficient sample size and too many model genes. Hence, we first acquired GRGs through the GGDB and tested their expressions from TCGA-GTEx joint database, which included 178 PC samples and 169 normal control samples. Based on the univariate and multivariate cox regression analysis, 5 genes associated with GRG were identified for prognostic modeling. In order to verify the reliability of the model, we randomly selected 60 pairs of pancreatic tumors and para-cancer samples from the first affiliated Hospital of Nanjing Medical University to validate the prognostic performance of the GRGs signature. Moreover, the five GRGs prognostic model were consisted of 3 elevated expressed (ALG1L2, B3GNT3, B3GNT8) and two decreased genes (HS6ST3, ST8SIA5). Notably, B3GNT8 was negatively correlated with risk score, however, the prognosis of PC patients was better. Hence, we chose B3GNT8 for functional experiments to further verify the accuracy and practicability of the risk score. B3GNT8 was initially cloned in 2004 [
35]. Subsequently, in 2005, a team of Japanese researchers led by Akira Seko demonstrated that B3GNT8 could form heterodimers with B3GNT2 in vitro, resulting in an enhancement of catalytic activity [
36]. Building on this, a study conducted in 2010 by Shen unveiled that B3GNT8 plays a role in regulating matrix metalloproteinase 2 (MMP-2) and tissue inhibitor of metalloproteinase 2 (TIMP-2) in gastric cancer cells [
37]. These findings propose that B3GNT8 might offer potential as a therapeutic target for gastric cancer. Based on the risk scores, we discovered the risk model has relationship with tumor infiltrating lymph cells, immune checkpoint, chemotherapy drug sensitivity, immune escape and tumor mutational load in PC. However, further verification is still required to establish the correlation between the risk model and tumor immune microenvironment.
PC is one of the most immune-resistant tumor types, exhibiting an immunologically "cold" tumor microenvironment (TME) [
38]. This indicates a lack or dysfunction of adaptive T cell immunity and resistance to checkpoint blockade [
23]. To date, monotherapy with immune modulators has been proven ineffective in PC clinical settings, indicating a multimodal approach targeting the immune therapy resistance mechanism [
39]. Personalized immunotherapy strategies based on PC genetic and phenotypic heterogeneity for individual patients may offer new insights [
40]. Therefore, to identify potential immune features, we performed multi-marker immunohistochemistry staining on 4 samples selected randomly from both high- and low-risk groups. We labeled tumor cells with panCK, CD8 and PD-L1 to observe the expression differences of these markers in PC tissues. The results showed more tumor cell and less CD8 + T cells were consisted in high-risk samples. Moreover, we performed a neighboring cell analysis, which was divided into cell count within a 100 μm range and analysis of the nearest cell-to-cell distance. We observed an interesting phenomenon that, although the infiltration of cytotoxic T cells was reduced, their distance to tumor cells was closer. Mature CD8 + T cells, also known as cytotoxic T cells, can recognize infected or damaged cells and trigger a death pathway through cytotoxic proteins. Under sustained or repeated stimulation, the immune system gradually would lose its normal function, leading to a decrease in immune cell infiltration and the occurrence of immune exhaustion. Hence, we found the cytotoxic T cells in the high-risk group were closer to the tumor cells in this study. Combined with the previous reports, we hypothesize that this phenomenon may facilitate the exhaustion or dysfunction of the infiltrating CD8 + T cells in PC microenvironment by potential mechanisms. The number and function of immune system cells are affected, leading to a decline in immune function, forming a malignant cycle that accelerates tumor progression. The specific mechanism underlying immune exhaustion in high-risk patients is currently unclear, and we will further investigate this phenomenon.
The present study also has some limitations. This study is retrospective, and the prognostic model need to be further validated by prospective data. The mechanistic studies of prognostic model genes are still lacking, which will be added in our next work. Furthermore, single omics data studies are still limited. The multi-omics and integrated analysis for high-throughput data from multiple levels and sources is our further research direction.