AUTHOR=Zhou Dongkai , Zhong Cheng , Yang Qifan , Cui Bijun , Wang Yizhi 

TITLE=Integration of bulk and single-cell transcriptomic data reveals a novel signature related to liver metastasis and basement membrane in pancreatic cancer

JOURNAL=Frontiers in Immunology

VOLUME=Volume 16 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/immunology/articles/10.3389/fimmu.2025.1671956

DOI=10.3389/fimmu.2025.1671956

ISSN=1664-3224

ABSTRACT=BackgroundPancreatic cancer (PC) is characterized by an exceptionally poor prognosis, primarily attributable to its aggressive metastatic behavior and high recurrence rates. Liver metastasis is the predominant distant metastasis model of PC. Moreover, invasion and metastasis of PC are closely associated with the remodeling or loss of basement membrane (BM). Consequently, identifying pivotal genes involved in PC liver metastasis (PCLM) and BM could pave the way for more effective and precise targeted therapies. This study aims to construct a prognostic model based on PCLM and BM-related genes, while also validating the association between this model and the immune microenvironment of PC, as well as its predictive value for the efficacy of chemotherapy and immunotherapy.MethodsTranscriptomic, mutation, and clinical data were retrieved from the TCGA, ICGC, and GEO databases. Core prognostic genes were identified through single-cell (sc) and bulk transcriptomic sequencing data combined with WGCNA analysis. The prognostic model was established using machine learning algorithms and multivariate Cox regression analyses. Specifically, the TCGA-PAAD cohort was utilized as the training set while the PACA-AU cohort served as the validation set. The performance of this model was assessed in both the training and validation sets. Additionally, the associations between the model and tumor mutation burden (TMB) as well as tumor immunity were evaluated using multiple immunity databases. Additionally, the predictive capacity of the model regarding the efficacy of chemotherapy, immunotherapy, and targeted therapy was also assessed. Finally, the expression of COL7A1 was knockdown in cancer-associated fibroblasts (CAFs) in PC to explore its role in PC progression.Results30 PCLM and BM-related prognostic genes were preliminarily identified integrating sc and bulk transcriptomic sequencing data. Through machine learning algorithms and multivariate Cox regression analysis, six signatures, including COL7A1, ITGA6, ITGA7, ITGB5, ITGB7 and NTN4, were subsequently utilized to construct a prognostic model. This model demonstrated superior prognostic performance compared with conventional clinicopathological variables. Immune analysis revealed that the infiltration levels of M0 macrophages and Treg cells were significantly elevated in the high-risk group, whereas the infiltration levels of CD8+T cells and γδT cells were significantly reduced. Moreover, the high-risk group exhibited higher TMB and poorer survival outcomes. Additionally, the high-risk group showed a higher TIDE and a lower IPS score, indicating less effective immunotherapy response. Furthermore, the high-risk group displayed significantly higher IC50 values for common PC chemotherapeutics, suggesting reduced chemotherapeutic efficacy. Notably, scRNA-seq analysis indicated that COL7A1, which has not been systematically investigated in PC previously, predominantly expressed in fibroblasts. Specifically, CAFs exhibited significantly higher expression levels of COL7A1 compared to normal pancreatic fibroblasts, and COL7A1 knockdown in CAFs markedly reduced the migratory capacity of PC cells while enhancing their chemosensitivity to gemcitabine.ConclusionThis study developed and rigorously validated an innovative prognostic model for PC. This model, incorporating pivotal genes of PCLM and BM, may also serve as potential tool for predicting the tumor immune microenvironment and therapeutic efficacy. Notably, COL7A1, which was demonstrated to be vital in PC metastasis in this study, warrants further investigation in future research.