Background Pancreatic cancer (PC) is a highly malignant tumor featured with high intra-tumoral heterogeneity and poor prognosis. Cell-in-cell (CIC) structures have been reported in multiple cancers, and their presence is associated with disease progression. Nonetheless, the prognostic values and biological functions of CIC-related genes in PC remain poorly understood. Methods The sequencing data, as well as corresponding clinicopathological information of PC were collected from public databases. Random forest screening, least absolute shrinkage, and selection operator (LASSO) regression and multivariate Cox regression analysis were performed to construct a prognostic model. The effectiveness and robustness of the model were evaluated using receiver operating characteristic (ROC) curves, survival analysis and establishing the nomogram model. Functional enrichment analyses were conducted to annotate the biological functions. The immune infiltration levels were evaluated by ESTIMATE and CIBERSORT algorithms. The expression of KRT7 (Keratin 7) was validated by quantitative real-time PCR (qRT-PCR), western blotting and immunohistochemistry (IHC) staining. The CIC formation, cell clusters, cell proliferation, migration and invasion assays were applied to investigate the effects of silencing the expression of KRT7. Results A prognostic model based on four CIC-related genes was constructed to stratify the patients into the low- and high-risk subgroups. The high-risk group had a poorer prognosis, higher tumor mutation burden and lower immune cell infiltration than the low-risk group. Functional enrichment analyses showed that numerous terms and pathways associated with invasion and metastasis were enriched in the high-risk group. KRT7, as the most paramount risk gene in the prognostic model, was significantly associated with a worse prognosis of PC in TCGA dataset and our own cohort. High expression of KRT7 might be responsible for the immunosuppression in the PC microenvironment. KRT7 knockdown was significantly suppressed the abilities of CIC formation, cell cluster, cell proliferation, migration, and invasion in PC cell lines. Conclusions Our prognostic model based on four CIC-related genes has a significant potential in predicting the prognosis and immune microenvironment of PC, which indicates that targeting CIC processes could be a therapeutic option with great interests. Further studies are needed to reveal the underlying molecular mechanisms and biological implications of CIC phenomenon and related genes in PC progression.

Pancreatic cancer (PC) is a highly lethal and aggressive disease with its incidence and mortality quite discouraging. A robust prognostic signature and novel biomarkers are urgently needed for accurate stratification of the patients and optimization of clinical decision-making. Since the critical role of immune microenvironment in the progression of PC, a prognostic signature based on seven immune-related genes was established, which was validated in The Cancer Genome Atlas (TCGA) training set, TCGA testing set, TCGA entire set and GSE71729 set. Furthermore, S100A14 (S100 Calcium Binding Protein A14) was identified as the gene occupying the most paramount position in risk signature. According to the GSEA, CIBERSORT and ESTIMATE algorithm, S100A14 was mainly associated with lower proportion of CD8 + T cells and higher proportion of M0 macrophages in PC tissue. Meanwhile, analysis of single-cell dataset CRA001160 revealed a significant negative correlation between S100A14 expression in PC cells and CD8 + T cell infiltration, which was further confirmed by tissue microenvironment landscape imaging and machine learning-based analysis in our own PUMCH cohort. Additionally, analysis of a pan-pancreatic cancer cell line illustrated that S100A14 might inhibit CD8 + T cell activation via the upregulation of PD-L1 expression in PC cells, which was also verified by the immunohistochemical results of PUMCH cohort. Finally, tumor mutation burden analysis and immunophenoscore algorithm revealed that patients with high S100A14 expression had a higher probability of responding to immunotherapy. In conclusion, our study established an efficient immune-related prediction model and identified the potential role of S100A14 in regulating the immune microenvironment and serving as a biomarker for immunotherapy efficacy prediction.