We obtained three pairs of surgically resected pancreatic cancers and corresponding peritumoral tissues from the Xi-jing Hospital and immediately stored them in liquid nitrogen. The patients that met the 2020 cancer diagnostic criteria signed informed consent before the experiment. The experiment was approved by the ethics committee of the Air Force Military Medical University.

Normal pancreatic cells (HPDE6-C7) and pancreatic cancer cell lines (PANC-1, MIAPaCa-2, PaTu8988T, SW 1990, Canpan-2, BXPC-3, HPAF-II, and Capan-1) were obtained from Procell Life Science and Technology Company (Wuhan, China) and BeNa Culture Collection (BNCC, Beijing, China). The cells were cultured in the recommended medium (10% fetal bovine serum) and kept in a CO₂-free humidifying incubator at 37°C.

Normal pancreatic cells and pancreatic cancer cell lines were lysed using lysis buffer (Applygen, Beijing, China) with a phosphatase and protease inhibitor cocktail (Roche, Branchburg, USA). Sodium dodecyl sulphate–polyacrylamide gel electrophoresis was used to fractionate the protein samples, and the protein bands were transferred onto a gel. After being blocked by using 5% nonfat milk for 3 h, polyvinylidene difluoride (PVDF) membranes were incubated using antibodies TFRC (1:1,000, 10084-2-AP, proteintech) and glyceraldehyde 3-phosphate dehydrogenase (1:1,0000, AC002, abcolone) overnight at 4°C. Then, we used horseradish peroxidase-conjugated secondary antibodies to incubate the PVDF membranes. In the end, the protein bands were detected using a Tanon-5200 hemiluminescent system (Shanghai, China).

Tumor tissue sections and peritumoral tissue sections were deparaffinized using xylene and then rehydrated using different concentrations of ethanol. We inhibited endogenous hydrogen peroxide activity with 3% hydrogen peroxide and immediately repaired the antigen using hot citric acid buffer. Then, 5% normal goat serum was used to block the tissue sections. The sections were subsequently washed in phosphate-buffered saline (PBS) and incubated overnight at 4°C with the antibody against TFRC (1:100, 10084-2-AP, Proteintech) and CD3 (1:6400, 60181-1-Ig, Proteintech). On the next day, after washing in PBS solution for 10 min, the peroxidase-conjugated secondary antibody was used to incubate the microarray and sections for 30 min. Then, these tissue microarray and tumor tissue sections were stained with diaminobenzidine for 3 min. Finally, the slides were counterstained with haematoxylin.

We downloaded the mRNA sequencing data and associated clinical information from the cancer genome atlas (TCGA) database, which involved 177 tumor tissues and four normal tissues. Furthermore, we downloaded other mRNA expression data from the GSE15471, GSE16515, GSE62452 and GSE45757 datasets from the gene expression omnibus (GEO) database. As these data are public network resources, the use of this particular data did not require any ethical approval or informed consent.

We used GEPIA (a web-based tool for analyzing the data provided by TCGA) to analyze the prognostic value of TFRC expression in pancreatic adenocarcinoma (PAAD) patients and calculate the hazard ratio (HR) with 95% confidence and log-rank p-value. The statistical package for the social sciences (SPSS) software was applied for multivariate and univariate analyses based on the TCGA data to determine whether it was an independent prognostic factor or not.

To further explore the TFRC-related molecular mechanisms in pancreatic cancer, we used the cBioPortal database (https://www.cbioportal.org/) to identify co-expressed genes and related miRNAs. Subsequently, we selected 10 genes and several miRNAs that were significantly correlated with TFRC for further analysis and verified their correlation with TFRC in the tumor immune estimation resource (TIMER) database.

We used STRING (a search tool for the retrieval of interacting genes) to construct the TFRC and related gene–protein interaction network with several aspects (interaction, co-expression, and co-localization). After the standard network diagram was constructed, the Cytoscape software was imported to beautify the network diagram.

We used the David (https://david.ncifcrf.gov/) online database to analyze the gene ontology (GO) and kyoto encyclopedia of genes and genomes (KEGG) pathways of TFRC and related genes. GO analysis includes biological processes (BP), molecular functions (MF), and cellular components (CC). Statistical significance was set at p < 0.05, while data visualization was executed using the GraphPad Prism5 software.

We evaluated the immune characteristics of TFRC in pancreatic cancer from four aspects: infiltration level of tumor infiltrating immune cells (TIICs), expression of inhibitory immune checkpoints, expression of immunomodulators, and activity of the cancer immunity cycle. The invasion level of TIICs in the tumor microenvironment was calculated using a variety of algorithms based on bulk RNA-seq data. Then, we examined the relationship between TFRC and common immune checkpoints (CTLA4, LAG3, PD1, and PDL1) in pan-cancer. We also collected information on 144 immunomodulators, including immune inhibitors, immune stimulators, MHC, chemokines, and receptors, and analyzed their relationship. Previous studies (Hu et al., 2021) have reported that a cancer immunity cycle reflects the anticancer immune response and comprises seven steps: release of cancer cell antigens (Step 1), cancer antigen presentation (Step 2), priming and activation (Step 3), trafficking of immune cells to tumors (Step 4), infiltration of immune cells into tumors (Step 5), recognition of cancer cells by T cells (Step 6), and killing of cancer cells (Step 7). We analyzed the correlation between TFRC and the cancer immunity cycle. We further evaluated the role of TFRC in immunotherapy by selecting some immune-checkpoint-relevant and immune-activity-relevant genes (Zhang et al., 2020) in PAAD. In addition, TIDE (Jiang et al., 2018) is a computational method to model two primary mechanisms of tumor immune evasion, we analyzed the correlation between TFRC and Interferon-gamma (IFNG) signature, microsatellite instability (MSI) signature, macrophage M2 type and TIDE score.

We used SPSS 20.0, GraphPad Prism 5 and R version 4.1.0 software to perform the data analysis and create plots and figures. We used the mean ± SD (standard deviation) to express the measurement characteristics. This is the basis for the analysis of the differential expression of TFRC in tumors and non-tumors of pancreatic cancer patients from the TCGA and GEO databases by Student’s t test. Moreover, the association between TFRC and clinical characteristic variables was analyzed using the Pearson chi-squared test. At the last step, the univariate and multivariate analyses of TFRC were conducted using Cox proportional hazards analysis.

We analyzed TCGA data related to TFRC expression levels in a variety of cancers (Figure 1A). Then, we investigated the difference in the expression levels of TFRC mRNA between pancreatic cancer and normal pancreas, and found that their levels were significantly higher in pancreatic cancer than that in normal pancreas (Figure 1B). Subsequently, we used the paired t-test to determine TFRC mRNA expression levels in pancreatic cancers and corresponding peritumoral tissues based on the data from the GEO database (Figures 1C–E). Then, we used immunohistochemistry to test the differences about TFRC protein levels in pancreatic tissues (Figure 1F). Besides, we found the same thing at the cellular level. The result of western blot showed that TFRC expression was significantly lower in normal pancreatic cell (HPDE6-C7) than that in pancreatic cancer cells (PANC-1, MIAPaCa-2, PaTu8988T, SW 1990, Canpan-2, BXPC-3, HPAF-II, Capan-1) (Figure 1G). Meanwhile, we verified TFRC mRNA levels in one pancreatic cell and 22 pancreatic cancer cell lines, which shows the TFRC mRNA levels were higher in most pancreatic cancer cell lines than in normal cell (Figure 1H).

Firstly, we use TCGA and GEO database to build ROC curve, and found the AUC = 0.856 (GSE16515) and 0.742 (TCGA and GTEx), which means the TFRC has diagnostic significance (Figure 2A). Then, we used GEPIA and found that the patients with high TFRC expression had poor overall survival (p = 0.036, group cutoff = median) and poor disease-free survival (p = 0.045, group cutoff = median) (Figure 2B). At the same time, the univariate analysis showed that grade (HR = 1.451, p = 0.011), tumor node metastasis classification (TNM) stage (HR = 1.279, p = 0.038), invasion depth (T) stage (HR = 1.614, p = 0.033), and lymph node (N) stage (HR = 1.598, p = 0.022), and TFRC expression (HR = 1.681, p = 0.043) influenced the overall survival of patients. (Figure 2C). However, the multivariate analysis did not help in defining the upregulated TFRC expression as an independent prognostic factor (Figure 2D).

To analyze the clinical value of TFRC in pancreatic cancer, we examined its relationship with the clinical factors of the patients. We used the median TFRC expression value to create a categorical dependent variable. The Pearson chi-squared test showed that TFRC expression was related to age (p = 0.001), pathological stage (Ⅰ vs. Ⅱ–Ⅳ, p = 5.97e-13), T stage (T1 vs. T2 vs. T3, T4, p = 0.047), and N stage (N0 vs. N1–Nx, p = 2.15e-05) (Table 1).

To further investigate the possible effect of TFRC in PAAD, the genes that were co-expressed with TFRC were identified using the STRING website with TCGA data (Figure 3A). The top ten significant genes correlated with TFRC were selected (Figure 3B). The results indicated that TFRC was significantly correlated with PGM2 (r = 0.6392, p = 7.537e−22), ECT2 (r = 0.5817, p = 1.671e-17), E2F8 (r = 0.581, p = 1.865e-17), MK167 (r = 0.5809, p = 1.9e−17), ESCO2 (r = 0.5662, p = 1.765e−16), DMAP1 (r = -0.5827, p = 1.431e−17), POMT1 (r = -0.5764, p = 3.813e-17), CIRBP (r = -0.5722, p = 7.218e-17), UCK1 (r = -0.5521, p = 1.374e−15) and R3HCC1 (r = -0.5465, p = 2.997e−15). Furthermore, the correlation between TFRC and these genes was verified using TIMER (Figure 3C). Besides, we used the database to analyze TFRC and these 10 genes in 46 pancreatic cell lines. (Figure 3D). Based on the above results, we found that TFRC was strongly correlated with ECT2 and E2F8.

Also, in order to analyze the relationship between TFRC and related miRNAs in PAAD and explore the molecular mechanism of TFRC, the STRING website analysis found that TFRC mainly interacts with mir-326, mir-378, mir-196b, mir-23a, mir-1247, mir-885, mir-30c-2, mir-2114, mir-488, and mir-190b (Figures 4A–C).

To analyze the relationship between TFRC and related proteins and to further study the molecular mechanism of TFRC, STRING analysis was conducted. It was found that TFRC mainly interacts with proteins such as ARF1, TGOLN2, LAMP1, VAMP2, and DNM2 to exert its function (as shown in Figure 5).

For better understanding of the molecular mechanism of TFRC, we performed GO (Table 2) and KEGG (Table 3; Figures 6G,H) enrichment analysis using the STRING website. In particular, the GO enrichment analysis divided the functional annotations of target genes into three categories: MF (Figures 6A,B), CC (Figures 6C,D), and BP (Figures 6E,F). In the KEGG and GO pathway analysis, some signaling pathways affected by TFRC are enriched during tumorigenesis. Cell adhesion molecule binding reflects the function of its surface position. Moreover, TFRC is associated with condensed chromosomes, chromosome segregation, and the cell cycle, so it may participate in cell division.

Firstly, we examined the relationship between TFRC and 144 immunomodulators, including immune inhibitors, immune stimulators, MHC, chemokines, and receptors in the TCGA cohort (Figure 7A). Then, we analyzed the immune activation of the high-TFRC and low-TFRC groups in the TCGA cohort. To this end, we selected several immune-related signatures, including immune-checkpoint-relevant signatures (CD274, CTLA4, HAVCR2, IDO1, LAG3, and PDCD1) and various immune-activity-related signatures (CD8A, CXCL10, CXCL9, GZMA, GZMB, PRF1, TBX2, and TNF). We used the Wilcoxon test to verify that the high-TFRC group had higher immune-related signatures, except for PDCD1, LAG3, and TBX2 (Figure 7B). Besides, we analyzed the activities of the cancer immunity cycles, which can evaluate the chemokine system and other immunomodulators. The high-TFRC group can activate several steps in the cycle, including the release of cancer cell antigens (Step 1), cancer antigen presentation (Step 2), trafficking of immune cells to tumors (Step 4) (CD8 T cells, MDSCs, neutrophils, TH1 cells, and TH22 cell recruiting), and infiltration of immune cells into tumors (Step 5) (Figure 7C). These steps may increase the number of infiltrating immune cells in the tumor microenvironment. Next, TFRC also enhances immunotherapy-positive gene signatures, including cell cycle, DNA replication, mismatch repair, P53 signaling pathway, and proteasome (Figure 7D). Subsequently, we used a butterfly heat map to determine the relationship between TFRC and the activities of the cancer immunity cycle (Figure 8A), immunotherapy-positive gene signatures (Figure 8B), immune cell infiltration (Figure 8C), and immune-relevant genes (Figure 8D) in PAAD. Overall, TFRC can used to predict immune phenotypes, molecular subtypes in PAAD, which can lay a foundation for new immunotherapies for pancreatic cancer.

After analyzing the correlation between TFRC and immunophenotype, to further explore the relationship between TFRC and immunotherapy, we firstly calculated the immune cells infiltrating the tumor microenvironment on the strength of bulk RNA-seq data by using several algorithms (TIMER, CIBERSORT, quantiseq, MCPcounter, xCell, and EPIC) (Figure 9A). As a result, we found that the high expression levels of TFRC means more infiltrating immune cells. To verify the results of data analysis, we selected a common immune cell marker CD3 for immunohistochemical tests on tissues from three pancreatic cancer patients with high TFRC expression and observed numerous inflammatory infiltrates (Figure 9B). For the current common immunotherapy methods, we also demonstrated that TFRC was co-expressed with various immune checkpoints in several cancers, including PAAD (Figures 9C–F). Besides, tumor mutation burden (TMB) has always been considered as one of the criteria for immunotherapy, so we analyzed the TMB and found that it was positively correlated with TFRC (Figure 10A). Interferon-gamma (IFNG) signature and microsatellite instability (MSI) signature are also used to predicted the immunotherapy (Fu et al., 2020), so we used them to predict and got the same conclusion (Figures 10B,C). In addition, Peng Jiang (Jiang et al., 2018)developed TIDE (a computational method to model two primary mechanisms of tumor immune evasion), we found tumor associated macrophage M2 type (Fu et al., 2020) and TIDE score were negatively correlated with TFRC (Figures 10D,E), which further demonstrates the relationship between TFRC and immunotherapy.