Pancreatic cancer (PC) is one of the most common and lethal cancers worldwide (Kaźmierczak-Siedlecka et al., 2020). In developed countries, PC is the fourth leading cause of cancer-related deaths, and it is ranked the seventh around the world (Ducreux et al., 2015). Sadly, the incidence rate of PC is gradually accelerating, and it will rank as the second leading cause of cancer-related mortalities in 2030 (Hasan et al., 2019). The treatments for early stage PC are surgery and chemotherapy. The treatments for advanced PC are chemotherapy and radiotherapy (Ducreux et al., 2015). The most used and acknowledged chemotherapy regimens for all stages of PC constitute cytotoxic drugs, such as, FOLFIRINOX and gemcitabine plus nab-paclitaxel (Conroy et al., 2011; Von Hoff et al., 2013). Disappointedly, patients are often resistant to these treatments and tend to have a poor prognosis (Jiang et al., 2023). For the minority of early stage patients with local disease, the 5-year survival rate can reach 36%, and the rate decreases to 12% in patients with lymph-node metastasis. Most patients who suffer from distant spread have the lowest 5-year survival, with 3% (Poruk et al., 2013; Bray et al., 2018). Therefore, it is necessary to find a novel and potent method to perform risk assessment to recognize high-risk patients in the early stage and provide them with proper treatment to avoid cancer progression.

The tumor microenvironment (TME) is an immunosuppressive niche that is formed in the process of tumor cells hijacking the transcriptional mechanisms of the stroma cells (Kleeff et al., 2016). The main components of the TME are cancer-associated fibroblasts (CAFs), extracellular matrix (ECM), endothelial cells, stroma-associated pancreatic stellate cells (PSCs), adipose cells, neural cells, and some immune cells, such as myeloid-derived suppressor cells (MDSCs), tumor-associated macrophages (TAMs), and regulatory T cells (Tregs) (Feig et al., 2012; Wolfgang et al., 2013). PC is notorious for its dense TME, which is enriched with the stroma, MDSCs, TAMs, CAFs, and many other cells (Farrow et al., 2008). As immunosuppressive TME characteristics, MDSCs block immune responses and release interleukin-10 (IL-10) and transforming growth factor-β (TGF-β) to induce an anti-inflammatory environment in PC (Huang et al., 2006; Sinha et al., 2007; Ostrand-Rosenberg et al., 2012; Pinton et al., 2016). CAFs fulfill the bi-function in PC, which is mostly anti-immune and partial immunosuppression (Belle and DeNardo, 2019; Elyada et al., 2019; Das et al., 2020). By promoting the epithelial–mesenchymal transformation (EMT), interacting with cancer stem cells (CSCs), inducing the apoptosis of T cells, and breaking local immune surveillance, the TAMs boost PC, lead to resistance of treatment, and result in poor prognosis (Zhang et al., 2022). Treg cells eliminate effector T cells or acquire antigen-presenting cells which compete with effector T cells against immunology (Jang et al., 2017). However, there are still some immune cells that play anti-tumor roles and offer promising prospects for survival in PC. Tertiary lymphoid structures, organized by tumor-infiltrating lymphocytes (TILs), which are often observed in cancer tissue, are considered to participate in the immune response to suppress cancer and positively impact prognosis (Balch et al., 1990; Zhang et al., 2003; Hiraoka et al., 2015). Infiltration of CD8⁺ lymphocytes was an independent factor for longer disease-free survival (DFS) and overall survival (OS) in PC (Lohneis et al., 2017). The roles of the TME are complex: some can be used to foresee the prognosis, while some are indexes for the sensitivity of immunotherapy of PC (Nomi et al., 2007; Samstein et al., 2019). However, there are no immune markers that can solve well all the problems at one times. Therefore, we established a TME-related risk model to predict survival and to test drug sensitivity in PC.

In this study, the expression patterns of TME-related genes in PC were comprehensively revealed, and we established a new but robust risk model to predict the prognosis, identify therapy targets, and foresee the treatment sensitivity of PC patients.

Recently, increasing evidence has shown a close relationship between PC and TME. The stroma and PC cells dynamically cooperate and promote all aspects of aggressiveness. From a genetic perspective, growth, metabolism, homeostasis and proliferation of cancer cells and the synthesis of cancer-related proteins are influenced by tumorigenesis, which is promoted by the mutation of genes in PC cells, such as KRAS, TP53, CDKN2A, and SMAD4 (Saiki et al., 2021). At the same time, PC is very heterogeneous because it contains abundant ECM and various stromal cells, such as TAMs, MDSCs, CAFs, immunocytes, and PSCs (Reyes-Castellanos et al., 2022). Interestingly, cancer cells can work together with the TME by rewiring metabolism, autophagy, and other mechanisms to serve as vital mediators of PC progression (Perera and Bardeesy, 2015; Qin et al., 2020; Reyes-Castellanos et al., 2020; Reyes-Castellanos et al., 2022). Generally, cancer cells and the TME can promote the development, invasion, metastasis, drug resistance, and evasion of immune surveillance in PC. Previous studies have developed histological or molecular classifications to predict the prognosis of PC, but their predictive ability for therapeutic management is poor (Reid et al., 2013; Bailey et al., 2016; Gutiérrez et al., 2021). Consequently, we analyzed the genetic characteristics involved in the TME and developed a new method to predict the prognosis and sensitivity of patients to treatments in PC.

By univariate and multivariate Cox and LASSO analyses, we established a robust risk model to predict the survival of PC patients and verified it in three cohorts. Cohorts TCGA-PAAD, GSE57495, and CPTAC were used to check the practicality and reliability. Overall, the survival analysis showed that our model is helpful in identifying patients who might suffer from poor prognosis. In cohort TCGA-PAAD, the best effect of the model was exhibited at prediction for 5-year survival rate, while in cohort GSE57495, it was at 1-year. This difference may be caused by the distinction of the constitution of each cohort of patients’ clinical features. For instance, cohort GSE57495 completely focused on the early stage of PC patients. In addition, in cohort CPTAC, no patient lived more than 5 years, which resulted in the absence of AUC at 5 years. Furthermore, along with the construction of the risk model, we spotted five genes that showed an intimate relationship with the prognosis of PC patients. MET is a well-known gene that acts as a growth factor under physiological conditions and can promote oncogenesis via the active mode (Comoglio et al., 2008; Nakamura et al., 2011). In head and neck squamous cell carcinoma, over half of the patients overexpress HGF which stimulates MET to induce the proliferation of cell cycle genes by activating STAT3 in the TME (Igelmann et al., 2019; Boschert et al., 2020; Raj et al., 2022). Our study found MET overexpressed in PC cells and tissues and high-expressions of MET was correlated with worse prognosis. A previous study has confirmed that MET facilitates stromal rewiring by upregulating tenascin-C (TNC) expression, which interacts with ECM components and is deeply involved in the metastasis of cancer, stimulating the proliferation and restraining the differentiation of CSCs in PC (Jones and Jones, 2000; Orend, 2005; Orend and Chiquet-Ehrismann, 2006; Modica et al., 2021). By proteolysis, MMP3, another high-expression gene in PC samples, destroyed various molecules, such as ECM and adhesion molecules, and enabled the tumor to be more aggressive (Sternlicht et al., 1999; Munhoz et al., 2010; Niland et al., 2021). Upregulated MMP3 participates in the progression of genomic instability in tumors (Sun et al., 2014). By manipulating the ECM, MMP3 is involved in oncogenesis, cancer cell proliferation, and invasion, and this explains the factor of poor survival in PC (Hadler-Olsen et al., 2013). Studies have also confirmed that the expression of MMPs is regulated by the ECM and immune system which are absent in cells in vitro (Conlon and Murray, 2019). This may explain the different expressions of MMP3 in PC cells and samples. Due to TGFβ signaling, the expression level of FERMT1 mRNA increases in several PC cell lines and promotes migration and invasion (Sin et al., 2011). CARD9 is critically involved in various inflammatory responses. By manipulating inflammatory cytokines, CARD9 is involved in adaptive immunity (Liu et al., 2022). In cancer, the cellular location of CARD9 is in tumor-infiltrating macrophages rather than in cancer cells (Zhong et al., 2018; Zhong et al., 2019), which explains the lower expression of CARD9 in PC cells and samples than in normal pancreatic ones. Furthermore, as determined by the biological state, macrophages, as a constituent of the TME, can boost or suppress the proliferation and metastasis of cancer (Pan et al., 2020; Xu et al., 2020; Zhou et al., 2020). This may account for the dual functions of CARD9 that is a tumor promoter and/or tumor inhibitor (Yang et al., 2007; Yang et al., 2008; Bergmann et al., 2017; Haas et al., 2017). By working with cytokines and ECM, five genes participate in the progress of PC. In addition, this close relationship of genes and the TME in cancer makes a promising prospective prediction of survival in PC patients, and a comparison with other models certified the efficiency of our model. Next, to predict survival in a more functional and simpler way, the nomogram was established. The result of the nomogram showed that although other clinical factors were taken into account, the risk score still played the leading role.

GSEA, immune cell infiltration, and single-cell analysis confirmed that low-risk patients have a strong relationship with the immune system. The pathway enrichment analysis showed that cancer-related pathways, such as pathways in cancer and the cell cycle, were significantly enriched in the high-risk group, illustrating poor survival. However, in the low-risk group, we noticed the enrichment of immune-related pathways and this was in accordance with the result that the anti-cancer immune cells, such as T cells, CD8⁺ T cells, CTL, and B cells, were mostly infiltrated in the low-risk group. A previous study has demonstrated that better DFS and OS were guaranteed in patients by a higher expression of T cells (Muller et al., 2022). CD8⁺ T cells attack tumor cells by recognizing the antigen peptides on their surface (Fukunaga et al., 2004; van der Leun et al., 2020). The role of B cells in PC is still ambiguous. Preclinical evidence have verified that immuno-suppressive B cells could promote cancer by suppressing the activity of CD8⁺ T cells and secreting cytokines (Gunderson et al., 2016; Pylayeva-Gupta et al., 2016). However, this kind of B cells only account for 10% in B cells in PDAC in humans (Delvecchio et al., 2022). Most B cells form TLS generate an inflammatory phenotype that facilitates the activation and recruitment of antigen-presenting cells and dendritic cells (DCs) (Ene-Obong et al., 2013; Watt and Kocher, 2013; Affara et al., 2014). Therefore, the pro-tumoral role of B cells may be overwhelmed by the anti-tumorigenic role. An intimate relationship with the immune mechanism may explain better outcomes in patients with low risk. In addition, drug sensitivity analysis manifested patients gotten low risk score might benefit from the immune treatment, while the counterpart might not.

Derived from the model, we identified several potential immune targets, such as LAG3, TIGIT, and CTLA-4 (Chihara et al., 2018; DeLong et al., 2019) that are highly expressed in the low-risk group. In PDAC, the CD155/TIGIT axis maintains immune evasion. Combining regimens, with inhibitors of TIGIT and PD-1 plus CD40 agonism, in preclinical models exhibited encouraging tumor suppression (Freed-Pastor et al., 2021). In the TME, T-cell exhaustion is caused by LAG-3 that cooperates with a pile of blockade receptors (Chihara et al., 2018; Edwards et al., 2018; Karlsson et al., 2020). In mice, treatment with anti-LAG-3 and anti-PD-1 antibodies also showed a strong anti-tumor effect (Woo et al., 2012). Although the mentioned targets are still stuck with preclinical trials, our model suggests that immune therapy could be a promising treatment for PC patients.

Mutation of KRAS is one of the most common alterations in both high- and low-risk groups, being present in approximately 90% of PC patients, and is regarded as the major genetic initiating event in oncogenesis by influencing the TME and cell proliferation, apoptosis, autophagy, and metabolism (Bailey et al., 2016; Chan-Seng-Yue et al., 2020; Pereira et al., 2022). Though there was no significant difference in sensitivity of anti-KRAS-G12C in the two groups, we noticed patients marked with high risk are more sensitive to this drug than those at low risk. In addition, the result of GSEA in hallmark data sets showed pathway “KRAS signaling up” was mostly enriched in the high-risk groups. Meanwhile, considering that the mutation of KRAS G12C is less frequent than the mutation of G12D in PC, KRAS G12D inhibitor may be a better choice for high-risk patients.

Another gene, ARID1A, was only mutated in the high-risk group. The function of ARID1A is complicated. In colorectal cancer (CRC), the proliferation of KRAS-mutated cancer cell rely on ARID1A (Sen et al., 2019). In gynecologic cancers, ARID1A suppresses cancer via co-operating with p53 (Guan et al., 2011). In PC, we found that high-risk patients exhibited more ARID1A mutation, but the reason for this phenomenon was not clear. Anti-ARID1A might give us a new therapeutic target for PC. Our study identified that co-mutations of TP53 and KRAS might cause worse survival. This is consistent with a former study (Shoucair et al., 2022). A preclinical study confirmed that mutated KRAS and TP53 could upregulate FOXA1 by stimulating CREB1 and finally exert an oncogenic effect (Kim et al., 2022).

Gemcitabine plus paclitaxel and FOLFIRINOX are the first recommended chemotherapy agents for PC chemotherapy, and using our model, doctors can choose sensitive chemotherapy for PC patients (Jameson et al., 2019; Robatel and Schenk, 2022). The compound regimen of other treatments and immunotherapies, which remodels the TME by adjusting the quantities and type of T cells, exhibits promising effectiveness (Kole et al., 2022). The results verified that low-risk patients would exhibit a better response to immunotherapy. The predictive value of drug responses was weakened by the rough risk characteristics classification. As the only distinctive standard, the assessment of the quantified risk scores for evaluating drug responses cannot conclude with a precise regime. Finally, we identified several new treatment targets, but these require further substantiation in future.

To a certain extent, the use of only online databases for data verification and histological validations at a single level caused validation limitation of our model. In future, local normal and cancer pancreatic tissues and clinical data are called for to verify the robustness of our model. Meanwhile, further experiments that include genomic or proteomic analysis are demanded to investigate the mechanism of FERMT1, IL20RB, MET, MMP3, and CARD9 in the progress of PC and verify potential treatment targets.

In this study, we comprehensively analyzed the expression and prognostic value of TME-related genes in PC. We established a risk model showing high-risk patients with worse prognostic tendencies. In addition, based on this model, multiomics methods were used to explore the immune and genetic conditions to define the traits of the TME, to identify novel treatment targets (LAG3, TIGIT, and ARID1A) and predict diverse treatment sensitivities (high-risk patients were more resistant to oxaliplatin, irinotecan, and immunotherapy).