Cpg Island Methylator Phenotype With Distinctly Different Prognosis And Molecular Features In Pancreatic Cancer Patient

CpG island methylator phenotype (CIMP), featured with concurrent and widespread hypermethylation of a cluster of CpGs, has been reported to play an important role in carcinogenesis. Limited studies have investigated the CIMP in pancreatic cancer (PC). The aim of this study was to explore the CIMP in PC patients and its clinical-pathological and genomic characteristics. Methods: DNA methylation, somatic mutation, mRNA and corresponding clinical data of PC patients were downloaded from TCGA (184 patients) and ICGC (264 patients). Univariate and multivariate regression analysis were used to identify prognosis related CpGs. Consensus clustering analysis was used for identication of the CIMP in PC patients. ESTIMATE and CIBORORT were used for estimation of tumor microenvironment (TME) in PC patients. checkpoint genes, such as PD-1, CTLA4, CD86, VTCN1 and LAG-3 were also found at patients of CIMP-H subgroup compared to that in patients of CIMP-L subgroup. In the end, we validated the CIMP status in PC patients of ICGC dataset. Three distinct CIMP subgroups of PC patients with distinct clinical characteristic, prognosis, gene mutation landscape and TME were identied and The CIMP may help to make an assertion to provide specic and ecient treatment options for patients of different subtypes. 6 (GraphPad Software) and R software (version 3.5.1) were used for statistical analysis and ploting graph. The association between CIMP status with clinical-pathological features were analyzed with the chi-square test. One-way ANOVA analysis was carried out to compare difference of TMB, stromal score, immune score, estimate score, tumor-inltrating immune cells, expression of chemokines and immune checkpoint genes among PC patients with different CIMP status. Univariate and multivariate Cox regression analysis were performed to analyze the prognostic value of CIMP status. Kaplan–Meier analysis with log-rank test was performed to analyze the difference of OS or PFS among patients with different CIMP status. P<0.05 was considered as statistically signicant. inactivated in about 50% of PC patients. Alterations in p53 and SMAD4 were late events in PC, as they often occurred in pancreatic patients with histologic grade 3 and high invasiveness (29). Consistent with the results in gene mutation, KEGG pathways, such as “P53 signaling pathway,” was found to be signicantly enriched in patients of CIMP-H subgroup. Moreover, CIMP may also act as a tumor promoter in PC carcinogenesis by inuencing mutation of major driver genes, and PC patients of different CIMP subtypes originated from precursor cells might have a distinct epigenetic background of the cell of origin.


Introduction
Pancreatic cancer (PC) is the fteenth most common cancers and the seventh leading cause of cancer-related death worldwide, which is reported to have poor prognosis and high mortality rate (1). Lots of efforts were put in understanding the underlying mechanisms of PC and new treatments were investigated, but a marginal improvement of survival was observed in PC (2). Neureiter et al. reported the median survival of PC patients to be 6 months and the 5-year survival rate to be as low as 6%, attributing the cause to be lack of reliable early diagnostic markers and the nature of high aggressiveness and drug resistance of the disease itself (3). Of note, only about 10% of PC patients were report with family histories (4), indicating the etiology of PC to be more likely sporadic than genetic and suggesting epigenetic alterations may take in part of the development and progression of PC.
DNA methylation is one of the main epigenetic modi cations and plays an important role in the development and progression of sorts of malignant tumors (5). During the process of oncogenesis, aberrant DNA methylation of CpG islands are located in the promoter regions of genes, such as hypermethylation surrounding the promoters of tumor suppressor genes (TSGs) and hypomethylation of promoters of oncogenes, resulting in transcriptional silence of TSGs and over-expression of oncogenes (6). Aberrant DNA methylation of non-promoter of genes also contributes to the formation of intra-tumoral heterogeneity (7). To date, many gene mutations with deregulated DNA methylation have been identi ed in PC patients. Treatments targeting these genes were reported to signi cantly improved the overall survival in PC mice models (8,9).
CpG island methylator phenotype (CIMP), rst discovered and validated in colorectal cancer, is featured with concurrent and widespread hypermethylation of a cluster of CpGs in distinct cancer subtypes. CIMP affects the intergenic regions of the whole genome and leads to chromosomal instability, which is thought to play an important role in carcinogenesis (10). To date, CIMP phenotype has been identi ed in many kinds of tumors, including PC (11). Subtypes with different CIMP pattern showed distinct epidemiological, clinical-pathological and genomic characteristics (12,13). However, the limitation of earlier studies on CIMP of PC patients was obvious for the limited patient samples or CpG sites used in these study (5,14). Little of these studies validated their ndings in independent cohorts and further explored the association of CIMP status with tumor microenvironment (TME) in PC patients (11). In the present study, we identi ed and validated three distinct methylation subgroups of PC patients, termed CIMP-L subgroup, CIMP-M subgroup, and CIMP-H subgroup, with the data downloaded from The Cancer Genome Atlas (TCGA) and International Cancer Genome Consortium (ICGC). Patients of different CIMP status showed distinct clinical characteristic, prognosis, gene mutation landscape and TME. As identi cation of clinically relevant cancer subtypes based on DNA methylation pattern is an important computational problem in medicine, CIMP may help to make an assertion to provide speci c and e cient treatment options for patients of different subtypes.

Ethics statement
All the data analyzed in the present study were got from TCGA and ICGC. Informed consents had already been obtained from the patients before the present study.
Data acquisition from TCGA Level-3 DNA methylation data of 184 PC patients and 10 normal controls were downloaded from the TCGA (https://cancergenome.nih.gov/, 2020-04-20). The methylation status of each CpG was expressed as a β value, which ranged from 0 to 1, where higher β values indicated higher methylation. We removed β values for those CpG probes that were either mapped against chromosomes X and Y, or missing in more than 20% of the samples. Besides, those probes mapped to SNP within 10bp of interrogated CpG sites were also removed (15).
The somatic mutation data of 177 PC patients measured by the whole exome sequencing were downloaded from TCGA and maftools package was used to analyze these data (16). Patients harboring missense mutations, nonsense mutations, multiple hits, splice-site mutations, frame shift insertions, frame shift deletions, in-frame insertions or in-frame deletions were considered as positive for mutation. Moreover, tumor mutational burden (TMB), regarded as a promising biomarker for immunotherapy responses, was also calculated with the method described in previous study (17). mRNA expression data of 178 PC patients was also obtained from TCGA. The averaged gene expression was adopted in case of duplicates and the gene expression data was normalized with the scale method (18). Besides, gene expression missed in more than 20% of the samples were removed.
Meanwhile, corresponding clinical-pathological data of 184 PC patients, including gender, age, T stage (T), N stage (N), M stage (M), TNM stage, overall survival (OS) status and time, progression free status (PFS) status and time were also downloaded.

Identi cation of the CIMP in PC patients
To evaluate the CIMP phenomenon in PC patients, rst, the differential CpGs between 184 PC patients and 10 normal samples were screened out. Then, univariate and multivariate Cox regression analysis were performed to further identify the OS-related CpGs. After that, consensus clustering analysis for unsupervised class of 184 PC patients based on expression similarity of OS-related CpGs was performed with consensusclusterplus packages. Besides, principal component analysis (PCA) was also performed to examine whether the clusters of PC patients were suitable with 'limma' package.
Estimation of stromal and immune cells in malignant tumor tissues using expression data (ESTIMATE) ESTIMATE was a tool used for predicting tumor purity, the presence of in ltrating stromal/immune cells in tumor tissues with mRNA expression data. Based on single sample Gene Set Enrichment Analysis (ssGSEA), ESTIMATE could generate three scores: stromal score (that captured the presence of stroma in tumor tissue), immune score (that represented the in ltration of immune cells in tumor tissue), and the estimate score (that negatively correlated with tumor purity) (19).

Cancer stemness indices of PC patients
Previously, Malta et al. extracted transcriptomic and epigenetic feature sets derived from non-transformed pluripotent stem cells and their differentiated progeny using an innovative one-class logistic regression machine learning algorithm (OCLR). As a results, they identi ed four cancer stemness indices, including mRNAsi, epigenetically regulated mRNAsi (EREG-mRNAsi), mDNAsi and EREG-mDNAsi for assessing the degree of oncogenic dedifferentiation (20). Based on their study, we could attain the four stemness indices of each PC patients in the TCGA database. To explore the underlying mechanism exploited by the CIMP status to in uence the prognosis of PC patients, GSEA analysis (Version: 4.2; http://software.broadinstitute.org/gsea/index.jsp) was performed. As a result, the difference of KEGG pathways between PC patients with distinct CIMP status were identi ed (22).

Validation of CIMP status in PC patients of ICGC dataset
To independently test the CIMP status in PC patients, DNA methylation data of 264 PC patients, somatic mutation data of 264 PC patients, mRNA expression data of 175 PC patients and clinical-pathological parameter of 264 PC patients were downloaded from ICGC (https://dcc.icgc.org/, 2020-04-20). Processing of methylation data, somatic mutation data and mRNA expression data were similar to that in TCGA.

Data analysis ow chart
To make our study to be better understood, a work ow of the study was depicted and was shown at Figure 1.

Statistical analysis
GraphPad Prism 6 (GraphPad Software) and R software (version 3.5.1) were used for statistical analysis and ploting graph. The association between CIMP status with clinical-pathological features were analyzed with the chi-square test. One-way ANOVA analysis was carried out to compare difference of TMB, stromal score, immune score, estimate score, tumor-in ltrating immune cells, expression of chemokines and immune checkpoint genes among PC patients with different CIMP status. Univariate and multivariate Cox regression analysis were performed to analyze the prognostic value of CIMP status. Kaplan-Meier analysis with log-rank test was performed to analyze the difference of OS or PFS among patients with different CIMP status. P<0.05 was considered as statistically signi cant.

Identi cation of the CIMP in PC patients
To identify the CIMP in PC patients, we rst screened out the differential CpGs between PC samples and normal samples with the DNA methylation data of 185 PC patients and 10 normal controls downloaded from TCGA. In total, 22,450 differential CpGs were identi ed, and among these CpGs, 12,937 were hypermethylated CpGs and 9,513 were hypomethylated CpGs. Next, univariate Cox analysis was performed to identify OS-related CpGs and 3102 CpGs were selected with p <0.05. Then, 1037 out of 3102 CpGs with p <0.01 were further used to identify the most OS-related CpGs by multivariate Cox analysis. Seventy-two CpGs were nally selected for unsupervised consensus clustering analysis with p<0.05 (Supplementary materials). Based on the unsupervised consensus clustering analysis, 184 PC patients were clustered into three distinct groups ( Next, the associations between CIMP status and clinical characteristics were analyzed. As were shown at Table 1, there were more patients with advanced T stage and TNM stage in the CIMP-M subgroup and CIMP-H subgroup compared to that in CIMP-L subgroup (all p<0.05, Table 1). Besides, Kaplan-Meier analysis showed that there were signi cant differences in OS and PFS among PC patients from different CIMP status. Patients of the CIMP-H subgroup had poorer OS and RFS while the patients of CIMP-L subgroup had better OS and RFS (all p<0.05, Figure 2D, 2E). Moreover, univariate Cox analysis indicated that CIMP status were signi cantly related with OS and PFS, and multivariate Cox analysis also suggested that CIMP status were independent prognostic factors for OS and PFS of PC patients after adjusting for gender, age, T state, N stage, M stage and TNM stage (all p<0.05, Table 2).

Mutational landscapes of PC patients with different CIMP status
A number of mutated genes with deregulated DNA methylation had been identi ed to be played important roles in the development and progression of PC (8). The association of CIMP status with gene mutations was analyzed with somatic mutation data of 177 PC patients downloaded from TCGA. As were shown at Figure Figure 3A-3C), which had been shown to be major driver genes in PC (23). Similarly, signi cant difference of tumor mutational burden (TMB), served as a biomarker of immunotherapy responses, were also found among patients with different CIMP status. Higher TMB were found at patients of CIMP-H subgroup, while lower TMB were observed at patients with CIMP-L status ( Figure 3D).

Landscape of TME in PC patients with distinct CIMP status
Consisted of cancer cells, stromal cells and extracellular components, the TME had been demonstrated to play indispensable roles in tumorigenesis, progression, metastasis, recurrence, and drug resistance of PC (24). The difference of TME in patients with distinct CIMP status was also analyzed. As were shown at Figure 4, signi cantly lower stromal score, immune score and estimate score were found at patients of CIMP-H subgroup, while signi cantly higher stromal score, immune score and estimate score were found at patients of CIMP-L subgroup ( Figure 4A-4C). Similarly, signi cantly higher tumor purity score was also observed in patients of CIMP-H subgroup ( Figure 4D).
Cancer stem cells (CSCs) were cancer cells that possessed the ability to give rise to all tumor cell types and CSCs were considered to be responsible for tumor growth, metastasis and recurrence, and resistant to chemotherapy and radiation therapy. The association of CIMP status with cancer stemness indices was explored. As expected, signi cantly higher tumor stemness indices, including mRNAsi score, mDNAsi score EREG-mDNAsi score were found at patients of CIMP-H subgroup compared to that in patients of CIMP-L subgroup ( Figure 5A, 5C, 5D).
Next, we analyzed the difference of tumor-in ltrating immune cells among patients with distinct CIMP status. As were shown at Figure  6, signi cantly lower total T cells, total B cells, naive B cells, CD8 T cells, CD4 T cells, resting memory CD4 T cells and activated memory CD4 T cells were found in patients of CIMP-H subgroup compared to that in patients of CIMP-L subgroup ( Figure 6A-6F).
Besides, signi cantly higher M0 macrophages was found in patients of CIMP-H subgroup and signi cantly higher regulatory T cells was found at patients of CIMP-M subgroup compared to that in patients of CIMP-L subgroup ( Figure 6G-6H). It had been reported that different kinds of immune cell subsets were recruited into the TME via interactions between chemokines and their chemokine receptors (25). We furthered analyzed the difference of expression of 58 kinds of chemokines among patients with different CIMP status. As were summarized at Table 3, in line with the results of tumor-in ltrating immune cells, 31 kinds of chemokines, such as CCL2, XCL2, CCR2, CCL5 and CCR5, were found to be overexpressed in patients of CIMP-L subgroup, while only 7 kinds of chemokines, such as CXCL14 and CXCL16, were found to be increased in patients of CIMP-M and CIMP-H subgroup. Taken together, these results suggested that patients in CIMP-H subgroup have a distinct TME, characterized by higher tumor purity and tumor stemness, and lower immune activation and immune in ltration.

Expression of immune checkpoint genes in PC patients with different CIMP status
The advent of immunotherapy, especially checkpoint inhibitor-based immunotherapy, had revolutionized cancer treatments, especially for patients with advanced tumor. These treatments functioned through blockade of immunosuppressive checkpoints, so the expression of these immune checkpoint genes were necessary for checkpoint inhibitor immunotherapy (26). The difference of expression in10 immune checkpoint genes (including PD-1, PD-L1, CTLA4, PD-L2, CD86, CD80, CD276, VTCN1, Tim-3 and LAG-3) in PC patients with different CIMP status were further analyzed. As were shown at Figure 7, the expression of PD-1, CTLA4, CD86, VTCN1 and LAG-3 of PC patients of CIMP-H subgroup were signi cantly lower than that of patients of CIMP-L subgroup ( Figure 7A, 7C, 7E, 7H-J). These results may indicate that checkpoint inhibitor immunotherapy to be less effective in patients of CIMP-H subgroup as they showed less expression of immune checkpoint genes.
Potential mechanism by which CIMP status to in uence the prognosis of PC patients GSEA analysis was performed to explore the underlying biological mechanism by which CIMP in uenced prognosis of PC patients. As was shown at Figure 8, KEGG pathways, such as "P53 signaling pathway", "notch signaling pathway", "calcium signaling pathway", "DNA replication" and "base excision repair" were found to be signi cantly enriched in patients of CIMP-H subgroup compared that in patients of CIMP-L patients ( Figure 8A). Similarly, "P53 signaling pathway", "base excision repair" and "proteasome" were found to be signi cantly enriched in patients of CIMP-M subgroup compared that in patients of CIMP-L subgroup ( Figure 8B). These results may suggest that the CIMP status may in uence the prognosis of PC patients by regulation of the aforementioned biological process.  (Table 4). Moreover, Kaplan-Meier analysis showed that there were signi cant differences in OS among PC patients with different CIMP status. Patients of the CIMP-H subgroup had poorer OS while the patients of CIMP-L subgroup had better OS (p=0.003, Figure 9A). Univariate and multivariate Cox analysis also suggested that the CIMP status were signi cantly related with OS, and was also an independent prognostic factor for OS of PC patients after adjusting for gender, age, T state, N stage, M stage and TNM stage (p=0.012, Table 5).  Figure 9B). Besides, higher TMB were also found at patients of CIMP-H subgroup, while lower TMB were observed at patients with CIMP-L status ( Figure   9B).
The landscape of TME among PC patients with distinct CIMP status was also explored. As expected, signi cantly lower stromal score, immune score and estimate score were found at patients of CIMP-L subgroup. Meanwhile, signi cantly lower total T cells, total B cells, CD8 T cells, memory CD4 T cells were found in patients of CIMP-H subgroup, but signi cantly higher follicular helper T cells were found in patients of CIMP-H subgroup ( Figure 9C).
Finally, the association of CIMP status with immune checkpoint genes was also analyzed. Only expression of 4 immune checkpoint genes, including PD-1, PD-L1, CD86 and CD276 were available. Similarly, lower expression of PD-1, PD-L1 and CD86 were found at patients of CIMP-H subgroup, but the difference was not statistically signi cant ( Figure 9D).

Discussion
Identi cation of clinically relevant cancer subtypes based on DNA methylation pattern is an important computational problem in medicine, which may help to provide speci c and effective treatment options for patients with different subtypes. Previously, DNA methylation pattern analysis of PC patients have been performed but were limited to low sample size and small number of CpGs sites (5,11,14). Sato et al. analyzed the genome-scale DNA methylation patterns in PC patients, but only 8 genes of methylation speci c site were used (14). Thompson et al. explored the association of DNA methylation patterns with survival of PC patients, but number of samples used for analysis was small, in which only 11 PC patients, 2 normal controls and 3 chronic pancreatitis patients were included (2). Recently, Nitish et al. identi ed three CIMP subtypes of PC patients with distinct clinical characteristic and gene mutation landscape by clustering of differentially methylated sites using the genome-scale methylome data of PC patients from TCGA; however, they did not further explored the relationship among CIMP status, prognosis of PC patients and TME. They also did not validate the three CIMP subtypes in independent PC cohorts (11). In the present study, we identi ed and validated three distinct CIMP subgroups (termed CIMP-L, CIMP-M, and CIMP-H subgroup) in 448 PC patients with the data downloaded from TCGA and ICGC. In agreement with previous studies, we observed that CIMP status was signi cant associated with clinical characteristic, OS and RFS in PC patients.
There were more patients with advanced T stage and TNM stage in the CIMP-M subgroup and CIMP-H subgroup. Patients of CIMP-H subgroup had the worst OS and RFS. Moreover, the CIMP status was also an independent prognostic factor for OS and DFS in PC patients.
Accumulation of somatic mutations in oncogenes and TSGs were common in the development and progression of cancer (27). Based on the mutation analysis, signi cantly higher somatic mutation burdens in KRAS, TP53, SMAD4, CDKN2A were observed in patients of CIMP-H subgroup, which had been shown to be major driver genes in PC (23). Encoding a small GTPase that involved in cellular proliferation, motility, and cytoskeletal remodeling, KRAS was the most frequently mutated oncogene in PC. More than 90% of PC patients showed somatic mutations in KRAS (28). CDKN2A encoded an essential cell-cycle regulator and was reported to be the most frequently mutated TSGs in PC. Similarly, more than 90% of PC patients exhibited function defect of CDKN2A because of gene mutation (28). Besides, studies have showed that somatic KRAS and CDKN2A mutations were early events of PC development as they were the earlier alteration genes in most low-grade pancreatic intraepithelial neoplasia (29). TP53 played a vital role in the cellular stress response. Somatic mutations in TP53 were also observed frequently in a wide range of tumor types, including PC. SMAD4 was found to mainly mediate signaling downstream of TGFβ receptor and was inactivated in about 50% of PC patients. Alterations in p53 and SMAD4 were late events in PC, as they often occurred in pancreatic patients with histologic grade 3 and high invasiveness (29). Consistent with the results in gene mutation, KEGG pathways, such as "P53 signaling pathway," was found to be signi cantly enriched in patients of CIMP-H subgroup. Moreover, CIMP may also act as a tumor promoter in PC carcinogenesis by in uencing mutation of major driver genes, and PC patients of different CIMP subtypes originated from precursor cells might have a distinct epigenetic background of the cell of origin.
Consists of cancer cells, stromal cells (mainly composed of broblasts and immune cells) and extracellular components, TME has been found to promote tumor progression, metastasis niche formation and therapeutic resistance in PC (24). In the present study, we also found that CIMP status was signi cantly associated with TME. On one hand, signi cantly higher tumor purity and cancer stemness indices were found in patients of CIMP-H subgroup. Researches have proven that loss of differentiated phenotype, acquisition of progenitor and stem cell-like features were the main hallmarks of cancer progression. It was speculated that cancer cells might arise from a cell population with self-renewal ability, which was thought to be CSCs. Pancreatic cancer stem cells (PCSCs) have been rst identi ed in 2007 and were reported to take part in the resistance to standard chemotherapy and radiation treatment as they could express multidrug resistant membrane transporters, aberrantly activate proliferation signaling pathways and increase the capability of repairing DNA (30). On other hand, lower stromal score and immune score were found at PC patients of CIMP-H subgroup.

Consistently, lower proportion of total T cells, total B cells, naive B cells, CD8 T cells, CD4 T cells, memory CD4 T cells and activated
memory CD4 T cells were also found at patients of CIMP-H subgroup, which indicated that patients of CIMP-H subgroup had a distinct immune phenotype, characterized by less immune cells in ltration, lower cytotoxic potential and immune activation. Moreover, many kinds of chemokines, such as CCL2, XCL2, CCR2, CCL5 and CCR5, were also found to be decreased at patients of CIMP-H subgroup, which may in part accounted for the lower in ltration of immune cells as many different kinds of immune cell were mainly recruited into the TME via interactions between chemokines and their chemokine receptors (25). Taken together, the higher tumor purity and cancer stemness indices, and lower in ltration of immune cells in patients of CIMP-H subgroup may contribute to their worse OS and DFS.
The advent of immunotherapy, especially checkpoint inhibitor-based immunotherapy, have revolutionized cancer treatments, especially for advanced cancer patients (26). Currently, the monoclonal antibodies, targeted for PD-1 and its ligands, have been successfully applied in clinical practice and have been approved for several cancers (such as melanoma, non-small cell lung carcinoma, renal cancer, and bladder cancer) (31). Besides, researches have showed that combination of anti-PD-1/PD-L1 antibody and CTLA-4 inhibitor can improve treatment effect of patients with advanced melanoma, which was approved by FDA in treating BRAF V600E wild-type patients with unresectable or metastatic melanoma (32,33). For example, a recent phase-II clinical trial have proved that combination of ipilimumab (a CTLA-4 inhibitor) and nivolumab (a PD-1 inhibitor) signi cantly improved treatment e cacy in advanced melanoma patients compared with monotherapy with ipilimumab (34). In our study, signi cant association of CIMP status with expression of immune checkpoint genes was found. Lower expression of PD-1, CTLA4, CD86, VTCN1 and LAG-3 were found at PC patients of CIMP-H subgroup. Consider the expression of immune checkpoint genes were necessary for checkpoint inhibitor immunotherapy. Effective immunotherapy of immune checkpoint inhibitors depended on the generation of neoantigen-speci c T-cells and its in ltration in theTME. Immunotherapies were less likely to be e cacious in patients of CIMP-H phenotype as lower expression of immune checkpoint genes and T cells in ltration were found.
As changeable and possibly heritable genetic alteration, epigenetic regulation may prove promising clues for the treatment of various kinds of diseases, including cancers (35). In-vitro studies have shown that 5-Aza-CdR, a DNA methyltransferase 1 (DNMT1) inhibitor, could induce cell death and apoptosis of pancreatic cancer cells by reactivation of RASSF1A and up-regulation of Bax genes (36). Han et al. also observed the synergistic effects of the combination of 5-Aza-CdR and suberoylanilide hydroxamic acid on the anticancer property of PC (37). More inspiring, phase I/II clinical trials of DNMT1 inhibitors (azacitidine, decitabine and guadecitabine) in PC patients were currently underway, the inhibitors of which exhibited potential treatment outcomes (38). It is a promising future in drugs designing for epigenetic targets, while CIMP may help to make an assertion to provide speci c and e cient treatment options for patients with different CIMP status.
However, there were some limitations to be addressed in our study. First, our analysis was performed on the basis of single-omics (DNA methylation). Patients with the same CIMP status might have heterogeneity due to the different characteristics in terms of other omics data. Second, our analysis was performed on the basis of retrospective cohort. Prospective studies with larger sample sizes should be performed to validate our ndings. Finally, the biological functions and molecular mechanisms of CIMP status to in uence the prognosis of PC patients should be further validated in-vitro experiment.
In conclusion, we identi ed and validated three distinct CIMP subgroups in PC patients. Patients of different CIMP subgroups showed distinct clinical characteristic, prognosis, gene mutation landscape and tumor microenvironment (TME). It may help to make an assertion to provide speci c and e cient treatment options for patients with different CIMP status.

Declarations Ethics approval
All the data analyzed in the present study were got from TCGA and ICGC. Informed consents had already been obtained from the patients before the present study.

Consent to participate
Not applicable.

Consent for publication
Not applicable.

Availability of data and materials
The data of the study are available from the corresponding web page link, including GDC Data portal (https://cancergenome.nih.gov/) and ICGC portal ICGC (https://dcc.icgc.org/).

Competing interests
The authors declare that the research was conducted in the absence of any commercial or nancial relationships that could be construed as a potential con ict of interest.   Figure 1 The work ow chart of the present study.