Identification of three immunogenic subtypes for PADC
In order to evaluate the ability of compiled 138 immune gene sets in pancreatic cancer classification, we performed ssGSEA on the RNA-seq data extracted from 146 pancreatic ductal adenocarcinoma (PADC) samples collected by TCGA cohort (across). We found that although different gene sets are relatively large in the enrichment score variation on a single sample, the enrichment scores of related gene sets are relatively consistent between multiple samples (Fig. 1A). At the same time, the enrichment scores of these gene sets are correlated within immunogenic modular components, especially the enrichment scores of immune cell tag gene sets are highly positively correlated. This analysis revealed that the relative abundance of many immune cell populations is correlated (Fig.1B), suggesting at least some degree of co-infiltration of the tumor. In short, these gene sets will provide profound and detailed information on the classification of immune subtypes of PDAC.
In order to define discrete pancreatic cancer categories built on the immune components of pancreatic cancer, we performed an unsupervised consensus cluster based on the expression of these 138 representative immune genes which were selected a priori. Calinski index was used to select the best number of clusters, which indicated that the best separation was a partial trivial solution reflected by dividing the queue into three clusters. These clusters reflected the different amplitudes of overall gene expression (K = 3; Fig. 1C). When the total sum of square is between 2000 and 3000, and the consistency index score is determined as 3, it is the best and the sample group could be effectively divided into three clusters in the cluster separation scatter plot (Fig. 1D). Before further excavating the patient population with the characteristics actually represented by the classification, we designated the samples as IS1 (N = 71), IS2 (N = 43) and IS3 (N = 32). IS1 is the most common, while IS3 is the least. We posited that different types to represent different immune response activations, which would be analyzed in more detail below.
To evaluate the tumor micro-environment of PDAC of different immuno-types, we wanted to identify the specific immunocytes subtypes activated among the immunogenic clusters. In particular, we deconvolved the gene-expression signatures of the IS1, IS2 and IS3 samples with CIBERSORT. Overall, in the three types of the immunogenic clusters, the majority types are NK cell resting, T cell CD4 memory activated, macrophage M0, plasma cells, B cell memory, dendritic cell resting, T cell regulatory, T cell CD8 and monocytes etc.[38, 39], which is consistent with the previous reports. IS3 tumors were enriched with B- and T-cell marker(Fig. 2A). However, less T-reg cells, macrophages, neutrophils, and DCs and more mast cells resting were obserced in the IS3 tumors compared with IS1 and IS2. Monocyte-related transcripts were also represent more in IS1 samples. Additionally, IS1 shows an enrichment in macrophages M2, while IS2 macrophages were inhibited in M0 stage. Different immune subtypes show different characteristics of immune cell infiltration, revealing different manifestations in anti-cancer immune function.
Since we have identified the significant associations and difference in immune-cell composition and abundance between three pancreatic cancer immune subtypes, we further investigated whether alterations in immune hub-genes expression were related to the presence of immune cells in the tumor microenvironment (TME). Analysis of a panel of immune-regulatory genes was confirmed, which included 5 types about immune activators, immune inhibitor, immune checkpoint resistance (ICR), major histocompatibility complex (MHC) and T regulatory (Treg) cell. The immune subtypes were significantly different among the three IS clusters, which showed a distinct association with the immune subtypes (Fig. 2B). Immune subtypes are part of a shared cluster with high expression of immunosuppressive agents, MHC-related genes, STAT1 and STAT3 transcription factors. Interestingly, from the heatmap we find the expression of PD-L1, CTLA-4, IDO1, STAT1 and STAT3was significantly higher in IS3 than other molecular subtypes.
Differential tumor biological events associated to immune-phenotypes
We next explored whether the immune-phenotypes related with clinical and pathological characteristics of the tumors. First we proceeded to compare the different immune subtypes in term of survival. In coherence with previous studies, we observed that the survival of patients showed an improvement trend in immune types IS3 at first (Fig. 3A). However, since the IS3 cluster was enriched in anti-tumor immune response, which is classically characterized by worse prognosis, prognosis of patients bearing IS3 immune phenotype was worse as compared with subjects bearing the other both immune phenotypes in the end (Fig. 3A) An intriguing exception was that the clinical outcome of pancreatic cancer is best in the case of IS2 (immune-phenotypes of lowest cytotoxicity), with the longest average survival age of the patients. These results suggest that anti-tumor immune response negatively promote the survival potential ability of patients and has higher survival time and rate of risk than tumor immune response.
In addition, we also observed that tumors at a higher pathological stage at the time of diagnosis were significantly associated with decreased cytotoxicity of immune subtypes in each immune cancer subtype (Figure 3B). These observations indicate that tumors preferentially grow when cytotoxic immune infiltration is weak. In contrast, tumors with a highly cytotoxic immune phenotype are partially controlled by the immune system and progress to more advanced stages less frequently. Finally, we analyzed the biological functional pathways enriched by differentially expressed genes of different immune subtypes (Figure 3C). The activation pathway of each immune subtype is complex and irregular, and it is impossible to find valuable clues that affect tumor occurrence and development.
Somatic mutations across immune subtypes drive invasive phenotype
To identify whether somatic mutations are specifically associated with the level of immune activations, we compared the mutations genes in each immune phenotypes (Fig. 4A). Several differentially mutated genes (DMGs) by the maftools are identified. The mutation frequency of five of these genes (e.g., KRAS, TP53, SMAD4, CDKN2A and TTN) was higher than expected by chance given background mutation processes and the missense and frame-shift mutations were in majority. Therefore, this situation were considered driver mutations. Then we observed the relationship between Ti/Tv in the overall mutation (Fig. 4B) and found that the overall level of C>T transformation is significantly higher than other transformations, indicating that the enzyme gene catalyzing the conversion of cytosine to thymine is inactivated, and the enzyme catalyzing the conversion of thymine to guanine is activated (Fig. 4B). This mutation conversion indicates that somatic mutations of immune subtypes may increase the invasiveness of normal pancreatic cells, and may be associated with poor prognosis of pancreatic cancer[40, 41].
In addition to SNP mutation, cancer mutation information also includes copy number variation and chromosome aberration. In order to investigate the difference in copy number variation in different immune subtypes, CISTIC2 was further adopted for analysis (Fig. 5A). From the results, it can be found that the copy number variation of IS3 is significantly different from the other two subtypes, and the copy number is missing at 1-2p, and increased at 4p, especially at 6-13p when compared with IS1 and IS2. The copy number changes little after 16p. IS2 has relatively more variation in amplified copy number while its delepted copy number variety is relatively few. (Fig. 5B). This may suggest that structural damage in the genome has an impact on the immune infiltration performance of tumors and further increases the mutation caused by the immune infiltration of tumor cell. However, there is no abnormal change between different subtypes when we count neoantigen, which is worth thinking and exploring when mutations between different subtypes are used to accelerate the cell death in the treatment of pancreatic cancer.
DDR defect may affect the production of immune subtypes
Since overall genomic and chromosomal instability are insufficient to explain the differences in the immune-cell composition within each PADC subtype, there were several studies reporting that DNA damage repair (DDR) is closely related to the tumor immune response. So, we aimed to investigate the effects of DDR expression in the anti-tumor activity in the TME (total mesorestal excision) and found the expression of DDR genes correlated with immune infiltration (Fig. 6A). Classical DDR related genes were selected to show their relationship with immunocyte content, such as B cell and CD4+/CD8+ T cell. These three kinds of cells are the most common immune cells derived from lymphocytes after the occurrence of cancer, , whose main function is to eliminate abnormal cells, protect the body and avoid proliferation[42, 43]. Then, we found the expression of DDR genes negatively correlated with the expression of immune-regulatory genes. For example, activation of ATM, RB1, and TP53 were part of the same group and showed high relationship with PDCD1/PD-L1 and CTLA4 (Fig. 6B). Among the seven genes, ATM was positively correlated with PDCD1, and the correlation between PALB2 and PDCD1 was the lowest (0.01). The correlation between CTLA4 and ATM was also the highest (0.499). It should be noted that CD274 was highly correlated with ATM, BRCA2, PALB2, Rb1(0.561, 0.593, 0.407, 0.533).
Another study showed that ATM, BRCA1/2, PALB2, RB1, and TP53 were linked to significant increase in immunogenic mutations. These results suggest the presence of cytolytic and regulatory cells was impacted by expression of DDR genes within the TME of PADC tumors[44]. These findings suggest that we may start with DDR genes within the TME of PADC tumors for the treatment of PCDA.