m5C regulatory gene mutations in PAAD
Thirteen m5C regulatory genes were investigated in 35 independent PAAD samples, with a total of 177 mutations found in gene sequencing data. Compared with eraser and reader genes, writer genes had a greater mutation frequency, as eraser gene mutations were not detected (Fig. a). Six of the samples revealed a higher mutation frequency in the writer gene DNMT3A. In 183 PAAD samples with CNV data, m5C regulatory genes showed a high frequency of CNV (Fig. 1b), although CNVs within NSUN1 and DNMT2 genes were undetectable. For example, the reader gene ALYREF had a CNV event frequency of 18.48%, followed by the writer gene NSUN4 with a frequency of 15.22%. The eraser gene TET2 had the lowest frequency at 6.42% (Supplementary Table 1). The writer gene DNMT3A showed the highest mutation level, which could ultimately lead to impaired biological function such as cyanosis since its protein product affects the transmission of m5C signals in the cell (Fig. 1c). The high level of DNMT3A mutations in PAAD tumor cells suggests that m5C activity could be abnormal.
These results showed that in 177 PAAD samples, two out of 13 m5C regulatory genes were hardly detected. Excluding these two genes, mRNA expression levels of the remaining 11 m5C regulatory genes significantly correlated with their CNV patterns. Increased copy numbers for eight genes, which were distributed across all m5C regulation categories, were associated with higher mRNA expression. Deletion mutants also resulted in decreased mRNA expression. Additionally, three out of 11 m5C regulatory genes had no significant relationship with CNV expression and were concentrated only in writers genes (Fig. 2). The expression levels of all eraser and reader genes were significantly related to CNVs. We next used seven functionally altered genes to predict PAAD patient survival, but only uncovered a small number of mutations without predictive statistical significance (Fig. 1d).
m5C regulatory gene changes are related to molecular clinicopathological characteristics
Since TP53, BRCA1, CDKN2A, and ATM genes play crucial roles in the pathogenesis of PAAD, we investigated their relationship with m5C regulatory genes. Interestingly, variations in m5C regulatory genes associated significantly with alterations to TP53, BRCA1, CDKN2A, and ATM (Table 1).
Table 1
Relationship between changes in m5C regulatory genes and high-frequency PAAD-related genes
| | | without SNV and CNV | with SNV and CNV | X2 | P |
ATM | | WT | 148 | 0 | 104.012609 | 2.0103E-24 |
n = 151 | Alternation | 0 | 3 |
BRCA1 | | WT | 149 | 0 | 84.1791333 | 4.5192E-20 |
n = 151 | Alternation | 0 | 2 |
CDKN2A | | WT | 150 | 1 | 18.3742163 | 1.815E-05 |
n = 151 | Alternation | 0 | 1 |
TP53 | | WT | 52 | 49 | 37.016707 | 1.1712E-09 |
n = 151 | Alternation | 0 | 50 |
| WT | 148 | 0 |
Association Between m5c Regulatory Genes And PAAD Survival
In the previous analysis, we observed a correlation between tumor clinical grade and PAAD patient prognosis (Fig. 3a). Based on the tumor staging system, we considered T1/T2 as low stage cases, and cases T3/T4 as high stage cases. We then analyzed and clustered m5C regulatory gene expression in different clinically graded cases (Fig. 3b). These results indicated that only one out of 11 m5C regulatory genes, DNMT3A, significantly correlated with the clinical grade of patients in that its high expression associated with high staging (Fig. 4). As previously described, DNMT3A also had the highest mutation rate and a positive correlation between CNV changes and expression level.
Given the relationship between clinical classification and patient prognosis, we believed that m5C regulatory gene expression would be related to patients’ prognosis. However, no significant correlation between m5C regulatory gene CNVs and patient prognosis was observed. In order to correlate the expression of some m5C regulatory genes with CNVs, we used single factor Cox regression to explore the relationship between m5C regulatory gene expression and patient prognosis. The expression of two genes showed a significant positive correlation with patient prognosis (P < 0.05), including DNMT3A. Among these two genes, the expression level of DNMT3A was significantly related to CNV changes (Table 2). Using multivariate Cox regression analysis, we next found that m5C regulatory gene expression can significantly predict patient risk using AUC at 1, 3, and 5 years, with all AUC values greater than 0.65 (Fig. 5a, b). These results demonstrated that the expression of m5C regulatory genes can be used as a prognostic marker for pancreatic cancer.
Table 2
Univariate Cox regression exploring the relationship between different m5C regulatory gene expression levels and patient prognosis.
| beta | HR | (95% CI for HR) | Wald test | p-value | CNV sig. |
DNMT3A | -0.31 | 0.74 | (0.57–0.95) | 5.6 | 0.018 | yes |
NSUN6 | -0.3 | 0.74 | (0.57–0.96) | 5.2 | 0.022 | no |
NSUN7 | -0.21 | 0.81 | (0.64-1) | 3.2 | 0.071 | no |
NSUN2 | 0.044 | 1 | (0.98–1.1) | 2.1 | 0.15 | yes |
NSUN3 | 0.21 | 1.2 | (0.76-2) | 0.71 | 0.4 | yes |
NSUN5 | 0.02 | 1 | (0.97–1.1) | 0.53 | 0.47 | no |
NSUN4 | -0.04 | 0.96 | (0.72–1.3) | 0.07 | 0.79 | yes |
TET2 | -0.045 | 0.96 | (0.68–1.4) | 0.06 | 0.8 | yes |
ALYREF | -0.0013 | 1 | (0.98-1) | 0.02 | 0.88 | yes |
DNMT3B | 0.031 | 1 | (0.69–1.5) | 0.02 | 0.88 | yes |
DNMT1 | -0.0011 | 1 | (0.89–1.1) | 0 | 0.99 | yes |
Based on the above results, we performed a LASSO technique on the 13 m5C regulatory genes to more precisely identify prognostic markers of PAAD. Through 1,000 LASSO regressions, we observed that LASSO results repeatedly appeared more than 900 times. Only DNMT3A exhibited a significant result for expression level and CNV by single factor Cox analysis, and a significant correlation between expression level and clinical grade (P < 0.05) (Table 3).
Table 3
Results of lasso analysis of m5C regulatory genes
duplicates | genes | CNV express sig. | functions | Survival sig. | Stage sig. |
985 | NSUN6 | No | writer | no | no |
965 | DNMT3A | Yes | writer | Yes | Yes |
939 | NSUN2 | Yes | writer | no | no |
923 | NSUN3 | Yes | writer | no | no |
234 | DNMT3B | Yes | writer | no | no |
200 | TET2 | Yes | eraser | no | no |
186 | DNMT1 | Yes | writer | no | no |
105 | NSUN7 | No | writer | no | no |
99 | NSUN4 | Yes | writer | no | no |
88 | NSUN5 | No | writer | no | no |
86 | ALYREF | Yes | reader | no | no |
Given that the DNMT3A gene is a writer gene involved in vital m5C regulatory functions and to predict patient risk based on DNMT3A gene expression, we analyzed the relationship between DNMT3A expression and PAAD patient prognosis. Notably, the results illustrated that low DNMT3A expression significantly correlated with poor prognosis (Fig. 5c). These results suggested that DNMT3A gene expression could be a clinically significant biomarker for PAAD patients.
Functional Enrichment Analysis Of DNMT3A Gene In Expression Level
Because the DNMT3A gene is a methylation writer gene, we next investigated the role of m5C dysfunction in PAAD pathogenesis. We examined enriched gene sets in samples with varying mRNA expression of the DNMT3A gene. Gene enrichment analysis by gene set enrichment analysis (GSEA) showed that low DNMT3A expression was related to ribosomal processing (Fig. 5d-f). Ribosomes are essential for protein synthesis, and because of the elevated metabolic activity in tumor cells, protein synthesis requirements are much greater than normal cells. These findings also confirmed that the low expression of DNMT3A was related to the poor prognosis of patients.
Validation of DNMT3A act as an important pancreatic cancer target tiller invalidation data set
Finally, we used a validation data set to analyze the relationship between DNMT3A gene expression and patient survival. Based on Cox regression analysis, low expression of DNMT3A was associated with poor prognosis (Fig. 6a). A schematic diagram showed the related mechanisms of m5C in PAAD cells (Fig. 6b), and these findings provided evidence of the role that m5C epigenetic regulation played in pancreatic cancer.