A m6A/m1A/m5C-related eight-gene signature predicts prognosis and correlates with immune microenvironment in pediatric acute myeloid leukemia

Introduction

Acute myeloid leukemia (AML) is a rare, life-threatening hematologic malignancy originating from myeloid stem-cell precursors, accounting for approximately 25% of pediatric leukemia cases (1,2). Substantial research efforts have advanced our understanding of pediatric AML, revealing its unique biological and clinical characteristics distinct from adult AML. Pediatric AML exhibits distinct molecular heterogeneity and clinical characteristics compared to adult AML, with a survival rate of around 70%, but a high relapse rate of up to 30% (3). Chemotherapy, the primary therapeutic approach, has limited effectiveness due to drug resistance (4). Intensive investigation on immune-checkpoint targets in leukemic cells has promoted the development of immunotherapy for AML in clinical practice (5). However, the exploration of immunotherapy targets is still in its fancy. Therefore, identifying effective prognostic biomarkers and exploring candidate molecular targets are of great clinical significance, as it can improve the outcomes of pediatric AML patients.

RNA methylation [N6-methyladenosine (m6A), 5-methylcytosine (m5C) and N1-methyladenosine (m1A)], a dynamic and reversible epigenetic modification, plays a pivotal role in regulating gene expression by modulating RNA metabolism, including translation, degradation, splicing, and stability (6). m6A/m5C/m1A are added to target RNAs by methyltransferases (referred to as “writers”), removed by demethylases (“erasers”), and recognized by binding proteins (“readers”). Through these processes, they participate in modulating RNA translation, degradation, splicing and other functions (7).

Mounting evidence has established the critical roles of m6A/m5C/m1A regulatory genes in tumor initiation and progression. For m6A, the most widespread and well-studied form of RNA modification, plays a significant role in AML tumorigenesis and progression (8,9). A m6A-related signature composed of YTH domain-containing protein 3 (YTHDF3, a reader) and alkB homologue 5 (ALKBH5, an eraser) can predict the survival of AML patients and is closely associated with the tumor immune microenvironment (10). m5C “writers”, such as DNA methyltransferases (DNMTs), and “readers”, such as YTHDF2, ALYREF and Y-box binding protein 1 (YBX1), are involved in cellular metabolism and motility, thereby influencing cancer metastasis and progression (11). An increasing amount of evidence indicates that m5C regulators are relevant to predicting the prognosis of ovarian cancer patients (12,13). In addition to its roles in tumor progression, m1A is involved in shaping the immune-microenvironment and has prognostic value in several cancers, including ovarian cancer (14) and oral squamous cell carcinoma (15). Nevertheless, the biological functions of m6A/m1A/m5C regulators in pediatric AML have acarely been elucidated.

While the prognostic significance of m6A/m5C/m1A regulators has been increasingly acknowledged in various adult malignancies (16,17), their functions in pediatric AML remain largely uncharted. Pediatric AML differs significantly from adult AML in genetic landscape, and tumor microenvironment (18). Adult-derived prognostic models are not directly applicable to pediatric AML due to divergent mutation frequencies (e.g., FLT3-ITD, NPM1, and DNMT3A) and disease origin, treatment response, long-term prognosis (19). Established risk stratification systems for pediatric AML, based on cytogenetics, molecular aberrations, and clinical features, guide treatment decisions but still face challenges in refining risk groups and predicting relapse (20). Given that pediatric AML-specific prognostic signatures combining these three RNA methylation types are currently lacking, we hypothesize that m6A/m5C/m1A regulatory genes may serve as valuable prognostic biomarkers and immunotherapy targets for pediatric AML.

Though epigenetic regulators like TET2 and DNMT3B have been studied in adult AML, their roles together with m6A/m1A/m5C modifications in pediatric AML remain unclear. Pediatric AML differs substantially from adult AML in genetics, immune microenvironment, and treatment response, requiring disease-specific prognostic tools. Most previous studies focused on a single RNA modification, while neglecting their coordinated effects in pediatric AML. Therefore, we constructed an integrated prognostic signature combining m6A, m5C and m1A to enable more stable and accurate risk stratification. In this research, we focused on screening for differentially expressed m6A/m1A/m5C genes [m6A/m1A/m5C-related differentially expressed genes (DEGs)] from the Xena database. Based on the prognostic m6A/m1A/m5C-related DEGs, a risk model and a nomogram were constructed to predict survival in pediatric AML patients. Additionally, we explored the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways associated with these signature genes, as well as their relationships with the tumor microenvironment (TME) and predicted immunotherapy response. The present study aimed to fill the gap by establishing an m6A/m1A/m5C-related 8-gene signature and a clinical nomogram for pediatric AML, with further exploration of immune microenvironment features and potential associations with immunotherapy response. We present this article in accordance with the TRIPOD reporting checklist (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0367/rc).

Methods

Data source

On October 15, 2022, RNA sequencing (RNA-seq) data and clinical information of 187 AML patients from The TARGET AML projects were downloaded from the Xena database (https://xenabrowser.net/datapages/). All samples were diagnostic bone marrow specimens. We selected 179 patients aged <18 years with survival data as the training set. In addition, the E-MTAB-1205 dataset (50 AML adolescents, mean age 11.39±4.54 years) (21) was downloaded from the BioStudies ArrayExpress database (https://www.ebi.ac.uk/biostudies/arrayexpress), and was used as an independent European validation set. Detailed clinical characteristics of the validation set are summarized in Table S1. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

RNA-seq expression data from the TARGET and E-MTAB-1205 datasets were processed as follows: raw counts were normalized per cohort using the Trimmed Mean of M-values (TMM) method in edgeR. No cross‑dataset batch correction was applied to preserve external validation validity, mimicking real‑world data processing. All data analyses were performed with pre-specified parameters and automated scripts, and investigators were blinded to sample group allocation during key analytical steps to minimize potential bias.

Screening m6A/m5C/m1A-related DEGs between alive and dead samples, and protein-protein interaction (PPI) network analysis

Based on the independently normalized expression data of the training set (22,23), all samples were divided into an alive group (patients who survived) and a dead group (patients who deceased). Known m6A/m1A/m5C writer, reader, and eraser genes were retrieved from published literatures (24,25). DEGs between the alive and dead groups were identified using Student’s t-test with false discovery rate (FDR) t <0.05. Then, correlations among the identified m6A/m1A/m5C-related DEGs were calculated using the cor function in R language (version 3.6.1). To minimize bias and avoid circularity, we applied a two‑step approach: DEGs first underwent univariate Cox regression (P<0.05) accounting for survival time and censoring; and only survival‑associated genes were retained for signature construction.

Subsequently, the proteins encoded by the identified DEGs searched in the STRING database (26) (version 11.0) were used to construct a PPI network, and visualized using Cytoscape software (version 3.9.0). Furthermore, gene ontology (GO) function enrichment analysis was performed for these m6A/m1A/m5C-related DEGs within the PPI network using clusterProfiler software (version, 4.4.4), with a P value <0.05 set as the cutoff.

Construction of a risk score model for survival prediction

Based on the patients’ survival information, the relationships between the DEGs and the survival time of AML patients were explored using univariate Cox regression analysis (P value <0.05) using the survival package (27) in R. Subsequently, with the lars package (version 1.2) in R software, the prognostic genes screened by univariate Cox regression analysis were further refined by least absolute shrinkage and selection operator (LASSO) cox regression analysis to screen the optimal prognostic genes. Ten-fold cross-validation (10-CV) was performed to select the optimal lambda (λ) value, which minimizes the mean squared error and avoids overfitting. Based on the optimal lambda value, the LASSO algorithm pinpointed the survival-related genes for constructing the final signature (28,29). By combining the LASSO coefficients and the expression data of the optimal prognostic genes, a risk score (RS) was calculated using the following formula:

RS=∑Coefgenes×Expgenes[1]

here, Coef_genes denotes LASSO coefficient of target gene, while Exp_genes denotes expression level of target gene.

Based on the median of RS values, patients were classified into a high-risk subgroup (above the median risk score) and a low-risk subgroup (below the median risk score). In survival package (version 2.41-1), Kaplan-Meier (KM) survival analysis, along with the log-rank test, was used to analyze the survival of patients in the two risk subgroups. The receiver operating characteristic (ROC) curve was applied to assess the predictive performance of the risk score model. Additionally, in the training set, samples were divided into recurrent and non-recurrent groups according to the recurrence status. The expression of m6A/m1A/m5C-related DEGs that were significantly associated with prognosis was then compared between the two groups using the Kruskal-Wallis test.

Nomogram construction

Using the data from the training set, clinical factors were subjected to univariate and multivariate Cox regression analyses to identify independent prognostic factors (log-rank P value <0.05) using the survival package in R. By combining the independent prognostic factors with the RS, a nomogram was built to predict 1-, 3- and 5-year survival probabilities using rms package (version 5.1-2) in R. Additionally, the Schoenfeld residuals test was conducted to verify the proportional hazards assumption of the nomogram. To comprehensively validate the nomogram, we employed multiple complementary strategies: (I) to evaluate prognostic performance of the nomogram, the concordance index (C-index) was calculated using the survcomp package (1.34.0) in R. A C-index score around 0.7 suggests good model performance.; (II) calibration curves were plotted to compare the predicted and actual overall survival (OS); (III) to evaluate the clinical utility of the nomogram, decision curve analysis (DCA) (30) was performed for each factor by calculating the net benefit using the rmda package (version 1.6) in R.

Comparative analysis of immune landscapes in different risk subgroups

It has become increasingly clear that a significantly altered immune environment in the bone marrow is a crucial factor in the progression of AML (31). The application of ESTIMATE and CIBERSORT is based on the evaluation of immune infiltration at expression levels, with the aim of conducting correlation analysis from immune levels, which is a common immune related approach (32). To characterize the TME features of different risk subgroups, the fractions of tumor infiltrating immune cells (TIICs) were compared between the risk subgroups using the CIBERSORT software (33). The Stromal Score, Immune Score, ESTIMATE Score, and Tumor Purity (34) were calculated using the estimate package in R. Tumor purity, which represents the proportion of cancer cells in tumor tissue, is regarded as a prognostic indicator for immunotherapy (34). The Stromal Score, Immune Score and ESTIMATE Score are related to tumor purity.

Immunotherapy response prediction and KEGG pathway enrichment analysis

The Tumor Immune Dysfunction and Exclusion (TIDE) score (35) was calculated for each sample in the training set using the TIDE algorithm to predict probability of response to immunotherapy (i.e., distinguish predicted responders from predicted non-responders). The distributions of the risk score in responders and non-responders within the high-risk and low-risk subgroups were analyzed using the Kruskal-Wallis test. Predicted responders and predicted non-responders were compared in terms of risk scores using Fisher’s Exact test. Besides, the expression of m6A/m1A/m5C-related DEGs significantly associated with prognosis was compared between these two groups using the Kruskal-Wallis test. In the survival package (version 2.41-1), KM survival analysis together with log-rank test, were also used to analyze the survival of patients in the two subgroups.

Based on the gene-profiling data of the training set, the Gene Set Enrichment Analysis (GSEA) (36) algorithm was employed to determine the significant KEGG pathways enriched in the identified high-risk and low-risk subgroups (P value <0.05). A higher normalized Enrichment score (NES) score and a smaller P value suggested stronger significance.

Results

Identification and PPI network analysis of 25 m6A/m5C/m1A-related DEGs

As detailed in the Method section, a total of 25 m6A genes, 14 m5C genes, and 13 m1A genes were screened (Table S2), and there were 90 dead samples and 89 alive samples in the training set. Using FDR <0.05 as the significance threshold, a total of 25 m6A/m5C/m1A-related DEGs were identified between the alive and dead samples (Figure 1A). The correlation heatmap (Figure 1B) showed the expression relationships among the 25 m6A/m5C/m1A-related DEGs. Most genes were positively correlated, while TRNA Methyltransferase 6 (TRMT6), DNA methyltransferase 3 beta (DNMT3B), Insulin Like Growth Factor 2 MRNA Binding Protein 2 (IGF2BP2), and fat mass and obesity-associated protein (FTO) were negatively correlated with the majority of other genes. These opposing patterns suggest potential antagonistic regulatory relationships among key RNA methylation enzymes in pediatric AML.

Figure 1 Differentially expressed m6A/m5C/m1A genes. (A) Expression levels of 25 m6A/m5C/m1A genes in alive and dead samples. (B) Correlations between expression levels of m6A/m5C/m1A genes. Color gradient corresponds to Pearson correlation coefficient ranging from 0 (dark blue) to 1 (dark red). m1A, N1-methyl-adenosine; m5C, 5-methylcytosine; m6A, N6-methyladenosine.

Subsequently, through the STRING database, we retrieved 124 PPIs of the proteins (with a combined score >0.4) encoded by the 25 identified DEGs and constructed a PPI network. As depicted in Figure 2A, DNMT3B exhibited a higher combined score and more interactions, suggesting it may function as a hub gene among these regulators. This result represents an in silico prediction and requires further experimental validation. Thereafter, the genes in the PPI networks were significantly enriched in 72 GO terms of biological processes, including methylation, RNA methylation, regulation of mRNA metabolic process, mRNA modification, mRNA methylation, and mRNA destabilization (Figure 2B).

Figure 2 A PPI network analysis. PPI network (A) and GO function annotations (B). The network includes 25 nodes (genes) and 124 links (interactions). Node size stands for combined score. Gene count stands for the number of genes significantly enriched in a GO term. FDR, false discovery rate; GO, Gene Ontology; PPI, protein-protein interaction.

Establishment of an m6A/m5C/m1A-related RS model for predicting pediatric AML prognosis

To confirm the prognostic relevance of the 25 DEGs identified via binary “alive vs dead” comparison, we further performed univariate Cox regression analysis (accounting for follow-up time and censoring status). Among these, 22 genes with P<0.05 were retained for prognostic screening. Then, these 22 genes were subjected to LASSO Cox regression with 10‑fold cross‑validation. After determining the optimal lambda value (Figure 3A,3B),8 survival-related genes with non‑zero regression coefficients were selected as optimal prognostic genes. These genes included Zinc Finger CCCH-Type Containing 13 (ZC3H13), Methyltransferase 16 (METTL16), tet oncogene family member 2 (TET2), TRNA Methyltransferase (TRMT)10C, NOP2/Sun RNA Methyltransferase 2 (NSUN2), IGF2BP2TRMT6 and DNMT3B. Based on the LASSO regression coefficients and the expression levels of these 8 genes, a RS model was calculated for each sample following the formula below: RS = −0.002541076 * Exp _ZC3H13 + 0.001146952 * Exp _METTL16 + 0.001241913 * Exp _TET2 + 0.02855901 * Exp _TRMT10C + 0.031737541 * Exp _NSUN2 + 0.032462861 * Exp _IGF2BP2 + 0.048860116 *Exp _TRMT6 + 0.119011606 * Exp _DNMT3B.

Figure 3 Prognostic performance of a risk score model. (A) Coefficients of m6A/m5C/m1A genes according to the log lambda value. (B) The log lambda value corresponding to mean-squared error. (C) KM overall survival curve (left), risk score distribution (upper-middle), survival time of patients (lower-middle), and ROC curve for TARGET dataset (right). (D) KM survival curve (left), risk score distribution (upper-middle), survival time of patients (lower-middle), and ROC curve for E-MTAB-1205 dataset (right). AUC, area under the curve; CI, confidence interval; HR, hazard ratio; KM, Kaplan-Meier; m1A, N1-methyl-adenosine; m5C, 5-methylcytosine; m6A, N6-methyladenosine; ROC, receiver operating characteristic.

The expression levels of the eight signature m6A/m5C/m1A-related genes in both the training and validation sets are provided in Table S3 and Table S4, respectively. In the training set (TARGET), patients were dichotomized into a high-risk subgroup and a low-risk subgroup based on the median RS values. As shown in Figure 3C and Figure S1, KM survival curves indicated that the low-risk patients had significantly higher overall survival and recurrence-free survival rates compared to the high-risk patients. The area under the curve (AUC) value was 0.861(0.820, 0.778), suggesting that the risk-score model had good predictive performance in the training set. Similarly, in the E-MTAB-1205 dataset, used as the validation set, patients were divided into two risk subgroups with significantly different survival rates, and the AUC value was 0.774 (0.625, 0.833) (Figure 3D). These results validate the accuracy of the model in predicting survival of pediatric AML patients.

Furthermore, we stratified the samples according to recurrence status and assessed the expression levels of eight survival-related genes in the recurrent and non-recurrent groups. It was found that the expression levels of ZC3H13, METTL16, TET2, TRMT10C, NSUN2, TRMT6, and DNMT3B were significantly elevated in the recurrent group compared to the non-recurrent group (Figure 4). However, the expression of IGF2BP2 did not differ significantly between the two groups (Figure 4).

Figure 4 The levels of the 8 m6A/m5C/m1A-related DEGs between the recurrent and non-recurrent groups. NS, not significant; *, P<0.05, **, P<0.01; ***, P<0.001. DEGs, differentially expressed genes; m1A, N1-methyl-adenosine; m5C, 5-methylcytosine; m6A, N6-methyladenosine.

Establishment of a nomogram model based on risk score, WT1 mutation and tumor recurrence

FLT3/ITD and Wilms’ tumor-1 (WT1) mutations are well‑known clinical prognostic factors in pediatric AML, and included in the analysis to screen independent prognostic factors for constructing an accurate nomogram. In the training set, FLT3/ITD mutation, WT1 mutation, recurrence and risk score were significantly correlated with OS time in uni-variate Cox regression analysis (Table 1). Furthermore, multivariate analysis confirmed that WT1 mutation (Figure 5A), recurrence (Figure 5B) and risk score were independent prognostic factors (Table 1), which were further integrated into a nomogram to improve prognostic prediction accuracy.

Figure 5 Nomogram model. (A) KM overall survival curves for patients with or without WT1 mutation. (B) KM overall survival curves for patients with or without recurrence. (C) Nomogram is constructed based on a combination of risk score, WT1 mutation, and tumor recurrence. (D) Calibration curves. (E) Decision curve analysis of the nomogram. “All” = all patients assumed to receive intervention; “None” = no patient receives intervention; “Full” = integrated nomogram combining risk score, WT1 mutation, and recurrence. The full model curve is above the reference lines, indicating good clinical utility. (F) Clinical impact curve. CI, confidence interval; HR, hazard ratio; KM, Kaplan-Meier; OS, overall survival.

Combining the risk score, WT1 mutation and tumor recurrence, a composite nomogram was built to predict 1-, 3- and 5-year survival probabilities (Figure 5C). Calibration curves showed that 1-, 3- and 5-year C-index was 0.7133, 0.7214, and 0.7173, respectively (Figure 5D), indicating good discriminative ability of the nomogram. The proportional hazards assumptions underlying the nomogram model were verified through Schoenfeld residual tests (Figure S2). DCA analysis demonstrated that the combination of all three factors had better clinical value than any single factor alone (Figure 5E). In the clinical impact curve, the predicted number of high-risk patients was in a good agreement with the actual number of high-risk patients (Figure 5F).

TME characteristics and immunotherapy responses in high-risk and low-risk patients

To characterize the signature‑associated immune landscape, patients were stratified by the 8‑gene risk score into high‑ and low‑risk subgroups; as well as TME scores were calculated to reflect tumor purity and immune infiltration status. Of note, transcriptome‑based analysis can only estimate immune cell infiltration proportions, but cannot define distinct immune infiltration patterns. Therefore, we focused on exploring the correlations between RNA methylation‑related genes and immune cell fractions as well as TME scores. Then, the proportions of 22 tumor‑infiltrating immune cells and TME scores were compared between the two risk groups (signature‑level analysis, Figure 6A,6B). Compared to the low-risk patients, the high-risk patients had significantly higher contents of naïve B cells, plasma B cells, resting memory CD4+ T cells, and activated NK cells, but lower contents of monocytes, M2 macrophages, and resting myeloid dendritic cells (Figure 6A). Low-risk patients exhibited significant increases in the Stromal Score, Immune Score, and ESTIMATE Score, together with a significant reduction in tumor purity in comparison with the high-risk patients (Figure 6B).

Figure 6 Characteristics of immune microenvironment and immunotherapy response for high-risk and low-risk subgroups. (A) Proportions of tumor infiltrating immune cells in high-risk and low-risk subgroups. (B) Stromal Score, Immune Score, ESTIMATE Score, and Tumor Purity in high-risk and low-risk subgroups. *, P<0.05; **, P<0.01; ***, P<0.001. (C) Correlations of 8 prognostic genes with 7 tumor-infiltrating immune cells, Stromal Score, Immune Score, ESTIMATE Score, and Tumor Purity. Color gradient represents the Spearman correlation coefficient (r), ranging from −0.4 (dark blue, negative correlation) to 0.4 (dark red, positive correlation). (D) Ratios of predicted responders (yellow) and predicted non-responders (blue) in high-risk and low-risk subgroups. (E) Comparison of predicted responders and predicted non-responders for risk score.

In addition, gene‑level correlations between individual signature genes and immune features were explored to dissect potential regulatory mechanisms (Figure 6C). Notably, ZC3H13, NSUN2, METTL16, TRMT10C and DNMT3B were positively correlated with tumor purity, B cell naive and T cell CD4+ memory resting; but negatively correlated with monocyte, macrophage M2 and myeloid dendritic cell resting (Figure 6C). In contrast, TRMT6 was negatively correlated with tumor purity (Figure 6C). These results suggest that RNA methylation genes may regulate the immune microenvironment and tumor purity in pediatric AML. Notably, the primary immune analysis was performed at the signature level (RS grouping), while gene-level correlations served as secondary mechanistic exploration.

Immunotherapy response prediction and KEGG pathway enrichment analysis

By applying the TIDE online algorithm to predict response to immune-checkpoint blockade therapy, patients in the training set were divided into responders and non-responders according to the TIDE score. The low-risk subgroup had 69 predicted responders and 20 predicted non-responders, while the high-risk subgroup included 60 predicted responders and 30 predicted non-responders. The distribution of predicted responders and predicted non-responders did not differ significantly between the two subgroups (Figure 6D). However, predicted non‑responders exhibited significantly higher risk scores than predicted responders in the training set (Figure 6E), consistent with the expected association between higher risk scores, worse prognosis, and poorer immunotherapy outcomes.

Additionally, in the training cohort, the expression levels of TRMT6, TET2, and DNMT3B were significantly elevated in predicted responders compared to predicted non-responders (Figure 7A). However, the expression levels of ZC3H13, IGF2BP2, NSUN2, METTL16, and TRMT10C did not show significant differences between the two groups (Figure 7A). KM survival curves showed a trend toward better overall survival and recurrence-free survival in predicted responders than in predicted non-responders, although these differences did not reach statistical significance (Figure 7B).

Figure 7 Immunotherapy response prediction. (A) The levels of the 8 m6A/m5C/m1A-related DEGs between predicted responders and predicted non-responders in the training set. NS, not significant; ***, P<0.001. (B) KM survival curves (including overall survival and recurrence free survival for predicted responders and predicted non-responders.

Finally, KEGG enrichment analysis (Table 2) showed 11 KEGG signaling pathways were significantly associated with the risk subgroups of the training set, including the WNT signaling pathway, RNA degradation, cell cycle, mismatch repair, and p53 signaling pathway. These pathways are closely involved in leukemic cell proliferation, survival, and RNA metabolism, implying that the m6A/m5C/m1A signature may affect pediatric AML progression through these biological processes.

Discussion

This study focused specifically on pediatric AML, addressing its unique biological and clinical features. Despite advances in treatment, pediatric AML remains associated with poor prognosis, necessitating improved risk stratification and personalized strategies (37). As key epigenetic regulators, m6A, m5C, and m1A RNA methylation are implicated in cancer progression (38), but most studies are limited to adult AML or single modification types. Here, our study exclusively focused on pediatric AML and systematically analyzed m6A, m1A, and m5C RNA modification regulators together. Specifically, we identified 25 m6A/m5C/m1A-related DEGs between the dead and alive patients, and then constructed an 8-gene prognostic siganture via LASSO regression. This signature model could effectively distinguish the high-risk patients from low-risk patients in both training and validation cohorts. Furthermore, a nomogram incorprating independent prognostic clinical factors and the RS was developed to predict survival probability. The sights gained from this study advance our knowledge on functional roles and molecular mechanisms of m6A/m1A/m5C genes in pediatric AML and provide a promising prognostic tool.

In this study, we adopted a two-step selection workflow: initial screening of alive/dead-related DEGs followed by univariate Cox validation, ensuring genes were both differentially expressed and prognostically relevant. Pediatric AML is distinct from adult AML and solid tumors, with a survival rate of approximately 70% but up to 30% relapse, as well as chemotherapy resistance is a core bottleneck for treatment failure (39). Existing RNA methylation signatures are largely adult-derived or focus on a single modification (e.g., ovarian cancer, oral squamous cell carcinoma) (25,40), limiting their applicability to pediatric AML. Our 8-gene signature (ZC3H13, METTL16, TET2, TRMT10C, NSUN2, IGF2BP2, TRMT6, and DNMT3B) was derived exclusively from pediatric patients and correlated with recurrence, aligning with clinical needs.

Existing RNA methylation signatures in other cancers mostly focus on a single modification (e.g., only m6A or m5C) or universal regulatory factors (e.g., YTHDF3, ALKBH5) (41,42). In contrast, this study is the first to integrate three core RNA methylation types (m6A/m1A/m5C). While TET2 and DNMT3B are known in adult AML, their roles within this combined signature remain undefined in children. Here, both emerged as key prognostic genes linked to tumor purity and immune infiltration. The remaining six genes (ZC3H13, METTL16, TRMT10C, NSUN2, IGF2BP2, TRMT6) are rarely reported in pediatric AML, underscoring the signature’s novelty. The eight genes span all three methylation systems, with no single type dominating. Their prognostic power derives from coordinated crosstalk, which jointly regulates RNA metabolism, cell cycle, DNA repair, and immune landscapes. These complementary, non-redundant effects support the superiority of a multi‑modification signature over single‑modification models.

TET2 is able to regulate DNA expression by facilitating the conversion of 5mC into 5-hydroxymethylcytosine (5-hmC) (43), and is frequently downregulated in AML, linking to adverse prognosis (44,45). DNMT3B, an m5C writer implicated in leukemogenesis (46), has been reported to be a potential prognostic gene in pediatric AML (47). The current study showed that DNMT3B had a high combined score and numerous interactions in the PPI network, indicating it may act as a hub gene among m6A/m5C/m1A regulators. Notably, this finding is an inferred bioinformatic result and lacks direct functional validation, which warrants further experimental investigation. The m6A reader IGF2BP2 acts as an oncogene in AML. associated with poor prognosis and metabolic dysregulation (48,49). Weng et al. discovered that high expression of IGF2BP2 in AML is associated with an unfavorable prognosis and plays a significant role in amino acid metabolism via its m6A modification, suggesting that targeting IGF2BP2 may represent a promising therapeutic strategy for AML (50). The remaining five genes (e.g. ZC3H13, TRMT6) remain elusive in AML. TRMT6 catalyzes m1A modification to regulate tRNA stability and translation, supporting tumor proliferation (51). Although direct pediatric AML evidence is limited, related m1A methyltransferases drive MYC and PD-L1 expression in other cancers (52). Combined with its overexpression in relapsed pediatric AML and poor prognosis, TRMT6 may promote disease progression via m1A-dependent immune regulation. Similarly, ZC3H13 stabilizes the m6A writer complex and regulates tumor progression via m6A (53). While its role in AML is unreported, METTL3 drives leukemogenesis through the mdm2/p53 pathway (54). ZC3H13 overexpression in relapsed pediatric AML suggests it may promote disease by enhancing METTL3-mediated m6A and disrupting cell cycle progression. Negative correlations of IGF2BP2 and FTO with most other RNA methylation regulators suggest possible antagonistic crosstalk in pediatric AML. Taken together, our study suggests these eight genes reflect pediatric AML-specific molecular features and serve as promising prognostic biomarkers in pediatric AML.

The 8‑gene signature performed better in the training set (AUC =0.861) than in the validation set (AUC =0.774), likely due to the smaller validation cohort (50 vs. 179 samples), heterogeneity (e.g., differences in clinical annotation, treatment strategies), and technical variation. Despite this, the validation set still demonstrated significant survival stratification (P=2.677e−02) and a clinically meaningful AUC (0.774), confirming the model’s core prognostic reproducibility, an acceptable attenuation for a preliminary hypothesis-generating study in rare pediatric AML. Unlike single RNA methylation-based models in other cancers (55,56), our integrated nomogram combines the 8-gene signature with two pediatric AML-specific clinical indicators (WT1 mutation, tumor recurrence) and achieved a C-index of 0.713–0.721 for 1–5-year survival prediction, with calibration/clinical impact curves supporting its preliminary predictive accuracy and research utility (clinical applicability pending validation) (57). This integrated model provides valuable risk stratification insights for pediatric AML and outperforms single‑marker models, though larger prospective cohorts are needed to validate its clinical utility.

TIICs are key components of the bone marrow microenvironment, shaping AML progression and treatment response (58). Pediatric AML exhibits a distinct immune landscape (e.g., monocyte infiltration, tumor purity) compared to adult AML and solid tumors (59). In the present study, low-risk patients had higher monocyte and dendritic cell infiltration, lower B/CD4+ T cell levels, and reduced tumor purity. Noticeably, monocytes had a remarkably higher proportion than other immune cell subsets. Monocytes are innate immune cells and have plasticity to perform pro- and anti-tumoral immune functions, contributing to the overall orchestration of immunity (60). It is speculated that higher level of monocytes may suppress tumor progression, thereby improving prognosis of pediatric patients with AML. Furthermore, immunotherapy has been considered as an effective treatment option for AML (61). Our study showed that the risk score of pediatric AML patients was significantly associated with predicted immunotherapy response (predicted responders had lower risk scores, P=0.025). These observations suggest that therapy resistance may contribute to the poor prognosis in high-risk patients, which is in concordance with a previous report (62). In our study, ZC3H13, NSUN2, METTL16, TRMT10C and DNMT3B exhibited stronger positive correlations with tumor purity, whereas TRMT6 exhibited stronger negative correlation with tumor purity. These findings suggest a link between the 8‑gene signature and immune checkpoint sensitivity, offering preliminary immunotherapy targets for pediatric AML and unique insights rarely addressed by other RNA methylation signatures.

In addition, ZC3H13, NSUN2, TET2, METTL16, TRMT10C, and DNMT3B were negatively correlated with monocytes, M2 macrophages, and resting myeloid dendritic cells, but positively correlated with naive B cells and resting CD4+ memory T cells. Consistent with these findings, ZC3H13 is linked to CD8+ T cell infiltration and immune activation (63); NSUN2 modulates immune cell infiltration (activated B cells, activated CD4+ T cells, and activated CD8+ T cells) and checkpoint gene expression (64,65); TET proteins, particularly Tet2, regulate B and T cell development and functions (66); and METTL16 are positively correlated with naive B cells and CD8+ T cells, while negatively correlated with M0 macrophages in pancreatic ductal adenocarcinoma (67). Collectively, these data suggest that m6A/m5C/m1A-related genes participate in immune regulation, though further experimental validation is needed. Notably, immunotherapy-related analyses remain exploratory. No definitive conclusions can be drawn, given non-significant survival differences between predicted responders and non-responders, and most mechanistic evidence is extrapolated from solid tumors, requiring further confirmation in pediatric AML.

Notably, all findings in this study are interpreted within the unique context of pediatric AML biology, ensuring that our conclusions are disease-specific rather than generic across cancer types. Our study also unraveled that the two risk subgroups were functionally associated with P53 and Wnt signaling pathways. P53 is one of the most well-defined tumor suppressor pathways in cancers including AML (68). the activation of Wnt signaling pathway plays an essential role in hematopoietic development and maintenance of leukemic stem cells (69,70). Collectively, these pathway (P53 and Wnt signaling pathways) associations further support the biological validity of our signature and highlight its potential to reflect core oncogenic programs driving pediatric AML progression.

However, our study has several limitations that need to be acknowledged. First, this study relies solely on public RNA-seq datasets without experimental validation or primary patient samples. Second, the validation cohort is small, limiting generalizability. Third, the TIDE algorithm used for immunotherapy prediction was developed in adult solid tumors and may be inaccurate in pediatric AML, which has a unique immune landscape and immature immunity. Fourth, blast percentage and residual normal hematopoiesis may confound TME and gene expression analyses, though our signature’s prognostic robustness mitigates this issue. Fifth, the initial DEG screening did not account for follow-up time, censoring and survival duration heterogeneity, as well as our analytical framework cannot fully distinguish RNA methylation from DNA epigenetic effects. Sixth, we did not explore subtype-specific associations or perform sensitivity analyses, an evaluation needed to confirm the signature’s robustness and independent prognostic value. Finally, our 8-gene signature was developed and validated exclusively in pediatric AML cohorts, so its prognostic performance in adult AML remains unclear. Additionally, although our study primarily focused on constructing an RNA methylation-based prognostic signature for pediatric AML, several genes in the signature, such as DNMT3B and TET2, are well-established AML-associated oncogenic/epigenetic factors. Due to the limitations of the public datasets used, we did not systematically analyze the correlation between the 8-gene signature and specific pediatric AML oncogenes or molecular subtypes. Future work should include in vitro/vivo functional assays, single-cell sequencing, and large prospective cohorts to validate our findings.

Conclusions

Our study elucidates the prognostic implications of m6A/m1A/m5C genes in pediatric AML and establishes a survival-prediction model using a signature of eight m6A/m1A/m5C genes (ZC3H13, METTL16, TET2, TRMT10C, NSUN2, IGF2BP2, TRMT6 and DNMT3B). This model could distinguish high-risk from low-risk patients in the training and independent validation cohorts, highlighting the preliminary prognostic value of the 8-gene signature tailored for pediatric AML. These prognostic genes are closely linked to the TME and hold potential as preliminary candidates for targeted immunotherapy research, though definitive validation is required. What sets our work apart from previous studies is that they mainly focused on adult cancers or combined different age groups; in contrast, our model captures the unique immunological and molecular features of pediatric AML. It provides a preliminary prognostic tool that offers insights for future risk stratification research in this susceptible population, though its clinical applicability and generalizability must be confirmed by larger, multi-center prospective studies.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0367/rc

Peer Review File: Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0367/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0367/coif). A.F.C. serves as an Editor-in-Chief of Translational Pediatrics from June 2025 to May 2027. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Pelcovits A, Niroula R. Acute Myeloid Leukemia: A Review. R I Med J (2013) 2020;103:38-40.
2. Pollyea DA, Bixby D, Perl A, et al. NCCN Guidelines Insights: Acute Myeloid Leukemia, Version 2.2021. J Natl Compr Canc Netw 2021;19:16-27. [Crossref] [PubMed]
3. Quessada J, Cuccuini W, Saultier P, et al. Cytogenetics of Pediatric Acute Myeloid Leukemia: A Review of the Current Knowledge. Genes (Basel) 2021;12:924. [Crossref] [PubMed]
4. Rautenberg C, Germing U, Haas R, et al. Relapse of Acute Myeloid Leukemia after Allogeneic Stem Cell Transplantation: Prevention, Detection, and Treatment. Int J Mol Sci 2019;20:228. [Crossref] [PubMed]
5. Tabata R, Chi S, Yuda J, et al. Emerging Immunotherapy for Acute Myeloid Leukemia. Int J Mol Sci 2021;22:1944. [Crossref] [PubMed]
6. Zhang L, Lu Q, Chang C. Epigenetics in Health and Disease. Adv Exp Med Biol 2020;1253:3-55. [Crossref] [PubMed]
7. Wang T, Kong S, Tao M, et al. The potential role of RNA N6-methyladenosine in Cancer progression. Mol Cancer 2020;19:88. [Crossref] [PubMed]
8. Zhong F, Yao F, Cheng Y, et al. m6A-related lncRNAs predict prognosis and indicate immune microenvironment in acute myeloid leukemia. Sci Rep 2022;12:1759. [Crossref] [PubMed]
9. Yankova E, Aspris D, Tzelepis K. The N6-methyladenosine RNA modification in acute myeloid leukemia. Curr Opin Hematol 2021;28:80-5. [Crossref] [PubMed]
10. Yuan S, Cong Z, Ji J, et al. Analysis of m6A-related signatures associated with the tumor immune microenvironment and predict survival in acute myeloid leukemia. Ann Transl Med 2022;10:902. [Crossref] [PubMed]
11. Zhang Q, Liu F, Chen W, et al. The role of RNA m5C modification in cancer metastasis. Int J Biol Sci 2021;17:3369-80. [Crossref] [PubMed]
12. Xu J, Liu X, Chen Y, et al. RNA 5-Methylcytosine Regulators Contribute to Metabolism Heterogeneity and Predict Prognosis in Ovarian Cancer. Front Cell Dev Biol 2022;10:807786. [Crossref] [PubMed]
13. Wang L, Gao S. Identification of 5-methylcytosine-related signature for predicting prognosis in ovarian cancer. Biol Res 2021;54:18. [Crossref] [PubMed]
14. Liu J, Chen C, Wang Y, et al. Comprehensive of N1-Methyladenosine Modifications Patterns and Immunological Characteristics in Ovarian Cancer. Front Immunol 2021;12:746647. [Crossref] [PubMed]
15. Gao L, Chen R, Sugimoto M, et al. The Impact of m1A Methylation Modification Patterns on Tumor Immune Microenvironment and Prognosis in Oral Squamous Cell Carcinoma. Int J Mol Sci 2021;22:10302. [Crossref] [PubMed]
16. He W, Cong Z, Niu C, et al. A prognostic signature based on genes associated with m6A/m5C/m1A/m7G modifications and its immunological characteristics in clear cell renal cell carcinoma. Sci Rep 2024;14:18708. [Crossref] [PubMed]
17. Liu HT, Rau CS, Liu YW, et al. Deciphering the Divergent Gene Expression Landscapes of m6A/m5C/m1A Methylation Regulators in Hepatocellular Carcinoma Through Single-Cell and Bulk RNA Transcriptomic Analysis. J Hepatocell Carcinoma 2023;10:2383-95. [Crossref] [PubMed]
18. Jones LM, Tarlock K, Cooper T. Targeted Therapy in Pediatric AML: An Evolving Landscape. Paediatr Drugs 2021;23:485-97. [Crossref] [PubMed]
19. Mevatee P, Tantiworawit A, Traisathit P, et al. FLT3-ITD, NPM1, and DNMT3A Gene Mutations and Risk Factors in Normal Karyotype Acute Myeloid Leukemia and Myelodysplastic Syndrome Patients in Upper Northern Thailand. Asian Pac J Cancer Prev 2017;18:3031-9. [Crossref] [PubMed]
20. Conneely SE, Stevens AM. Acute Myeloid Leukemia in Children: Emerging Paradigms in Genetics and New Approaches to Therapy. Curr Oncol Rep 2021;23:16. [Crossref] [PubMed]
21. Beesley AH, Firth MJ, Anderson D, et al. Drug-gene modeling in pediatric T-cell acute lymphoblastic leukemia highlights importance of 6-mercaptopurine for outcome. Cancer Res 2013;73:2749-59. [Crossref] [PubMed]
22. Sun Y, Luo J, Qian C, et al. The Value of Nutritional Status in the Prognostic Analysis of Patients with AIDS-Related Lymphoma. Infect Drug Resist 2021;14:1105-13. [Crossref] [PubMed]
23. Yan LP, Liu ZB, Wu M, et al. Effect of lncRNA MALAT1 expression on survival status of elderly patients with severe pneumonia. Eur Rev Med Pharmacol Sci 2020;24:3959-64. [Crossref] [PubMed]
24. Wang E, Li Y, Ming R, et al. The Prognostic Value and Immune Landscapes of a m6A/m5C/m1A-Related LncRNAs Signature in Head and Neck Squamous Cell Carcinoma. Front Cell Dev Biol 2021;9:718974. [Crossref] [PubMed]
25. Xue C, Chu Q, Zheng Q, et al. Role of main RNA modifications in cancer: N(6)-methyladenosine, 5-methylcytosine, and pseudouridine. Signal Transduct Target Ther 2022;7:142. [Crossref] [PubMed]
26. Szklarczyk D, Gable AL, Nastou KC, et al. The STRING database in 2021: customizable protein-protein networks, and functional characterization of user-uploaded gene/measurement sets. Nucleic Acids Res 2021;49:D605-12. [Crossref] [PubMed]
27. Wang P, Wang Y, Hang B, et al. A novel gene expression-based prognostic scoring system to predict survival in gastric cancer. Oncotarget 2016;7:55343-51. [Crossref] [PubMed]
28. Goeman JJ. L1 penalized estimation in the Cox proportional hazards model. Biom J 2010;52:70-84. [Crossref] [PubMed]
29. Liu Z, Li M, Hua Q, et al. Identification of an eight-lncRNA prognostic model for breast cancer using WGCNA network analysis and a Cox‑proportional hazards model based on L1-penalized estimation. Int J Mol Med 2019;44:1333-43. [Crossref] [PubMed]
30. Lan Y, Pan C, Qiu X, et al. Nomogram for Persistent Post-Stroke Depression and Decision Curve Analysis. Clin Interv Aging 2022;17:393-403. [Crossref] [PubMed]
31. Vago L, Gojo I. Immune escape and immunotherapy of acute myeloid leukemia. J Clin Invest 2020;130:1552-64. [Crossref] [PubMed]
32. Gu W, Wu Y, Jiang R, et al. A 12-gene immune signature predicts prognosis and identifies KRT6B as a therapeutic target in lung adenocarcinoma. Front Immunol 2026;17:1693469. [Crossref] [PubMed]
33. Chen B, Khodadoust MS, Liu CL, et al. Profiling Tumor Infiltrating Immune Cells with CIBERSORT. Methods Mol Biol 2018;1711:243-59. [Crossref] [PubMed]
34. Deng Y, Song Z, Huang L, et al. Tumor purity as a prognosis and immunotherapy relevant feature in cervical cancer. Aging (Albany NY) 2021;13:24768-85. [Crossref] [PubMed]
35. Ding C, Shan Z, Li M, et al. Characterization of the fatty acid metabolism in colorectal cancer to guide clinical therapy. Mol Ther Oncolytics 2021;20:532-44. [Crossref] [PubMed]
36. Reimand J, Isserlin R, Voisin V, et al. Pathway enrichment analysis and visualization of omics data using g:Profiler, GSEA, Cytoscape and EnrichmentMap. Nat Protoc 2019;14:482-517. [Crossref] [PubMed]
37. Elgarten CW, Aplenc R. Pediatric acute myeloid leukemia: updates on biology, risk stratification, and therapy. Curr Opin Pediatr 2020;32:57-66. [Crossref] [PubMed]
38. Xie S, Chen W, Chen K, et al. Emerging roles of RNA methylation in gastrointestinal cancers. Cancer Cell Int 2020;20:585. [Crossref] [PubMed]
39. Rasche M, Zimmermann M, Steidel E, et al. Survival Following Relapse in Children with Acute Myeloid Leukemia: A Report from AML-BFM and COG. Cancers (Basel) 2021;13:2336. [Crossref] [PubMed]
40. Wu X, Tang J, Cheng B. Oral squamous cell carcinoma gene patterns connected with RNA methylation for prognostic prediction. Oral Dis 2024;30:408-21. [Crossref] [PubMed]
41. Han Y, Sun J, Yao M, et al. Biological roles of enhancer RNA m6A modification and its implications in cancer. Cell Commun Signal 2025;23:254. [Crossref] [PubMed]
42. Wen L, Fu J, Wang Z, et al. Regulatory mechanisms of m6A RNA methylation in esophageal cancer: a comprehensive review. Front Genet 2025;16:1561799. [Crossref] [PubMed]
43. Cong B, Zhang Q, Cao X. The function and regulation of TET2 in innate immunity and inflammation. Protein Cell 2021;12:165-73. [Crossref] [PubMed]
44. Zhang TJ, Zhou JD, Yang DQ, et al. TET2 expression is a potential prognostic and predictive biomarker in cytogenetically normal acute myeloid leukemia. J Cell Physiol 2018;233:5838-46. [Crossref] [PubMed]
45. Nabil R, Hassan NM, Abdellateif MS, et al. The prognostic role of C-KIT, TET1 and TET2 gene expression in Acute Myeloid Leukemia. Mol Biol Rep 2023;50:641-53. [Crossref] [PubMed]
46. Bi L, Zhou B, Li H, et al. A novel miR-375-HOXB3-CDCA3/DNMT3B regulatory circuitry contributes to leukemogenesis in acute myeloid leukemia. BMC Cancer 2018;18:182. [Crossref] [PubMed]
47. Liu Z, Elcheva I. A six-gene prognostic signature for both adult and pediatric acute myeloid leukemia identified with machine learning. Am J Transl Res 2022;14:6210-21.
48. He X, Li W, Liang X, et al. IGF2BP2 Overexpression Indicates Poor Survival in Patients with Acute Myelocytic Leukemia. Cell Physiol Biochem 2018;51:1945-56. [Crossref] [PubMed]
49. Liu C, Ma Y, Wang R, et al. LINC00987 knockdown inhibits the progression of acute myeloid leukemia by suppressing IGF2BP2-mediated PA2G4 expression. Anticancer Drugs 2022;33:e207-17. [Crossref] [PubMed]
50. Weng H, Huang F, Yu Z, et al. The m6A reader IGF2BP2 regulates glutamine metabolism and represents a therapeutic target in acute myeloid leukemia. Cancer Cell 2022;40:1566-1582.e10. [Crossref] [PubMed]
51. Wang B, Niu L, Wang Z, et al. RNA m1A Methyltransferase TRMT6 Predicts Poorer Prognosis and Promotes Malignant Behavior in Glioma. Front Mol Biosci 2021;8:692130. [Crossref] [PubMed]
52. Li S, Feng T, Liu Y, et al. m1A inhibition fuels oncolytic virus-elicited antitumor immunity via downregulating MYC/PD-L1 signaling. Int J Oral Sci 2024;16:36. [Crossref] [PubMed]
53. Qu Y, Gao N, Zhang S, et al. Role of N6-methyladenosine RNA modification in cancer. MedComm (2020) 2024;5:e715.
54. Sang L, Wu X, Yan T, et al. The m(6)A RNA methyltransferase METTL3/METTL14 promotes leukemogenesis through the mdm2/p53 pathway in acute myeloid leukemia. J Cancer 2022;13:1019-30. [Crossref] [PubMed]
55. Yin J, Fu J, Xu J, et al. Integrated analysis of m6A regulator-mediated RNA methylation modification patterns and immune characteristics in Sjögren's syndrome. Heliyon 2024;10:e28645. [Crossref] [PubMed]
56. Ding Y, Bajpai AK, Wu F, et al. 5-methylcytosine RNA modification regulators-based patterns and features of immune microenvironment in acute myeloid leukemia. Aging (Albany NY) 2024;16:2340-61. [Crossref] [PubMed]
57. Jiang Q, Hu H, Liao J, et al. Development and validation of a nomogram for breast cancer-related lymphedema. Sci Rep 2024;14:15602. [Crossref] [PubMed]
58. Menter T, Tzankov A. Tumor Microenvironment in Acute Myeloid Leukemia: Adjusting Niches. Front Immunol 2022;13:811144. [Crossref] [PubMed]
59. Perzolli A, Koedijk JB, Zwaan CM, et al. Targeting the innate immune system in pediatric and adult AML. Leukemia 2024;38:1191-201. [Crossref] [PubMed]
60. Olingy CE, Dinh HQ, Hedrick CC. Monocyte heterogeneity and functions in cancer. J Leukoc Biol 2019;106:309-22. [Crossref] [PubMed]
61. Aureli A, Marziani B, Sconocchia T, et al. Immunotherapy as a Turning Point in the Treatment of Acute Myeloid Leukemia. Cancers (Basel) 2021;13:6246. [Crossref] [PubMed]
62. Witkowski MT, Lasry A, Carroll WL, et al. Immune-Based Therapies in Acute Leukemia. Trends Cancer 2019;5:604-18. [Crossref] [PubMed]
63. Xu Q, Xu H, Deng R, et al. Landscape of Prognostic m6A RNA Methylation Regulators in Hepatocellular Carcinoma to Aid Immunotherapy. Front Cell Dev Biol 2021;9:669145. [Crossref] [PubMed]
64. Santin AD, Bellone S, Buza N, et al. Regression of Chemotherapy-Resistant Polymerase ε (POLE) Ultra-Mutated and MSH6 Hyper-Mutated Endometrial Tumors with Nivolumab. Clin Cancer Res 2016;22:5682-7. [Crossref] [PubMed]
65. Sun G, Ma S, Zheng Z, et al. Multi-omics analysis of expression and prognostic value of NSUN members in prostate cancer. Front Oncol 2022;12:965571. [Crossref] [PubMed]
66. Li J, Li L, Sun X, et al. Role of Tet2 in Regulating Adaptive and Innate Immunity. Front Cell Dev Biol 2021;9:665897. [Crossref] [PubMed]
67. Lu L, Zheng D, Qu J, et al. METTL16 predicts a favorable outcome and primes antitumor immunity in pancreatic ductal adenocarcinoma. Front Cell Dev Biol 2022;10:759020. [Crossref] [PubMed]
68. Liu X, Feng Y, Lai X, et al. Virgibacillus halodenitrificans ST-1 for fermentation of shrimp paste and hydrolysates of its protease. Food Sci Nutr 2020;8:5352-61. [Crossref] [PubMed]
69. Cardona-Echeverry A, Prada-Arismendy J. Deciphering the role of Wnt signaling in acute myeloid leukemia prognosis: how alterations in DNA methylation come into play in patients' prognosis. J Cancer Res Clin Oncol 2020;146:3097-109. [Crossref] [PubMed]
70. Gruszka AM, Valli D, Alcalay M. Wnt Signalling in Acute Myeloid Leukaemia. Cells 2019;8:1403. [Crossref] [PubMed]