Mitochondrial RNA helicase DDX28 promotes cell cycle and DNA repair programs and shapes an immunosuppressive microenvironment in acute myeloid leukemia Authors: Zi Wang, Yunli Liu, Xinzhu Guan, Miaoqing Ye, Xinxin Xu Published: 2026 ABSTRACT Highlights•Identifies DDX28 as a n prognostic biomarker linked to immune evasion in AML.•Reveals DDX28 remodels an immunosuppressive TIME via M2 macrophage and Treg enrichment.•Links DDX28-driven metabolic reprogramming to DNA repair and chemoresistance.•Maps DDX28 to malignant blasts and exhausted T cells using single-cell transcriptomics.•Demonstrates targeting DDX28 suppresses proliferation, suggesting a new therapeutic strategy. Introduction Acute myeloid leukemia (AML) stands as the predominant leukemia type in adults, marked by the rampant growth of immature myeloid cells in the bone marrow [1]. Over recent years, trends show a consistent uptick in both its incidence and associated mortality. This malignancy reveals striking heterogeneity, where affected individuals exhibit diverse genetic makeups and clinical profiles [1]. At its root, AML progression stems from recurring genetic changes that disrupt metabolic functions, signaling routes, and epigenetic oversight, alongside disruptions to apoptosis and cytokine signaling. Notably, somatic mutations occur in 97.3% of AML patients [2], correlating tightly with particular clinical and pathological attributes. Advances in targeted sequencing have unveiled critical prognostic mutations in genes including FLT3, KIT, CEBPA, and TP53 [3]. Consider, for instance, regimens incorporating venetoclax [4,5], as well as precision therapies directed at FLT3, IDH1/2, and BCL-2. That said, leukemic stem cells (LSCs) commonly drive early resistance or eventual relapse, presenting formidable barriers to durable remission [6]. In relapsed cases, AML further complicates matters through elaborate clonal and genomic evolutions, such as amplified DNA repair and metabolic shifts [7]. It follows, then, that deciphering the molecular drivers of relapsed AML remains vital for advancing more potent interventions. DDX28 is an ATP-driven RNA helicase belonging to the DEAD-box family that can influence cancer formation, inflammatory signaling and RNA metabolism [8]. DEAD-box family members participate in almost all RNA-associated processes, such as transcriptional regulation, protein synthesis, and mRNA decay, through ATP hydrolysis-driven remodeling of RNA and ribonucleoprotein (RNP) complexes [9]. In particular, DDX28 has been reported to facilitate hypoxic migration and support growth and invasion in glycolytically dependent 3D culture systems [10]. DDX28 is located in the mitochondria and interacts with 16S rRNA and assembly intermediates, facilitating the biogenesis of the mitochondrial large ribosomal subunit [9,11]. Furthermore, DDX28-dependent translation regulates mitochondrial oxidative phosphorylation (OXPHOS), which is related to AML. Hence, the potential involvement of DDX28 in AML pathogenesis requires further investigation. This study aimed to clarify the clinical relevance, epigenetic regulation and tumor-promoting functions of DDX28 in AML. We employed an integrated approach involving multi-database bioinformatics and in vitro experimental assays to investigate the underlying mechanisms. Materials and methods Dataset collection analysis Leukemia transcriptome dataset and clinical information were obtained from the TCGA database (TCGA-LAML project), including RNA-Seq gene expression profiles, patient survival data (overall survival, vital status), and baseline clinical characteristics (age, sex, ethnicity). We used the GEPIA2 database to evaluate DDX28 expression across multiple cancer types. Because TCGA-LAML generally lacks matched normal bone marrow or peripheral blood controls, tumor–normal expression comparisons used platform-integrated normal reference samples (for example, GTEx-based or GEPIA2/TIMER-defined normal references) rather than paired TCGA-LAML specimens. Patients were divided into high- and low-expression groups using the survival and survminer R tools based on the median expression value of DDX28. Survival durations and event outcomes were integrated to compute and visualize Kaplan-Meier curves, with statistical significance of inter-group differences determined by Log-rank testing. Using the timeROC program, a time-dependent ROC analysis was performed to determine the AUC for survival prediction at 1, 3, and 5 years. In order to evaluate relationships with patient survival, DDX28 was first subjected to univariate Cox proportional hazards regression. This was followed by multivariate Cox modeling that included clinical factors (age and sex) in addition to DDX28 expression, reflecting the availability of clinical variables in the analyzed cohort. Proportional hazards assumptions were assessed using Schoenfeld residual-based diagnostics. Using the limited cubic splines methodology, the association between DDX28 expression and overall survival was further investigated. Promoter methylation analysis of Illumina 450 K data was used to identify prognostic CpG sites, which were then displayed using lollipop plots. Analysis of epigenetic regulation and functional enrichment We employed the Limma package to identify differentially expressed genes (DEGs), contrasting cohorts stratified by high and low DDX28 expression. P-values were adjusted for multiple testing using the Benjamini–Hochberg procedure, and DEGs were defined as |log2 FC| > 1.5 and FDR < 0.05. Building on this, functional enrichment for Gene Ontology (GO) and KEGG pathways, along with Gene Set Enrichment Analysis (GSEA), proceeded through the clusterProfiler package, with significance reported using adjusted p-values or FDR as appropriate. For GSEA specifically, we deemed results significant based on |NES| > 1, FDR < 0.25, and adjusted p < 0.05. Quantification of the immune microenvironment We carried out immune infiltration analysis on the TCGA-LAML expression matrix through the IOBR package in R. Immune cell enrichment scores were estimated using ssGSEA implemented in IOBR with its built-in immune gene signatures, and associations with DDX28 expression were evaluated by Spearman correlation. Intercellular correlation heatmaps were generated using the corrplot package. To evaluate distinctions between the high and low DDX28 expression groups, we relied on Wilcoxon rank-sum tests. Analyses were performed within the same TCGA-LAML expression matrix; when platform-integrated normal references were used for visualization, the data source was explicitly labeled to avoid cross-platform effect-size misinterpretation. Assessment of drug sensitivity and immune evasion We applied the pRRophetic algorithm to estimate IC50 values for 198 compounds sourced from the GDSC2 database across TCGA specimens. When contrasting groups based on high and low DDX28 expression, we turned to the Wilcoxon test to assess drug sensitivity, adopting a rigorous threshold of p < 0.001. In addition, to measure aspects of immune evasion, we drew on the TIDE platform, which provided scores reflecting T cell dysfunction, T cell exclusion, and infiltration by cancer-associated fibroblasts (CAFs). Analysis of single-cell transcriptomes Seurat (v4.0) was used to integrate and handle public single-cell datasets for cell clustering annotation, dimensional reduction (UMAP), and normalization. Public AML scRNA-seq datasets were retrieved from GEO, and accession IDs, sample numbers, and key preprocessing parameters. Quality control filtering removed low-quality cells based on commonly used thresholds, including minimum detected genes per cell and maximum mitochondrial transcript fraction; doublets were excluded when applicable according to the original dataset processing or standard doublet-detection workflows. Datasets were integrated using an anchor-based strategy in Seurat, and cell-type annotation relied on canonical marker genes and published annotations. DDX28 expression patterns were visualized across malignant blasts, exhausted T cells, and hematopoietic stem cells (HSCs) via feature expression plots and violin plots. Cellular functional assays The human AML cell line HEL cell was maintained at 37 °C in an atmosphere of 5% CO2, utilizing RPMI-1640 medium supplemented with 10% FBS (Gibco, USA). Following was siRNA sequences that designed to silence DDX28:siRNA-1: 5ʹ-UUGACAACCAGCUCCACACCA-3ʹsiRNA-2: 5ʹ-UUCGUCGGGACUGCGGACCGU-3ʹ siRNA-1: 5ʹ-UUGACAACCAGCUCCACACCA-3ʹ siRNA-1: 5ʹ-UUGACAACCAGCUCCACACCA-3ʹ siRNA-2: 5ʹ-UUCGUCGGGACUGCGGACCGU-3ʹ siRNA-2: 5ʹ-UUCGUCGGGACUGCGGACCGU-3ʹ Cells were subjected to reverse transfection with siRNA oligonucleotides as recommended, and subsequently collected after 48 h for downstream assays. Western blot analysis The protein was extracted and the concentration was assessed via bicinchoninic acid assay method by measuring absorbance at 562 nm, and aliquots were stored at −80 °C. Following electrophoretic separation on an SDS-polyacrylamide gel, proteins were immobilized on PVDF membranes. After blocking with 5% skimmed milk, the membranes were exposed overnight at 4 °C to specific primary antibodies against DDX28 (Proteintech, China) and GAPDH (CST, USA). After being visualized using an ECL detection system, the expression of DDX28 was analyzed and quantified with ImageJ software and normalized to GAPDH. Assessment of cellular proliferation To investigate the proliferative capacity of HEL cells, we employed both the CCK-8 assay and an EdU staining kit, with cultures maintained independently over 24, 48, and 72 h. At each designated interval, 100 µL of complete medium combined with CCK-8 (in a 10:1 ratio) was introduced, followed by a one-hour incubation period within the incubator. We then assessed absorbance at 450 nm using a microplate reader. For a more comprehensive evaluation of proliferation, the EdU assay came into play as well. Here, cells were initially plated at a density of 1 × 10^5 per well in 24-well plates and allowed to proliferate for 24 h. Next, 10 µM EdU was added to the medium, and the setup incubated at 37 °C for an additional two hours. From there, the cells received a PBS wash, fixation via 4% paraformaldehyde, and permeabilization through 0.5% Triton X-100. Subsequently, we applied a click reaction cocktail to the fixed cells for 30 min at room temperature, ensuring protection from light. Nuclei counterstaining with DAPI followed for 10 min. Images were ultimately captured under fluorescence microscopy and subjected to quantitative analysis using ImageJ software. Analysis of cell invasion The Transwell assay was performed. HEL cells were resuspended in RPMI-1640 medium (without serum), with the bottom chamber were added with 600 µL of RPMI-1640 medium containing FBS. The medium was carefully removed after a 24-hour incubation period. Cell counts were obtained by direct microscopic observation and analyzed using ImageJ software. Statistical analysis Statistical analyses were conducted through R software (v4.1.0). Quantitative in vitro data are displayed as mean ± standard error of the mean (SEM). The number of independent biological replicates (n) and the exact statistical tests are stated in the corresponding figure legends. Two-tailed Student's t-test (or non-parametric alternatives when appropriate) was used for two-group comparisons. For analyses involving multiple comparisons, p-values were adjusted using the Benjamini–Hochberg false discovery rate (FDR) procedure, and adjusted q-values are reported where applicable; a threshold of q < 0.05 was considered statistically significant. Results Elevated expression of DDX28 in AML and its clinical relevance Analysis of DDX28 expression across pan-cancer and within LAML dataset was performed using the GEPIA2 database. As shown in Fig. 1A, a marked upregulation of DDX28 across a spectrum of cancers, including BLCA, CHOL, COAD, ESCA, HNSC, KICH, and LAML (AML). Fig. 1B-C illustrates DDX28 expression in LAML tumor versus normal tissues. Further comparison of immune infiltration distributions based on somatic copy number alteration (sCNA) status across TCGA cancer types (Fig. 1D) revealed deep deletions and arm-level deletions in AML.Fig. 1DDX28 is overexpressed in AML and correlates with poor clinical outcomes. (A) Pan-cancer DDX28 expression in tumor (red) and normal tissues (blue) from TIMER2.0. (B) DDX28 expression in TCGA-LAML tumors versus platform-integrated normal reference samples (for example, GTEx-based normal references) in GEPIA2. (C) DDX28 expression in tumors versus platform-defined normal reference samples across TCGA cancers. (D) Distribution of somatic copy-number alterations (sCNAs) for DDX28 in TCGA cancers. *p< 0.050, **p< 0.010, and ***p< 0.001 indicates significant differences between the specified groups.Fig 1 dummy alt text DDX28 is overexpressed in AML and correlates with poor clinical outcomes. (A) Pan-cancer DDX28 expression in tumor (red) and normal tissues (blue) from TIMER2.0. (B) DDX28 expression in TCGA-LAML tumors versus platform-integrated normal reference samples (for example, GTEx-based normal references) in GEPIA2. (C) DDX28 expression in tumors versus platform-defined normal reference samples across TCGA cancers. (D) Distribution of somatic copy-number alterations (sCNAs) for DDX28 in TCGA cancers. *p< 0.050, **p< 0.010, and ***p< 0.001 indicates significant differences between the specified groups. Prognostic significance of DDX28 expression In exploring the clinical implications of DDX28 expression, we conducted a Cox regression analysis via the TIMER platform. As illustrated in Fig. 2A, a heatmap presents the normalized infiltration scores across various cancer types, highlighting notable disparities particularly in AML and prostate adenocarcinoma (PRAD). From this, it became evident that such differences hold substantial weight. To gauge the prognostic value of DDX28 for AML patient outcomes, we applied Kaplan–Meier survival analysis. Across the LAML cohort and in broader pan-cancer assessments, elevated DDX28 levels consistently tied to diminished survival rates and worse clinical prognoses (Fig. 2B and Figure S1A–B). Moreover, univariate Cox regression underscored a strong link between DDX28 expression and survival metrics (Figure S1C, p < 0.01).Fig. 2DDX28 as an independent prognostic biomarker in AML. (A) Multivariate Cox regression heatmap. Red boxes indicate significant risk associations (HR > 1). (B) Kaplan-Meier curve for TCGA-LAML cohort (log-rank p = 0.00189, HR = 1.41). (C) Time-dependent ROC curves for 1/3/5-year survival. (D) Univariate Cox regression (HR = 2.187). (E) Multivariate Cox regression (HR = 2.018). (F) Forest plot of univariate Cox regression for DDX28 in pan-cancers. (G) Nomogram-predicted survival probability for specified clinical profile. (H) Restricted cubic spline (RCS) analysis of DDX28 expression and overall survival (p-overall = 0.029, p-nonlinear = 0.097). (I) DDX28 expression across AML molecular subtypes (p< 0.001).Fig 2 dummy alt text DDX28 as an independent prognostic biomarker in AML. (A) Multivariate Cox regression heatmap. Red boxes indicate significant risk associations (HR > 1). (B) Kaplan-Meier curve for TCGA-LAML cohort (log-rank p = 0.00189, HR = 1.41). (C) Time-dependent ROC curves for 1/3/5-year survival. (D) Univariate Cox regression (HR = 2.187). (E) Multivariate Cox regression (HR = 2.018). (F) Forest plot of univariate Cox regression for DDX28 in pan-cancers. (G) Nomogram-predicted survival probability for specified clinical profile. (H) Restricted cubic spline (RCS) analysis of DDX28 expression and overall survival (p-overall = 0.029, p-nonlinear = 0.097). (I) DDX28 expression across AML molecular subtypes (p< 0.001). Beyond that, we evaluated the prognostic potential of DDX28 through receiver operating characteristic (ROC) curves [12]. These curves offer a robust approach to gauging the precision of survival prediction models. As seen in Fig. 2C, DDX28 exhibited solid forecasting power for survival durations, boasting area under the curve (AUC) values exceeding 0.6 at both 1- and 3-year marks. It is worth highlighting that its performance shone especially at the 1-year horizon, registering an AUC of 0.71, which points toward meaningful clinical applicability. Building on this, variate Cox regression affirmed DDX28 expression as a key determinant of patient survival (Fig. 2D). In line with these observations, forest plots illustrated how elevated DDX28 tied to heightened mortality risks across multiple cancers, with a pronounced effect in LAML (Fig. 2F). To probe its standalone prognostic value, multivariate Cox models integrated factors like age and sex, assessing DDX28′s link to survival (Fig. 2E and Figure S1D). From these, DDX28 emerged as a notable risk indicator in AML, carrying a hazard ratio (HR) of 1.93. Time-dependent ROC assessments then scrutinized the biomarker's reliability for predicting survival at 1, 3, and 5 years, factoring in demographics such as sex and age alongside other clinical traits (Fig. 2G). With a composite score of 174 points, the framework projected survival odds at 37.8% for 1 year, 9.94% for 3 years, and 4.12% for 5 years—all holding statistical weight. We further dissected the tie between DDX28 expression and overall survival (OS) via restricted cubic splines (RCS), uncovering a straightforward linear impact on risk among LAML cases (Fig. 2H). On top of that, DDX28 levels varied markedly by molecular subtype (p < 0.001), with the AML.1 group showing substantially elevated expression relative to the notably subdued levels in AML.5 (Fig. 2I). DDX28-driven remodeling of an immunosuppressive microenvironment To investigate the link between DDX28 expression and infiltrating immune cells in AML, a Spearman's rank correlation test was conducted. Infiltration levels were quantified using ssGSEA scores. As Fig. 3A shown, in the high DDX28 expression group, decreases in CD4+ T cells and increases in M2 macrophages and Treg cells were statistically significant, suggesting immune exhaustion and transition toward a tumor-promoting microenvironment, correlating with the pro-cancer role of DDX28. Correlation heatmaps showed that DDX28 expression in AML patients was positively correlated with B cells, NK cells, M2 macrophages, and eosinophils, and negatively correlated with CD4+ T cells (Fig. 3B). Additionally, DDX28 was associated with immune escape. Wilcoxon tests were used to compare differences in microsatellite instability (MSI) status, cancer-associated fibroblast (CAF) infiltration score, and PD-L1 expression between DDX28 high and low groups (Fig. 3C-E). Our analysis detected a statistical association between elevated DDX28 expression and higher MSI-related scores in the TCGA-LAML expression-based metrics. Because clinically actionable MSI-high status is uncommon in AML, we interpret this finding cautiously as a potential surrogate of mismatch repair or replication stress signatures in a subset of cases, rather than as evidence of classical MSI-driven neoantigen generation. The tumor microenvironment was transformed into a pro-tumor phenotype by increased CAF infiltration and higher DDX28 expression, which enabled immunological escape through cytokine release and immune cell inhibition. High DDX28 expression did not, however, significantly increase PD-L1, indicating that immune escape can take place via alternative mechanisms including CAF infiltration or Treg enrichment.Fig. 3DDX28 promotes an immunosuppressive tumor microenvironment. (A) Differential immune cell proportions between the high and low DDX28 expression groups (p < 0.05 for highlighted subsets). (B) Correlation heatmap of DDX28 with immune infiltration. (C) MSI-related scores in the high and low DDX28 expression groups. (D) Cancer-associated fibroblast (CAF) scores. (E) CD274 (PD-L1) expression (ns, no significance). Data are shown as mean ± SEM; n indicates independent experiments or samples as described in Methods. *p< 0.050, **p< 0.010, and ***p< 0.001 indicates significant differences compared to control.Fig 3 dummy alt text DDX28 promotes an immunosuppressive tumor microenvironment. (A) Differential immune cell proportions between the high and low DDX28 expression groups (p < 0.05 for highlighted subsets). (B) Correlation heatmap of DDX28 with immune infiltration. (C) MSI-related scores in the high and low DDX28 expression groups. (D) Cancer-associated fibroblast (CAF) scores. (E) CD274 (PD-L1) expression (ns, no significance). Data are shown as mean ± SEM; n indicates independent experiments or samples as described in Methods. *p< 0.050, **p< 0.010, and ***p< 0.001 indicates significant differences compared to control. Mechanisms of therapeutic resistance are revealed by functional enrichment AML patients were stratified into high and low DDX28 expression groups based on the median level of DDX28 expression. GO and KEGG enrichment studies revealed high enrichment in nuclear division, immunoglobulin complex, antigen binding, and pathways associated to RNA binding, cell cycle, and apoptosis (Fig. 4A-C). Similarly, DDX28 expression was positively correlated with the functional activity scores for cell cycle, DNA damage, and DNA repair pathways in LAML, according to Gene Set Variation Analysis (GSVA) (Figure S2A). Furthermore, DDX28 expression exhibited consistent changes with cell cycle progression, indicating a potential correlation (Fig. 4F). GSEA revealed that gene clusters were predominantly enriched in pathways linked to leukemia and apoptosis, including Wnt/β-catenin and PI3K/AKT/mTOR signaling (Fig. 4D-E). The effect of DDX28 on drug sensitivity was also investigated. Box plots showed that the group with high DDX28 expression had much lower sensitivity to 5-fluorouracil, linking high DDX28 expression to pro-tumor effects and 5-fluorouracil resistance (Fig. 4G). Scatter plots showed that DDX28 expression and 5-fluorouracil sensitivity were negatively correlated (R = −0.56, p = 6.9e-14) (Fig. 4H-I). Higher DDX28 expression was linked to weaker tumor suppression by 5-fluorouracil, according to the negative correlation coefficient, which served as a foundation for clinical categorization. Using clustering analysis, a drug sensitivity heatmap showed response commonalities across samples and medicines (Fig. 4J). In an independent public perturbation dataset in which DDX28 was silenced by shRNA, DDX28 levels were markedly reduced relative to controls (p = 0.01), confirming successful gene silencing in this dataset (Figure S2B). Using the TF-Target Finder (TFTF) database, THAP11 was identified as the most significantly and positively correlated transcription factor for DDX28 (R = 0.62, p < 2.2e-6), suggesting THAP11 as a potential transcriptional regulator of DDX28 (Figure S2C). Because 5-fluorouracil is not a standard induction or salvage agent for AML, we present this association as a hypothesis-generating signal that may reflect broader vulnerabilities linked to nucleotide metabolism and DNA damage responses.Fig. 4DDX28 is associated with therapy resistance-related programs through cell cycle and apoptosis pathways. (A) GO enrichment bar plot. (B) GO enrichment dot plot. (C) KEGG pathway heatmap. (D) GSEA enrichment profile. (E) GSEA ranked list plot. (F) DDX28 expression during cell cycle progression. (G) Predicted drug sensitivity to 5-Fluorouracil in the high and low DDX28 expression groups (computational inference). (H) Scatter plot of predicted 5-Fluorouracil IC50 and DDX28 expression (R = −0.56, p < 0.0001). (I) Volcano plot of DDX28 correlation with predicted drug sensitivity. (J) Heatmap of significant predicted drug sensitivity profiles. Data are shown as mean ± SEM; n indicates independent experiments or samples as described in Methods. *p< 0.050, **p< 0.010, and ***p< 0.001 indicates significant differences compared to control.Fig 4 dummy alt text DDX28 is associated with therapy resistance-related programs through cell cycle and apoptosis pathways. (A) GO enrichment bar plot. (B) GO enrichment dot plot. (C) KEGG pathway heatmap. (D) GSEA enrichment profile. (E) GSEA ranked list plot. (F) DDX28 expression during cell cycle progression. (G) Predicted drug sensitivity to 5-Fluorouracil in the high and low DDX28 expression groups (computational inference). (H) Scatter plot of predicted 5-Fluorouracil IC50 and DDX28 expression (R = −0.56, p < 0.0001). (I) Volcano plot of DDX28 correlation with predicted drug sensitivity. (J) Heatmap of significant predicted drug sensitivity profiles. Data are shown as mean ± SEM; n indicates independent experiments or samples as described in Methods. *p< 0.050, **p< 0.010, and ***p< 0.001 indicates significant differences compared to control. Epigenetic and transcriptional regulation of DDX28 overexpression Using the MEXPRESS database, a thorough examination of the methylation status of several DDX28 CpG sites and their association with prognosis in AML patients was carried out. A total of 16 CpG methylation sites were identified within DDX28, among which three were associated with poor prognosis (cg11289251, cg06140118, cg13665184) and one with favorable prognosis (cg04528069) (Fig. 5A-B). A significant negative correlation was observed through Spearman’s analysis between DDX28 transcript levels and promoter hypermethylation (Fig. 5C-D). The distribution of each methylation site in LAML was shown in Fig. 5E. The CpG site cg04528069, which demonstrated the most significant inverse association with high expression, is illustrated in a scatter plot depicting the relationship between its methylation and transcript levels (Fig. 5F). The relationship between tumor mutation burden (TMB) and DDX28 expression levels was also examined. A thorough mutational profile, including variant classification, variant type, DNA class, distribution of mutations per sample, and proportional representation of mutation types, is shown in Figure S3A. Nevertheless, DDX28 expression and TMB did not significantly correlate (p > 0.05), indicating that DDX28 may not affect tumor growth via controlling the total mutation load (Figure S3B-C).Fig. 5Epigenetic regulation and candidate transcriptional control of DDX28. (A) Genomic distribution of CpG methylation sites in the DDX28 promoter region. (B) Correlation between DDX28 expression and methylation levels. (C) MEXPRESS validation of DDX28 expression and promoter methylation. (D) Methylation–expression correlation profile across CpG sites. (E) Methylation β-values distribution at key CpG sites. (F) Negative correlation of promoter methylation with DDX28 expression (Spearman R = −0.12, p < 0.05). Data are shown as mean ± SEM; n indicates independent samples as described in Methods. *p< 0.050, **p< 0.010, and ***p< 0.001 indicates significant differences compared to control.Fig 5 dummy alt text Epigenetic regulation and candidate transcriptional control of DDX28. (A) Genomic distribution of CpG methylation sites in the DDX28 promoter region. (B) Correlation between DDX28 expression and methylation levels. (C) MEXPRESS validation of DDX28 expression and promoter methylation. (D) Methylation–expression correlation profile across CpG sites. (E) Methylation β-values distribution at key CpG sites. (F) Negative correlation of promoter methylation with DDX28 expression (Spearman R = −0.12, p < 0.05). Data are shown as mean ± SEM; n indicates independent samples as described in Methods. *p< 0.050, **p< 0.010, and ***p< 0.001 indicates significant differences compared to control. Single-cell transcriptomic validation of pro-tumor functional states Single-cell transcriptomic analysis demonstrated that DDX28 is primarily expressed in malignant cells and immature blasts (Fig. 6A-D). Multiple external leukemia GEO datasets showed that DDX28 expression was mainly found in malignant phenotypic cells, exhausted T cells, proliferating T cells, and undifferentiated immature cells, with low expression in HSCs and CD8+ T cells. This suggests an association between DDX28 and malignant phenotypes and immune exhaustion at the single-cell level (Fig. 6E). Similarly, validation in other datasets indicated that DDX28 was expressed in exhausted immune cells like Tprolif, as well as in erythroid precursors and other primitive or poorly differentiated malignant cells (Fig. 6F-G).Fig. 6Single-cell atlas reveals DDX28 enrichment in malignant and proliferative cells. (A-D) scRNA-seq mapping of DDX28 in TCGA-AML, Malignant blasts(A) Immature progenitor cells (B) Pro-liferating T cells (Tprolif) (C) Exhausted CD8+ T cells (D). (E) Cross-dataset validation (GEO) of DDX28 enriched in malignant phenotypes. (F-G) External AML scRNA-seq confirmation, co-expression in Tprolif/exhausted T cells (F) Cell-type specificity scores(G).Fig 6 dummy alt text Single-cell atlas reveals DDX28 enrichment in malignant and proliferative cells. (A-D) scRNA-seq mapping of DDX28 in TCGA-AML, Malignant blasts(A) Immature progenitor cells (B) Pro-liferating T cells (Tprolif) (C) Exhausted CD8+ T cells (D). (E) Cross-dataset validation (GEO) of DDX28 enriched in malignant phenotypes. (F-G) External AML scRNA-seq confirmation, co-expression in Tprolif/exhausted T cells (F) Cell-type specificity scores(G). DDX28 knockdown inhibits proliferative, migratory, and invasive capacities in HEL cells We knocked down DDX28 in HEL cells, which were chosen for their high endogenous DDX28 levels, using plasmid transfection to study its role in AML. Successful knockdown was subsequently confirmed by western blot analysis, as shown in Fig. 7A. The CCK8 results revealed that DDX28 knockdown considerably reduced HEL cell growth (Fig. 7B). The EdU incorporation experiment further supported this antiproliferative effect (Fig. 7C, E). Furthermore, we evaluated the impact of DDX28 on metastatic behaviors using Transwell assays, which revealed that its inhibition simultaneously suppressed invasive and migratory abilities. Experimental results demonstrated that DDX28 inhibition significantly inhibited the migratory and invasive abilities of HEL cells (Figs. 7D, F).Fig. 7DDX28 knockdown inhibits AML proliferation and metastasis. (A) DDX28 expression in HEL cells transfected with DDX28 siRNA-1 or siRNA-2 was detected by Western blotting. (B) CCK8 assay showed that knockdown of DDX28 inhibited the proliferation of HEL cells. (C) EdU assay indicated that knockdown of DDX28 inhibited the proliferation of HEL cells. Scale bar = 100 μm. (D) Transwell assays indicated that knockdown of DDX28 inhibited the migration and invasion ability of HEL cells. Scale bar = 100 μm. (E) Quantitative analysis of EdU assay. (F) Quantitative analysis of Transwell assays. Data are shown as mean ± SEM; n indicates independent biological replicates as described in Methods. *p< 0.050, **p< 0.010, and ***p< 0.001 indicates significant differences compared to control.Fig 7 dummy alt text DDX28 knockdown inhibits AML proliferation and metastasis. (A) DDX28 expression in HEL cells transfected with DDX28 siRNA-1 or siRNA-2 was detected by Western blotting. (B) CCK8 assay showed that knockdown of DDX28 inhibited the proliferation of HEL cells. (C) EdU assay indicated that knockdown of DDX28 inhibited the proliferation of HEL cells. Scale bar = 100 μm. (D) Transwell assays indicated that knockdown of DDX28 inhibited the migration and invasion ability of HEL cells. Scale bar = 100 μm. (E) Quantitative analysis of EdU assay. (F) Quantitative analysis of Transwell assays. Data are shown as mean ± SEM; n indicates independent biological replicates as described in Methods. *p< 0.050, **p< 0.010, and ***p< 0.001 indicates significant differences compared to control. Discussion Through multi-omic analysis and experimental validation, this study confirms that DDX28 is overexpressed in acute myeloid leukemia (AML) and is closely associated with poor clinical prognosis. Multivariate Cox regression confirmed its status as an independent predictor of poor prognosis, suggesting that DDX28 may play a central role in the pathogenesis of AML. Current research on the role of DDX28 in tumors remains limited, although the functions of DDX family members in tumorigenesis and progression have attracted considerable interest [13]. As members of the RNA helicase family, DDX proteins are involved in several stages of RNA processing, including transcription, splicing, translation, and degradation [[14], [15], [16], [17], [18]]. Dysregulated expression of DDX family proteins is correlated with poor prognosis in various malignancies, including colorectal, hepatocellular, pancreatic, and breast cancers [[19], [20], [21], [22], [23]]. In addition, DDX proteins may play a critical role in hematologic malignancies [24]. For example, DDX18 is established as essential for hematopoiesis after allele hi1727 insertion perturbs its function, and its loss triggers depletion of myeloid and erythroid lineages through p53-dependent G1 arrest and enhanced apoptotic rates [25]. The present study proved the evidence that elevated DDX28 expression in AML and significantly correlated with poor patient prognosis, confirming its clinical value as an independent prognostic biomarker and providing new evidence for the functional role of DDX28 in hematologic malignancies. DDX28 may promote AML progression through multiple mechanisms. Enrichment analysis indicated that elevated DDX28 expression was mainly enriched in cell cycle regulation, DNA repair, and RNA binding pathways. Notably, gene set enrichment analysis showed a positive correlation between DDX28 expression and cell cycle progression, and GSEA indicated its regulation of leukemia-related pathways such as Wnt/β-catenin and PI3K/AKT/mTOR, which are known to play crucial roles in leukemia development and treatment resistance . The Wnt/β-catenin pathway is involved in the regulation of cell proliferation, survival and differentiation of hematopoietic cells . Due to its direct role in hematological regulatory mechanisms, alterations in the PI3K/AKT/mTOR pathway, which are seen in more than 50% of acute leukemia patients, cause aberrant hematopoiesis and, when overactivated, drive the development of leukemia . Furthermore, DDX28 expression is strongly impacted by epigenetic control; its mRNA overexpression is adversely linked with poor methylation status at particular CpG sites in the promoter region. According to bioinformatics studies, DDX28 expression may be positively regulated by the transcription factor THAP11, which would accelerate the development of AML. THAP11 serves as a transcription regulator that attaches to DNA and modulates the expression of genes vital for metabolic processes and cellular proliferation. Neurodevelopmental problems and cancer are linked to its dysregulation. When THAP11 was overexpressed, the transcript levels of several transcription factors crucial for hematopoiesis, such as GATA-2, c-Myc, Fli1, and c-Myb, underwent alterations . Our findings showed that DDX28 knockdown dramatically reduced the ability of HEL cells for proliferation and migration in vitro, which is comparable with the anti-tumor effects of targeting RNA helicases in other malignancies. DDX28 expression showed no significant association with overall tumor mutation burden in this cohort. This observation is consistent with a prominent contribution of epigenetic and transcriptional regulation to DDX28 overexpression, but it does not exclude roles for specific driver lesions or context-dependent genetic interactions. Linking DDX28 to metabolism requires careful framing. DDX28 localizes to mitochondria and participates in mitoribosome assembly and mitochondrial translation, which supports oxidative phosphorylation and the anabolic capacity required for rapid proliferation. In AML, proliferative expansion and activation of DNA repair are energetically and biosynthetically demanding processes. Therefore, the enrichment of cell cycle and genome maintenance programs in the high DDX28 group can be mechanistically compatible with altered mitochondrial activity, because mitochondrial translation influences ATP production, redox balance, and one-carbon and nucleotide biosynthesis that feed replication and repair. Although our present study did not include direct metabolic flux assays, this framework provides a biologically plausible connection between DDX28 mitochondrial functions and the transcriptional programs observed here, and it motivates future experiments measuring oxygen consumption, glycolytic rate, and nucleotide pools after DDX28 perturbation in AML models. Regarding predicted drug response, the association with 5-fluorouracil should be interpreted as a computational signal rather than a direct clinical recommendation, because 5-fluorouracil is mainly used in solid tumors. In the GDSC resource, 5-fluorouracil sensitivity captures perturbation of thymidylate synthesis and RNA and DNA damage pathways, which overlap with the replication stress and repair signatures enriched in high DDX28 AML samples. We therefore highlight 5-fluorouracil to illustrate a potential link between DDX28 expression and nucleotide stress responses, while acknowledging that AML-relevant agents, such as cytarabine or anthracyclines, should be prioritized in future validation. The MSI-related result also warrants caution. Clinically established MSI-high disease is rare in AML, and expression-based MSI surrogates can reflect mismatch repair activity or replication stress rather than true microsatellite length instability. Accordingly, we consider the observed MSI score association as hypothesis-generating for a subset of cases with heightened repair pathway engagement, and we avoid over-interpreting it as evidence for widespread neoantigen-driven immunity in AML. Our functional validation used the HEL AML cell line, selected because of its relatively high endogenous DDX28 expression and reliable transfection efficiency. Nevertheless, AML is molecularly heterogeneous, and HEL represents only one genetic context. These in vitro results should therefore be viewed as proof of principle, and future studies should extend validation to additional AML cell lines and primary patient samples representing distinct molecular subtypes. The therapeutic feasibility of targeting DEAD-box RNA helicases requires balanced consideration. Because these enzymes participate broadly in RNA metabolism, systemic inhibition may carry toxicity risks in normal hematopoietic and other proliferative tissues. Potential strategies to improve selectivity include targeting cancer-specific dependencies, exploiting differential mitochondrial stress thresholds, or developing approaches that modulate DDX28-associated complexes rather than globally blocking helicase activity. DDX28 may promote AML progression through multiple mechanisms. Enrichment analysis indicated that elevated DDX28 expression was mainly enriched in cell cycle regulation, DNA repair, and RNA binding pathways. Notably, gene set enrichment analysis showed a positive correlation between DDX28 expression and cell cycle progression, and GSEA indicated its regulation of leukemia-related pathways such as Wnt/β-catenin and PI3K/AKT/mTOR, which are known to play crucial roles in leukemia development and treatment resistance [[26], [27], [28]]. The Wnt/β-catenin pathway is involved in the regulation of cell proliferation, survival and differentiation of hematopoietic cells [29]. Due to its direct role in hematological regulatory mechanisms, alterations in the PI3K/AKT/mTOR pathway, which are seen in more than 50% of acute leukemia patients, cause aberrant hematopoiesis and, when overactivated, drive the development of leukemia [30,31]. Furthermore, DDX28 expression is strongly impacted by epigenetic control; its mRNA overexpression is adversely linked with poor methylation status at particular CpG sites in the promoter region. According to bioinformatics studies, DDX28 expression may be positively regulated by the transcription factor THAP11, which would accelerate the development of AML. THAP11 serves as a transcription regulator that attaches to DNA and modulates the expression of genes vital for metabolic processes and cellular proliferation. Neurodevelopmental problems and cancer are linked to its dysregulation [32]. When THAP11 was overexpressed, the transcript levels of several transcription factors crucial for hematopoiesis, such as GATA-2, c-Myc, Fli1, and c-Myb, underwent alterations [33,34]. Our findings showed that DDX28 knockdown dramatically reduced the ability of HEL cells for proliferation and migration in vitro, which is comparable with the anti-tumor effects of targeting RNA helicases in other malignancies. However, DDX28 expression showed no significant association with TMB, suggesting that its pro-tumor effects primarily depend on transcriptional or post-transcriptional regulation rather than gene mutations. AML leukemogenesis, progression, and treatment resistance are all dependent on the tumor immune microenvironment [33]. Our work found that DDX28 overexpression was inversely correlated with frequency, but positively associated with the enrichment of immunosuppressive M2 macrophages and Tregs within the AML immune microenvironment. These changes in the ratio of immune cells promote tumor growth and immune evasion, thereby favoring the development of an immunosuppressive microenvironment [35]. Furthermore, the degree of infiltration of cancer-associated fibroblasts (CAFs) correlates significantly with DDX28 expression, suggesting that this gene exerts its effect by modulating the tumor microenvironment. It is noteworthy that the progression of acute myeloid leukemia (AML) is exacerbated by the activation of immune evasion mechanisms, such as the infiltration of cancer-associated fibroblasts and the accumulation of regulatory T cells [36]. Consequently, therapeutic strategies targeting the tumor immune microenvironment could represent an effective approach to improving the prognosis of AML patients. Although our research has demonstrated the central role of DDX28 in AML and its potential as a prognostic biomarker, certain limitations remain. First, further studies using functional assays are needed to fully elucidate the precise molecular signaling pathways through which DDX28 regulates the immunological microenvironment. Furthermore, DDX28′s promise as a therapeutic target is currently unsupported by in vivo experimental results. In order to confirm the precise mechanisms of action of DDX28 in AML and its viability as a therapeutic target, further research will increase the sample size and include in vivo trials. Although our research has demonstrated the potential importance of DDX28 in AML, several limitations should be acknowledged. First, the bioinformatic associations for immune features, MSI-related scores, and predicted drug response are based on retrospective datasets and require prospective validation. Second, our functional experiments were performed in a single AML cell line and were limited to in vitro assays. Third, in vivo evidence and deeper mechanistic dissection of the signaling and mitochondrial pathways downstream of DDX28 are still needed. Future work incorporating additional AML models, primary samples, and animal studies will be important for confirming causality and evaluating therapeutic windows. Overall, the increased DDX28 levels in AML are strongly associated with poor clinical outcomes. Its overexpression, which is caused by THAP11 transactivation and promoter hypomethylation, increases cell cycle progression, DNA repair, immunosuppressive milieu, and confers resistance to therapy. In vitro knockdown inhibited proliferation, migration, and invasion, while single-cell transcriptomics identified DDX28 enrichment in malignant blasts and fatigued T cells. Collectively, DDX28 represents both a prognostic indicator and a potential treatment target in AML. Funding statement The author(s) received no specific funding for this study. Availability of data and materials Analytical datasets utilized in this study were acquired from TCGA repository, an open-access resource that operates under rigorous ethical frameworks mandating written informed consent from all enrolled subjects. All data acquisition and dissemination procedures strictly adhered to these institutional protocols. Ethics approval Not applicable. CRediT authorship contribution statement Zi Wang: Writing – original draft, Investigation, Formal analysis, Conceptualization. Yunli Liu: Writing – original draft, Investigation, Formal analysis, Conceptualization. Xinzhu Guan: Writing – review & editing, Supervision, Resources, Conceptualization. Miaoqing Ye: Visualization, Validation, Formal analysis. Xinxin Xu: Writing – review & editing, Supervision, Resources, Conceptualization. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.