Decitabine Induces Subtype-Specific Epigenomic Remodeling and Perturbs Age-Associated Regulatory CpGs in Breast Cancer

Abstract

Objectives:

Breast cancer (BC) subtypes such as HR+, HER2+, and triple-negative (TNBC) show distinct molecular features, treatment responses, and outcomes. DNA methylation is a key, targetable epigenetic regulator in BC. This study examined whether the DNA methyltransferase inhibitor decitabine (DAC) produces subtype-specific epigenomic and transcriptional effects in breast cancer cell lines representing distinct molecular subtypes.

Methods:

Gene expression and DNA methylation data from DAC-treated and untreated T-47D (Luminal-A) and JIMT-1 (HER2-amplified, trastuzumab-resistant with a TNBC-like phenotype) breast cancer cell lines were obtained from a published dataset. Differential expressions were assessed using limma, and methylation changes were defined using β-value thresholds. Integrated epigenomic–transcriptional analysis, functional enrichment, Horvath clock CpG evaluation, and survival analysis were performed in the METABRIC and TCGA cohorts.

Results:

In JIMT-1, DAC caused hypomethylation at 1195 CpG sites and upregulation of 187 genes, including TFAP2E, an age-associated locus selectively hypomethylated after DAC. In T-47D, DAC induced hypomethylation at 1937 CpGs and upregulated 248 genes. Amongst these, KRT20 was upregulated despite promoter hypermethylation, indicating a subtype-specific regulatory architecture. DAC-responsive genes in JIMT-1 were enriched for cytokine signaling and piRNA-mediated epigenetic silencing, whereas T-47D showed enrichment for extracellular matrix organization, collagen dynamics, and piRNA processing pathways. Horvath clock CpG analysis showed selective perturbation of age-associated sites. Survival analysis identified 114 DAC-responsive genes associated with overall survival in ER/PR-positive BC and 8 in the JIMT-1-derived gene set.

Conclusion:

DAC induces subtype-dependent epigenomic and transcriptional remodeling, selectively disrupts age-associated regulatory programs, and underscores the need for subtype-stratified evaluation of epigenetic therapies in breast cancer.

Introduction

Breast cancer (BC) is a heterogeneous disease with distinct molecular subtypes that differ in incidence, therapy response, and outcomes. Population studies show that subtype distribution and survival vary by age, race, and reporting practices, emphasizing the need for subtype-specific research and biomarker development.^1,2 Clinically, breast cancers are classified as hormone receptor–positive (HR+), HER2-positive, or triple-negative (TNBC). HR+ tumors respond to endocrine therapy, whereas TNBC, lacking ER, PR, and HER2, shows higher recurrence and limited targeted options. Beyond genetics, epigenetic dysregulation is a major non-genetic driver of tumor behavior and therapeutic resistance. DNA methylation is a key epigenetic mechanism regulating tumor suppressors, lineage programs, and cellular plasticity. Subtype-specific methylation landscapes reflect regulatory states and tumor evolution. ³ Because epigenetic changes are reversible, they are attractive therapeutic targets.^{4 -6} Recent work has explored combining epigenetic drugs with immunotherapy or targeted therapy to overcome resistance.^{5 -7} However, subtype-specific effects remain poorly understood. Decitabine (DAC, 5-aza-2′-deoxycytidine), a DNA methyltransferase inhibitor, induces genome-wide demethylation and gene reactivation. It is well established in hematologic malignancies and under study in solid tumors, including BC. However, its effects depend on baseline methylation, chromatin state, and replication dynamics, which differ across subtypes. Buocikova et al ⁸ reported that DAC alters methylation and expression in BC cell lines, reactivating tumor suppressors and oncogenes with responses shaped by metabolic factors. These findings support detailed subtype-specific analysis of DAC-induced epigenomic remodeling and transcriptional outcomes.

Demethylation can produce both favorable and adverse effects by reactivating tumor suppressors or oncogenes.^{5 -8} Functional interpretation, therefore, requires identifying methylation changes associated with biologically meaningful shifts in expression. Gene ontology and pathway analyses help reveal affected processes. Loci such as TFAP2E, regulated by AP-2 transcription factors, illustrate this complex interplay between methylation and transcription in BC. ⁹ DNA methylation also reflects aging. Epigenetic clocks, such as Horvath’s, capture CpG patterns associated with biological age that overlap with regulatory networks influencing cancer risk.^{10 -15} Accelerated epigenetic aging has been linked to BC risk, survival outcomes, and racial differences. Whether DAC preferentially alters age-related CpGs, and whether this varies by subtype, remain unclear. Clock CpGs should be viewed not as markers of aging but as components of age-linked regulatory systems. Prior studies implicate cytokine signaling, extracellular matrix remodeling, and piRNA-mediated epigenetic silencing in aggressive BC phenotypes,^{16 -19} providing context for DAC-associated pathway changes.

Integrating epigenomic findings with clinical data strengthens translational relevance. The Nottingham Prognostic Index (NPI) provides risk stratification, ²⁰ while datasets such as METABRIC define subtype-specific molecular correlates of survival.^21,22 Incorporating DAC-responsive signatures into these cohorts can highlight prognostic markers, especially in TNBC, where immune, metastasis, and resistance pathways drive outcomes. In HR+ tumors, markers such as CAV1, KRT5, and CALML3 are linked to stromal and hormonal signaling.^{23 -25} This study presents a multi-omic, subtype-stratified analysis of DAC-treated JIMT-1 (HER2-amplified, trastuzumab-resistant) and T-47D (Luminal-A) cells from Buocikova et al. ⁸ We integrated promoter methylation and expression data to identify coordinated epigenomic–transcriptional responses and examined pathways linked to cytokine signaling, extracellular matrix dynamics, and epigenetic silencing.^{16 -19} We also analyzed Horvath clock CpGs as regulatory components of age-associated programs^{10 -15} and evaluated the prognostic significance of DAC-responsive genes in METABRIC and TCGA cohorts stratified by NPI.^20,21 This approach clarifies how DNMT inhibition reshapes the epigenome across subtypes, intersects with age-linked programs, and identifies gene signatures with potential clinical relevance in breast cancer.

Methods

Data Acquisition and Study Design

Gene expression and DNA methylation data for decitabine (DAC)–treated and untreated breast cancer cell lines were retrieved from Buocikova et al. ⁸ Two models were analyzed: JIMT-1 (HER2-amplified, trastuzumab-resistant with a TNBC-like phenotype) and T-47D (Luminal-A). Normalized expression matrices and β-value methylation data from the original authors were used directly to maintain consistency in preprocessing. β-values represent the ratio of methylated to total probe intensity with variance stabilization applied. All analyses used matched expression and methylation datasets for integrated epigenomic–transcriptional assessment. The reporting of the observational components of this study conforms to the STROBE statement. ²⁶ To evaluate clinical relevance, patient-level RNA-seq data for TNBC were obtained from TCGA (Genomic Data Commons; log2(FPKM + 1)), and microarray data for Luminal-A samples were obtained from the METABRIC cohort via cBioPortal. Both datasets were used in their pre-processed form without additional normalization.

Differential Gene Expression Analysis

Differential expression between DAC-treated and control samples was analyzed separately for each cell line using limma (v3.62.2) in R (v4.4.3). Linear models were fitted with lmFit and moderated with eBayes to enhance variance estimation. Genes with absolute log2 fold change (FC) > 1 and false discovery rate (FDR) < 0.05 were defined as significantly differentially expressed.

DNA Methylation Analysis

Methylation β-value matrices were processed using minfi (v1.52.1). Probes with detection P-value >.01 were removed. CpG sites were defined as DAC-hypomethylated if baseline β > .5 and Δβ < −.05. A baseline β > .5 indicates a predominantly methylated state, where most alleles carry a methyl group at the queried cytosine. A Δβ threshold of −.05, while modest in absolute terms, has been previously validated as biologically meaningful in Illumina array-based methylation studies, where it corresponds to detectable shifts in chromatin accessibility and has been associated with changes in transcription factor binding and gene expression at affected promoters. This threshold prioritizes specificity by excluding stochastic probe-level variation while capturing functionally relevant demethylation events, particularly at promoter-proximal CpGs, where even moderate methylation changes can influence transcriptional output. ²⁷ Promoter-associated CpGs meeting these criteria were mapped to their corresponding genes using platform annotation files.

Integration of Methylation and Expression Data

Genes showing both significant promoter hypomethylation and transcriptional upregulation (log2FC > 1, adjusted P < .05) were considered epigenetically responsive to DAC. This integration identified coordinated methylation–expression responses within each subtype.

Functional Enrichment Analysis

Biological functions of DAC-responsive genes were examined using clusterProfiler (v4.14.6). Gene Ontology (Biological Process, Cellular Component, Molecular Function) and KEGG pathway enrichment analyses were performed, with significance set at adjusted P < .05. Results were visualized using dotplot and cnetplot functions in clusterProfiler with ggplot2 (v3.5.2).

Analysis of Age-Associated Epigenetic Clock CpGs

DAC effects on age-related methylation were assessed using Horvath’s epigenetic clock (353 CpGs). Minfi-processed methylation data from JIMT-1 and T-47D were used to evaluate changes at these loci. For TNBC TCGA samples, the proportion of clock CpGs with β > .2 was calculated to estimate age-program engagement. Multiple testing corrections were applied using the Benjamini–Hochberg method.

Survival and Clinical Association Analyses

Prognostic relevance of DAC-responsive genes was assessed using overall survival data from METABRIC and TCGA. Kaplan–Meier and log-rank tests were performed using survival (v3.5-7), survMisc (v0.5.6), and survminer (v0.5.0) in R. Patients were stratified by median expression or methylation values. Associations with tumor stage were examined using the Nottingham Prognostic Index. TFAP2E promoter methylation and expression were specifically analyzed due to its reported prognostic role in other cancers.

Results

Subtype-Specific Epigenomic and Transcriptional Alterations Induced by Decitabine

DAC induced extensive DNA methylation and transcriptional changes in both cell lines, with subtype-specific methylation–expression coupling. In JIMT-1 (HER2-amplified, trastuzumab-resistant), 1195 hypermethylated CpGs (β > .5) became hypomethylated (Δβ < −.05), and 187 genes were upregulated (|log2FC| > 1, FDR < 0.05), with many showing promoter hypomethylation consistent with coordinated epigenomic–transcriptional responses (Figure 1). In JIMT-1, TFAP2E showed marked promoter hypomethylation and upregulation (log2FC > 1), localizing to the hypomethylated–upregulated quadrant of the integrated Δβ–log2FC plot, indicating concordant epigenetic–transcriptional regulation. In contrast, T-47D (Luminal-A) exhibited a larger epigenomic response with 1937 hypermethylated CpGs (β > .5) becoming hypomethylated (Δβ < −.05) and 248 genes upregulated (|log2FC| > 1, FDR < 0.05), but greater methylation–expression heterogeneity (Figure 2). In T-47D, methylation status was not uniformly predictive of transcription: many upregulated genes showed promoter hypomethylation, but others, including KRT20, were induced despite promoter hypermethylation, suggesting alternative regulation via distal elements or transcription factors. Overall, DAC induced subtype-specific remodeling coherent hypomethylation–activation coupling in JIMT-1 (HER2-amplified, trastuzumab-resistant) versus heterogeneous patterns in T-47D (Luminal-A), reflecting distinct regulatory architectures.

Figure 1.

Integrated DNA methylation and gene expression changes following decitabine treatment in JIMT-1 (HER2-amplified, trastuzumab-resistant) cells. (A) Volcano plot showing differentially expressed genes in DAC-treated versus untreated JIMT-1 cells (|log2FC| > 1, FDR < 0.05) and (B) Integrated analysis of promoter DNA methylation change (Δβ) and gene expression change (log2FC). Genes are classified into the categories hypomethylated–upregulated, hypomethylated–downregulated, hypermethylated–upregulated, and hypermethylated–downregulated. TFAP2E is highlighted as an example of coordinated hypomethylation and transcriptional upregulation.

Figure 2.

Subtype-specific methylation–expression relationships in Luminal-A T-47D cells following decitabine treatment. (A) Volcano plot of differentially expressed genes in DAC-treated versus untreated T-47D cells (|log2FC| > 1, FDR < 0.05) and (B) Integrated promoter DNA methylation (Δβ) and gene expression (log2FC) analysis showing heterogeneous methylation–expression coupling. KRT20 is highlighted as an example of transcriptional upregulation despite promoter hypermethylation.

Pathway Enrichment Analysis

In JIMT-1 (HER2-amplified, trastuzumab-resistant) cells, KEGG analysis of DAC-responsive genes identified cytokine–cytokine receptor interaction as the most strongly enriched pathway, followed by cytoskeleton in muscle cells, viral protein interaction with cytokine and cytokine receptor, transcriptional misregulation in cancer, cadherin signaling, rheumatoid arthritis, neutrophil extracellular trap formation, alcoholism, malaria, and Cushing syndrome (Figure 3A). GO biological process enrichment revealed piRNA processing, male meiotic nuclear division, retrotransposition, smooth muscle cell proliferation, transposition, regulation of smooth muscle cell proliferation, gene silencing by piRNA-directed DNA methylation, transposable element silencing by piRNA-mediated DNA methylation, transposable element silencing, and piRNA-mediated heterochromatin formation (Figure 3B). These results highlight cytokine signaling and piRNA-mediated epigenetic silencing as key features of DAC responses in JIMT-1 cells.

Figure 3.

Pathway enrichment analysis of DAC-responsive genes in JIMT-1 (HER2-amplified, trastuzumab-resistant) cells. (A) KEGG pathway enrichment analysis showing significant enrichment of cytokine signaling, cadherin, and immune-related pathways. (B) GO biological process enrichment analysis highlighting piRNA processing, transposable element silencing, and smooth muscle cell proliferation. Bar length represents gene count, and color indicates adjusted P-value.

In contrast, pathway enrichment analysis in the Luminal-A T-47D cell line revealed a distinct biological profile. Reactome pathway enrichment showed significant overrepresentation of extracellular matrix organization, collagen formation, assembly of collagen fibrils and other multimeric structures, PIWI-interacting RNA biogenesis, degradation of fibrin clot, degradation of the extracellular matrix, metabolism of amine-derived hormones, collagen cross-linking, gene silencing by RNA, and collagen degradation (Figure 4A). GO biological process enrichment identified piRNA processing, gene silencing by piRNA-directed DNA methylation, transposable element silencing by piRNA-mediated DNA methylation, piRNA-mediated heterochromatin formation, pattern specification process, male meiotic nuclear division, regeneration, retrotransposition, regulatory ncRNA processing, and transposition (Figure 4B). Together, these findings indicate that DAC induces distinct pathway-level responses across breast cancer subtypes, with preferential activation of cytokine signaling and piRNA-mediated silencing in JIMT-1 cells and extracellular matrix remodeling, collagen dynamics, and piRNA-related epigenetic silencing programs in Luminal-A T-47D cells.

Figure 4.

Pathway enrichment analysis of DAC-responsive genes in Luminal-A T-47D cells. (A) Reactome pathway enrichment analysis identifies extracellular matrix organization, collagen formation, and piRNA biogenesis pathways. (B) GO biological process enrichment analysis showing enrichment of piRNA processing, gene silencing, transposable element silencing, and pattern specification. Bar length represents gene count, and color indicates adjusted P-value.

DAC Selectively Perturbs Age-Associated Epigenetic Clock CpGs

In the JIMT-1 (HER2-amplified, trastuzumab-resistant) cell line, DAC treatment resulted in significant hypomethylation and transcriptional upregulation of TFAP2E, a CpG locus included in the Horvath epigenetic clock. TFAP2E methylation is positively correlated with chronological age, and its hypomethylation following DAC exposure suggests antagonism of methylation patterns that typically accumulate with aging. Consistent with this observation, analysis of Horvath clock CpGs demonstrated that DAC preferentially induced hypomethylation at CpG sites that showed a positive age correlation, indicating selective disruption of age-associated methylation patterns rather than uniform demethylation (Figure 5A and B). In contrast, in the Luminal-A T-47D cell line, KRT20 was transcriptionally upregulated despite increased promoter methylation, indicating discordant methylation–expression coupling at this locus. This finding suggests that age-associated CpG regulation is subtype dependent and that DNA methylation is not uniformly predictive of transcriptional output across breast cancer subtypes. To assess whether age-associated epigenetic programs exhibit heterogeneity in patient tumors, Horvath clock CpG methylation was evaluated in TNBC samples from TCGA. The proportion of clock CpGs with methylation levels exceeding β > .2 showed a right-skewed distribution, with most samples clustering between 5% and 15% methylated CpGs (Figure 6). This distribution indicates substantial inter-patient variability in the engagement of age-associated epigenetic programs, suggesting that such regulatory CpGs are variably methylated across TNBC tumors rather than uniformly altered.

Figure 5.

DAC-induced perturbation of age-associated epigenetic clock CpGs. (A) Lollipop chart showing methylation change (Δβ) at Horvath clock CpGs matched in JIMT-1 DAC-treated differentially methylated positions (TFAP2E, SLC28A2, C10orf99, DKK3). All 4 CpGs that are positively correlated with age show DAC-induced hypomethylation. (B) Paired dot plot showing promoter β-values before (control) and after DAC treatment, demonstrating consistent demethylation across matched clock CpGs.

Figure 6.

Inter-patient variability in Horvath clock CpG methylation in TNBC. Histogram showing the distribution of the percentage of Horvath clock CpGs with methylation levels > β 0.2 across TNBC patient samples, demonstrating heterogeneous engagement of age-associated epigenetic programs.

Survival Analysis

Survival Analysis in ER/PR/HER Positive BC Samples

To evaluate the clinical relevance of DAC-responsive genes, survival analysis was performed in ER/PR-positive breast cancer samples using the METABRIC cohort. Among the differentially expressed genes identified in Luminal-A cells, 114 genes were significantly associated with overall survival by log-rank testing (Table 1). Several genes demonstrated strong statistical associations with patient outcome, including CALML3 (p = 1.15 × 10⁻⁸), KRT5 (p = 1.83 × 10⁻⁶), CAV1 (p = 2.27 × 10⁻⁷), and ABCB1 (p = 3.69 × 10⁻⁷). Additional survival-associated genes included RNF213, TRAPPC9, PCDH17, SORBS2, LAMB3, LGR6, and RND1, among others (Table 1). Collectively, these findings indicate that a substantial fraction of DAC-responsive genes in Luminal-A breast cancer are statistically associated with patient survival in retrospective cohorts. These associations are observational in nature and should not be interpreted as evidence that DAC treatment directly modulates survival outcomes. Prospective validation in DAC-treated clinical cohorts is required before any causal or therapeutic conclusions can be drawn.

Table 1.

Differentially expressed genes significantly associated with overall survival in ER/PR-positive (Luminal-A) breast cancer. Top 10 of 114 significant genes shown (log-rank test, P < .05). Full list available as Supplemental Data.

Gene	P-value	Direction
CALML3	1.15 × 10−8	↑
TRAPPC9	2.03 × 10−9	↑
CAV1	2.27 × 10−7	↑
ABCB1	3.69 × 10−7	↑
RNF213	3.87 × 10−6	↑
KRT5	1.83 × 10−6	↑
PCDH17	3.13 × 10−7	↑
SORBS2	1.12 × 10−6	↑
LAMB3	4.26 × 10−6	↑
LGR6	7.62 × 10−5	↑

Direction: ↑ = DAC-upregulated gene. Genes ranked by P-value (log-rank test).

Survival Analysis in TNBC Samples

Survival analysis was then conducted in TNBC patient samples to assess whether genes identified in the JIMT-1 dataset showed prognostic relevance in clinical TNBC cohorts. In contrast to ER/PR-positive breast cancer, a smaller subset of genes was associated with survival in TNBC. Specifically, 8 genes demonstrated significant associations with overall survival. Among these, F2RL1 (p = 1.97 × 10⁻⁴), GNG2 (p = 3.48 × 10⁻⁴), APOBEC3G (p = 2.87 × 10⁻⁴), and HLA-G (p = 5.61 × 10⁻³) showed the strongest associations with patient outcome. Additional survival-associated genes included ABCC2, CA12, ACHE, and SH2D3C (Table 2). These analyses demonstrate that DAC-responsive genes show subtype-specific statistical associations with survival in retrospective cohorts, with a larger number identified in ER/PR-positive breast cancer and a smaller, distinct set in the JIMT-1-derived gene set. These findings are hypothesis-generating and require prospective validation in DAC-treated patient cohorts before clinical translation. Given the previously reported prognostic relevance of TFAP2E methylation in other malignancies, TFAP2E methylation levels were compared between breast cancer subtypes. TFAP2E promoter methylation differed significantly between ER/PR-positive breast cancer and TNBC samples (Figure 7). However, despite subtype-specific TFAP2E methylation differences, neither methylation nor expression was associated with TNBC survival (Figure 8). Overall, DAC-responsive genes showed subtype-specific prognostic relevance: broadly linked to survival in ER/PR-positive BC but limited and distinct in TNBC, highlighting divergent prognostic architectures. These findings emphasize that DAC-induced epigenomic/transcriptional changes have context-dependent clinical relevance and require subtype-stratified analyses to evaluate epigenetic therapy.

Table 2.

Differentially expressed genes are significantly associated with overall survival in triple-negative breast cancer (TNBC). Genes derived from the JIMT-1 (HER2-amplified, trastuzumab-resistant) dataset and evaluated in clinical TNBC cohorts (TCGA).

Gene	P-value	Direction
F2RL1	1.97 × 10−4	↑
APOBEC3G	2.88 × 10−4	↑
GNG2	3.48 × 10−4	↑
HLA-G	5.62 × 10−3	↑
ABCC2	8.33 × 10−3	↑
CA12	4.28 × 10−2	↑
SH2D3C	1.77 × 10−3	↑
ACHE	2.00 × 10−2	↑

Direction: ↑ = DAC-upregulated gene. Genes ranked by P-value (log-rank test). These associations are observational and derived from retrospective cohorts.

Figure 7.

Differential TFAP2E methylation between ER/PR-positive breast cancer and TNBC. Boxplot showing TFAP2E promoter DNA methylation (β values) in ER/PR-positive breast cancer and triple-negative breast cancer (TNBC) samples. TFAP2E methylation levels differ significantly between subtypes (Wilcoxon test, p = 3.30 × 10⁻³).

Figure 8.

Association of TFAP2E methylation and expression with overall survival in TNBC Kaplan–Meier survival curves for TNBC patients stratified by TFAP2E promoter methylation (left) and TFAP2E expression (right). No significant differences in overall survival were observed between high and low methylation groups (log-rank P = .32) or between high and low expression groups (log-rank P = .92).

Association of Survival-Associated Genes with Tumor Stage

Using the Nottingham Prognostic Index (NPI), TNBC and Luminal-A samples were stratified into early-stage (NPI < 3.4) and late-stage (NPI ⩾ 3.4) groups. In TNBC, among the survival-associated genes, APOBEC3G (P = .287; Figure 9A) and CA12 (P = .023; Figure 9B) showed higher expression in late-stage tumors, with GNG2 showing the strongest association (p = 5.21 × 10⁻⁴; Figure 9C). HLA-G trended higher without significance (P = .147; Figure 9D), while SH2D3C showed significant stage-associated expression (P = .007; Figure 9E). In Luminal-A BC, multiple genes showed significant stage-dependent differences in expression. Among the top 5 most significant, FCGBP (p = 2.20 × 10⁻¹¹; Figure 10A), LRP8 (p = 6.63 × 10⁻¹¹; Figure 10B), P3H2 (p = 1.44 × 10⁻¹³; Figure 10C), PHYHD1 (p = 2.49 × 10⁻²¹; Figure 10D), and SORBS2 (p = 1.83 × 10⁻¹⁷; Figure 10E) showed the largest stage-associated expression changes. Additional genes, including RNF213, NDRG1, TRAPPC9, SLC17A9, MTHFD1L, and others, also showed significant differences (adjusted P < .05). These results reveal subtype-specific stage associations: limited in TNBC, broader in Luminal-A.

Figure 9.

Stage-associated expression of JIMT-1-derived survival-related genes in TNBC. Boxplots showing expression levels in TNBC samples stratified by tumor stage (early vs late) based on the Nottingham Prognostic Index: (A) APOBEC3G, (B) CA12, (C) GNG2, (D) HLA-G, and (E) SH2D3C. Statistically significant P-values are indicated.

Figure 10.

Stage-associated expression of survival-related genes in Luminal-A breast cancer. Boxplots showing expression levels of the top 5 differentially expressed survival-associated genes in Luminal-A breast cancer samples stratified by tumor stage (early vs late) based on the Nottingham Prognostic Index: (A) FCGBP, (B) LRP8, (C) P3H2, (D) PHYHD1, and (E) SORBS2. All genes show significant differential expression between early- and late-stage tumors.

Discussion

Breast cancer is a heterogeneous disease in which molecular subtype determines tumor biology, therapy response, and prognosis.^1,2 Although inherited susceptibility contributes to breast cancer risk, genetic variation alone does not fully account for disease diversity.^28,29 Epigenetic regulation, particularly DNA methylation, adds a reversible layer that controls lineage identity, tumor suppressor silencing, and therapy adaptation. ³ DNA methyltransferase inhibitors (DNMTis) such as decitabine (DAC) have therefore been explored as epigenetic therapies.^{4 -7} Yet their effects are context dependent, shaped by baseline methylation, transcription factor networks, and chromatin state. ³

We addressed this by analyzing subtype-specific DAC-induced methylation and transcriptional remodeling in JIMT-1 (HER2-amplified, trastuzumab-resistant) and Luminal-A (T-47D) cells. It is important to note that while JIMT-1 has been widely used as a model of aggressive breast cancer, it does not fulfill the classic definition of TNBC. JIMT-1 harbors HER2 gene amplification but exhibits intrinsic resistance to trastuzumab due to alterations in downstream signaling pathways, resulting in a TNBC-like phenotype characterized by therapeutic resistance and aggressive behavior. Consequently, direct extrapolation of JIMT-1 epigenomic data to clinical TNBC cohorts from TCGA should be interpreted with caution, as HER2 amplification and trastuzumab resistance may modulate epigenetic responses to DAC independently of ER/PR status. Future studies comparing JIMT-1 responses with HER2-positive non-responder clinical cohorts would strengthen the translational relevance of these findings.

TFAP2E illustrated this context dependence: it showed clear promoter demethylation and transcriptional activation in JIMT-1 cells, consistent with the known role of AP-2 factors in breast cancer signaling, ⁹ but no prognostic association in patient data. Thus, DAC responsiveness does not always imply survival relevance. Conversely, in Luminal-A cells, KRT20 was upregulated despite promoter hypermethylation, reflecting complex regulation by enhancers or transcription factor control rather than promoter methylation alone. ³⁰ These results align with prior findings that luminal tumors, governed by hormone-responsive transcriptional networks, exhibit more flexible methylation–expression coupling. ³

An important mechanistic consideration that may partly explain the observed subtype-specific transcriptomic and epigenomic differences is the differential metabolic activation of DAC between JIMT-1 and T-47D cells. DAC is a nucleoside analog prodrug that requires cellular uptake and subsequent phosphorylation by deoxycytidine kinase (DCK) to its active triphosphate form, which is then incorporated into DNA during S-phase replication, forming covalent adducts with DNA methyltransferases (DNMTs) that trap and deplete them. Consequently, the magnitude and specificity of DAC-induced demethylation are critically dependent on the basal expression and enzymatic activity of DCK within each cell type. Notably, Buocikova et al, ⁸ the source of the cell line data used in this study, specifically reported that DCK overexpression modulates decitabine-induced transcriptomic reprograming in breast cancer cell lines, highlighting DCK as a key determinant of DAC responsiveness. It is therefore plausible that differences in basal DCK expression and activity between JIMT-1 and T-47D cells contribute to the observed differences in the number of hypomethylated CpGs (1195 vs 1937) and differentially expressed genes (187 vs 248). Whether the subtype-specific epigenomic responses identified here reflect purely epigenetic regulatory differences or are partly driven by pharmacokinetic variability in DAC metabolism warrants further investigation, ideally through direct measurement of DCK expression and DAC incorporation efficiency in both cell lines. ⁸

A further observation that requires mechanistic explanation is the paradoxical hypermethylation at certain loci despite DAC treatment, ³¹ a phenomenon we term “HyperUp” when accompanied by transcriptional upregulation. While DAC is primarily regarded as a demethylating agent, hypermethylation under DNMT inhibition is not without biological precedent and may reflect several non-mutually exclusive mechanisms. First, DAC incorporation into DNA during S-phase generates covalent DNA-DNMT adducts that constitute genotoxic lesions, triggering activation of DNA damage response pathways including homologous recombination (HR) and nucleotide excision repair (NER). These repair processes involve de novo chromatin remodeling and may paradoxically recruit de novo methyltransferases such as DNMT3A and DNMT3B to repair sites, resulting in compensatory or aberrant methylation at specific loci. Second, at higher doses, DAC can exert dose-dependent cytotoxicity beyond epigenetic reprograming, inducing replicative stress and global chromatin reorganization that may redistribute methylation marks nonuniformly. Third, transcriptional upregulation at hypermethylated loci, such as KRT20, in T-47D cells likely reflects regulation by distal enhancer elements or pioneer transcription factors that act independently of proximal promoter methylation status. Collectively, the HyperUp profile observed in this study should be interpreted not solely as a direct epigenetic regulatory mechanism but also as a potential consequence of genotoxic stress responses and DNA repair-associated chromatin remodeling induced by DAC treatment. Future studies incorporating genome-wide chromatin accessibility assays such as ATAC-seq alongside methylation profiling would help distinguish these mechanisms.^25,31

In JIMT-1 cells, DAC-activated genes were enriched for cytokine signaling, cadherin pathways, and piRNA-mediated epigenetic silencing (Figure 3A and B), suggesting the engagement of immune and transposable-element regulatory mechanisms. In Luminal-A cells, enriched pathways involved extracellular matrix organization, collagen dynamics, and piRNA processing (Figure 4A and B),^{16 -19} indicating structural remodeling and epigenetic silencing programs following DAC exposure. These findings support the idea that epigenetic modulation engages subtype-specific signaling axes and can potentially sensitize tumors to targeted or combination therapies.^{4 -7,32 -35} Additionally, integrating DAC responses with age-associated CpGs showed that DAC selectively demethylates sites that typically gain methylation with age, rather than globally reversing “biological aging.” Clock CpGs, including those at TFAP2E, are better viewed as elements of regulatory aging programs that DNMT inhibition can disrupt.^{10 -15} Variation in the methylation of these loci across TNBC tumors suggests differing baseline “age-program” activation and variable responsiveness to epigenetic therapy. This heterogeneity may inform biomarker-based patient selection for DNMTi treatment. ³⁶

Linking molecular changes to clinical outcomes revealed that DAC-responsive genes showing survival associations differed by subtype, though these relationships are observational and derived from retrospective cohorts. In ER/PR-positive disease, survival-linked genes such as CALML3, KRT5, and CAV1 align with known roles in hormonal and stromal signaling,^{23 -25} but their association with survival in METABRIC does not imply that DAC-induced upregulation of these genes directly improves prognosis. Similarly, in the JIMT-1-derived gene set, outcome-associated genes, including GNG2, F2RL1, APOBEC3G, and HLA-G, implicate immune and signaling pathways, ²² yet these findings are correlative and require functional validation in DAC-treated in vivo models before therapeutic conclusions can be drawn. TFAP2E itself showed subtype-specific methylation differences but no association with survival, further illustrating that epigenetic responsiveness does not necessarily confer prognostic value and highlighting the disconnect between in vitro drug responses and clinical outcomes. These findings should be interpreted in the context of the study’s limitations, which are discussed in detail in Section 5, including the use of single cell lines per subtype, reliance on publicly available datasets, and the retrospective nature of the survival analyses. Nonetheless, these results highlight the subtype-specific nature of DAC’s action and suggest that integration with other therapies could enhance effectiveness.^8,37

Limitations

Several limitations of this study warrant consideration. Firstly, each breast cancer subtype was represented by a single cell line, such as JIMT-1 for the HER2-amplified, trastuzumab-resistant model and T-47D for Luminal-A, which limits the generalizability of the epigenomic findings to the broader heterogeneity of clinical subtypes. Intra-subtype epigenetic variation, which is substantial in both TNBC and Luminal-A breast cancers, cannot be captured by single-cell-line analyses. Ideally, validation in additional cell lines such as MDA-MB-231 for TNBC and MCF-7 for Luminal-A would strengthen the robustness of these findings. In the absence of such experimental validation, the translational claims of this study should be interpreted with caution, and the identified DAC-responsive gene signatures should be regarded as hypothesis-generating rather than as definitive subtype-specific epigenetic responses. Secondly, this study relied entirely on publicly available cell line and patient-level datasets, without independent experimental validation of key findings. The absence of enhancer-level methylation data further limits mechanistic interpretation, as gene regulation by distal regulatory elements cannot be assessed from promoter-focused methylation arrays alone. Thirdly, the survival associations identified in the METABRIC and TCGA cohorts are retrospective and observational. These associations do not imply causality and should not be interpreted as evidence that DAC treatment directly modulates patient survival. Prospective studies and DAC-treated clinical cohorts are required to validate the prognostic relevance of the identified gene signatures.

Conclusion

DAC induces broad but subtype-dependent epigenomic and transcriptional reprograming in breast cancer. TNBC cells display coherent promoter hypomethylation–activation coupling, while Luminal-A cells show heterogeneous regulation. Subtype-distinct pathways, such as cytokine signaling and piRNA-mediated silencing in JIMT-1 cells and extracellular matrix remodeling and collagen dynamics in Luminal-A, reflect how epigenetic therapy interacts with tumor-specific signaling. DAC also selectively alters age-associated CpGs, indicating targeted disruption of regulatory aging programs rather than overall epigenetic rejuvenation. Integrating these findings with patient data revealed divergent prognostic associations across subtypes, emphasizing that epigenetic responsiveness and clinical significance are not equivalent. These insights support a model in which DNMT inhibition remodels the cancer epigenome through context-specific networks and reinforce the need for subtype-aware evaluation, functional validation, and biomarker-guided translation of epigenetic therapies in breast cancer.^{4 -7,37}

Supplemental Material

sj-docx-1-cix-10.1177_11769351261452670 – Supplemental material for Decitabine Induces Subtype-Specific Epigenomic Remodeling and Perturbs Age-Associated Regulatory CpGs in Breast Cancer

Supplemental material, sj-docx-1-cix-10.1177_11769351261452670 for Decitabine Induces Subtype-Specific Epigenomic Remodeling and Perturbs Age-Associated Regulatory CpGs in Breast Cancer by Areez Shafqat, Itika Arora, Arshiya Akbar, Mohammed Alfuwais, Safiah Aldubaisi, Mohammad Imran Khan, Ahmed Abu-Zaid, Firoz Ahmed and Ahmed Yaqinuddin in Cancer Informatics