Distinct epigenetic alterations and accelerated mitotic aging define second primary breast cancers Authors: Andrés F Bedoya-López, Javier I J Orozco, Pere Llinàs‑Arias, Miquel Ensenyat-Mendez, Betsy J Valdez, Sandra Iñiguez‑Muñoz, Mar Morote-Llabrés, Chikako Matsuba, Matthew P Salomon, Diego M Marzese, Melanie Goldfarb Published: 2026 ABSTRACT AbstractAs cancer survivorship is steadily increasing, second primary breast cancers (SPBCs) are becoming a more common disease linked to poorer clinical outcomes compared to first primary breast cancers (FPBCs). While the molecular features of FPBCs are well characterized, the biological drivers of SPBCs remain largely unexplored. We performed genome-wide DNA methylation profiling and analysis of clinically and demographically paired ER-positive, HER2-negative SPBC (n = 13) and FPBC (n = 29). We validated our findings by integrating genomic, transcriptomic, proteomic, and epigenomic data from paired tumors from The Cancer Genome Atlas. Chromatin Immunoprecipitation sequencing maps from ENCODE were used to examine epigenetic alterations of transcription factor binding sites (TFBSs) linked to accelerated aging and malignancy traits in SPBC. Epigenetic aging was assessed using the mitotic clock epiTOC. Here, we show that SPBC tumors are associated with poorer prognosis compared to FPBC. Epigenetic profiling identified 11,321 differentially methylated CpG sites compared to FPBCs, with hypermethylation enriched at gene promoters involved in morphogenesis, hormone signaling, and cell adhesion processes. Concordantly, transcriptomic analysis confirmed a coordinated downregulation of hormone signaling pathways, including estrogen, progesterone, and androgen cascades. Beyond promoter-level changes, SPBC tumors exhibit altered DNA methylation at key TFBSs involved in accelerated aging and cancer traits and significantly elevated mitotic age. SPBC tumors exhibit recurrent molecular scars, defined by suppressed endocrine signaling and accelerated mitotic aging. These epigenetic hallmarks may reflect the cumulative effects associated with prior malignancy and its clinical context, potentially contributing to more aggressive disease behavior. Our findings provide a foundation for developing personalized risk stratification and therapeutic approaches for survivors at risk of SPBC.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-54795-4.Subject terms: Biomarkers, Cancer, Genetics, Molecular biology, Oncology Introduction With approximately 18.6 million cancer survivors currently living in the United States, a number expected to rise in the coming decades, there is a growing recognition of the long-term consequences of cancer survivorship, including the development of a second primary malignancy1. This risk is particularly relevant among younger cancer survivors, who have a longer post-treatment life expectancy and often receive intensive therapies such as chemotherapy2 and radiation3 for their primary malignancies. Studies of large cohorts of childhood cancer survivors reported a 4- to 6-fold increased risk of second primary malignancies compared to the general population4–7, with breast cancer being the most common8. As survival rates improve, understanding the clinical and molecular characteristics of second primary breast cancers (SPBCs) has become increasingly important9. Population-based studies have demonstrated that SPBCs are associated with worse overall and breast cancer-specific survival compared to age-matched patients with their first primary breast cancer (FPBC)10–13. In addition, women with SPBC also exhibit higher cardiovascular risk14, reflecting the complexity of the underlying complications in long-term survivors. In contrast to FPBCs, where molecular determinants of prognosis and prediction of therapy response are well studied15–19, data on the molecular alterations of SPBC remain scarce. Prior studies in SPBC after Hodgkin’s lymphoma have demonstrated increased microsatellite alterations and loss of heterozygosity in several chromosomes compared to patients with FPBC20, but comprehensive molecular characterization is still lacking. These alterations, along with increased aging of patients undergoing SPBCs, may result in a landscape of women with accumulated genetic mutations, prolonged exposure to hormones, and changes in breast tissue, which ultimately contribute to an accelerated aging process21. Epigenetic modifications have been found to have a strong relationship with aging, leading to the development of epigenetic clocks, which are mathematical models that estimate chronological age based on DNA methylation22–24. However, as shown by Horvath and others, these clocks primarily correlate with aging when measured in blood and are not universally accelerated in breast cancer25–28. In contrast, mitotic epigenetic clocks, based on CpG sites prone to progressively become methylated after each cell division, track the cumulative number of cell divisions and reveal accelerated mitotic aging in both breast tumors and adjacent normal tissue. This supports Vogelstein and Tomasetti’s hypothesis that increased mitotic age is a major cancer risk factor29,30. In an integrated analysis using transcriptomic and epigenomic data from the Cancer Genome Atlas (TCGA), we previously showed that patients with estrogen-receptor (ER)-positive, HER2-negative lobular and ductal invasive SPBC harbor unique molecular alterations compared to FPBC31. However, the biological basis underlying these differences remains unclear. Given the poorer clinical outcomes of SPBC and the lack of mechanistic understanding behind its distinct behavior, a deeper exploration of its epigenetic landscape is urgently needed to uncover drivers of tumor progression and potential therapeutic vulnerabilities. In this study, we profiled genome-wide DNA methylation in clinically well-annotated BRCA1/2–wild-type patients with ER-positive, HER2-negative SPBCs and FPBCs. We integrated our data with epigenomic, transcriptomic, proteomic, and genomic data from a clinically and demographically paired cohort of SPBC and FPBC patients enrolled in TCGA. We show that SPBC tumors have a worse prognosis than FPBCs and are characterized by recurrent and reproducible DNA methylation differences, marked by epigenetic alterations at promoter and enhancer-associated CpG sites of genes involved in hormone signaling. These changes are accompanied by concordant transcriptional repression of estrogen receptor, progesterone receptor (PR), and androgen receptor (AR) signaling cascades, independently of somatic mutations. Additionally, SPBC tumors also exhibit methylation changes at transcription factor binding sites (TFBSs) of factors linked to aging and cancer traits, as well as epigenetic marks associated with elevated mitotic age, indicating a more extensive proliferative history. Together, our findings suggest that SPBC tumors exhibit characteristic epigenetic features that may contribute to endocrine therapy resistance and aggressive disease behavior. Methods Patient selection criteria Given histologically confirmed invasive ductal carcinoma, female patients with ER-positive/HER2-negative tumors were selected from the Providence Saint John’s Health Center database. Only tumors from patients with BRCA1/2 wild-type were included. Exclusion criteria were male patients, receipt of neoadjuvant therapy, or ER-negative and/or HER2-positive disease. Regarding disease classification, patients were classified as having a first primary breast cancer (FPBC) or a second primary breast cancer (SPBC) according to the SEER Multiple Primary and Histology Coding Rules and the Maastricht Delphi Consensus32. Briefly, SPBC was defined as a new primary breast tumor meeting at least one of the following criteria: occurrence in the contralateral breast; a tumor with distinct histology compared with the first primary; a non-contiguous tumor in the same breast not attributable to direct extension; or an invasive carcinoma diagnosed more than 60 days after a prior in situ lesion in the same breast. The final cohort included 42 breast tumor samples, 13 SPBC, and 29 FPBC (Supplementary Table 1). Ethics oversight All clinical-demographic data and patient-derived samples were collected under research protocols approved by the Institutional Review Board of Providence Health and Services (IRB Protocol: #JWCI-18-0401). The experiments were performed following the World Medical Association Declaration of Helsinki and the National Institutes of Health Belmont Report. Tissues were de-identified following the Health Insurance Portability and Accountability Act (HIPAA) regulations. Tissue processing and DNA purification Collected formalin-fixed paraffin-embedded (FFPE) tumor blocks were reviewed by a breast pathologist to identify tumor-enriched areas. After deparaffinization, serial 10 to 15 μm-thick tissue sections were hematoxylin-stained, and tumor tissue was needle microdissected. Genomic DNA was isolated using the Quick-DNA FFPE Miniprep kit (#D3067; Zymo Research, CA.). Genomic DNA was quantified by Qubit 3.0 Fluorometer (Q33216; Thermo Fisher Scientific, CA). Genomic DNA was converted with Sodium bisulfite using EZ DNA Methylation-Direct (#D5021, Zymo Research, CA), following the manufacturer’s recommendations. DNA methylation processing Infinium MethylationEPIC 850 K BeadChip arrays (Illumina, Inc.) were used to generate DNA methylation profiles according to the manufacturer’s protocol. β-values represent the ratio of the intensity of the methylated (M) bead types to the total intensity of both methylated and unmethylated (U) bead types at each CpG site. They can be calculated using the equation β = [M / (M + U)]. Standard normalization and quality control procedures were applied to the DNA methylation data using the R package ChAMP (v.2.32.0)33. Raw intensity data were imported and further processed using default settings. Beta-Mixture Quantile (BMIQ) normalization was applied to normalize the array data, and the ComBat function was used to correct batch effects. Quality control showed 38,968 probes with a detection P-value above 0.01, which were therefore removed. The non-CG labeled probes (n = 2,250) or probes containing single-nucleotide polymorphisms (n = 93,335) were removed from downstream analyses34. Additionally, 65 probes targeting chromosome Y were also discarded. Tumor purity was assessed using the leukocyte unmethylation for purity (LUMP) method described by Aran et al.35. The obtained score ranges from 0 to 1, with values closer to 1 indicating higher tumor purity. TCGA-BRCA data access and normalization Matched clinical and molecular data from the TCGA Breast Cancer project (TCGA-BRCA) were obtained from the National Cancer Institute Genomic Data Commons portal using the TCGAbiolinks R package (v.2.28.4)36. These datasets included DNA methylation data (HumanMethylation450 array, Illumina), transcriptome profiling (RNA sequencing), protein expression (Reverse Phase Protein Array), copy number alterations (Genotyping array), and single-nucleotide variations (SNV; Whole-exome DNA sequencing). To identify SPBC cases, we filtered patients whose barcodes were associated with prior malignancies. We further refined the selection to only include ER-positive/HER2-negative invasive ductal tumors. Samples with low tumor purity (CPE < 0.5) were excluded. Given the limited number of SPBC patients in the TCGA-BRCA cohort and to ensure a robust comparison with FPBC cases, we employed the MatchIt (v.4.5.5) R package to select a subset of FPBC cases with clinical characteristics similar to the SPBC cohort. Random matching was performed based on age at diagnosis, tumor size, grade, AJCC stage, and tumor purity, maintaining a 1:2 SPBC-to-FPBC ratio mimicking our cohort. This resulted in a matched cohort of 14 SPBC cases and 28 FPBC cases (Supplementary Table 2). Cellular deconvolution via MethylCIBERSORT Estimated cell subpopulation proportions were derived from normalized DNA methylation β-values using the methylCIBERSORT (v.0.2.0) R package and the breast_v2_Signature as the DNA methylation signature matrix37. Analysis parameters included disabling quantile normalization and 1000 permutations per run. DNA methylation analyses Relative average β-values of each CpG site were compared between the SPBC and FPBC groups. All CpG sites with a differential mean β-value ≥ 15% and a P-value < 0.05 were considered differentially methylated sites (DMS). Given tumor heterogeneity, limited sample size, and probe intercorrelation in high-density DNA methylation arrays, we used nominal P-values to define DMS, an approach aimed at maximizing biological interpretability and reproducibility. For TCGA analyses, CpG sites with a differential mean β-value ≥ 10% and a P-value < 0.05 were considered DMS. CpG island terms and genomic positions were explored in each probe using the improved DNA methylation Illumina EPIC probe annotation released by the National Cancer Institute34. Identified differentially methylated regions (DMRs) were analyzed using the DMRcate R package (v.2.16.1) with default parameters38. Regions were considered DMRs if they showed an adjusted P-value < 0.05, corrected for multiple testing using the Benjamini–Hochberg False Discovery Rate method. To ensure biological relevance, we further required that each DMR contained at least one CpG with an absolute differential β-value > 15%. Identified DMRs were subsequently annotated to genomic features, using the ChIPseeker R package (v.1.38.0) together with the TxDb.Hsapiens.UCSC.hg19.knownGene annotation database (v.3.2.2)39. Networks of expression–methylation quantitative trait loci (emQTLs) linked to estrogen response–associated processes were obtained from previously published studies conducted by the Fleischer et al. laboratory40,41. The missMethyl R package (v. 1.36.0) was used to test Gene Ontology (GO) pathways enriched in epigenetic changes42. GO biological process gene sets (n = 7,608) were sourced from the MSigDB database via the msigdbr R package (v.7.5.1). Functional enrichment analyses were then simplified and clustered into key functional categories using the simplifyEnrichment R package (v.1.12.0)43. Briefly, GO semantic similarity matrices were computed for pathways enriched for hypermethylated DMS using the GO_similarity function with default parameters. The resulting GO similarity scores were subsequently clustered into primary functional groups using the Markov Clustering algorithm (MCL). Gene expression, single-nucleotide variation, and copy number analyses RNA sequencing data from the TCGA-BRCA were processed using the DESeq2 R package44. Genes with 10 reads or fewer were discarded. Differentially expressed genes (DEGs) were defined as an absolute log2 fold change > 0.5 and an adjusted P-value < 0.05. The gene enrichment analyses were conducted using the clusterProfiler R package (v.4.8.3)45. Copy number variations were categorized into gain (copy number > 2), diploid (copy number = 2), or loss of heterozygosity (copy number = 1). AR copy number data were not available for analysis. The maftools R package (v.2.18.0) was employed to analyze SNV data46. Tumor mutational burden (TMB) was calculated as the number of mutations per Mb sequenced. Survival analyses Survival analyses were performed using the TCGA-BRCA cohort to compare SPBC and FPBC tumors. Analyses were conducted with the survival R package (v3.5.8), using relapse-free survival (RFS) as the primary endpoint. Estimation of DNA methylation-based epigenetic-mitotic age score To estimate epigenetic and mitotic age in breast cancer tissues, we employed the Epigenetic Timer Of Cancer (epiTOC) model29, a DNA methylation-based, age-correlative model that approximates a mitotic clock in cancer tissues. The epiTOC/pcgtAge score reflects the lifetime accumulation of DNA methylation replication errors from the ground fetal state, providing an estimate of mitotic history. The score was adjusted for age at diagnosis by fitting a linear model regressing the pcgtAge score on age, and the resulting residuals were used as age-adjusted epiTOC values. Analysis of DNA methylation at transcription factor binding sites We used Chromatin Immunoprecipitation sequencing (ChIP-seq) data from the ENCODE project47 to explore methylation patterns at transcription factor binding sites, which have been previously associated with aging and cancer processes48. All data analyzed were retrieved from experiments using the ER-positive/HER2-negative MCF-7 cell line, except for EZH2, for which data from human mammary epithelial cells (HMEC) were used due to the unavailability of MCF-7-derived data. We obtained the ChIP-seq peak tracks from the ENCODE portal using the following identifiers: EP300 (ENCFF167FQT), FOXA1 (ENCFF465LTH), FOS (ENCFF282FWZ), JUN (ENCFF242UOB), JUND (ENCFF450KFZ), SUZ12 (ENCFF739TYI), and EZH2 (ENCFF224GAI). The intersection between ChIP-seq peaks and DNA methylation data for each CpG site was performed using the GenomicRanges R package (v.1.54.1)49. Peaks with scores lower than 200 were excluded from further analysis to minimize background signal arising from weak reproducible peaks. SUZ12 analysis was inconclusive due to the limited number of CpGs that met these criteria, with fewer than 200 CpG sites satisfying the conditions. Data processing and visualization Heatmap was used to visualize hierarchical cluster analyses based on Euclidean distance using the R package gplots (v.3.1.3.1). Phylogenetic trees with Spearman ρ distances between DNAm profiles were generated using the FigTree software version 1.4.4 (Andrew Rambaut Lab, University of Edinburgh). Principal component analysis (PCA) was performed using the R package M3C (v.1.24.0). For other data representation, the pheatmap (v.1.0.12), ggplot2 (v.3.5.1), and ggpubr (v.0.6.0) R packages were used. The UpSetR (v.1.4.0) package was used to represent the intersection between the enriched pathways in our cohort and the TCGA-BRCA comparisons50. Kaplan–Meier survival curves were generated and visualized using the survminer (v.0.5.0) R package. For data management, the collection of R packages grouped on tidyverse (v.2.0.0) was used. Statistics and reproducibility Categorical variables were compared using Fisher’s exact test. Continuous variables were assessed for normality using the Shapiro-Wilk test; variables following a normal distribution were compared using the two-sided Student’s t-test, whereas non-normally distributed variables were analyzed using the two-sided Wilcoxon rank-sum test. P-values adjusted for multiple comparisons are corrected for multiple testing using the Benjamini-Hochberg False Discovery Rate (FDR) method and indicated with “adjusted”; all other P-values are not adjusted for multiple comparisons. Given the exploratory nature of the study and sample availability, formal a priori power calculations were not performed. Pairwise Euclidean distances in PCA space were compared between groups using a permutation test (10,000 permutations, two-sided) based on the difference between mean inter-group and intra-group distances. Most variable sites were calculated based on the standard deviation of each CpG site. Odds ratios (ORs) with 95% confidence intervals (CIs) were calculated using the questionR (v.0.7.8) R package, with derived P-values obtained using Fisher’s exact test. For genomic enrichments, each set of probes included on the Infinium MethylationEPIC 850 K BeadChip array was used as the background, thereby correcting the inherent coverage bias of the platform toward specific genomic regions. Kaplan–Meier curves were estimated and compared using the log-rank test. Unless otherwise specified, all computational analyses were performed using R software (v.4.3.3). Results Baseline clinicopathologic features of the second and first primary breast cancer cohorts We evaluated tumor samples from a total of 42 female patients with BRCA1/2 wild-type, and ER-positive, HER2-negative invasive ductal carcinomas. Among these, 29 (69.1%) patients had no history of previous malignancy (first primary breast cancer; FPBC), and 13 (30.9%) patients had a prior tumor before developing their second primary breast cancer (SPBC). Overall, the median age (interquartile range) at diagnosis was 66.5 years (14.5), patients predominantly had a normal body mass index (BMI) with a median of 22.8 kg/m2 (5.1), and the majority of tumors were histological grade 2 (61.9%). The two cohorts were generally comparable, with no significant differences in age (P-value = 0.20), BMI (P-value = 0.79), tumor size (P-value = 0.74), tumor grade (P-value = 0.79), or lymph node positivity (P-value = 1.00), indicating similar clinicopathological features between groups (Table 1, Supplementary Table 1). The prior malignancies among the SPBC group were distributed as follows: 8 patients (61.5%) had a previous history of breast cancer, 4 patients (30.8%) had endometrial cancer, and 1 (7.7%) had peritoneal cancer. Table 1Comparison of demographic and clinicopathologic features of the Saint John’s Cancer Institute cohort according to history of previous malignancy status.All patientsFPBCSPBCP-value(n = 42)(n = 29)(n = 13)Age at diagnosis (years)0.20 2Median (IQR)66.5 (52–81)62.0 (46–78)68.0 (61–75)Body mass index (kg/m2)0.79 2Median (IQR)22.8 (17.7–28.0)23.0 (16.5–29.4)22.7 (19.5–26.0)Grade0.79 3I9 (21.4%)6 (20.7%)3 (23.1%)II26 (61.9%)17 (58.6%)9 (69.2%)III6 (14.3%)5 (17.2%)1 (7.7%)NA1 (2.4%)1 (3.4%)0 (0.0%)Positive lymph nodes1.00 3030 (71.4%)20 (69.0%)10 (76.9%)110 (23.8%)7 (24.1%)3 (23.1%)22 (4.8%)2 (6.9%)0 (0.0%)Tumor size (cm)0.74 1Mean (SD)1.51 (0.5)1.53 (0.4)1.46 (0.7)Tumor purity0.55 2Median (IQR)0.68 (0.54–0.82)0.69 (0.59–0.79)0.68 (0.47–0.89)(1) Student’s t-test.(2) Wilcoxon Rank-Sum Test.(3) Fisher’s Exact Test. Comparison of demographic and clinicopathologic features of the Saint John’s Cancer Institute cohort according to history of previous malignancy status. (1) Student’s t-test. (2) Wilcoxon Rank-Sum Test. (3) Fisher’s Exact Test. Genome-wide DNA methylation profiling reveals distinct epigenetic landscapes in second primary breast cancers After confirming similar tumor purity between SPBC and FPBC tumors (P-value = 0.55; Table 1), we next evaluated whether slight differences in immune or stromal composition could influence DNA methylation patterns. Deconvolution of cell-type proportions revealed largely comparable immune and stromal landscapes between FPBC and SPBC tumors (Supplementary Fig. 1). While significant but modest differences were detected, including lower proportions of regulatory T cells, fibroblasts, and neutrophils, and higher CD4⁺ effector T cells in SPBC (P-value < 0.05), these changes were quantitatively small and did not indicate broad shifts in cellular composition. Comparing genome-wide epigenomic profiling, we identified a total of 11,321 differentially methylated sites (DMS) between the groups, including 6,990 hypomethylated and 4,331 hypermethylated regions in SPBC (P-value < 0.05; Fig. 1a). Among these, 1,743 DMS remained significant after FDR correction (P-adjusted < 0.05; Supplementary Table 3). DNA methylation differences between SPBC and FPBC tumors are not randomly distributed. (a) Volcano plot showing the differential mean β-value and the inverse log10 P-value. 11,331 DMS were identified in SPBC compared to FPBC, including 4,341 hypermethylated and 6,990 hypomethylated sites. (b) Heatmap of hierarchical clustering based on DNA methylation patterns of the 11,331 differentially methylated sites (DMS). The color-coded annotation indicates the tumor group (FPBC or SPBC) and, within SPBC, the prior history of carcinoma. (c) Forest plot displaying the odds ratio with 95% confidence intervals for hypomethylated and (d) hypermethylated DMS in SPBC versus FPBC across different CpG genomic regions. Enrichment was calculated relative to the full distribution of probes represented on the Illumina MethylationEPIC array, which was used as the background for all comparisons. Unsupervised hierarchical clustering and phylogenetic analyses based on nominal-DMS separated SPBC from FPBC (Fig. 1b), supporting the presence of distinct epigenomic signatures. Notably, SPBC tumors formed mixed clusters regardless of the site of the first cancer (Fig. 1b, Supplementary Fig. 2a). When using the 5,000 most variable CpG sites genome-wide, PCA revealed a tendency for SPBC tumors to segregate from FPBCs (Supplementary Fig. 2b). Restricting the visualization to SPBC cases showed no strong separation between tumors arising after prior breast cancer and those after non–breast malignancies, as confirmed by permutation testing of Euclidean distances derived from the first five principal components (Supplementary Fig. 2c-d), suggesting comparable methylation features between these SPBC subgroups. To further increase the robustness of our findings, we expanded the analysis to the regional level. Comparison between SPBC and FPBC tumors identified 817 differentially methylated regions (DMR; P-adjusted < 0.05), of which 326 showed biologically relevant methylation changes, including 108 hypomethylated and 218 hypermethylated regions (Supplementary Table 4). Overall, these findings suggest that the SPBCs exhibit comparable DNA methylation patterns independently of the type of first primary tumor, supporting a shared epigenetic phenotype rather than tumor-specific inheritance. Next, we aimed to assess whether these differences were enriched in specific genomic regions. Notably, we found that hypomethylated DMS were primarily found in low-density CpG regions, including regions with low CpG content known as “open sea” and intergenic regions, while being depleted at CpG islands and gene promoters (Fig. 1c, Supplementary Fig. 2e). In contrast, hypermethylated CpG sites followed the inverse trend, with a predominant enrichment at high-CpG-density regions, particularly within gene promoters and the first exon of transcripts, while they were less frequent in open sea regions (Fig. 1d, Supplementary Fig. 2f). These findings indicate that the observed DNA methylation changes are not randomly distributed but instead follow specific genomic patterns. Identification of aberrantly methylated biological networks in second primary breast cancers Given the distinct DNA methylation patterns observed between SPBC and FPBC tumors, we investigated their potential molecular differences. Because methylation in CpG sites around gene promoters is usually involved in transcriptional regulation, we focused specifically on these sites. Therefore, we conducted GO analyses to identify enriched pathways and gene networks that may be affected by these epigenetic alterations. Using the subset of FDR-corrected DMS (n = 1,743), no pathways remained significant after multiple-testing correction (P-adjusted < 0.05). Therefore, to identify potentially affected networks, we employed the DMS selected based on the nominal P-value (n = 11,321). As expected, based on the promoter-enriched distribution of hypermethylated DMS, the analysis revealed a substantial number of affected pathways: 147 biological processes were significantly enriched (P-adjusted < 0.05) in SPBC tumors. The top enriched pathways were mainly involved in cell signaling and neuron-related processes (Supplementary Fig. 3a). To uncover the entire landscape, we grouped these enriched terms using the MCL algorithm, which identified three major biological themes: morphogenesis and developmental processes, hormone-related activity, and cell adhesion (Fig. 2a, Supplementary Table 5). Most pathways selected using this setup were grouped within the morphogenesis cluster, followed by hormonal processes, while cell adhesion was less represented. Accordingly, promoter-associated DMRs, which result from multiple-testing correction, captured the core pathway-level signal, predominantly overlapping genes within these clusters (Supplementary Fig. 3b). In this context, nominal DMS pathway enrichment serves as a complementary layer, revealing consistent patterns that align with the DMR-based results. Hypermethylation of hormone-related processes reduces gene expression and hormonal response in SPBC tumors. (a) Heatmap displaying the 147 biological processes enriched across hypermethylated DMS found in SPBC tumors (P-adjusted < 0.05). Based on their similarity index, these clusters were grouped into three main categories: cell adhesion, hormone and secretion, and morphogenesis and development. (b) Relapse-free survival rates for TCGA-BRCA SPBC and FPBC matched tumors. Statistical differences were assessed using the log-rank test. (c) Upset plot displaying the intersection of pathways enriched in CpG sites with higher methylation levels in SPBC tumors from our cohort (SJCI) and the TCGA dataset. (d) Barplot showing biological pathways enriched for hypermethylated regions that are also significantly downregulated at the transcriptomic level in the same SPBC tumors. (e) Gene set enrichment analysis (GSEA) revealing a significantly attenuated transcriptional response to estrogen, progesterone, and androgen signaling pathways in SPBC tumors. By contrast, and consistent with the observation that hypomethylation was largely excluded from promoter regions, the analysis of nominal DMS revealed fewer affected pathways, with only 9 biological processes significantly enriched (P-adjusted < 0.05). These included sensory perception and G protein-coupled receptor signaling pathways (Supplementary Fig. 3c). Together, these findings suggest that the SPBC epigenome exhibits distinct DNA methylation patterns that may influence its intrinsic molecular programs. Cross-cohort validation reveals epigenetic attenuation of hormone signaling in second primary breast cancers To validate our findings, we analyzed a matched cohort of 42 samples from the TCGA-BRCA cohort (28 FPBC, 14 SPBC) with similar clinicopathological traits. Importantly, despite this careful matching, patients with SPBC tumors exhibited significantly worse relapse-free survival compared with FPBC patients (Fig. 2b). This difference remained evident even when using the extended TCGA-BRCA cohort (Supplementary Fig. 4a), reinforcing the clinical importance of studying the molecular determinants behind the poor prognosis of SPBCs. By comparing the methylome of matched samples, we identified 4,455 DMS, including 3,711 hypomethylated and 744 hypermethylated CpG sites, likely reflecting a lower array coverage in the number of CpG sites (Supplementary Fig. 4b). Among these, only 93 of the 337,641 shared CpGs replicated directionally, indicating limited CpG-level concordance across platforms. Analysis of CpG sites with higher methylation levels in SPBC tumors revealed that while some pathways were uniquely enriched in our cohort, others remained consistently dysregulated across both datasets. Notably, cluster 2, primarily associated with hormone-related processes, remained significantly enriched across both cohorts (Fig. 2c). To evaluate the impact of these alterations on gene expression, we performed transcriptomic analysis using the same samples, identifying 86 downregulated and 137 upregulated genes in SPBC compared to FPBC (P-adjusted < 0.05; absolute log2 fold change > 0.5) (Supplementary Fig. 4c). Interestingly, GO analysis of downregulated genes depicted patterns consistent with DNA hypermethylation, showing a direct impact of these methylation marks on diminishing pathways related to hormonal regulation (P-value < 0.05; Fig. 2d). Conversely, upregulated genes were not enriched in any biological process, reinforcing our previous findings. With this evidence, we thoroughly analyzed our study cohort and observed an extensive gene network affected by DNA promoter hypermethylation, involving key mediators in steroid hormone signaling and metabolism. This network was directly linked to central regulators of hormone-driven, hormone-receptor-positive breast cancer, including ESR1, PGR, and AR (Supplementary Fig. 4d). These observations led us to hypothesize that SPBC tumors, despite being hormone-receptor-positive by immunohistochemistry (IHC), may exhibit attenuated transcriptional responses to steroid hormone signaling. Using TCGA transcriptomic data, we confirmed that SPBC patients exhibited a significant depletion in genes involved in response to estrogen, progesterone, and androgen (Fig. 2e). Leveraging clinical reports, we found that this was not cofounded by differences in menopausal status (P-value = 0.19; Supplementary Table 2) or ER and PR IHC staining percentages (Supplementary Fig. 5a). Consistently, analysis of proteomic data revealed no significant depletion of ER, PR, or AR protein levels, nor of their activated forms, in SPBC tumors (Supplementary Fig. 5b), indicating that altered hormone signaling in SPBCs is unlikely to be explained by reduced receptor abundance. To further explore the epigenetic basis of this hormonal response depletion, we specifically interrogated CpG sites previously defined as estrogen-responsive expression quantitative trait methylation loci (emQTLs)40,41. Strikingly, while direct CpG-to-CpG overlap across platforms was limited, CpGs belonging to this group consistently exhibited hypomethylation in SPBC tumors compared to FPBCs in both cohorts (Supplementary Fig. 6a). This, accompanied by our previous results, suggests a model in which hormonal-responsive regulatory networks remain aberrantly modulated by DNA methylation, thereby impairing hormonal-related transcriptional activity in SPBC carcinomas. In fact, we detected repression of multiple canonical hormone-responsive target genes associated with these emQTLs, including well-established downstream targets of estrogen and androgen receptor signaling, such as AREG and MSMB, respectively. In line with this observation, several luminal identity and hormone-responsive genes were significantly downregulated in SPBC tumors, including TSPAN1, IL20, GATA2, SLC30A8, BMPR1B, and GPER1. SPBCs also displayed coordinated deregulation of genes associated with epithelial plasticity, including KRT15, ZIC2, and GLI3, which, together, point towards a more plastic, hormone-unresponsive cellular state (Supplementary Fig. 6b). With these findings, we then explored if these differences were driven by somatic mutations51. By examining SNVs from the same cohort, we found no significant differences between SPBC and FPBC samples in any gene (Supplementary Table 6), as consistently shown by analyses of the top recurrently mutated genes and the hormone receptor genes ESR1, PGR, and AR (Supplementary Fig. 7a-b). Then, we assessed whether mutational differences might converge at the pathway level. Although several pathways displayed differences between SPBC and FPBC tumors (P-value < 0.05; Supplementary Table 7), including processes related to MAPK and protein kinase regulation, these alterations were not accompanied by mutations or emerging trends in key hormonal pathways or in known downstream mediators of estrogen, progesterone, or androgen signaling (Supplementary Fig. 6c). Likewise, comparing the widespread distribution of coding mutations between the groups, we found no differences as reflected by similar TMB (P-value = 0.32; Supplementary Fig. 7d). Ultimately, assessment of gene copy number data did not reveal systematic differences affecting ESR1 (P-value = 0.63) and PGR (P-value = 1.00) genes between FPBC and SPBC tumors (Supplementary Fig. 7e). Together with comparable mutation frequencies and tumor mutational burden, these data indicate that repression of hormone-related transcriptional programs in SPBC is unlikely to be driven by underlying genomic alterations. In the absence of major genetic alterations or receptor-level protein differences, these findings suggest that DNA methylation dysregulation may disrupt hormone-related signaling cascades, potentially contributing to reduced hormone responsiveness and endocrine resistance in SPBC tumors. Taken together, this integrative evidence indicates that cross-cohort validation is primarily supported at the regulatory and transcriptomic levels, rather than at the level of individual CpG sites. Methylome of second primary breast tumors reveals accelerated aging and increased tick rate We previously observed global hypomethylation in open sea regions from SPBC tumors (Fig. 1c). Based on our results and considering that these patients had undergone prior treatments, including chemotherapy, endocrine therapy, and/or radiotherapy (Supplementary Table 1), we hypothesized that it may reflect residual epigenetic scars from earlier therapeutic exposures or from the first primary tumor itself. Dysregulation of DNA methylation levels at a subset of transcription factor binding sites (TFBSs) has been identified as a readout of both aging and cancer progression48. In particular, TFBSs of differentiation and proliferation-associated transcription factors such as AP-1 (JUN, JUND, FOS), as well as pioneer and co-activator factors including FOXA1 and EP300, typically undergo progressive hypomethylation in cancer and aging, reflecting chromatin accessibility and transcriptional plasticity. Conversely, TFBSs associated with Polycomb-mediated repression, including those bound by EZH2, tend to accumulate DNA methylation marks in similar contexts. Based on this framework and consistent with our hypothesis, we observed distinct methylation patterns at the binding sites of most of these transcription factors, with SPBC tumors closely mirroring epigenetic shifts associated with cancer progression and advanced aging (Fig. 3a, b). Specifically, methylation patterns at TFBSs for the AP-1 subunits JUND, JUN, and FOS were significantly enriched for hypomethylated CpG sites. Similarly, FOXA1 and EP300 also exhibited pronounced hypomethylation. In contrast, EZH2 showed an increase in methylation at its TFBSs (Supplementary Fig. 8a). These differences remained evident even when using different thresholds of significance for the peak of the TFBSs, ensuring the robustness of the results (Supplementary Table 8). Altogether, these TFBS-specific methylation shifts closely resemble well-documented patterns observed in tumorigenesis and age-related processes, further supporting the notion that SPBC tumors recapitulate key molecular features of both processes. SPBC displays DNA methylation patterns associated with accelerated mitotic aging. (a) The average level of methylation at TFBSs for six transcription factors associated with aging and cancer phenotypes for both SPBC and FPBC tumors. Samples are ranked according to the average methylation level for each CpG site within TFBSs. Only CpGs with a P-value < 0.05 and an absolute differential methylation > 5% between SPBC and FPBC are represented. (b) Boxplots illustrating the Δβ distributions of FPBC and SPBC samples within TFBSs. The same criteria as above were used to select CpGs for representation. (c) Boxplot showing the pcgtAge score of FPBC and SPBC samples obtained using the epigenetic mitotic clock epiTOC. The Wilcoxon rank-sum test was used to test statistical differences. (d) Scatter plot with linear trend lines showing that SPBC samples exhibit higher pcgtAge levels across all ages compared to FPBC samples. To uncover the age-related epigenetic shift, and given that epigenetic clocks are not specifically designed for tumor tissue, and their potential remains suboptimal in solid tissue25–28, we applied a mitotic age-correlative epigenetic clock, “epiTOC” (Epigenetic Timer Of Cancer), which captures the cumulative number of cell divisions over time (pcgtAge)29. Interestingly, SPBC tumors showed significantly higher pcgtAge scores compared to FPBC tumors (P-value = 0.028; Fig. 3c). This difference remained after adjustment for patient age (P-value = 0.026; Supplementary Fig. 8b), indicating that cells from patients developing a second primary carcinoma may have undergone a higher number of mitotic divisions, resulting in a more aged epigenetic state and increased replicative history. Interestingly, when we examined the relationship between chronological and DNA methylation-derived age, we found that SPBC tumors consistently exhibited greater age acceleration across the entire age spectrum (Fig. 3d). Together, these findings suggest that SPBC tumors may harbor DNA methylation patterns reflective of accelerated biological aging and increased mitotic activity, potentially contributing to their more aggressive behavior and poorer outcomes. Discussion In this study, we demonstrate that second primary breast cancers (SPBCs), despite sharing ER-positive, HER2-negative histology with first primary breast cancers (FPBCs), possess distinct epigenetic and transcriptional features that may underline their more aggressive clinical behavior. Our data suggest that the development of a prior tumor, treatment exposure, and other related host- or tissue-level processes may have collectively contributed to shaping the SPBC epigenetic landscape. By integrating genome-wide DNA methylation profiling with genomic, transcriptomic, and proteomic analyses, we reveal that SPBCs are characterized by epigenetic dysregulation of hormone signaling pathways and accelerated mitotic aging. Importantly, all tumors included in our cohort were confirmed to be wild type for BRCA1/2. This strengthens the interpretation that the molecular alterations observed in SPBC reflect acquired epigenetic reprogramming rather than inherited genetic factors. A key finding of our study is the extensive promoter hypermethylation, which was not random but enriched in biologically meaningful pathways, particularly at genes involved in morphogenesis, hormone signaling, and cell adhesion processes, suggesting functional consequences for gene expression. These findings were validated using two complementary frameworks at both the single-site and regional levels. This was accompanied by epigenetically altered emQTLs, predominantly affecting distal regulatory elements associated with hormone-responsive transcription networks. Supporting this observation, gene expression data revealed repression of hormone receptor–driven signaling, including downstream targets of ER, PR, and AR. Importantly, this repression was not accompanied by differences at the protein level or IHC staining of these receptors and, in the absence of somatic mutations or copy number alterations in their respective genes, points to epigenetic dysregulation as a potential mechanism of endocrine resistance in SPBC. This observation is particularly relevant in light of the limited molecular knowledge of these tumors in the literature. Our findings have clinical implications since SPBC patients, many of whom have undergone prior cancer therapies, may harbor tumors with altered hormone responsiveness that are not captured by conventional IHC. This could partially explain the worse prognosis observed in SPBC, as these tumors may exhibit reduced sensitivity to standard endocrine treatments despite being classified as hormone receptor–positive10–13. Indeed, our results validated that SPBCs are associated with significantly poorer relapse-free survival, reinforcing the clinical relevance of distinguishing these tumors from FPBCs. Additionally, we acknowledge that additional factors, including shared genetic susceptibility and lifestyle-related exposures, may influence tumor susceptibility and the biological behavior of SPBCs. Interestingly, previous studies in MCF-7 cells showed that DNA methylation plays a role during tamoxifen resistance, mainly through pathways related to morphogenesis, development, and sensory perception. These findings align with our patient-derived data, suggesting these genes may contribute to hormone therapy resistance52. In any case, our study emphasizes the need for refined biomarker strategies in this population, potentially including an integration of gene expression, protein expression, and DNA methylation profiling assays to better predict therapeutic response. In addition to transcriptional silencing of hormone pathways, we observed that SPBC tumors exhibit widespread hypomethylation in intergenic and open sea regions, patterns commonly associated with genomic instability and aging53,54. DNA methylation changes at several TFBSs have been proposed as sensitive readouts of long-term chromatin remodeling, capturing the cumulative effects of cellular proliferation, differentiation, and stress signaling, as proposed by Pérez RF and colleagues48. In this context, our TFBS-centered analysis provides mechanistic insight into how aging- and cancer-associated epigenetic programs are dysregulated in SPBC tumors. Specifically, the pronounced hypomethylation observed at AP-1 binding sites in SPBC tumors is particularly noteworthy. The loss of DNA methylation at these sites becomes altered in aging55. Similarly, hypomethylation at FOXA1 and EP300 TFBSs supports a model in which enhancer reprogramming contributes to the altered transcriptional landscape of SPBC tumors. FOXA1, in particular, has been implicated in lineage plasticity and endocrine resistance, and its more accessible binding sites may facilitate transcriptional rewiring despite similar ER/PR/AR protein levels56,57. Conversely, the increased methylation detected at regions bound by EZH2aligns with the well-established role of Polycomb complexes in enforcing stable gene repression in cancer58. Taken together, these results suggest that SPBC tumors may accumulate epigenetic alterations not only due to tumor progression but also as a long-term effect of prior malignancy and/or the treatments used to manage it, mirroring epigenetic features typically observed in more aged cellular states. In addition, our application of the epiTOC mitotic clock29 revealed that SPBC tumors have significantly higher mitotic age compared to FPBCs, providing evidence of increased cumulative cell divisions throughout cancer development. This supports the hypothesis that SPBC tumors arise in a more aged, proliferative cellular environment, which may contribute to their distinct epigenetic landscape and aggressive clinical behavior. Moreover, our data do not exclude a model in which prior treatments such as chemotherapy or radiotherapy contribute to persistent epigenetic drift beyond mitotic age, further shaping the SPBC epigenetic landscape. Although this is one of the initial studies shedding light on the epigenomic landscape of SPBC tumors, our study has limitations that should be noted to better interpret the findings. The sample size of our institutional cohort, while carefully curated, remains modest, and the number of SPBC cases in public datasets is very limited. Nevertheless, the consistency of findings across two independent cohorts strengthens the validity of our conclusions. Additionally, while our analyses suggest functional repression of hormone signaling, in vitro and eventually in vivo validation, complemented by detailed IHC assessment of major hormonal receptors as well as clinical variables such as menopausal status, will be necessary to clarify the biological and clinical impact of the observed DNA methylation changes in SPBC tumors. In addition, although treatment information is available for our cohort, its completeness, cases per treatment category, and uniformity are limited, which constrains our ability to draw firm conclusions regarding the contribution of prior therapies to the SPBC epigenetic profile. Finally, while SPBCs were classified using standard clinical and pathological criteria, the lack of high-resolution genomic profiling does not exclude possible residual misclassification of metastases from the prior tumor. Together, these observations point to biological divergence, yet larger and more comprehensive datasets incorporating paired genomic sequencing will be necessary to define the molecular landscape of SPBC tumors. In conclusion, our findings highlight that SPBCs have a worse prognosis than clinicopathologically comparable FPBCs, characterized by silenced hormone signaling pathways and accelerated mitotic aging. These features may reflect the long-term consequences associated with prior cancer, providing new insights into SPBC pathobiology. However, alternative interpretations, including genetic high-risk or lifestyle factors, may also influence both cancer initiation and prognosis. Future studies are needed to shed light on these mechanisms by exploring whether DNA methylation-based biomarkers can improve risk stratification and guide endocrine therapy decisions in cancer survivors at risk for or diagnosed with SPBC. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Author contributions J.I.J.O, M.G., and D.M.M. conceived the study idea. M.G., J.I.J.O., and B.J.V. were responsible for cohort identification and tissue acquisition. J.I.J.O., D.M.M., and B.J.V. stratified patients into groups based on the clinical and demographic variables. J.I.J.O. and B.J.V. processed tissues. A.F.B.L. performed the molecular data integration and data analysis. M.E.A., P.L.A., S.I.M., and M.M.L. contributed with code curation, data analysis, and results interpretation. A.F.B.L., M.P.S., and C.M. performed the biostatistics analyses and results contextualization. D.M.M. and M.G. designed the study, oversaw the data analysis, and wrote the manuscript. A.F.B.L. and J.I.J.O. contributed equally to this work and shared the first authorship. All the authors reviewed, revised, and approved the final version of the manuscript. Funding This work was supported by Instituto de la Salud Carlos III (ISCIII) AES2022 project (#PI22/01496), Sara Borrell project (#CD22/00026) co-funded by the European Union, the postdoctoral contract Vicenç Mut (#POSTDOC001) funded by the Government of the Balearic Islands (CAIB), within the 2024 postdoctoral call, and co-funded by the European Social Fund Plus (ESF+), the Servei d’Ocupació de les Illes Balears (SOIB) Jove Qualificats program, the SYNERGIA program (SYN22/04) Institut d’Investigació Sanitària de les Illes Balears (IdISBa), and the Spanish Association Against Cancer (PRDPM246418BEDO). Data availability TCGA multi-omics data can be accessed from the National Cancer Institute GDC Data Portal (https://portal.gdc.cancer.gov/). The newly generated epigenetic data from this study have been uploaded to the NCBI-GEO (GSE302010). Code availability No specific code was generated for this project. All code used for the R scripts was adapted from the standard scripts provided in the packages mentioned in the Methods section. Declarations Competing interests The authors declare no competing interests. Ethics approval and consent to participate All clinical-demographic data and patient-derived samples were collected under research protocols approved by the Institutional Review Board of Providence Health and Services (Protocol #: JWCI-18-0401). The experiments were performed following the World Medical Association Declaration of Helsinki and the National Institutes of Health Belmont Report. Tissues were de-identified following the Health Insurance Portability and Accountability Act (HIPAA) regulations. Validation data for this cross-sectional study were gathered from the publicly available dataset, The Cancer Genome Atlas (TCGA). This study followed the Strengthening of the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Andrés F. Bedoya-López and Javier I. J. Orozco. Contributor Information Diego M. Marzese, Email: diego.marzese@idisba.es Melanie Goldfarb, Email: Melanie.Goldfarb@providence.org. References Associated Data Supplementary Materials Supplementary Material 1 Supplementary Material 2 Data Availability Statement TCGA multi-omics data can be accessed from the National Cancer Institute GDC Data Portal (https://portal.gdc.cancer.gov/). The newly generated epigenetic data from this study have been uploaded to the NCBI-GEO (GSE302010). No specific code was generated for this project. All code used for the R scripts was adapted from the standard scripts provided in the packages mentioned in the Methods section.