Bioinformatic analysis reveals the potential association of ESRP1 with the splicing of cytoskeleton-associated genes in doxorubicin-resistant MCF7 breast cancer cells Authors: Jing Zhang, Libin Chen, Qingling Song, Li Xu, Yinghua Li, Wen Li, Jie Wang Published: 2026 ABSTRACT AbstractBackgroundBreast cancer remains one of the most prevalent malignancies among women, with doxorubicin resistance posing a significant challenge that undermines treatment success and survival outcomes. Aberrant alternative splicing (AS), driven by dysregulation or mutations in splicing factors (SFs), is implicated in cancer initiation, progression, and drug resistance. This study aims to investigate the association of the epithelial cell-specific splicing factor ESRP1 with doxorubicin resistance in breast cancer, focusing on how ESRP1 deficiency correlates with AS changes that promote chemoresistance.MethodsWe analyzed RNA-sequencing (RNA-seq) data from doxorubicin-resistant (MCF7-DR) and parental (MCF7) breast cancer cell lines to identify enhanced alternative splicing events (ASEs) and changes in ESRP1 expression; we further leveraged The Cancer Genome Atlas (TCGA)-BRCA cohort to construct an SF–RASE correlation network for screening core SFs (including ESRP1). An integrative analysis combining crosslinking immunoprecipitation (CLIP-seq) data and The Cancer Genome Atlas (TCGA) database was performed to validate ESRP1 binding targets and assess the association between ESRP1-related splicing and cytoskeleton organization.ResultsWe observed extensive AS changes and significantly downregulated ESRP1 expression in MCF7-DR cells. Integrative analysis identified 61 high-confidence ASEs that correlate with ESRP1 expression. Further bioinformatic integration suggests that ESRP1 expression is associated with the splicing patterns of SPTBN1, MAP2K7, FGFR3, and CYB561A3—four genes involved in cytoskeleton organization—though direct experimental verification to confirm a causal regulatory relationship between ESRP1 and the splicing of these genes is still pending.ConclusionsOur findings suggest that ESRP1 expression is closely associated with doxorubicin resistance in breast cancer cells, with concomitant alterations in key ASEs linked to cytoskeletal remodeling that correlate with ESRP1. Exploring the ESRP1-related splicing network may offer new strategies to overcome chemoresistance and improve patient outcomes. However, the small cell line sample size (n = 2 per group) constrains the robustness of ASE and SF–ASE correlation findings, and these results should be interpreted with caution and require further validation with larger sample cohorts.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1007/s12672-026-05253-0.Keywords: ESRP1, Alternative splicing, Doxorubicin resistance, Breast cancer, Cytoskeleton Introduction Breast cancer is a disease characterized by abnormal growth of breast cells that are out of control and form tumors [1]. Breast cancer is one of the most common malignant tumors in women [2]. In 2022, 2.3 million women worldwide were diagnosed with breast cancer, and 670,000 died from it [3]. Chemotherapy resistance is a major obstacle in the treatment of breast cancer, severely affecting the long-term survival rate of patients [4]. Doxorubicin is a key treatment for breast cancer [5, 6], but its common resistance leads to recurrence and metastasis, poor prognosis and low survival [7]. The mechanisms of doxorubicin resistance include the overexpression of MDR transporters, DNA damage repair, epithelial-mesenchymal transition (EMT), and changes in the tumor microenvironment (TME) [8–10]. Thus, exploring the molecular mechanisms of doxorubicin resistance in breast cancer cells is of utmost importance for developing new therapeutic strategies to overcome or prevent drug resistance. Alternative splicing (AS) is critical for the regulation of human gene expression, significantly contributing to the diversity of functional proteins [11, 12]. In chemotherapy-resistant tumors, extensive and specific RNA splicing abnormalities can alter tumor cell responses to treatment. For instance, in breast cancer, AS of the Bcl-x gene leads to the production of pro-apoptotic and anti-apoptotic isoforms, influencing the survival of cancer cells in response to chemotherapeutic agents. Additionally, splicing of the CD44 gene generates various isoforms that enhance tumor invasiveness and metastatic potential, further complicating treatment outcomes [13]. AS has been shown to play a pivotal role in the prognosis, survival, and drug resistance of breast cancer [14–16], with specific mechanisms still requiring further exploration. Splicing factors (SFs) play pivotal roles in the recognition of splice sites and the assembly of spliceosomes during AS [17]. Studies have highlighted the regulatory role of SFs such as SRSF1 and hnRNPA1 in modulating alternative splicing events (ASEs) associated with doxorubicin resistance, underscoring the importance of understanding these pathways to develop targeted therapies. Epithelial splicing regulatory protein 1 (ESRP1) is an RNA-binding protein (RBP) and SF that regulates AS of mRNA, which plays an important role in maintaining cell epithelial characteristics and regulating tumor progression [18, 19]. ESRP1 mediates aberrant splicing in cancer cells, and plays a role in cell proliferation and chemoresistance through the regulation of apoptosis and autophagy [20]. In breast cancer, previous study has shown that ERα-ESRP 1/2 axis plays a role in its occurrence and development by controlling the splicing patterns of related genes [21]. Additional research has demonstrated that ESRP1 can target the splicing of α6 integrin mRNA to increase the sensitivity of triple-negative breast cancer cells to paclitaxel [22]. These results imply the significant role of ESRP1 in chemoresistance of breast cancer. However, the specific regulatory mechanisms are not clear. In this study, we aim to investigate the association of ESRP1 with doxorubicin resistance of breast cancer and to explore the AS network linked to ESRP1. We explored SFs and ASEs in doxorubicin-resistant MCF7 (MCF7-DR) human breast cancer cells using bioinformatics and expression analysis methods based on previously reported RNA-sequencing (RNA-seq) dataset (GSE174152); from the cell line-derived differentially expressed SFs, we further utilized The Cancer Genome Atlas (TCGA)-BRCA cohort to construct an SF–RASE correlation network, which helped narrow down 9 core SFs (including ESRP1) closely linked to doxorubicin resistance. We demonstrated that the expression of ESRP1 is reduced in doxorubicin-resistant breast cancer cells. The functions of ESRP1 as biomarker have also been verified in TCGA database, providing the potential for the prevention and intervention of doxorubicin resistance in breast cancer. Then, we explored the possible binding sites of ESRP1 using the crosslinking immunoprecipitation-high-throughput-sequencing data (CLIP-seq) dataset (GSE233931). Further analysis reveals a potential link between ESRP1 and the AS of SPTBN1, MAP2K7, FGFR3, and CYB561A3, which could be associated with cytoskeleton organization pathways (Fig. 1A). Our findings highlight the association between ESRP1 and doxorubicin resistance in breast cancer, potentially offering new molecular targets and a theoretical foundation for the treatment of breast cancer. Global profiles of AS in parental MCF7 and MCF7-DR cells. A Analysis workflow. RNA-seq data from MCF7 and MCF7-DR were used to identify RASEs via SUVA. DEGs and SFs were then integrated to construct an SF–RAS co-disturbed network, focusing on ESRP1. Additional validation was performed using TCGA BRCA dataset and ESRP1’s CLIP-seq binding profile (GSE233931). B Distribution of the five SUVA-defined ASE types (alt3p, alt5p, contain, olp, ir) detected in MCF7 vs. MCF7-DR cells. The y-axis shows the number of RASEs in each category. C Annotation of the SUVA-detected RASEs to classical ASE types, including A5SS, A3SS, cassette exon, mutually exclusive exons (MXE), and intron retention, among others. The bar plot indicates how many RASEs correspond to each classical AS category. D Bar plot showing the number of RASEs at different thresholds of pSAR (proportion of Splicing event-specific Reads among all reads in that region). Events with higher pSAR values have a greater relative abundance in the respective splicing site. E PCA of RASEs for which pSAR is greater than 50%. The x-axis (Dim1) explains 99.2% of the variance in splicing ratios, clearly separating doxorubicin-resistant samples from controls, while the y-axis (Dim2) explains 0.4%. F Heatmap of splicing ratios for RASEs (pSAR > 50%), illustrating hierarchical clustering of the four samples (two control and two doxorubicin-resistant). Red and blue represent relative differences (z-scores) in splicing ratios. G GO enrichment analysis of RASGs. The bubble color denotes the statistical significance (p-value), and the bubble size reflects the number of RASGs enriched in each biological process (BP) term. H Network diagram connecting the top five enriched GO terms (center) with the associated RASGs (peripheral nodes). Node size indicates the gene count within each GO term, and edges signify membership of a gene in a particular biological process Materials and methods Access to and processing of public data We used the transcriptome sequencing data (RNA-seq) of parental and MCF7-DR human breast cancer cells. The accession numbers of Gene Expression Omnibus (GEO) database were GSE174152. Public sequencing data were obtained from the Sequence Read Archive (SRA). SRA Run data files were transformed into fastq format with NCBI SRA Tool fastq-dump. Raw RNA-seq data were processed using fastp (v0.23.4) [23] to remove sequencing adapters and filter out low-quality bases. Cleaned reads were aligned to the human reference genome (GRCh38) using the STAR aligner (v2.7.10b) [24]. Uniquely mapped reads were retained for further analysis, including read count and expression level calculations, reported as fragments per kilobase of exon per million mapped reads (FPKM). Splicing analysis using SUVA SUVA (splicing site usage variation analysis) focused on different usages of each splice site [25]. It redefined splicing formula as clustered simple junction pairs to overcome the complex splicing patterns, with its filter’s core principle being to screen and quantify ASEs based on “splice site usage variation”. Specifically, SUVA takes splice junction (SJ) reads and non-junction reads from the mapping result of STAR (as input data for analysis) and simplifies complex splicing patterns by means of clustered simple junction pairs, avoiding misjudgment of complex splicing structures in traditional analyses. For the five defined ASE types (alt3p, alt5p, contain, olp, and ir), SUVA calculates the splicing ratio for each splice site. Conceptually, this splicing ratio is analogous to the Percent Spliced In (PSI) metric, representing the relative usage of a specific alternative splice site. SUVA then identifies statistically significant regulated alternative splicing (RAS) events by evaluating the differences in these splicing ratios between comparative sample groups. To ensure biological robustness, we subsequently implemented a strict filtering threshold of pSAR (proportion of Splicing event-specific Reads) ≥ 50%. This rigorous threshold was specifically applied to exclude low-abundance, stochastic background splicing noise, thereby guaranteeing that our downstream analyses focused exclusively on dominant, high-confidence ASEs with strong potential for functional relevance. Co-expression analysis of splicing ratio of RAS and expression of DE SFs The software DEseq2 [26], which is specifically used to analyze the differential expression of genes, was applied to screen the raw count data for differentially expressed genes (DEGs). The results were analyzed based on the fold change (FC ≥ 2 or ≤ 0.5) and false discovery rate (FDR ≤ 0.05) to determine whether a gene was differentially expressed. Then expression profile of differentially expressed SFs were filtered out from all DEGs according to a catalogue of 603 SFs of human was retrieved from Gene Ontology (GO) database and previous report [27]. To investigate the relationships between differentially expressed SFs and RAS, we reconstructed an SF–RAS correlation network using the TCGA-BRCA cohort. For each of the nine SFs, we extracted normalized SF expression levels and splicing ratios of RAS across TCGA-BRCA samples and calculated pairwise Pearson correlation coefficients. P values from the correlation tests were adjusted for multiple comparisons using the Benjamini–Hochberg method, and SF–RAS pairs with ∣r∣ ≥ 0.2 and FDR ≤ 0.05 were retained as edges in the network. The resulting network was visualized in Cytoscape, and GO biological process (GO-BP) enrichment of the RASE-associated genes was performed. GO functional enrichment analysis To identify functional categories of genes, we employed the clusterProfiler package (v4.6.2) [28], which enabled us to determine GO terms and KEGG pathways. TCGA-breast cancer data analysis For the analysis of ESRP1-regulated ASEs in breast cancer, we retrieved the TCGA BRCA dataset containing both gene expression profiles and clinical data from the UCSC XENA database (https://xenabrowser.net/datapages/). We selected samples based on ESRP1 expression levels. Samples were classified into two groups: those with high ESRP1 expression (top 15 samples) and those with low ESRP1 expression (bottom 15 samples). To assess the overlap between the RAS identified from TCGA and ESRP1-targeted RAS detected in the MCF7-DR cell line data, we performed a comparative analysis. ESRP1–RAS regression analysis in the TCGA-BRCA cohort We performed an event-level regression analysis to assess the association between ESRP1 expression and regulated ASEs in the TCGA-BRCA cohort. For each tumour sample, ESRP1 expression was obtained from normalized RNA-seq data and transformed as log₂(TPM + 1). Splicing ratio values for candidate RAS were extracted from the SUVA-based TCGA-BRCA splicing matrix. For each RAS, we fitted a multivariable linear model of the form where is the splicing ratio of event jin sample i, is the continuous ESRP1 expression, and denotes covariates including PAM50 subtype, ER/PR/HER2 status, pathologic stage, tumour purity, and neoadjuvant/radiation therapy where available. Models were fitted separately for each event using ordinary least squares after removing samples with missing values. P-values were adjusted across all tested events using the Benjamini–Hochberg procedure to obtain FDR. RAS with FDR ≤ 0.01 were defined as ESRP1-associated events. The distribution of ESRP1 effects across all RAS was visualized as a volcano plot of regression coefficient versus − log₁₀(FDR). Analysis of CLIP-seq data of ESRP1 We used the CLIP-seq for ESRP1. The accession numbers of GEO database were GSE233931. After reads were aligned onto the genome, only uniquely mapped reads were used for the following analysis. Reads with at least 1 bp overlap were clustered as peaks. For each gene, computational simulation was used to randomly generated reads with the same number and lengths as reads in peaks. The outputting reads were further mapped to the same genes to generate random max peak height from overlapping reads. The whole process was repeated for 500 times. All the observed peaks with heights higher than those of random max peaks (p-value < 0.05) were selected. The immunoprecipitation (IP) and input samples were analyzed by the simulation independently, and the IP peaks that have overlap with Input peaks were removed. The target genes of IP were finally determined by the peaks and the binding motifs of IP protein were called by HOMER software (v5.1) [29]. To further explore the functional significance of the identified binding sites, we cross-referenced the ESRP1-targeted genes with the RAS genes (RASGs) identified in the MCF7-DR cell line data and TCGA. Statistical analysis and data visualization If not specified, all bioinformatics analyses were performed with R software (v4.3.2) to compute statistics and generate plots throughout this manuscript. Specifically, differential gene expression was assessed using the Wald test in DESeq2. Splicing ratio comparisons between two groups in boxplots were evaluated using the Wilcoxon rank-sum test. For all multiple-testing scenarios, P-values were adjusted using the Benjamini-Hochberg procedure to control the false discovery rate (FDR). Principal component analysis (PCA) was performed using the R package factoextra (https://cloud.r-project.org/package=factoextra) to display the clustering pattern of samples based on the top two components. The sequencing data and genomic annotations were visualized with a script we developed in-house by normalizing the reads by the tags per million (TPM) of each gene. The pheatmap package in R was used for clustering based on Euclidean distance. Results Transcriptome analysis identifies differential ASEs in control and doxorubicin-resistant breast cancer cells AS produces complex and dynamic variations in protein isoforms that play a certain role in chemoresistance in breast cancer. For in-depth analysis of the changes in ASEs associated with doxorubicin resistance in breast cancer, we downloaded the transcriptome dataset GSE174152, which includes two MCF7-DR samples and two parental breast cancer cell (MCF7) samples. Five different types of ASE model defined by SUVA according to splicing site usage variation [25]. We first compared the differential ASEs between the two groups of cell samples as a whole. As shown in Fig. 1B, RASEs identified by SUVA were mainly alternative 5’ splice site (alt5p) and alternative 3’ splice site (alt3p). The splicing events were corresponding to classical splicing events, in which alternative 5’ splice site (A5SS) and cassette exon events accounted for a large proportion (Fig. 1C). Since a splicing event involves two transcripts, and these two transcripts may account for a very small proportion of the overall gene expression, we aim to find the more dominant spliced transcripts. We filtered the splicing events based on the proportion of all reads in the region (pSAR) that are different for the variable splicing events. Splicing events with pSAR less than 50% were filtered out, and we selected 1,021 events with pSAR ≥ 50% for further analysis (Fig. 1D, Table S1). The control and doxorubicin-resistant group can be clearly distinguished in the first principal component by using the splicing ratio of RASEs for PCA, indicating a significant increase in the heterogeneity of splicing patterns with the onset of chemotherapy resistance (Fig. 1E). This change may reflect fundamental changes in gene expression regulation mechanisms linked to chemotherapy resistance, particularly an increase in complexity at the post-transcriptional level. To analyze the RAS regulation patterns in the two groups, we used a heatmap to display the distribution of ratios for all differential splicing events (Fig. 1F). GO-BP enrichment analysis indicated that these RASGs were significantly enriched in cytoplasmic microtubule organization, regulation of actin filament-based processes, and regulation of actin cytoskeleton organization (Fig. 1G). Further, we presented the top 5 enriched pathways and the network diagram of genes included in the RAS (Fig. 1H). Overall, our findings indicate that a large number of ASEs are associated with doxorubicin resistance in breast cancer. These events may modulate key functional pathways, and these preliminary observations require validation in larger cohorts. Splicing factor ESRP1 is correlated with doxorubicin-resistance-related ASEs in breast cancer SFs are a class of RBPs that play pivotal roles in the recognition of splice sites and the assembly of spliceosomes during AS [30, 31]. Aberrant AS by either dysregulation or mutations of SFs contributes to cancer initiation and progression [32]. To thoroughly investigate the potential links of SFs in AS associated with doxorubicin resistance in breast cancer, we analyzed the DEGs and SFs in the MCF7-DR and MCF7 cell samples of the dataset GSE174152, identifying a total of 36 differentially expressed SFs (Fig. 2A). Scatter plot of the 36 overlapping SFs shows their expression changes in doxorubicin-resistant versus control cells. Among them, ESRP1 (highlighted) was the most significantly altered, exhibiting the smallest FDR and a substantial fold change (Fig. 2B). Its expression decreased sharply from an average FPKM of 47 in control cells to 0.6 in resistant cells. To screen core SFs closely linked to doxorubicin resistance, we leveraged the TCGA-BRCA cohort (large clinical sample size) and performed event-level regression analysis to evaluate associations between the 36 cell-line-derived differentially expressed SFs and RASEs functionally relevant to doxorubicin resistance. The potential mechanism likely involves their association with RASEs that participate in cytoskeleton-related pathways, such as actomyosin structure organization, cytoplasmic microtubule organization, actin filament organization, and regulation of actin cytoskeleton organization (Fig. 2C). To validate ESRP1 expression, we integrated transcriptome datasets of two doxorubicin-resistant breast cancer cell lines. ESRP1 expression was significantly lower in the doxorubicin-resistant group than in the control group in both MCF7 (GSE126365; Fig. 2D) and MDA-MB-231 (GSE264197; Fig. 2E) cells, with inherently low basal ESRP1 expression in MDA-MB-231 cells. Identification of key SFs in parental MCF7 and MCF7-DR cells. A Venn diagram showing the overlap between 5,184 DEGs and 603 curated SF genes in MCF7 vs. MCF7-DR cells. A total of 36 SFs are identified as DEGs (in the overlapping region). B Scatter plot of differentially spliced factors (SFs) between two control (MCF7) and two doxorubicin-resistant (MCF7-DR) samples. The x-axis shows the log₂ fold change in expression, and the y-axis represents the statistical significance as − log₁₀ (FDR). C TCGA-BRCA–derived SF–RAS interaction network. Nine differentially expressed SFs (orange triangles) are connected to RASEs whose splicing ratios are significantly correlated with the corresponding SF across the TCGA-BRCA cohort (|r| ≥ 0.2, FDR ≤ 0.05). D-E Box plots of ESRP1 expression (log₂(TPM + 1)) in control (NC) and doxorubicin-resistant (DR) groups from the MCF7 (GSE126365, D) and MDA-MB-231 (GSE264197, E) cell line datasets, respectively TCGA data analysis supports an association between ESRP1 and a wide range of ASEs in breast cancer To further investigate the potential mechanism by which ESRP1 inhibits doxorubicin resistance in breast cancer, we downloaded breast cancer-related data from the TCGA database, and selected 15 samples with the highest and 15 samples with the lowest expression of ESRP1 for RAS analysis (Fig. 3A). Using the SUVA software, we found that the main types of differential RASEs were alt5p and alt3p (Fig. 3B). To focus on more functionally relevant splicing events, we retained a total of 1,904 events with pSAR ≥ 50% for subsequent analysis (Fig. 3C). A heatmap was generated to visualize the distribution of splicing ratios of these RASEs across the two sample groups, suggesting differences in splicing patterns between ESRP1-high and ESRP1-low groups (Fig. 3D). PCA based on the splicing ratios of these events further confirmed that samples with high ESRP1 expression and low ESRP1 expression could be significantly separated, indicating a strong correlation between ESRP1 expression levels and global splicing profiles (Fig. 3E). GO-BP enrichment analysis was performed on genes harboring these RASEs (RASGs). The results showed that these RASGs were mainly enriched in biological pathways such as regulation of small GTPase-mediated signal transduction and regulation of cell-matrix adhesion (Fig. 3F). To assess potential confounding factors, we compared the baseline characteristics of the ESRP1-high and ESRP1-low groups. We observed significant imbalances in PAM50 subtypes, pathologic T stage, and tumor purity (Figure S1), underscoring the necessity of the multivariate regression analysis performed subsequently to adjust for these covariates. Accordingly, we conducted an event-level regression analysis in the TCGA-BRCA cohort to rigorously assess the association between ESRP1 expression and RASEs. A total of 2,109 RASEs met the predefined significance threshold (FDR ≤ 0.01) and were designated as ESRP1-associated events (Fig. 3G). We then performed an overlap analysis to integrate three sets of ESRP1-associated RASEs: RASEs identified by regression analysis in TCGA-BRCA, differential RASEs between ESRP1-high and ESRP1-low groups in TCGA-BRCA, and ESRP1-associated RASEs previously identified from the MCF7/MCF7-DR cell line system. This integration yielded 61 overlapping RASEs that were consistently associated with ESRP1 across both clinical cohorts (TCGA-BRCA) and cell line models (MCF7/MCF7-DR), representing high-confidence ESRP1-associated splicing events (Fig. 3H). GO-BP enrichment analysis of the 61 RASGs revealed that these ESRP1-modulated splicing targets were primarily involved in cytoplasmic microtubule organization and regulation of cell-substrate junction organization (Fig. 3I)—processes closely related to cell structure maintenance and mechanical property regulation, which may underlie ESRP1’s role in breast cancer doxorubicin resistance. ESRP1-RASEs in the TCGA BRCA dataset. A Box plot comparing ESRP1 expression in two groups of breast cancer patients from the TCGA cohort (ESRP1-low vs. ESRP1-high). The y-axis shows normalized ESRP1 expression, and each point represents an individual sample. B Distribution of five SUVA-defined AS event categories (alt5p, alt3p, contain, olp, ir) in TCGA samples. The x-axis indicates the number of RASEs in each category. C Bar plot illustrating the number of RASEs across varying thresholds of pSAR in TCGA samples. Higher pSAR implies greater abundance of the corresponding AS event. D Heatmap depicting splicing ratios (z-scores) for RASEs with pSAR > 50% in individual TCGA samples. Samples are hierarchically clustered into high-ESRP1 (orange) and low-ESRP1 (blue) groups. E PCA of the same RASEs (pSAR > 50%), showing clear separation between high-ESRP1 (red) and low-ESRP1 (blue) samples. Dim1 and Dim2 represent the two principal components capturing the most variance in splicing patterns. F GO enrichment analysis of RASGs between the high- and low-ESRP1 expression groups. The bubble size reflects the number of genes, whereas bubble color indicates the statistical significance (adjusted p-value). G Volcano plot summarizing ESRP1–RAS associations in the TCGA-BRCA cohort. Each dot represents a single RAS event; the x-axis shows the ESRP1 regression coefficient from multivariable linear models, and the y-axis shows − log₁₀(FDR). Orange dots indicate RASEs that pass the predefined effect-size and significance thresholds (FDR ≤ 0.01), while grey dots denote other events. H Venn diagram integrating ESRP1-associated RASEs identified by three independent approaches: TCGA-BRCA regression analysis, ESRP1 high- versus low-expression comparison in TCGA-BRCA, and the ESRP1-centered SF–RAS correlation network derived from TCGA-BRCA (Fig. 2B). The central overlap (n = 61) denotes high-confidence RASEs consistently associated with ESRP1 across all three approaches. I GO-BP enrichment analysis of the 61 overlapping ESRP1-associated RASGs. The x-axis shows the q-score (− log₁₀ adjusted p-value). Bubble size represents the number of genes annotated to each term, and bubble color reflects the significance level (adjusted p-value) CLIP-seq identifies putative ESRP1 binding sites To identify potential ESRP1 binding sites, we downloaded the CLIP-seq dataset (GSE233931), which established SGC7901 gastric cancer cell line with stable overexpression of ESRP1 and conducted CLIP seq for ESRP1. In two independent IP replicates, there was a difference in the number of raw binding peaks (1,745 peaks in IP-1 vs. 1,180 peaks in IP-2). This is a normal phenomenon attributed to inherent characteristics of technical replicates, such as antibody batch effects and fluctuations in CLIP-seq peak screening thresholds. The dataset remains usable because we excluded random errors by retaining 576 binding peaks common to both replicates (Fig. 4A). ESRP1 mainly bound to intronic regions and coding sequences (CDS) (Fig. 4B). Core candidate target genes were further identified via functional screening to enhance the interpretability of these results. GO analysis showed that genes with ESRP1 binding peaks were primarily enriched in RNA splicing and protein-RNA complexes assembly (Fig. 4C). The motif enrichment analysis of ESRP1-binding peaks identified the top five enriched nucleotide sequences, with the most prominent ESRP1-associated motif being 5’-CGUUGCU-3’ (Fig. 4D). By performing an overlap analysis genes with ESRP1 binding peaks (from both IP-1 and IP-2 replicates) and the RASGs detected in the previous section, we identified four shared genes—SPTBN1, MAP2K7, FGFR3, and CYB561A3—which we define as putative ESRP1 binding candidates in the context of AS regulation (Fig. 4E). These results suggest that ESRP1 may be associated with doxorubicin resistance in breast cancer, and its potential mechanism may involve direct regulatory effects on the AS of the above four genes—providing more precise candidates for further exploring ESRP1-mediated splicing regulatory networks in chemoresistance. Identification of ESRP1 binding sites and potential target genes using CLIP-seq. A Venn diagram illustrating the overlap of ESRP1-binding peaks identified in two independent immunoprecipitation (IP) replicates (IP_1 and IP_2). A total of 1,169 peaks are specific to IP_1, 604 are specific to IP_2, and 576 peaks are shared by both. B Genomic distribution of ESRP1-binding peaks. The majority of peaks (49.74%) are located in intronic regions, followed by CDS (19.2%), 3′ UTR (13.44%), noncoding exons (10.3%), and other genomic features, indicating that ESRP1 predominantly associates with introns and coding sequences. C GO enrichment analysis of ESRP1-bound target genes. The enriched GO terms primarily include RNA splicing, protein–RNA complex assembly, and mRNA processing events, suggesting that ESRP1 plays a significant role in post-transcriptional gene regulation. D Top four RNA-binding motifs identified in ESRP1-bound peaks. Motif logos and associated statistics (p-value, log p-value, and target coverage) are shown, highlighting consensus sequences that may mediate ESRP1’s splicing regulation. E Venn diagram comparing RASGs (blue circle) with ESRP1-binding genes from IP_1 (green circle) and IP_2 (orange circle). Four genes (SPTBN1, MAP2K7, FGFR3, and CYB561A3) overlap among all three sets, indicating potential ESRP1 binding to these RASGs and potential roles in ESRP1-related splicing ESRP1 is associated with altered AS patterns of key target genes in doxorubicin-resistant breast cancer To characterize the splicing features of ESRP1-associated target genes in the context of doxorubicin resistance, we analyzed the AS patterns of SPTBN1, MAP2K7, FGFR3, and CYB561A3 in doxorubicin-resistant and parental breast cancer cells, with validation in relevant datasets. For SPTBN1, genome browser visualization and splicing ratio quantification showed that the doxorubicin-resistant group exhibited a preferential shift toward the distal 5’ splice site, and the proportion of this splicing event was significantly reduced in MCF7-DR cells (GSE174152, GSE126365), TCGA samples with low ESRP1 expression, and MDA-MB-231-DR cells (GSE264197) (Fig. 5A). CYB561A3 and FGFR3 showed a clear proximal 5’ splice site bias in doxorubicin-resistant cells, while MAP2K7 in the doxorubicin-resistant group had marked exon skipping; the proportions of these corresponding splicing events were increased in MCF7-DR cells (GSE174152, GSE126365) and TCGA samples with low ESRP1 expression, but the increasing trend was not significant in MDA-MB-231-DR cells (GSE264197) (Fig. 5B-D). Collectively, these data demonstrate that ESRP1 expression is associated with distinct AS pattern changes in SPTBN1, MAP2K7, FGFR3, and CYB561A3 in doxorubicin-resistant breast cancer, supporting a potential link between ESRP1 and AS-driven doxorubicin resistance. Direct regulatory effects and functional impacts of these AS changes, however, require further experimental validation. It is important to note that the CLIP-seq data was derived from ESRP1-overexpressing gastric cancer cells, so the binding candidates identified here may not fully reflect ESRP1 binding in breast cancer or drug-resistant states. Thus, these candidates require further validation in breast cancer and doxorubicin-resistant contexts to confirm their biological relevance. AS analysis of SPTBN1, MAP2K7, FGFR3, and CYB561A3 in different groups. A Left: Genome browser-style visualization of read coverage and splice junctions for the SPTBN1 locus in control (control_1, control_2) and doxorubicin-resistant (DR_1, DR_2) cells. Blue regions indicate normalized read coverage, arcs show junction-spanning reads, and the red boxes highlight the critical splice site where usage differs between control and doxorubicin-resistant cells. Transcript annotations (ENSEMBL IDs) are shown below. Right: Box plots of the splicing ratio for the same SPTBN1 event in MCF7 vs. MCF7-DR (left), TCGA BRCA samples with high vs. low ESRP1 expression (upper right), and MDA-MB-231 vs. MDA-MB-231-DR (lower right). Schematic diagrams above the plots illustrate the differing exon usage resulting from altered 5’ splice site selection. B Similarly, for CYB561A3; the orange bar below the tracks indicates the ESRP1 binding region identified by CLIP-seq. C Similarly, for MAP2K7. D Similarly, for FGFR3; the orange bar indicates the ESRP1 binding region. Statistical significance for the splicing ratio comparisons was determined using the Student’s t-test (* P < 0.05, ** P < 0.01, *** P < 0.001) Discussion In this study, we explored the association of AS and the splicing factor ESRP1 with doxorubicin resistant breast cancer, using MCF7/MCF7-DR as primary models. Notably, the initial discovery phase was based on only two biological replicates per group (MCF7 and MCF7‑DR). Key findings were further validated across multiple datasets—including additional cell line data (GSE126365, GSE264197) and clinical samples from TCGA-BRCA. Our findings indicate an association between ESRP1 and the splicing regulation of key genes including SPTBN1, MAP2K7, FGFR3, and CYB561A3, implying that ESRP1 may potentially regulate these genes’ splicing; such AS changes may influence doxorubicin resistance through cytoskeleton-related pathways. These cross-validated results from bioinformatics analyses support ESRP1 as a potential therapeutic target linked to doxorubicin resistance in breast cancer, though its regulatory mechanisms and relevance in other subtypes await further verification. Our study revealed that the ASEs identified in MCF7/MCF7-DR cells (GSE174152) were significantly enriched in pathways related to cytoplasmic microtubule organization, regulation of actin filament-based processes, and regulation of actin cytoskeleton organization. These pathways are crucial for maintaining cell shape, motility, and intracellular transport—key processes for tumor progression and metastasis. The enrichment of these splicing events suggests that alterations in the splicing landscape may enhance the invasive potential of breast cancer cells, thereby contributing to doxorubicin resistance. Previous studies have documented similar associations between cytoskeletal dynamics and chemoresistance [33–36], indicating that changes in cell morphology and motility can affect drug uptake and efficacy. Our findings expand upon this literature by demonstrating that specific splicing alterations in key regulatory genes can directly influence these critical pathways. However, it is critical to clarify that the link between ESRP1-mediated splicing and cytoskeletal function is derived solely from pathway enrichment analyses and isoform expression patterns, not from direct phenotypic assays. Since we have not experimentally validated changes in cytoskeletal function, the current conclusion regarding a “cytoskeleton-related mechanism” remains an indirect inference based on omics data, rather than a directly confirmed functional phenotype. The implications of our results suggest potential therapeutic applications, where targeting the splicing machinery—specifically ESRP1 and its downstream effects—may offer a novel strategy to counteract chemoresistance in MCF7-derived doxorubicin-resistant cells. However, it remains a hypothesis that this strategy could be effective in other breast cancer cell types or patient-derived models; future research should explicitly test this by validating the role of ESRP1-mediated splicing in diverse breast cancer models. In our investigation of AS in doxorubicin-resistant breast cancer cells, we identified several SFs that may play critical roles in mediating chemoresistance in the MCF7/MCF7-DR system. Among these SFs, NOVA1 exhibits splicing regulatory activity in natural breast tumors [37], and it has also been reported to mediate doxorubicin resistance of hepatocellular carcinoma (HCC) cells by acting as a sponge for miR-3129-5p [38]. MSI1 is associated with resistance to cancer therapy, as its depletion leads to reduced expression of the catalytic subunit of DNA-PK, resulting in increased DNA damage [39]. CELF2 has shown antitumor activity in a variety of cancer models, including breast cancer, lung cancer, gastric cancer, and ovarian cancer, and loss of CELF2 can promote drug resistance in squamous cell carcinoma (SCC) [40]. Loss of ESRP2 leads to the mesenchymal subtype expression of genes associated with EMT—a process that participates in tumor progression [41, 42]. IGF2BP3 promotes the progression and cisplatin resistance of laryngeal squamous cell carcinoma (LSCC) through the UBA2-PI3K pathway [43]. KHDRBS3 can alter CD44 isoform expression, enhancing the stemness of basal-like breast cancer cells [44]. RBM47 inhibits the progression and metastasis of breast cancer by regulating the splicing and stabilization of its target mRNA [45, 46]. Circ_RBM23 promotes sorafenib resistance, malignant proliferation, migration, and invasion in HCC by regulating the miR-338-3p/RAB1B axis [47]. Apart from the above SFs, ESRP1 stands out as a particularly significant candidate. ESRP1 is known for its involvement in maintaining epithelial characteristics and regulating splicing patterns of genes implicated in tumor progression [19, 48]. Existing studies have indicated that ESRP1 is closely related to tumor chemoresistance, and its role in breast cancer chemoresistance remains unclear [21, 49, 50]. Our study emphasizes ESRP1’s unique contribution in the MCF7-DR context, as its downregulation in resistant cells correlates with a shift towards splicing patterns that favor invasive phenotypes and enhance cell motility. The choice to focus on ESRP1 was driven by its dual role in regulating AS and maintaining epithelial integrity, both of which are crucial in the context of breast cancer. By establishing an association between ESRP1 expression and splicing events linked to cytoskeletal organization and cellular dynamics, we provide new insights into how SFs may correlate with drug sensitivity in breast cancer. These findings are consistent with existing literature that links aberrant splicing to cancer progression, and extend the discussion by identifying ESRP1 as a key player associated with chemotherapy resistance in MCF7-DR cells. Looking ahead, further research should explore the potential mechanistic links between ESRP1 and alternative splicing (AS) in the MCF7-DR model, as well as the association between this splicing signature and chemotherapeutic agent efficacy in this system. Additionally, investigating ESRP1 or its associated splicing events may represent a promising therapeutic strategy to overcome drug resistance in in MCF7-derived breast cancer—though the applicability of this strategy to other breast cancer subtypes remains a hypothesis to be validated in additional preclinical models. Our integrated analysis utilizing TCGA, MCF7/MCF7-DR cell line data, and CLIP-seq results identified several ASEs that are potentially regulated by ESRP1 and associated with chemotherapy resistance. SPTBN1 encodes the β II subunit of spectrin (a cytoskeletal protein) to maintain cell morphology and is involved in the regulation of DNA damage repair, angiogenesis, and stemness maintenance [51]. Existing studies have confirmed that SPTBN1 is underexpressed in breast cancer, and its deletion promotes EMT [52]—a key inducer of tumor cell chemoresistance. It has also been reported that SPTBN1 plays a significant role in the pathogenesis, progression, and chemoresistance of many types of cancer [53–56]. MAP2K7 encodes MEK7, a key upstream activator of the JNK pathway that mediates signals for cell stress, apoptosis, and proliferation [57]. In breast cancer, it can activates the JNK pathway to regulate AP-1 (e.g., c-JUN), induce senescence in ER+ breast cancer cells to enhance sensitivity to endocrine therapy combined with CDK4/6 inhibitors, and its deletion mediates resistance to this combination therapy in the tumor [58]. It is known that the JNK pathway regulates the polymerization/depolymerization of cytoskeletal components such as actin and microtubules, as well as the activity of related regulatory proteins [59]. Thus, MAP2K7 may indirectly regulate cytoskeletal dynamic balance through this pathway. FGFR3 encodes fibroblast growth factor receptor 3, whose expression product mediates signal transduction such as cell proliferation and differentiation; it is often abnormally activated in breast cancer and involved in regulating tumor initiation and progression, invasion and metastasis, as well as therapeutic resistance [60]. Studies have shown that FGFR3 mediates breast cancer resistance to PARP inhibitors by phosphorylating the Y158 site of PARP1, recruiting BRG1, prolonging the chromatin-bound state of MRE11, promoting homologous recombination repair, and reducing PARP trapping efficiency [61]. Meanwhile, FGFR3 can regulate cytoskeletal remodeling through downstream signaling pathways, affecting cell proliferation, migration, and morphological maintenance, and serves as a key regulator of cytoskeletal dynamic balance [62]. However, no studies have clearly established its association with the cytoskeleton in the context of tumor chemoresistance. Additionally, research on CYB561A3 is limited. As a ferritinophagy-related gene, it exhibits differential expression in hepatocellular carcinoma (LIHC) and may be closely associated with the tumor immune microenvironment [63]. Presently, there is a lack of studies confirming that CYB561A3 is correlated with tumor chemoresistance. In our study, integrating data from TCGA, MCF7/MCF7-DR cell lines, and CLIP-seq, we made a key observation: significant alterations in the splicing patterns of SPTBN1, MAP2K7, FGFR3, and CYB561A3 in both MCF7-DR cells and breast cancer patient samples (from TCGA). Given their established roles, we further hypothesize that the longer SPTBN1 isoforms in drug-resistant cells are formed via the selection of specific 5’ splice sites, and might enhance the repair of doxorubicin-induced DNA damage or modulate processes like EMT that indirectly contribute to chemoresistance. The exon-skipping event in MAP2K7 could generate a truncated, functionally impaired isoform that disrupts JNK pathway activation—reducing doxorubicin-induced apoptotic signaling and altering cytoskeletal dynamics to enhance cell motility and evade drug-induced stress. For FGFR3, the shift toward proximal 5’ splice site usage may produce a constitutively active isoform that drives downstream signaling cascades to promote cytoskeletal remodeling, increasing cell invasive potential and reducing intracellular drug accumulation. Additionally, the altered splicing of CYB561A3 might disrupt iron ion homeostasis, which in turn affects iron-dependent post-translational modifications of cytoskeletal proteins, impairing the ability of doxorubicin to target microtubules and induce cell death. Collectively, these findings suggest that these four genes are potential key downstream targets of ESRP1-mediated AS. Their altered splicing patterns in doxorubicin-resistant settings point to ESRP1’s likely pivotal role in driving chemoresistance by modulating cytoskeleton-related pathways. However, direct evidence confirming their identity as ESRP1 downstream targets remains to be established, and their functional relevance to ESRP1-regulated chemoresistance requires further experimental validation. These observations nonetheless provide new clues into the molecular mechanisms of doxorubicin resistance in breast cancer and offer a foundation for deepening understanding of how aberrant AS contributes to cancer treatment failure. In conclusion, our study highlights the critical role of AS in mediating doxorubicin resistance in the MCF7-DR breast cancer cell model, with a specific focus on the splicing factor ESRP1. We identified key ASEs potentially associated with ESRP1 in the MCF7/MCF7-DR system, and these events may link to pathways related to tumor progression and drug resistance, such as cytoskeletal organization. These findings not only enhance our understanding of the molecular mechanisms underlying chemoresistance in this specific model but also suggest potential therapeutic targets for improving treatment outcomes in breast cancer. However, this study is not without limitations. First, a prominent limitation is the small sample size of the core MCF7/MCF7-DR RNA-seq dataset (GSE174152), with only two biological replicates per group. This severely limits statistical power to distinguish true biological differences from random variation in ASE discovery and SF–ASE correlation analyses at the cell line level, as variance is poorly estimated. Consequently, PCA separations and heatmaps based on these samples are not persuasive, and strong claims regarding ASE patterns or SF–ASE associations derived solely from this small dataset should be avoided. Although subsequent SF narrowing via TCGA-BRCA cohort-based network analysis and back-verification of core SF expression in doxorubicin-resistant cells helped optimize the candidate pool, the inherent constraints of the initial cell line data cannot be fully mitigated. Second, reliance on cell line models, even cross-validated ones, fails to capture in vivo tumor complexity (e.g., TME, inter-patient heterogeneity, in vivo drug metabolism) that could influence ESRP1-mediated splicing and doxorubicin resistance. Third, our TCGA-BRCA analyses, which compared ESRP1 high- and low-expression samples to validate splicing associations, did not account for potential confounding factors. These include patient clinical characteristics, prior treatment history, and tumor pathological features—all of which could independently modulate AS patterns and thus introduce bias into the observed correlations between ESRP1 and target gene splicing. Fourth, a key limitation specific to our ESRP1 target identification lies in the source of the CLIP-seq dataset (GSE233931), which was generated using SGC7901 gastric cancer cells with stable overexpression of ESRP1, rather than the breast cancer cells (MCF7/MCF7-DR) focused on in this study. While ESRP1 likely has some conserved binding motifs across cell types, the binding repertoire can be context-dependent. This dependence arises from cell-type-specific differences in transcriptional programs, RNA secondary structure, and chromatin accessibility, which can alter ESRP1’s ability to recognize and bind target RNA sequences. As a result, the four overlapping targets (SPTBN1, MAP2K7, FGFR3, and CYB561A3) identified by crossing CLIP-seq-derived ESRP1-binding genes with breast cancer-associated RASGs are designated as “putative” ESRP1 binding targets in our results, and their status as direct ESRP1-regulated targets in breast cancer (especially in the context of doxorubicin resistance) remains unvalidated. Finally, because this study relies exclusively on computational data, the functional significance of these findings must be interpreted with strong caution. Direct causal relationships to chemoresistance require independent experimental validation, specifically via RT-PCR of key splice events (SPTBN1, MAP2K7, FGFR3, CYB561A3) and assessing doxorubicin sensitivity following ESRP1 knockdown (KD) or overexpression (OE) interventions. Future research should aim to explore these mechanisms in more diverse and clinically relevant models, as well as investigate the potential of splicing modulators as therapeutic strategies to overcome chemoresistance in breast cancer. Specifically, for expanding model validation, future studies will focus on three core directions: First, establishing acquired doxorubicin-resistant strains of multiple breast cancer subtype cell lines to verify ESRP1 expression patterns and downstream target splicing changes across subtypes, and conducting ESRP1 rescue experiments to clarify subtype-specific regulatory effects. Second, collecting paired primary and doxorubicin-resistant tumor tissues from clinical breast cancer patients, to analyze ESRP1 expression and target gene splicing patterns in clinical samples. Third, constructing patient-derived organoid (PDO) models from clinical tumor tissues, inducing doxorubicin resistance in PDOs, and correlating ESRP-related splicing features between PDOs, cell lines, and clinical tissues, to systematically evaluate the consistency of findings across different experimental and clinical models. By addressing these limitations, we can further elucidate the therapeutic implications of our findings and advance the field of precision oncology. Supplementary Information Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Acknowledgements The authors would like to thank Chao Cheng and Qingqing Yin (Center for BioBigData Analysis, Wuhan Biosalt Biotechnology Co., Ltd, Wuhan, Hubei, China) for their guidance during data analysis. Author contributions Conception: JW and LC.Interpretation or analysis of data: JZ, WL, QS, and LX.Preparation of the manuscript: WL, JZ, YL, and LC.Revision for important intellectual content: WL, JZ, and JW.Supervision: LC. Funding This research received no specific grant from any funding agency in the public, private or not-for-profit commercial sectors. Data availability The core algorithm SUVA is available on GitHub (https://github.com/ablifedev/SUVA/), and the other scripts are deposited on Figshare (https://doi.org/10.6084/m9.figshare.30353272). The datasets analyzed in this study are available in GEO under the following accessions: GSE174152, GSE233931, GSE126365, GSE264197. GDC TCGA Breast Cancer (BRCA) dataset containing both gene expression profiles and clinical data from the UCSC XENA database (https://xenabrowser.net/datapages/). Splice junction data in BED format from TCGA were downloaded from Genomic Data Commons Data Portal (GDC) for alternative splicing identification. Declarations Ethics approval and consent for participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Jing Zhang and Libin Chen have contributed equally to this work. Contributor Information Wen Li, Email: 61624612@qq.com. Jie Wang, Email: 1605574199@qq.com. References Associated Data Supplementary Materials Supplementary Material 1 Supplementary Material 2 Supplementary Material 3 Data Availability Statement The core algorithm SUVA is available on GitHub (https://github.com/ablifedev/SUVA/), and the other scripts are deposited on Figshare (https://doi.org/10.6084/m9.figshare.30353272). The datasets analyzed in this study are available in GEO under the following accessions: GSE174152, GSE233931, GSE126365, GSE264197. GDC TCGA Breast Cancer (BRCA) dataset containing both gene expression profiles and clinical data from the UCSC XENA database (https://xenabrowser.net/datapages/). Splice junction data in BED format from TCGA were downloaded from Genomic Data Commons Data Portal (GDC) for alternative splicing identification.