Introduction

Breast cancer (BC) is the world’s most widespread cancer among women [1]. The majority of BC patients are estrogen receptor-positive (ER+), i.e., Cancer cells have estrogen receptors (ESR1) [2, 3]. When the ER interacts with estrogen, it acts as a transcription factor (TF) and triggers cell proliferation and differentiation [4]. Hence, finding ER-DNA interaction genomic regions can help identify genes that are regulated by ER. Specifically, investigating these genomic regions can help to understand ER modes of action in cancer progression.

Chromatin immunoprecipitation sequencing (ChIP-seq), as a next-generation sequencing technique, is applied to find the transcription factor binding sites (TFBSs) [5]. In the present era, several ChIP-seq datasets have been generated to find differentially bound sites (DBSs) in BC patients [6,7,8]. Each study has results unique or commonalities with other related studies, which could be due to known and unknown batch effects [6,7,8]. On the other hand, gathering relevant studies and performing meta-analysis supplies more exact results [9]. Few studies have been performed on the meta-analysis of ChIP-seq data. A meta-analysis study was conducted by Kolmykov and colleagues on ChIP-seq datasets through the rank aggregation approach, and the significant TFBSs in BC were identified [10]. Some complexities with ChIP-seq data should be considered for meta-analysis, including: (1) it is necessary to consider the same cell line, dose, and treatment period (known batch effects), (2) selecting the appropriate treatment period, i.e., after which treatment period, treated and untreated samples show the most differentiation, (3) removing the effects of differences laboratory conditions (unknown batch effects) seems to be vitally important to obtain more precise meta-analysis results, (4) since each study may have been analyzed with different platforms and genomic references, it is necessary to re-analyze the samples with the same tools and references, and (5) it is noteworthy that even if the same platform is applied, a different number of peaks and regions are obtained for each sample. Therefore after combining the datasets, they must be prepared to obtain the same number, regions, and scores of TFBSs for all samples. Next, the unknown batch effects must be removed before meta-analysis. Despite the necessities mentioned above, no meta-analysis study has already been performed on integrated TFBSs to obtain significant binding sites in BC.

In the current study, to obtain meta-differentially bound sites in estrogen receptor-positive breast cancer cell lines, we employed an innovative workflow for the meta-analysis of ChIP-seq data. Given the ER's essential role in BC, MCF7 and T47D cell lines stimulated with E2 were collected to find genomic regions directly interacting with ER. MCF7 and T47D cell lines, representing luminal A subtype, are extensively utilized as experimental models in breast cancer research, particularly for the study of hormone-dependent BC. These two ER + BC cell lines serve as valuable tools in understanding the molecular mechanisms and exploring potential therapeutic interventions for BC, particularly luminal A subtype. In the MCF7 cell line, concerning the same cell line, dose, and treatment period (removing known batch effects), public ChIP-seq datasets were selected and re-analyzed. Next, samples were integrated, and the same number, regions, and scores of TFBSs were obtained in doses of 10 nM and 100 nM separately. Finally, unknown batch effects were removed, and several ChIP-seq datasets were meta-analyzed based on TFBSs scores. Hence, we take advantage of the term meta-differentially bound sites (meta-DBSs) for the first time. For more precision, the meta-analysis results were compared with the results of T47D cell lines, and the intersection of both cell lines was obtained. By enriching the shared meta-DBSs and their associated genes, TFs and pathways that had not been previously reported in BC were identified. Moreover, the results of the ChIP-seq data meta-analysis were confirmed by comparing the results of the RNA-seq data meta-analysis.

Materials and methods

After collecting data from SRA-NCBI [11] and ENA-EBI [12] databases, nine MCF7 and T47D cell line-associated ER alpha-ChIP-seq datasets from eight studies were selected. Differential analysis for each dataset was performed, and DBSs were obtained following data pre-processing and mapping. Next, the meta-analysis was performed, and meta-DBSs were identified in the MCF7 cell line treated with 10 nM and 100 nM E2 separately. Then, Peak Annotation was applied for DBSs and meta-DBSs, and the results of both cell lines were compared. Peak set functional enrichment analysis (PSFEA) and ChIP enrichment analysis (ChEA) were implemented for the common meta-DBSs and their associated genes, respectively. Also, the meta-analysis results of both ChIP-seq and RNA-seq data were compared. The general steps of the present study are summarized in Fig. 1.

Fig. 1
figure 1

An overview of the meta-analysis steps performed in the present study. ChIP-seq datasets were retrieved from SRA-NCBI and ENA-EBI databases. Pre-processing and re-analyzing steps of datasets were conducted. Then, batch effects removal and meta-analysis were performed. Subsequently, peak set functional enrichment analysis (PSFEA) and ChIP Enrichment Analysis for meta-DBSs-associated common genes were performed using the Cistrome-GO database and ChEA3 database, respectively. The packages and methods employed are in bold form. TFBSs Transcription factor binding sites, DBSs Differentially bound sites, Meta-DBSs Meta-differentially bound sites

Full size image

Data collection

A comprehensive search was conducted on SRA-NCBI [11] and ENA-EBI [12] databases to find ChIP-seq datasets in BC (Fig. 2). Some criteria were considered in the selection of the ChIP-seq datasets: (1) association with ER+ cell lines; MCF7 and T47D cell lines, (2) availability of both treated with vehicles (as controls) and treated with E2 samples, (3) datasets without any knocked out genes, and (4) datasets with the same and appropriate dose and period of treatment. Consequently, GSE94023, GSE99626, GSE67295, GSE115607, GSE80367, GSE23893, GSE54855, and GSE59530, including MCF7 and T47D cell lines stimulated with 10 nM and 100 nM E2 for 40 or 45 min were selected. The details of the datasets are described in Table 1.

Fig. 2
figure 2

The flowchart to select datasets. A total of 351 datasets from SRA-NCBI and ENA-EBI were evaluated. Finally, based on the four criteria described, eight studies on MCF7 and T47D cell lines treated with E2 were used in the present study

Full size image
Table 1 Characteristics of the selected ER-ChIP-seq datasets and their experimental design
Full size table

Pre-processing and data analysis

The selected ChIP-seq datasets were re-analyzed under the Galaxy platform (https://galaxyproject.org) [13] to homogenize studies. The data quality control and trimming were performed using FastQC (version 0.11.5) [14] and Trimmomatic (version 0.38) [15], respectively. The human reference genome (hg38) was utilized for data mapping by HISAT2 (version 2.1.0) [16].

Analysis of individual datasets to identify DBSs

Within each individual study, DBSs were identified in MCF7 and T47D cell lines treated with E2 compared to treated with vehicle (as control) through MACS2 [22] (version 2.1.1.20160309.6) (-log10 (q−value) > 2; fold-enrichment > 2). Then, the ChIPseeker (version 1.26.2) [23] package was applied for peak annotation. The identified DBSs associated genes of individual studies were further utilized to compare with the results of the ChIP-seq meta-analysis. During peak calling with MACS2, the input samples were also considered to reduce noise. The Blacklist regions were also considered to improve the signal-to-noise ratio. The maximum tags were set to keep one tag at the same location (–keep-dup = 1). In the individual dataset analysis, replicates were combined using the rmspc (Multiple Sample Peak Calling) package [24].

Typically, analysis of treatment, vehicle, and input samples on ChIP-seq data is performed in the following steps:

  • A. ER-ChIP treated with E2 vs. input = corrected ER-ChIP treated with E2.

  • B. Vehicle vs. input (for vehicle) = corrected vehicle.

  • C. Corrected ER-ChIP treated with E2 vs. corrected Vehicle = detection of enriched regions in ER-ChIP treatment E2 (using MACS2).

Meta-analysis to identify meta-DBSs

To find meta-DBSs, a meta-analysis workflow was utilized to compare cells of treated with vehicle versus treated with E2 at the TFBSs score levels (Fig. 3). First, the same number, regions, and scores of TFBSs were obtained utilizing the DiffBind package (Version 3.0.15) [25]. All replicates were considered as independent samples to produce the binding affinity matrix. Then, the ARSyNseq (ASCA Removal of Systematic Noise for sequencing data) method [26], implemented in the NOIseq package [26], was applied to remove unknown batch effects (Fig. 1). In ARSyNseq, the TMM (Trimmed Mean of M values) method [27] was utilized for between-sample normalization on TFBSs scores. Then, the metaSeq package [28] was employed on all samples to find meta-DBSs and set the statistical threshold as -log10 (q−value) > 2 to select meta-DBSs. Finally, the ChIPseeker package was utilized for peak annotation. All packages were implemented in R software. The parameters in the ARSyNseq and the metaSeq packages were set as follows:

mydata2corr1 = ARSyNseq (mydata2, factor = “batch”, batch = TRUE, norm = “tmm”) result <- meta.oneside.noiseq (cds, k=0.5, norm = “n”, replicates = “biological”, factor = flag1, conditions = c(1,0), studies = flag2)

Fig. 3
figure 3

A workflow for integration and meta-analysis of ChIP-seq TFBSs scores from several studies. Datasets were selected based on the same criteria and re-analyzed with the same platform. Then samples were integrated, normalized, and meta-analyzed. This process was performed for MCF7 cell lines, and meta-DBSs were obtained. Next, meta-DBSs-associated genes were identified with peak annotation. The packages used in each step are marked in blue. TFBSs Transcription factor binding sites, DBSs Differentially bound sites, Meta-DBSs Meta-differentially bound sites

Full size image

Peak set functional enrichment analysis and ChIP enrichment analysis

Gene ontology (GO) and pathway enrichment analyses were performed with Cistrome-GO (the updated version of 2019) (http://go.cistrome.org/) [29] to determine the functional importance of the identified meta-DBSs. For such analysis, we employed peak set functional enrichment analysis (PSFEA) term for the first time. The GO annotation was performed at three levels, including biological process (BP), molecular function (MF), and cellular component (CC). Furthermore, the Kyoto Encyclopedia of Genes and Genomes (KEGG) (the updated version of 2019) [30] enriched pathways were determined. The adjusted P-value ≤ 0.05 was applied to select significant GO terms and KEGG pathways. Also, a comprehensive ChIP Enrichment Analysis (ChEA) was performed on meta-DBSs-associated common genes using ChEA3 (the updated version of 2019) (https://amp.pharm.mssm.edu/ChEA3) [31], which contains six libraries, including Literature, Enrichr, ARCHS4—Coexpression, ENCODE, ReMap, and GTEx—Coexpression. Also, two overall results were obtained: (1) integrated_topRank, and (2) integrated_meanRank; in both of them TFs were ranked based on estimated scores in libraries. Adjusted P-value ≤ 0.05 was considered significant.

Results

The meta-analysis was performed on the ChIP-seq data based on TFBSs scores to identify meta-DBSs. Here, known batch effects were first removed. Since the treatment period with E2 can affect the results, several treatment periods were investigated to select the appropriately treated cell lines. In the GSE94023 study, there were various periods of treatment with E2, including 0 (mock treated as control) and 5, 10, 20, 40, 80, 160, 320, 640, and 1280 min (cases). An appropriate sampling time reveals the maximum distinction between treated and untreated samples. Hence, a heat map correlation matrix was drawn for all samples. According to the matrix, the samples after 40 min showed the most significant difference compared to the control (see Additional file 1: Figure S1). Thus, samples treated with 10 nM and 100 nM E2 for 40 or 45 min were selected. The minimum peak length was considered equal to 150-bp as the significant peak cutoff.

Identification of DBSs

After selecting and pre-processing data, DBSs were obtained from nine individual ChIP-seq datasets (Table 2; see Additional file 2: Tables S1-S9) and compared with the meta-analysis results.

Table 2 Identified DBSs/ meta-DBSs through individual/ meta-analysis on eight studies of the MCF7 and T47D cell lines treated with 10 nM and 100 nM E2
Full size table

Identification of meta-DBSs

To identify meta-DBSs in MCF7 cell lines, datasets were integrated based on stimulation with 10 nM or 100 nM E2 separately. Next, DiffBind was applied to obtain the same number, regions, and scores of TFBSs (see Additional file 2: Tables S10 and S11). Following unknown batch effect removal (see Additional file 1: Figures S2-S5), ChIP-seq data were meta-analyzed, and the meta-DBSs (-log10 (q−value) > 2) were identified (see Additional file 2: Tables S12 and S13). The details of DBSs, meta-DBSs, and associated genes are described in Table 2.

Shared significant TFBSs between meta-DBSs and DBSs

Meta-DBSs were compared with DBSs obtained from individual datasets for more precision. Based on our findings, 617 genes were common among meta-DBSs- and DBSs- associated genes, 35 of which were TFs (Fig. 4). Remarkably, nine of those TFs were also identified among the top 50 TFs obtained from ChEA3 in Section 3.5.1 (Fig. 6). There were 282 genes associated with peaks identified through meta-analysis of MCF7 cell lines treated with 10 nM E2 but not in initial individual datasets. Six of them were POU5F1B, ZNF662, ZNF442, KIN, ZNF410, and SGSM2 TFs (Fig. 4A). Moreover, most of the top 50 TFs were also identified among the meta-analysis results and some initial individual datasets (Figs. 4 and 6). PCGF2, HNF1B, and ZBED6 TFs predicted by ChEA3 were not identified through meta-analysis or initial individual datasets.

Fig. 4
figure 4

Shared meta-DBSs between meta-analysis and individual studies based on TFBSs associated genes in MCF7 and T47D cell lines treated with E2. Among the peaks associated genes, only TFs are displayed. Nine of these TFs, which are also among the top 50 TFs obtained from ChEA3, are shown in purple. The six TFs only identified through meta-analysis are shown in orange. The marked blue TFs are mitochondrial TFs. TFs Transcription factors, DBSs Differentially bound sites, Meta-DBSs Meta-differentially bound sites

Full size image

Genomic occupancy of ER binding sites

The genomic locations of 7,308 ER-meta-DBSs correlated with 617 common genes were annotated using the ChIPseeker. According to the genome-wide annotation of TFBSs, it was found that there were 1,534 binding sites located within 10 kb of a transcription start site (TSS) (Fig. 5A). 1,531 of these sites were situated in the proximal to the TSSs (promoter region). Also, peaks were 3,327 in introns, 2,104 in intergenic, 218 in exons, 13 in 5' untranslated regions (UTRs), three in TSSs (downstream region), and 112 in 3' UTRs (Fig. 5B and C). Because some of the annotations overlap, the complete annotations and their overlaps are illustrated in Fig. 5D. These results indicated that most ER binding sites are located in introns, followed by intergenic regions, promoter-TSSs, and exons to a lesser degree in 3' UTRs, TSSs, and 5' UTRs.

Fig. 5
figure 5

Genome-wide annotation of 7,308 meta-DBSs correlated with 617 common genes and response elements of ER between MCF7 and T47D cell lines treated with E2. A Distribution of ER-meta-DBSs relative to the nearest TSS across the human genome. B Pie plot of the ER-meta-DBSs percentages according to peak location across different genomic regions of the human genome. C Visualization of ER-meta-DBSs obtained from the UCSC genome browser (version hg38). D Venn pie of annotations and their overlap. TSS transcription start site

Full size image

Enrichment analysis

ChIP enrichment analysis

The 617 meta-DBSs-associated common genes were enriched using the ChEA3 database to predict significant TFs that regulate gene expression in breast cancer. The results of six libraries and two overall results are presented in Additional file 2: Tables S14-S21. In accordance with the integrated_meanRank (see Additional file 2: Table S21), 1632 TFs were ranked, and the top 50 TFs are shown in Fig. 6. Remarkably, TRPS1, FOXA1, TFAP2C, GLIS3, ELF3, and ESR1 TFs have the highest rank score. Also, PCGF2, HNF1B, and ZBED6 TFs were predicted as potential key regulators.

Fig. 6
figure 6

ChIP Enrichment Analysis (ChEA) for 617 meta-DBSs-associated common genes. The bar graph integrated_meanRank for the top 50 TFs using ChEA3 in MCF7 and T47D cell lines treated with E2. Each color represents a library, and each bar's length indicates the weight of that TF in each library. TFs transcription Factors

Full size image

Peak set functional enrichment analysis

Peak set functional enrichment analysis, including GO and KEGG pathways for 7,308 meta-DBSs (correlated with common 617 genes) was performed. By enriching meta-DBSs and considering adjusted P-value ≤ 0.05, 94 BPs, 3 CCs, and 12 MFs for GO terms and 9 KEGG pathways were detected (see Additional file 2: Tables S22 and S23). GO terms regarding BPs were annotated in cornification (GO:0070268), tissue development (GO:0009888), regulation of anatomical structure morphogenesis (GO:0022603), anterior/posterior pattern specification (GO:0009952), phosphatidylcholine metabolic process (GO:0046470), negative regulation of transcription by RNA polymerase II (GO:0000122), response to metal ion (GO:0010038), negative regulation of RNA biosynthetic process (GO:1902679), embryonic organ morphogenesis (GO:0048562), and response to inorganic substance (GO:0010035).

Also, enriched CCs terms were linked to the cortical actin cytoskeleton (GO:0030864), cortical cytoskeleton (GO:0030863), and cytoskeleton (GO:0005856). Significant GO terms were found for MFs with regulatory region nucleic acid binding (GO:0001067), glucose binding (GO:0005536), lipid binding (GO:0008289), RNA polymerase II regulatory region DNA binding (GO:0001012), activating transcription factor binding (GO:0033613), DNA-binding transcription factor binding (GO:0140297), and transcription coregulator activity (GO:0003712) (Fig. 7A).

Fig. 7
figure 7

Peak set functional enrichment analysis (PSFEA) for 7,308 meta-DBSs, including GO and KEGG pathways. (A) The chart of the top 20 GO terms and (B) nine KEGG pathways was obtained from 7,308 meta-DBSs in MCF7 and T47D cell lines treated with E2. Also, TFs that enriched GO terms and KEGG pathways are shown along with the number of peaks. TFs Transcription factors, DBSs Differentially bound sites, Meta-DBSs Meta-differentially bound sites

Full size image

The nine enriched KEGG pathways included the Estrogen signaling pathway (hsa04915), Rap1 signaling pathway (hsa04015), Tight junction (hsa04530), Neurotrophin signaling pathway (hsa04722), Fluid shear stress and atherosclerosis (hsa05418), MAPK signaling pathway (hsa04010), Thyroid cancer (hsa05216), Bladder cancer (hsa05219), and Pathways in cancer (hsa05200) (Fig. 7B). Moreover, the top 50 TFs obtained from ChEA3 were examined. The TFs that enriched the Top 20 GO terms and nine KEGG pathways with the number of peaks in each route is shown in Fig. 7.

A general classification of pathways obtained from the enrichment of meta-DBSs indicated that most of those pathways play a role in cancer, environmental information processing, and organismal systems (Fig. 8).

Fig. 8
figure 8

Sankey diagram for categories of KEGG pathways

Full size image

Evaluation of ChIP-seq meta-analysis results along with RNA-seq meta-analysis results

Recently we performed a meta-analysis of RNA-seq data on both MCF7 and T47D cell lines treated with 10 nM E2 for 8 h [32]. We investigated the expression of the top 50 TFs predicted by ChEA3 (Table 3). Comparing meta-analysis results of ChIP-seq and RNA-seq data showed that many TFs were up-regulated in RNA-seq meta-analysis or individual datasets.

Table 3 Differential express of top 50 TFs using ChEA3 for 617 meta-DBSs-associated common genes [32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52]
Full size table

Discussion

Breast cancer is the most incident cancer among women worldwide. The largest group of patients with BC is ER+. The estrogen receptor plays the role of a transcription factor when exposed to estrogen and triggers cellular proliferation and differentiation [2, 3]. Hence, evaluating the expression of TFs directly affected by ER can help identify and understand ER action's mechanism in cancer progression. Remarkably, combining more samples and performing meta-analysis will produce more stringent results than single experiments [9]. To the best of our knowledge in the current study, for the first time, we applied the workflow for meta-analysis of ChIP-seq data based on TFBSs in the MCF7 cell lines treated with 10 nM and 100 nM E2. First, samples were integrated, and known and unknown batch effects were removed. Next, the meta-analysis approach was conducted on TFBSs scores, and meta-DBSs were obtained. Our results showed significant peaks (specifically for TFs affected by ER) identified through the meta-analysis but not found in individual studies. For the strict selection of meta-DBSs, MCF7 cell lines meta-analysis results were compared with the results of T47D cell lines and initial individual studies, and common genes and the associated meta-DBSs were determined. With the enrichment of meta-DBSs correlated with common genes, several pathways with presumably clinical importance in BC were identified. Meta-DBSs-associated common genes were also enriched, and TFs were identified that have been predicted as potential key regulators in the progression of BC. In addition, we compared the meta-analysis results of both ChIP-seq and RNA-seq data.

Some of the differentially expressed genes (DEGs) identification methods in the RNA-seq context, such as DEseq2 [53] and edgeR [54], are also exploited to discover DBSs. However, the NOIseq package [26, 28] performs relatively better for obtaining DEGs [28]. Moreover, packages were developed for meta-analysis of RNA-seq data based on those methods [28, 55, 56]. To find meta-DBSs, we utilized the metaSeq package [28], in which NOIseq has been used. Before performing the meta-analysis, we faced two problems: (1) none of the mentioned methods removed unknown batch effects, and (2) the number of peaks and their associated genomic start and end intervals were different among samples, and batch effects could not be removed. Therefore, sample preparation was required. To this end, we obtained the same number, regions, and scores of TFBSs using the DiffBind package.

Herein, POU5F1B, ZNF662, ZNF442, KIN, ZNF410, and SGSM2 TFs were found in the meta-analysis of MCF7 cell lines treated with 10 nM E2 but not in individual studies. POU5F1B, a processed pseudogene highly homologous to OCT4, was recently shown to be transcribed in ER+ BC [57]. POU5F1B is specifically expressed in mammalian totipotent embryonic stem and germ cells and has a crucial role in regulating and maintaining pluripotency and self-renewal [57]. The Zinc finger protein family often functions in transcriptional regulation through sequence-specific DNA binding [58]. ZNF662, ZNF442, and ZNF410 were involved in the regulation of transcription by RNA polymerase II [59]. It has shown that low expression of G protein-coupled estrogen receptor 1 (GPER1) is significantly associated with adverse survival of BC patients [60]. ZNF662 was identified as one of the unique factors related to GPER-DNA binding [60]. ZNF442 plays a role in the strategy adopted by ER+ BC and triple-negative breast cancer (TNBC) cell lines for maintaining zinc homeostasis [61]. ZNF410 uniquely activates CHD4, one of the catalytic components of the Nucleosome Remodeling and Deacetylase (NuRD) complex [62]. NuRD has been shown to have opposing effects on cancer. For instance, promoting and inhibiting tumor growth and metastasis in different tissues such as ER+ BC [63, 64]. Recently, it has been shown that the Knockdown of DNA/RNA‑binding protein KIN17 (KIN) promotes apoptosis of TNBC [65]. Small G protein signaling modulator 2 (SGSM2) is involved in ER+ BC metastasis by enhancing migrator cell adhesion via interaction with E-cadherin [66].

With enrichment analysis of meta-DBSs-associated common genes by ChEA3, TFs including PCGF2, HNF1B, and ZBED6 were predicted as regulatory factors involved in BC progression. Based on our findings, no peaks associated with those TFs were found in individual studies and meta-analysis results. Maybe it is because all TFs do not always regulate their own expression. Polycomb group proteins (PcG) play a critical role in cancer development, proliferation, senescence, and carcinogenesis [67]. PCGF2 serves as a tumor suppressor in BC, gastric cancer, and colon cancer, probably for the negative regulation of Akt activation [68]. Recent studies have shown that methylation of homeobox genes, such as the HNF1B, plays a critical role in BC's insurgence or progression [69]. Zinc Finger BED-Type Containing 6 (ZBED6) has been shown to repress insulin-like growth factor 2 (IGF2) transcription [59]. The role of ZBED6 has only been reported in TNBC but is not well understood in ER+ BC so far [70].

Most TFs indicated in Fig. 4 were shared among results of individual studies, meta-analysis, and the top 50 TFs obtained from ChEA3 (Fig. 6). The ESR1 and PGR, which play an essential role in the development of BC, were ranked as top-ranking TFs (Fig. 6). Because ChEA3 ranks TFs only based on a gene list without any assumptions, it confirms the correctness of our results. It verifies that removing batch effects and performing a correct meta-analysis can produce more exact results. POLRMT and TFAM are human mitochondrial TFs [71]. As an essential mitochondrial DNA (mtDNA)-binding protein, TFAM functions in genome maintenance such as determining the abundance of the mitochondrial genome. Emerging evidence indicates that altered TFAM levels or mtDNA copy numbers may impact mitochondrial homeostasis in Alzheimer's and other neurodegenerative diseases [71]. Several studies suggested that TFAM TFs are potential biomarkers for the prognosis of BC [72]. POLRMT is a mitochondrial RNA polymerase that requires TFAM for mitochondrial transcription initiation [73]. Targeting both those factors is a promising strategy for sensitizing BC cells to chemotherapy [72, 74]. Although most of the datasets analyzed in the present study have used the chip-anti-ER (Santa Cruz HC-20, cat# SC-543), the tagging approach could alter the binding preferences of the protein, and therefore the experimental results.

Interestingly, the significant GO terms and KEGG pathways obtained from PSFEA in the present study were enriched by many predicted TFs by ChEA3 (Figs. 6 and 7). Therefore, meta-analysis led to the recognition of TFs involved in the development of BC. Cornification (GO:0070268) [75], anterior/posterior pattern specification (GO:0009952) [76], embryonic organ morphogenesis (GO:0048562) [77], response to inorganic substance (GO:0010035) [78], and negative regulation of transcription by RNA polymerase II (GO:0000122) [79] have been previously reported in ER+ BC. Tissue development (GO:0009888) [80] was enriched in BC. Cortical actin cytoskeleton (GO:0030864) [81], and cortical cytoskeleton (GO:0030863) [81] in pancreatic cancer, and cytoskeleton (GO:0005856) [82, 83] in ovarian cancer were enriched but not reported in ER+ BC previously. Regulation of anatomical structure morphogenesis (GO:0022603) [84] in colorectal cancer and phosphatidylcholine metabolic process (GO:0046470) [85] in hepatocellular carcinoma were enriched. Response to metal ion (GO:0010038) [86] in chronic myeloid leukemia cells, and negative regulation of RNA biosynthetic process (GO:1902679) [87] in Alzheimer’s Disease, were enriched but not reported in BC before. Regulatory region nucleic acid binding (GO:0001067) [88], lipid binding (GO:0008289) [89], RNA polymerase II regulatory region DNA binding (GO:0001012) [90], activating transcription factor binding (GO:0033613) [91], DNA-binding transcription factor binding (GO:0140297) [92], transcription coregulator activity (GO:0003712) [93] have been previously reported in ER+ BC. Glucose binding (GO:0005536) [94] was enriched in multiple myeloma but not reported in BC before.

The nine enriched KEGG pathways include Pathways in cancer (hsa05200) [95], Estrogen signaling pathway (hsa04915) [95], Rap1 signaling pathway (hsa04015) [96], Neurotrophin signaling pathway (hsa04722) [97], and Fluid shear stress and atherosclerosis (hsa05418) [98] have been previously reported in ER+ BC. MAPK signaling pathway (hsa04010) [99] was enriched in MCF7 cell lines of the resistance to treatment. Tight junction (hsa04530) [100], and Bladder cancer (hsa05219) [101, 102] have already been reported tumor BC. A study has shown that Thyroid cancer (hsa05216) [103] is related to metabolic pathways which were enriched in TNBC.

According to the cases mentioned, GO terms and pathways previously approved concerning BC were enriched (Fig. 7). Moreover, many GO terms and pathways identified were associated with gastrointestinal cancers, including colorectal and hepatocellular carcinoma. In particular, our findings showed that many TFBSs associated with mitochondrial genes are directly affected by ER, including POLRMT and TFAM TFs. Especially TFAM increased expression has been reported in a meta-analysis of MCF7 cell lines treated with E2 [32]. This may clarify why the metabolism pathways were identified along with the estrogen signaling pathway in BC (Fig. 7 and see Additional file 2: Table S22). Although some relationships are found between BC and hepatocellular carcinoma, more studies are needed on these TFs and their associated pathways.

The comparison of ChIP-seq and RNA-seq data meta-analysis results showed that many TFs predicted by ChEA3 were up-regulated in RNA-seq meta-analysis results. There can be several reasons why some TFs are down-regulated. FOXA1 is a determinant of drug resistance in breast cancer cells, and it has different functions in response to treatment in ER+ and TNBC cell lines [33] (Table 3). Investigation of primary regulatory regions has shown that there is plasticity in ER binding, with distinct ER binding profiles associated with clinical outcome. These differential ER binding profiles appear to be mediated by FOXA1 [104].

SPDEF also is with both oncogenic and tumor-suppressor functions in BC [105]. Therefore, according to its role in the cell, it can have a different expression. Furthermore, the treatment period of RNA-seq data can be effective in the gene expressions, so it is possible that at one time, the expression of the gene is increased, and at another time, it is down-regulated.

Our study was constrained by the number of datasets with the same conditions. Although the results were remarkable and substantial, they could be more precise and reliable if more data were combined. In addition, despite generating large volumes of datasets related to BC in databases, the data with the same conditions is negligible. Therefore, developing datasets with more uniform and focused conditions is could help to detect more precise meta-analysis results.

Conclusion

In the current study, we applied a workflow for integrating and performing meta-analysis of several ChIP-seq studies based on TFBSs scores in ER+ BC. The same number, regions, and scores of TFBSs were obtained for all samples. Then, the known and unknown batch effects were removed, and the meta-analysis was performed to obtain meta-DBSs. Some TFs were significantly identified in the meta-analysis but not in individual studies. Also, with meta-DBSs enrichment, many GO terms and pathways were identified that were not previously reported in BC. Finally, by enriching the meta-DBSs-associated genes, some TFs were prioritized and predicted as potential regulators. Those TFs were not found in the results of the meta-analysis and individual studies. Moreover, the results of the ChIP-seq meta-analysis were confirmed by comparing them with the meta-analysis of the RNA-seq data. The results showed that many TFs predicted by ChEA3 were up-regulated in RNA-seq meta-analysis or individual datasets results. As a suggestion, this workflow can be applied to other types of BC, such as HER2+ and other diseases, provided that there are sufficient datasets with the same conditions. It can also be performed on other TFs involved in BC that cooperate with ER as a cofactor, revealing new insights into ER mechanism of action.