Research

Identification of genomic features that uniquely impact estrogen receptor alpha binding and its effects on gene expression in endometrial cancer

    • 1Huntsman Cancer Institute, University of Utah, Salt Lake City, Utah 84112, USA;
    • 2Department of Biomedical Engineering, University of Utah, Salt Lake City, Utah 84112, USA;
    • 3Department of Oncological Sciences, University of Utah School of Medicine, Salt Lake City, Utah 84112, USA;
    • 4Department of Biochemistry, University of Utah School of Medicine, Salt Lake City, Utah 84112, USA;
    • 5Howard Hughes Medical Institute, University of Utah School of Medicine, Salt Lake City, Utah 84112, USA
Published July 17, 2026. Vol 36 Issue 9, pp. 1771-1784. https://doi.org/10.1101/gr.281792.125
Download PDF Cite Article Permissions Share
Alerts Please log-in to or register for your personal account in order to apply an alert.Log-in/register
cover of Genome Research Vol 36 Issue 9
Current Issue:
September 2026, Vol. 36, No. 9
Focus view

Abstract

Estrogen receptor 1 (ESR1, also known as estrogen receptor alpha or ER) is an established oncogenic transcription factor in breast and endometrial cancer; however, more is known about the mechanisms controlling ER behavior in breast cancer, and therapies targeting ER have been much more successful in breast cancer. To address this disparity, we characterize the genomic features that control ER in endometrial cancer and determine to what extent these factors differ from those in breast cancer. We focus on the locations of estrogen response elements (EREs), ER's preferred DNA-binding motif, throughout the human genome. To identify factors that predict ER genomic binding and effects on target gene expression, we apply machine learning to genomic data for each ERE in Ishikawa cells (ER-positive endometrial cancer) and T-47D cells (ER-positive breast cancer). Many of these factors, such as chromatin accessibility and histone modifications, are predictive of ER activity in both cell lines. However, the transcription factors that predict ER activity are cell type specific, including FOXA1 and GATA3 in T-47D cells and ETV4 and SOX17 in Ishikawa cells. In addition, the features that predict ER binding and effects on gene expression differ, with transcription at EREs in the absence of estrogen being predictive of ER regulatory activity. A CRISPR knockout screen in Ishikawa cells, as well as follow-up experiments, confirms the discovery that SOX17 controls ER activity in endometrial cancer cells. These results identify important genomic features of ER binding and regulatory activity and how these features differ between endometrial cancer and breast cancer cells.

Estrogen receptor 1 (ESR1, also known as estrogen receptor alpha or ER) is an established oncogene in the hormone-dependent forms of breast and endometrial cancers (Jordan and Brodie 2007; Paterni et al. 2014; Rodriguez et al. 2019). Although ER signaling has been extensively studied in breast cancer, leading to the development of targeted therapies such as selective estrogen receptor modulators (SERMs) and degraders (SERDs) (Jordan 2006; Lee et al. 2017), similar therapeutic advancements have not been achieved in endometrial cancer. For instance, tamoxifen, a SERM that functions as an ER antagonist in breast tissue, exerts agonistic effects in the endometrium, preventing its use as an endometrial cancer treatment (Hu et al. 2015). Despite its established clinical significance, ER and the related molecular mechanisms contributing to endometrial cancer progression remain poorly understood. With the incidence and mortality of endometrial cancer rising year after year (Siegel et al. 2025), there is an urgent need to understand the molecular mechanisms of endometrial cancer in order to identify therapeutic targets and improve patient outcomes.

One likely contributor to the limited translatability of the ER-targeting therapies of breast cancer to endometrial cancer is the tissue type–specific behavior of ER. Fewer than 10% of ER genomic binding sites are shared between the Ishikawa (ER+ endometrial cancer) and the T-47D (ER+ breast cancer) cell lines (Gertz et al. 2013), and ER binding site patterns in patient samples segregate distinctly by tumor type (Droog et al. 2016; Rodriguez et al. 2020). Cofactors of ER also differ between breast and endometrial cancer. FOXA1 and GATA3 are essential for regulating ER binding in the breast cancer context (Hurtado et al. 2011; Kong et al. 2011; Theodorou et al. 2013), and their corresponding motifs are uniquely enriched at ER binding sites in breast cancer patient samples compared with endometrial cancer patient samples (Ross-Innes et al. 2012; Rodriguez et al. 2020). However, GATA3 is rarely expressed in endometrial cancer, and FOXA1 has minimal overlap with ER genomic binding in endometrial cancer (Droog et al. 2016; Rodriguez et al. 2020). ETV4 has been shown to regulate ER binding in endometrial cancer cells but not in breast cancer cells, and the ETS and SOX motifs are uniquely enriched at ER binding sites in endometrial cancer patient samples compared with breast cancer patient samples (Rodriguez et al. 2020).

The preferred DNA-binding motif of ER is the estrogen response element (ERE) (Klein-Hitpass et al. 1988), which is important for recruiting ER to gene expression enhancers (Carleton et al. 2020). Like other transcription factors, ER does not bind to all EREs within the genome (Gertz et al. 2013), and disruption of ER-bound EREs does not always alter gene expression (Carleton et al. 2017; Ginley-Hidinger et al. 2019; Abewe et al. 2025). This suggests that the presence of an ERE alone is insufficient to predict ER occupancy or transcriptional regulatory activity. The locations of EREs across the genome are consistent across breast and endometrial cancer cells, providing the opportunity to make direct comparisons of their regulatory potential between tissue types.

Machine learning approaches provide a powerful tool for characterizing the ER-mediated molecular mechanisms in endometrial cancer and uncovering how these mechanisms differ from those of breast cancer. Many large-scale data sets have been generated for breast and endometrial cancer (Ginley-Hidinger et al. 2024; Abewe et al. 2025), and machine learning allows us to uncover important features of potential regulatory regions. Gregoricchio et al. (2025) successfully used machine learning to identify an enhancer mutation that alleviates transcriptional repression of ER during endometrial tumorigenesis. In our previous work, we successfully used Boruta (Kursa and Rudnicki 2010), a feature selection algorithm, to identify predictors of timing, noise, and expression levels of the transcriptional response to estrogen (Ginley-Hidinger et al. 2024) and to identify predictors of changes in 3D genomic interactions in response to estrogen (Abewe et al. 2025). Other researchers have also used Boruta to identify gene variants that may lead to disease states (Ragunthar and Selvakumar 2019). Here, we apply Boruta to the study of EREs that bind ER and impact gene expression. We also performed a CRISPR knockout (KO) screen of epigenetic and transcription factors to identify important contributors to ER activity in endometrial cancer cells.

Results

Very few EREs in the genome are bound by ER and affect gene expression

To evaluate the usage of EREs by ER, we first identified 82,534 significant full ERE matches in the human female genome (see Methods). We considered a bound ERE to be one that overlapped with an ER ChIP-seq peak from the corresponding cell line (Fig. 1A). We considered an ERE that affects gene expression, termed “regulatory,” to be one that is located near the transcription start site (TSS) of a gene that changes expression in response to estrogen, using a 100 kb distance cutoff based on our previous analysis (Fig. 1A; Carleton et al. 2017). In Ishikawa cells, 2154 (2.6%) EREs are bound by ER, and 935 (43.4% of bound EREs) EREs affect expression (Fig. 1B). In T-47D cells, 2329 (2.8%) EREs are bound, and 1005 (43.2% of bound EREs) EREs affect expression (Fig. 1B). In Ishikawa cells, ∼20% of regulatory EREs are associated with both up- and downregulated genes, 62% are only associated with upregulated genes, and 18% are only associated with downregulated genes. In T-47D cells, ∼12% of regulatory EREs are associated with both up- and downregulated genes, 67% are only associated with upregulated genes, and 21% are only associated with downregulated genes. Ishikawa and T-47D cells have been shown to have roughly equal levels of ER expression, which matches their similar levels of bound and regulatory EREs (Gertz et al. 2013). Between the cell lines, 895 of the bound EREs (24.9% of all bound EREs) and 335 of the regulatory EREs (20.09% of all regulatory EREs) are shared in both cell lines. With the vast majority of EREs not bound by ER in a given cell line, we evaluated whether the number of EREs in the human genome was mostly owing to random chance or if there were signs of evolutionary pressure to maintain a high number of EREs in the genome. We counted the EREs on Chromosome 1 and then shuffled the sequence of Chromosome 1 10 times and recounted the EREs. Chromosome 1 contains 6922 EREs compared with an average number of 5005 EREs in the shuffled chromosomes, with a standard deviation of 87. Therefore, the number of EREs on Chromosome 1 is 22 standard deviations higher than the mean from the shuffled chromosomes (Fig. 1C), suggesting that there is some evolutionary pressure to preserve the high number of EREs in the human genome, even if most remain unbound by ER within a cell type.

Figure 1.

Most EREs are unbound by ER and do not affect gene expression. (A) Illustration shows the definition of bound EREs (left) and regulatory EREs (right). (B) Pie chart on the left shows the number of EREs that are bound in both cell lines, are bound only in Ishikawa, are bound only in T-47D, and are unbound. Pie charts on the right show, out of the EREs that are bound in Ishikawa (top) and T-47D (bottom), the number of regulatory EREs unique to that cell line, regulatory EREs shared between cell lines, and nonregulatory EREs. (C) Bar chart shows the number of EREs found in Chromosome 1 compared with the number of EREs in 10 different random shufflings of Chromosome 1. (D) Scatter plot displays the percentage of bound EREs shared with the Ishikawa cell line (y-axis) and with the T-47D cell line (x-axis) from endometrial cancer patient samples (blue circles) and breast cancer patient samples (purple squares). All patient samples share more bound EREs with their corresponding cell line than with their noncorresponding cell line.

1771f01
Open in new tabDownload PowerPointLink to figure

Analysis of ER ChIP-seq data from breast and endometrial cancer patient samples (Rodriguez et al. 2020) was used to determine how many EREs were bound in the patient samples and how many bound EREs they shared with the Ishikawa and T-47D cell lines (Fig. 1D). The average percentage of bound EREs in the endometrial cancer patient samples was 1.63% with a standard deviation of 1.22%, and the average percentage of bound EREs in the breast cancer patient samples was 1.59% with a standard deviation of 1.36%. The bound EREs in the endometrial cancer patient samples matched more closely with the bound EREs in the Ishikawa cell line (Wilcoxon signed-rank test; P-value = 9.766 × 10−4), and the bound EREs in the breast cancer patient samples matched more closely with the bound EREs in the T-47D cell line (Wilcoxon signed-rank test, P-value = 1.526 × 10−5). For our further analyses, we focus on Ishikawa and T-47D cell lines, because they represent the broad differences observed in ER binding between endometrial and breast cancer specimens.

Several factors predict ERE binding in a cell type–specific manner

To identify factors that can predict whether an ERE will be bound by ER, we compiled the following data sets: the relative affinity of the ERE for ER (determined by the sequence of the ERE, see Methods), ChIP-seq for multiple transcription factors and histone modifications, PRO-seq, DNase-seq, ATAC-seq, and HiChIP to determine the number of 3D connections associated with each ERE (Supplemental Table S1). These data sets were previously published, with the exception of the SOX17 ChIP-seq data in Ishikawa cells. We previously found that the SOX motif is enriched at ER binding sites in endometrial cancer specifically (Rodriguez et al. 2020), and SOX17 is the most highly expressed SOX factor in Ishikawa cells. Our data sets differed somewhat between Ishikawa cells and T-47D cells owing to differences in data availability. We focused only on data sets taken from cells grown in the absence of estrogen to ensure that the data represented a state preceding ER binding rather than a state caused by ER binding.

To identify which factors are predictive of an ERE's ability to bind ER, we used the Boruta feature selection algorithm (Kursa and Rudnicki 2010). In the Ishikawa cells, 22 factors were in the initial data set tested in Boruta. Of those, 19 were confirmed important in predicting ER binding, one was tentative, and two were rejected (Fig. 2A; Supplemental Table S2). In the T-47D cells, 21 factors were in the initial data set tested in Boruta. Of those, 18 were confirmed important in predicting ER binding, and three were rejected (Fig. 2B; Supplemental Table S2). For all factors in both cell lines that were confirmed as important, their average value was higher for bound EREs than for unbound EREs, suggesting that all predictors are positively associated with ER binding (Fig. 2A,B), which is consistent with a focus on factors related to active gene regulation. Because there is a large imbalance between the number of bound and unbound EREs, we also performed 100 Boruta iterations on subsamples of our data set with equal numbers of bound and unbound EREs. These results largely aligned with the results from running Boruta on the full data set, with a few factors, such as CTCF, changing their designation (Supplemental Fig. 1; Supplemental Table S2).

Figure 2.

Bound EREs are predicted by sequence, chromatin accessibility, histone modifications, and cell line–specific transcription factors. (A,B) Boxplots show the ranked factor importance for predicting bound EREs by Boruta for Ishikawa (A, left) and T-47D (B, left). Colors indicate Boruta determination of importance; blue is confirmed, light gray is tentative, red is rejected, and dark gray is a Boruta attribute. Heatmaps show the normalized average value of each factor at bound and unbound EREs for Ishikawa (A, right) and T-47D (B, right). (C,D) Scatter plots show the relationship between the log of relative affinity for ER of an ERE and the percentage of EREs bound by ER for Ishikawa (C) and T-47D (D). (E,F) Bar charts show the ROC AUC of each factor individually in predicting bound EREs in Ishikawa (E) and T-47D (F). Bar colors correspond with Boruta determination of importance. The dotted line shows the value 0.5, indicating an ROC AUC from random prediction. (G) Scatter plot shows the comparison of Boruta importance scores in predicting bound EREs between cell lines from the experimental data, with a Pearson correlation coefficient of 0.647. (H) Scatter plots show the comparison of ROC AUCs in predicting bound EREs between cell lines from the experimental data set, with a Pearson correlation coefficient of 0.464.

1771f02
Open in new tabDownload PowerPointLink to figure

For both cell lines, the top scoring factor in predicting ER binding was relative affinity for ER (Fig. 2A,B). Further analysis showed that relative affinity is positively correlated with the percentage of EREs bound by ER (Fig. 2C,D), indicating that the closer an ERE sequence is to the optimal ERE, the more likely it is to be bound. Relative affinity is higher on average in bound EREs shared between the cell lines than in EREs that are uniquely bound in one cell line or the other (Wilcoxon rank-sum test, P-value = 1.2116 × 10−42). Despite the high importance of relative affinity, optimal EREs, EREs with the highest possible relative affinity, were only bound by ER 40% of the time, indicating that ER binding affinity is mostly determined by other factors. For optimal EREs, 46% are in open chromatin in Ishikawa cells, and 11% are in T-47D cells. Of these optimal EREs in open chromatin, 60% are bound by ER in Ishikawa cells and 79% in T-47D cells. The percentage of EREs in open chromatin bound by ER drops as relative affinity drops. For EREs scoring in the lowest 10% of relative affinity, 21% are in open chromatin in Ishikawa cells and 6% in T-47D cells. Of these low-scoring EREs in open chromatin, 3% are bound in Ishikawa cells and 8% in T-47D cells.

Additional factors found to be predictive in both cell lines included chromatin accessibility, as measured by DNase-seq and ATAC-seq, and activating histone modifications, including H3K4me1. For T-47D cells, FOXA1 and GATA3 binding were found to be predictive, which supports the validity of Boruta, as FOXA1 and GATA3 have been established to control ER binding in the context of breast cancer (Hurtado et al. 2011; Kong et al. 2011; Theodorou et al. 2013). ETV4, which we previously identified as a contributor to ER binding in endometrial cancer (Rodriguez et al. 2020), was identified as predictive of ER binding in the Ishikawa cells (Fig. 2A). The Boruta results are supported by the receiver operator characteristic curve (ROC) analysis, in which factors mentioned above exhibited areas under the curve (AUC) substantially greater than 0.5 (Fig. 2E,F).

Motif analysis reveals more cell line–specific predictors of ER binding

To expand our analysis beyond experimentally assayed transcription factors, we used the HOMER motif database (Heinz et al. 2010), which provides the genomic locations of the cognate DNA-binding motifs for 450 DNA-binding proteins. We first performed clustering of the 450 motifs to remove redundancy, which created 270 motif clusters (Supplemental Table S3), and then performed Boruta on these 270 motif clusters. For this data set, each factor was binary: Either an ERE had one of the motifs in the cluster within 100 bp of the ERE, or it did not.

In Ishikawa cells, 103 clusters were confirmed as important in predicting ER binding, with 32 clusters being designated as tentative and 135 clusters being rejected as unimportant in predicting ER binding (Supplemental Table S3). In T-47D cells, 98 clusters were confirmed as important in predicting ER binding, with 26 clusters being designated as tentative and 146 clusters being rejected in predicting ER binding (Supplemental Table S3). However, many of these confirmed motif clusters contained motifs that are very similar to the ERE motif. Therefore, we removed any motifs containing at least an ERE half site from further investigation. To narrow down our focus, we further analyzed two groups of clusters: (1) clusters that had an importance score higher than 10 and (2) clusters that were confirmed in only one cell line.

All the motif clusters that had an importance score above 10 in either cell line were confirmed as important for predicting binding in both cell lines. Eight clusters in Ishikawa and 12 clusters in T-47D had an importance score greater than 10 (Supplemental Fig. 2A). We observed some notable motifs in the clusters that were confirmed in only one cell line (Supplemental Fig. 2B,C). The compound motif of NF1:FOXA1 was confirmed only in T-47D, along with a cluster containing the GATA3 motif. The cluster containing the SOX17 motif was only confirmed in Ishikawa. These results support our findings from experimental data sets while identifying some additional factors that may contribute to ER binding.

Factors that contribute to ER binding are mostly cell type specific

For both cell lines, relative affinity for ER had the highest Boruta importance score (Fig. 2G). It also had high predictive ability in ROC AUC analysis in both cell lines. The relative affinity of ERE is the exact same in both cell lines; therefore, ERE strength does not explain why different sets of EREs are bound in the two cell lines. Of the genomic features assayed in both cell lines, we found several factors that are better at predicting ER binding in one cell line over the other. In both Boruta and ROC AUC analyses, FOXA1 and JUN were better predictors in T-47D cells, whereas ATAC-seq, LSD1, and SIN3A had higher predictive ability in Ishikawa cells (Fig. 2G,H). Most motif clusters had similarly poor predictive ability in both cell lines, but there were a handful of outliers for one cell line or the other (Supplemental Fig. 2D). In addition to the features assessed in both cell lines, transcription factors that were predicted to be important for ER binding, GATA3, ETV4, and SOX17, were only examined in one cell line because of cell type–specific expression (Supplemental Fig. 3A; Gertz et al. 2013). The cell type–specific expression of these factors, as well as FOXA1, is also observed in patient tumor data from The Cancer Genome Atlas (TCGA) (Supplemental Fig. 3B–E; The ICGC/TCGA Pan-Cancer Analysis of Whole Genomes Consortium 2020). Overall, these results are consistent with previous findings that ER binding is controlled by different factors in breast and endometrial cancer cells.

Predictive multivariable model of ERE binding

With the results from Boruta and the ROC AUC analyses, we sought to create multivariable logistic regression models that could predict, with high sensitivity and specificity, whether any given ERE would be bound by ER. For both cell lines, we ran two-variable logistic regression for all possible combinations using our nonmotif data set and downsampling while creating the training groups to account for the unbalanced data set. We then ranked the variables based on the average ROC AUC across their two-variable logistic regression models (Supplemental Fig. 4A,B). These ranks correlated well with the Boruta importance score ranks in both cell lines, with Pearson correlation coefficients of 0.671 and 0.592 for Ishikawa and T-47D, respectively. As a baseline for more complex models, we started with the single variable relative affinity because of its high Boruta importance score and because it performed well in all two-variable analyses. For Ishikawa EREs, this model can predict which EREs will be bound with a sensitivity of 63.5% and a specificity of 75.9% (ROC AUC = 0.741) (Supplemental Fig. 4C). For T-47D EREs, the model has a sensitivity of 60.4% and a specificity of 74.9% (ROC AUC = 0.720) (Supplemental Fig. 4D). These models can be improved upon, as relative affinity is the same between the cell lines, and a different set of EREs is bound in each cell line. To do this, we added ATAC-seq, a measure of chromatin accessibility, and ChIP-seq of H3K4me1, a histone modification associated with active enhancers. Even though these data sets are predictive of ER binding in both cell lines, they can explain the cell line differences, as the genomic locations of accessible chromatin and H3K4me1 differ between T-47D and Ishikawa cells (Gertz et al. 2013). For Ishikawa cells, adding ATAC-seq and H3K4me1 ChIP-seq improved the sensitivity to 78.5%, the specificity to 87.4%, and the ROC AUC to 0.907 (Supplemental Fig. 4C). For T-47D cells, it improved the sensitivity to 73.0%, the specificity to 84.9%, and the ROC AUC to 0.862 (Supplemental Fig. 4D). Another way to improve the model is adding cell type–specific transcription factors. For Ishikawa cells, having the variables of relative affinity, ETV4 ChIP-seq, and SOX17 ChIP-seq creates a model with 75.1% sensitivity, 83.5% specificity, and an ROC AUC of 0.869 (Supplemental Fig. 4C). For T-47D cells, having the variables of relative affinity, FOXA1 ChIP-seq, and GATA3 ChIP-seq creates a model with 75.6% sensitivity, 87.1% specificity, and an ROC AUC of 0.894 (Supplemental Fig. 4D). These results suggest that the sequence of the ERE, accessible chromatin, and histone modifications associated with active enhancer activity, such as H3K4me1, are important determinants of ER binding potential, regardless of the cell line. However, sequence-specific transcription factors that determine ER binding potential are cell line specific, with FOXA1 and GATA3 determining ER binding potential in T-47D cells and ETV4 and SOX17 determining ER binding potential in Ishikawa cells.

To validate our findings in another endometrial cancer model, we performed ATAC-seq, H3K4me1 ChIP-seq, and SOX17 ChIP-seq in HCI-EC-23 (Rush et al. 2022) to recreate our logistic regression models for predicting ER binding. We were unable to perform ETV4 ChIP-seq for technical reasons. HCI-EC-23 has lower ER activity compared with Ishikawa cells, as reflected by it having only 386 bound EREs. Relative affinity alone has a sensitivity of 63.4% and a specificity of 75.9% (ROC AUC = 0.742) (Supplemental Fig. 4E). Adding SOX17 increased the sensitivity to 70.7% and the specificity to 80.3% (ROC AUC = 0.833) (Supplemental Fig. 4E). When we added ATAC-seq and H3K4me1 ChIP-seq to relative affinity, the model improved to a sensitivity of 78.0%, a specificity of 87.5%, and an ROC AUC of 0.905. A model including all four variables has a sensitivity of 78.8%, a specificity of 87.7%, and an ROC AUC of 0.908 (Supplemental Fig. 4E). The performance of these models is very similar to Ishikawa cells, suggesting that the same features are important for ERE binding across endometrial cancer models.

The models are only slightly improved by combining all five variables. For Ishikawa cells, a model with all five variables has a 79.5% sensitivity, 87.9% specificity, and an ROC AUC of 0.913 (Supplemental Fig. 4C). For T-47D, a model with all five variables has a 79.4% sensitivity, 88.4% specificity, and an ROC AUC of 0.913 (Supplemental Fig. 4D). Similarly, when all possible variables are used, the Ishikawa model has an 80.4% sensitivity, 88.0% specificity, and an ROC AUC of 0.917, and the T-47D model has an 80.5% sensitivity, 90.0% specificity, and an ROC AUC of 0.924.

Transcription and chromatin looping predict regulatory EREs

To identify factors that are predictive of a bound ERE's ability to affect gene expression, we analyzed the same data sets as were used when analyzing binding. We also added the factor “target gene 3D loop,” which is a binary classifier that indicates whether an ERE has a 3D genomic interaction with the TSS of an estrogen-regulated gene, based on HiChIP data. We first compared all regulatory EREs against all nonregulatory bound EREs. For the Ishikawa cells, 15 factors were confirmed important in predicting whether an ERE would affect gene expression, two were tentative, and six were rejected (Fig. 3A; Supplemental Table S2). For T-47D cells, 16 factors were confirmed, one was tentative, and five were rejected (Fig. 3B; Supplemental Table S2). For both cell lines, the top two scoring features were target gene 3D loop and PRO-seq. This suggests that 3D genomic interactions and active transcription at promoter distal EREs in the absence of estrogen, as measured by PRO-seq, are both important in distinguishing regulatory EREs from nonregulatory EREs. Once again, the average value of each factor confirmed important by Boruta was higher for regulatory EREs than for nonregulatory EREs, indicating that our data set is biased toward factors that promote ER activity (Fig. 3A,B). We also performed Boruta analysis comparing only promoter-distal regulatory EREs to nonregulatory EREs, only regulatory EREs of upregulated genes to nonregulatory EREs, and only regulatory EREs of downregulated genes to nonregulatory EREs. The results from these analyses were very similar to the analyses including all regulatory EREs (Supplemental Table S4; Supplemental Fig. 5).

Figure 3.

Regulatory EREs are enriched for 3D genomic interactions and active transcription. (A,B) Boxplots show the ranked factor importance for predicting regulatory EREs by Boruta for Ishikawa (A, left) and T-47D (B, left). Colors indicate Boruta determination of importance; blue is confirmed, light gray is tentative, red is rejected, and dark gray is a Boruta attribute. Heatmaps show the normalized average value of each factor at regulatory and nonregulatory EREs for Ishikawa (A, right) and T-47D (B, right). (C,D) Bar charts show the ROC AUC of each factor individually in predicting regulatory EREs in Ishikawa (C) and T-47D (D). Bar colors correspond to Boruta determination of importance. The dotted line shows the value 0.5, indicating an ROC AUC from random prediction. (E) Scatter plot shows the comparison of Boruta importance scores in predicting regulatory EREs between cell lines from the experimental data, with a Pearson correlation coefficient of 0.843. (F) Scatter plot shows the comparison of ROC AUCs in predicting regulatory EREs between cell lines from the experimental data set, with a Pearson correlation coefficient of 0.628.

1771f03
Open in new tabDownload PowerPointLink to figure

We also calculated ROC AUCs for this data set, and target gene 3D loop and PRO-seq were both highly ranked in Ishikawa and T-47D. However, overall, the ROC AUCs for predicting regulatory EREs were lower than those for predicting bound EREs (Fig. 3C,D). In Ishikawa cells, the average ROC AUC for predicting bound EREs was 0.663, and the average ROC AUC for predicting regulatory EREs was 0.557 (Wilcoxon rank-sum test, P-value = 4.62 × 10−7). In T-47D cells, the average ROC AUC for predicting bound EREs was 0.689, and the average ROC AUC for predicting regulatory EREs was 0.579 (Wilcoxon rank-sum test, P-value = 4.31 × 10−6).

We next determined which motifs are predictive of an ERE's ability to affect gene expression. Very few of the 270 motif clusters were found to be important; nine were confirmed important in the Ishikawa data set and 11 in the T-47D data set, whereas seven were tentative for Ishikawa and three for T-47D (Supplemental Fig. 6A,B; Supplemental Table S3). Only one cluster was confirmed in both cell lines: the MAZ motif, which is very GC rich. It should also be noted that relative affinity for ER scored very poorly for predicting regulatory EREs (Fig. 3A,B). In general, sequence-based features are less predictive of regulatory EREs than experimental data sets such as ChIP-seq, PRO-seq, and HiChIP.

Similar features predict regulatory EREs across cell lines

When we directly compare Ishikawa and T-47D regulatory ERE-associated factors, we observed several similarities. For both cell lines, interaction with an estrogen-responsive gene via looping and transcription prior to estrogen treatment at the location of the ERE are the top predictors of a bound ERE's ability to affect gene expression. This is shown both with Boruta importance scores and ROC AUCs (Fig. 3E,F). We observed high correlation coefficients in both Boruta importance scores (r = 0.843) and ROC AUCs (r = 0.628). JUN ChIP-seq is a better predictor in T-47D cells, and SIN3A ChIP-seq is a better predictor in Ishikawa cells, just as they were for predicting binding. Supplemental Figure 6C shows the comparisons of importance scores from the Boruta analysis using motif data. These data show very little concordance between the cell lines; however, because motif data overall was not found to be very effective in predicting regulatory EREs, the lack of correlation is expected.

Predictive multivariable model of regulatory EREs

As we did with ERE binding, we sought to create a multivariable logistic regression model that could identify regulatory EREs with high sensitivity and specificity. We performed logistic regression using all possible two-variable combinations from the nonmotif data sets and ranked variables by their mean ROC AUC (Supplemental Fig. 7A,B). As with ER binding, these ranks correlated well with the Boruta importance score ranks in both cell lines, with Pearson correlation coefficients of 0.872 and 0.671 for Ishikawa and T-47D, respectively. As a baseline, we used the single variable model for target gene 3D loop, because it had the highest importance score in both cell lines in the Boruta analysis and performed well in all two-variable models. For Ishikawa cells, that model has a sensitivity of 34.8%, a specificity of 99.1%, and an ROC AUC of 0.670 (Supplemental Fig. 7C). For T-47D cells, that model has a sensitivity of 42.0%, a specificity of 89.9%, and an ROC AUC of 0.660 (Supplemental Fig. 7D). To expand the model for Ishikawa cells, we used target gene 3D loop, PRO-seq, and SIN3A ChIP-seq, which returned a sensitivity of 40.1%, a specificity of 94.9%, and an ROC AUC of 0.743 (Supplemental Fig. 7C). For the T-47D multivariable model, we used target gene 3D loop, PRO-seq, and JUN ChIP-seq, which returned a sensitivity of 44.4%, a specificity of 88.7%, and an ROC AUC of 0.723 (Supplemental Fig. 7D). From these models and the results of the Boruta analyses, we can conclude that 3D genomic interactions and active transcription at promoter distal enhancers are important determinants of a bound ERE's regulatory potential, along with the presence of SIN3A in Ishikawa cells and the presence of JUN in T-47D cells. However, our models achieved high specificity only and had low sensitivity, indicating that these factors can accurately identify a regulatory ERE but that most regulatory EREs are missed.

Different factors associate with bound EREs and regulatory EREs

The factors that predict whether an ERE will bind ER and the factors that predict whether a bound ERE will affect gene expression are largely different. In both Ishikawa and T-47D cells, relative affinity is only predictive of whether an ERE will bind ER, and PRO-seq is more predictive of whether a bound ERE will affect gene expression than it is of whether an ERE will bind ER (Fig. 4A,B). Differences between the cell lines are more prevalent in predicting binding than in predicting effects on expression. There is higher correlation between the Boruta importance scores shown in the scatter plot for predicting regulatory EREs (r = 0.843) (Fig. 3E) than there is shown in the scatter plot for predicting bound EREs (r = 0.647) (Fig. 2G). There are strong differences in estrogen-mediated gene regulation between breast cancer cells and endometrial cancer cells, and these results suggest that these differences arise more prominently from where ER binds within the genome than from how ER regulates gene expression once bound to an ERE.

Figure 4.

Different factors are predictive of bound EREs and regulatory EREs. Scatter plots show Boruta importance scores of each factor in experimental data for predicting bound EREs versus predicting regulatory EREs in Ishikawa (A) and T-47D (B).

1771f04
Open in new tabDownload PowerPointLink to figure

CRISPR KO screen reveals factors important for ER activity in Ishikawa cells

To functionally evaluate the importance of different factors in mediating ER activity, we performed a CRISPR KO screen targeting 2500 genes encoding epigenetic and transcription factors that could influence ER activity. To measure ER activity, we created an Ishikawa line with an integrated GFP reporter driven by three consecutive EREs. The reporter line was transduced with the Human Epigenetic Knockout lentiviral library (Williams et al. 2020). After selection, growth, and 72 h of estrogen treatment, cells were sorted into four bins based on their GFP levels (negative, low, medium, and high). Genes whose guide RNAs were enriched in the negative GFP bin compared with the other three bins were determined to be ER-enabling genes, and genes whose guide RNAs were enriched in the high GFP bin compared with the other three bins were determined to be ER-blocking genes (Supplemental Table S5). We identified 18 significant genes (Fig. 5A,B), including ESR1, the gene encoding ER. In general, the ER-enabling genes supported conclusions made from the Boruta and ROC AUC analyses. As expected, there were several significant genes encoding histone-modifying enzymes, including ASH1l, KAT6A, KMT2A, and BRPF1, all of which are associated with transcriptional activation. Two ER-blocking genes were identified: CSDE1 and RUVBL1. CSDE1 has been shown to both enable and block gene expression in the cancer context as an RNA-binding protein (Ciocia et al. 2024). RUVBL1 encodes an H4 and H2A acetyltransferase, which is typically associated with transcriptional activation (Doyon et al. 2004); therefore, it may indirectly block ER activity.

Figure 5.

CRISPR KO screen identifies important genes for ER activity in Ishikawa cells. (A) Volcano plot shows the ER-enabling and ER-blocking genes with their log2 fold changes and adjusted P-values. (B) Stacked bar plot shows the gRNA counts for each significant gene in the four GFP bins. The gRNAs for the ER-enabling genes are enriched in the negative GFP bin (black), and the gRNAs for the ER-blocking genes are enriched in the high GFP bin (darkest green).

1771f05
Open in new tabDownload PowerPointLink to figure

SOX17 is important for ER activity in Ishikawa cells

SOX17 was identified as an ER-enabling gene in the CRISPR KO screen, which agrees with our computational analyses, suggesting that SOX17 regulates ER activity in endometrial cancer. In support of this role, SOX17 is more highly expressed in endometrial tumor tissue than in normal tissue or breast cancer (Supplemental Fig. 3E). SOX17 expression is also positively correlated with ESR1 expression in endometrial cancer, with a Spearman's correlation of 0.556 (Supplemental Fig. 8A). Based on DepMap data (Arafeh et al. 2025), endometrial cancer models are more dependent on SOX17 compared with breast cancer models (Wilcoxon signed-rank test, P-value = 6.5 × 10−4) and models from other cancer types (Wilcoxon signed-rank test, P-value = 0.0017), with the exception of ovarian cancer models (Supplemental Fig. 8B). Together, our results suggest that SOX17 could be an important factor in promoting ER activity in endometrial cancer.

To further investigate the role of SOX17 in regulating ER activity in endometrial cancer cells, we created SOX17 KO lines derived from Ishikawa cells using CRISPR and individually cloned guide RNAs. We also created ESR1 KO Ishikawa cells as a positive control. We verified the absence of SOX17 and ER protein in the respective KOs (Supplemental Fig. 9A–E). To assess effects on ER activity, we performed an ERE reporter assay, similar to the one used in the CRISPR screen, and found that ER activity was decreased from 40-fold, over no estrogen controls, to 10-fold in SOX17 KO lines; ESR1 KO drove ERE levels back to no estrogen control levels as expected (Fig. 6A). To investigate the effects on estrogen-mediated gene expression, we performed RNA-seq with and without estrogen treatment in SOX17 KO lines and control lines. ESR1 expression itself was affected, decreasing from an average of 6899 normalized RNA-seq counts in control lines to an average of 1472 normalized counts in SOX17 KO lines (Fig. 6B). In control lines, the expression of 3835 genes was significantly impacted by estrogen treatment, whereas only 317 genes changed expression in response to estrogen treatment in the SOX17 KO lines. The magnitude of expression responses, for both upregulated and downregulated genes, was significantly reduced in SOX17 KO cells (Fig. 6C,D; Supplemental Fig. 9F).

Figure 6.

SOX17 KO decreases ER activity. (A) Bar chart shows ERE luciferase reporter activity relative to no estrogen treatment in SOX17 KO lines compared with a control and an ESR1 KO line. (B) Bar graph shows decreased ESR1 mRNA normalized counts in SOX17 KO lines compared with a control. (C,D) Aggregate plots show Z-scores comparing DMSO-treated and estrogen (E2)-treated samples from the RNA-seq data for SOX17 KO lines and control lines for E2 upregulated genes (C) and E2 downregulated genes (D). (E) Browser track shows decreased ER binding at estrogen-responsive genes CISH and PGR. (F) Heatmap shows decreased ER genomic binding signal in SOX17 KO lines compared with control lines across all ER bound sites. (G) Browser track shows six SOX17 binding sites near ESR1 gene locus. (H) Venn diagram shows overlap of binding sites between ER and SOX17 in parental Ishikawa cells.

1771f06
Open in new tabDownload PowerPointLink to figure

ER genomic binding was also reduced in SOX17 KO lines at estrogen-responsive genes such as CISH and PGR (Fig. 6E) and across the genome (Fig. 6F). ER ChIP-seq in SOX17 KO lines produced an average of 429 peaks compared with 2550 peaks in wild-type cells (Fig. 6F). The reduction in ER activity may be owing to two separate causes: (1) SOX17 regulates ER transcription by binding to six sites near the ESR1 gene locus (Fig. 6G), consistent with the fivefold decrease in ESR1 expression levels in the SOX17 KO cells (Fig. 6B), and (2) SOX17 binding overlaps with ER at many genomic sites, potentially aiding in ER-driven transcription regulation. In parental Ishikawa cells, ER has 8621 binding sites in the presence of estrogen, SOX17 has 3970 in the absence of estrogen, and they overlap at 1184 sites (30% of SOX17 bound sites) (Fig. 6H), suggesting that SOX17 binding at these sites may help facilitate ER binding. Overall, we found that the presence of SOX17 significantly affects ER protein levels, activity, and genomic binding in endometrial cancer cells.

Discussion

Estrogen signaling, working through ER, plays an oncogenic role in breast and endometrial cancer. However, the genomic binding sites and target genes of ER differ significantly between the cancer types. To explore the causes of cell type specificity, we decided to use EREs as a consistent anchor between breast and endometrial cancer cells. In cell lines and tumor samples from both cancer types, very few EREs (<3%) contribute to genomic binding or gene regulation. Of the bound EREs or regulatory EREs, about one-fourth are shared between breast cancer and endometrial cancer cells. Despite the low level of ERE utilization, there are significantly more EREs in the human genome than can be expected by random chance, suggesting that EREs are under evolutionary pressure to be conserved. It is possible that these EREs are involved in gene regulation in other tissues or in different phases of development. The highly selective usage of EREs provided an opportunity to evaluate the factors that contribute to ER genomic binding and gene regulation and how they differ between cell types.

To identify important features of EREs that bind ER in endometrial cancer and breast cancer cells, we compiled several data sets in Ishikawa and T-47D cells. Boruta, which assigns importance scores to features based on a random forest algorithm, revealed several factors that associate with ER binding. The highest scoring factor for both Ishikawa and T-47D cells was relative affinity, meaning the closer an ERE's sequence is to the optimal ER binding sequence, the more likely it is to be bound by ER. However, only 40% of EREs with the highest possible relative affinity are bound, suggesting other features are key in controlling ER binding. Other features found to be enriched at bound EREs in both the Ishikawa and T-47D lines were features correlating with accessible chromatin and histone modifications associated with active enhancer regions, both of which are found at active transcription factor binding sites (Field and Adelman 2020; Panigrahi and O'Malley 2021). Patterns of open chromatin are cell type specific (Gertz et al. 2013), supporting the cell type specificity of ER binding, but these chromatin patterns are controlled by sequence specific transcription factors.

Our analysis uncovered several transcription factors associated with bound EREs, many of which were expected based on previous studies of ER control in breast and endometrial cancer. FOXA1 and GATA3 were both found to be highly predictive of bound EREs in the T-47D breast cancer cell line. FOXA1 and GATA3 act as pioneer factors and facilitate ER binding in the breast cancer context by associating with chromatin ahead of ER recruitment (Hurtado et al. 2011; Theodorou et al. 2013; Jiang et al. 2019). ETV4 was found to be predictive of ER binding in the Ishikawa endometrial cancer cell line, which is supported by our previous work showing that ETV4 is essential for ER activity in endometrial cancer cells by binding with ER at many of its genomic sites and helping to localize ER to the nucleus (Rodriguez et al. 2020). With these results, we built both a transcription factor–focused model using cell type–specific transcription factors and a chromatin-focused model using ATAC-seq and H3K4me1; each could predict binding with reasonable sensitivity and specificity. For endometrial cancer, we were able to verify the predictive power of our models in a second cell line. Even though HCI-EC-23 has less ER activity compared with Ishikawa and less bound EREs, we were still able to construct a model that predicted bound EREs with similar sensitivity and specificity to the Ishikawa models for predicting ERE binding.

To expand the search beyond previous experimental data, we analyzed motif instances, and the results generally supported our findings from the experimental data. The FOXA1-containing motif cluster was more predictive of ER binding in T-47D than in Ishikawa cells, the GATA3 motif cluster was predictive of ER binding in T-47D cells only, and the SOX17 motif cluster was predictive of ER binding in Ishikawa cells only. Other findings from the motif analyses warrant further investigation. For example, the ZIC2, GLIS3, and various KLF motifs scored highly in both cell lines. ZIC2 has been proposed as a tumor suppressor in breast cancer (Liu et al. 2020) but has not been explicitly linked to estrogen signaling. Similarly, GLIS3 is implicated in breast cancer but not in ER+ breast cancer (Rami et al. 2016). KLF5 has been shown to suppress ER in breast cancer (Guo et al. 2010) but has not been studied in endometrial cancer. The KLF9 and PAX8 motifs were found to be predictive of ER binding in Ishikawa cells only, and both have been shown to be implicated in endometrial cancer but have not been connected to ER activity specifically. KLF9 has been shown to alter cell morphology, proliferation, and differentiation in an ER− endometrial cancer cell line (Simmen et al. 2008), and high PAX8 protein levels are associated with poorer outcomes in endometrial cancer patients (Mhawech-Fauceglia et al. 2012).

Predicting regulatory EREs proved to be more challenging than predicting bound EREs. Fewer factors were found to be important, and the predictive models exhibited lower accuracy than those created to predict bound EREs. Despite the lower predictive ability, some strong patterns emerged. Although transcription factor occupancy, ERE strength, and chromatin accessibility were the top predictors of ERE binding, chromatin looping and transcription were more predictive of regulatory EREs. Promoter-distal EREs that are engaged in transcription prior to ER binding are more likely to impact the expression of target genes. This is consistent with previous studies showing that transcription is a hallmark of active enhancer regions and that enhancer RNA production mirrors that of its target promoter (Field and Adelman 2020; Panigrahi and O'Malley 2021). Looping of an ERE both to a target gene promoter and other regulatory regions was also predictive of ERE regulatory activity. This is consistent with our previous study (Abewe et al. 2025), in which we found that estrogen-regulated genes with increased 3D genomic interactions after estrogen treatment had larger transcriptional responses, and most ER bound sites were anchors for chromatin loops. ER-mediated loops have also been identified in breast cancer and have been shown to change as the cancer progresses (Fullwood et al. 2009; Zhou et al. 2019; Van Den Brand et al. 2025). Overall, this leads to a model in which bound EREs with high levels of transcription prior to estrogen treatment and many 3D genomic interactions are very likely to be regulatory; however, most regulatory EREs cannot be identified in this manner, suggesting that important features could be missing.

To experimentally evaluate the factors important for ERE activity, we used CRISPR KO to test the importance of thousands of epigenetic factors. We focused on Ishikawa cells because similar CRISPR-based techniques have been used to study ER control in breast cancer, and their results corroborate our findings (Xiao et al. 2018; Samuels et al. 2025). We encountered many histone-modifying enzymes, which supported our computational findings. ASH1L encodes an H3K36me3 methyltransferase, which is related to transcription (An et al. 2011), KAT6A and BRPF1 encode members of a histone acetylation complex associated with gene activation (Huang et al. 2016), and KMT2A encodes an H3K4me3 methyltransferase (Nakamura et al. 2002), a histone modification confirmed important by Boruta in both cell lines for predicting both bound and regulatory EREs. BRPF1 has also been identified as a key component of estrogen signaling in ER+ breast cancer; silencing of BRPF1 inhibits ER transactivation capability and causes cell proliferation arrest and activation of apoptotic mechanisms (Salvati et al. 2024). CITED2 has been identified as a transcriptional coactivator that modulates ER in breast cancer (Lau et al. 2013), indicating that some transcriptional coactivators are shared between breast and endometrial cancer. EZH2 and SUZ12 were both identified as ER enabling, which is unexpected as they are members of PRC2, a transcriptional repressive complex. It is possible that PRC2 represses genes that repress ER, making it an indirect supporter of ER activity; however, the role of polycomb proteins in cancer is controversial. EZH2 and other members of PRC2 have been implicated as tumor suppressors in multiple hematopoietic cancers (Pasini and Di Croce 2016), but they have also shown oncogenic potential in multiple murine leukemia models and are being investigated as potential therapeutic targets (Scelfo et al. 2015). We identified HDAC1 as an ER-enabling gene even though it is usually considered a corepressor (Ng and Bird 2000). However, it has been positively associated with transcriptional activation caused by glucocorticoid receptor (Qiu et al. 2006), another steroid hormone receptor, so it could be associated with ER in a similar way and promote ER activity in endometrial cancer cells.

Previous work identified the SOX motif as uniquely enriched at ER binding sites in endometrial cancer cells (Rodriguez et al. 2020). Although there are many SOX family members, we chose SOX17 for further investigation as it is the highest expressed SOX factor in endometrial cancer cell lines, is essential for murine endometrial gland development, is involved in key aspects of endometrial biology such as cycling and implantation (Guimaraes-Young et al. 2015; Wang et al. 2018; Kinnear et al. 2019), and has been identified as a potential biomarker for endometrial and ovarian cancers (Shaker et al. 2023). Because of this, we performed ChIP-seq on SOX17 and were able to establish that SOX17 binding is predictive of bound EREs in Ishikawa cells. These results were supported by our identification of SOX17 as an ER-enabling gene in the Ishikawa CRISPR KO screen.

By performing SOX17 KO, we were able to validate that SOX17 is important for ER activity in endometrial cancer cells. SOX17 KO caused significant reductions in ER reporter activity, ER ChIP-seq signal, and gene expression responses to estrogen. SOX17 likely regulates ER activity in two ways. SOX17 plays a role in activating ESR1 transcription and driving higher protein levels by binding near the gene. SOX17 also binds to thousands of ER binding sites across the genome in the absence of estrogen, potentially aiding in ER genomic binding. In endometrial cancer patient samples, SOX17 mRNA expression is positively correlated with ESR1 mRNA expression, and SOX17 is overexpressed in tumor samples compared with normal samples, partially through genomic amplification in 4.7% of tumors. SOX17 is altered in ∼11%–15% of endometrial tumors, although these alterations do not seem to have an effect on overall expression levels, and the tumors included in these studies are not all ER-positive (Levine 2013; Walker et al. 2017). SOX17 has also been identified as a tumor suppressor in endometrial cancer (Zhang et al. 2016; Walker et al. 2017), indicating a complex role for SOX17 in endometrial cancer that may go beyond its effects on estrogen signaling.

Although this study has uncovered interesting features of ER genomic binding and gene regulation, there are limitations that should be mentioned. For the machine learning analysis, we considered only one ERE at a time, despite our previous work showing that EREs often work together to regulate gene expression (Ginley-Hidinger et al. 2019; Carleton et al. 2020; Ginley-Hidinger et al. 2024). We focused solely on full EREs even though half sites have been shown to be present at ∼38% of ER binding sites in T-47D and Ishikawa cells (Gertz et al. 2013), and these sites can functionally impact gene regulation (Carleton et al. 2020). ERE half sites are 6 bp long and would be expected to occur more than 700,000 times in the human genome by random chance, making it computationally infeasible to evaluate these sites in the same way that we evaluated the full ERE sites. We classified bound EREs within 100 kb of an estrogen-responsive gene's TSS as regulatory based on our previous analysis; however, we know that several of these sites do not impact gene expression (Carleton et al. 2017; Abewe et al. 2025). Functional studies of EREs in their genomic context would be the ideal method of identifying regulatory EREs; however, this has not been done at scale. Indicative of this issue, similar features were associated with regulatory EREs for both upregulated and downregulated genes, which is likely owing to most regulatory EREs associated with downregulated genes being associated with upregulated genes as well. In addition, we were unable to achieve high sensitivity when predicting regulatory EREs. The data used in our machine learning analysis to compare Ishikawa and T-47D cells came from multiple sources. It is possible that differences discovered between the cell lines were owing to technical artifacts as opposed to underlying biological differences. For example, the ATAC-seq data came from two different studies, and our analyses showed that ATAC-seq was a better predictor of bound EREs in Ishikawa cells than it was in T-47D cells. This is likely a technical artifact, as previous literature has established that the chromatin environment is important for ER binding in both cell lines (Gertz et al. 2013). The results from the CRISPR KO screen provided valuable information to support our computational results, but it also has limitations. With the reporter readout, we are unable to disentangle factors that affect ER binding and factors that impact ER's regulatory activity. GFP reporter expression levels were already high in response to estrogen, making it difficult to identify genes that decrease ER activity. Although the CRISPR screen and the machine learning analyses identified interesting factors, further studies are needed to determine the underlying mechanisms and functional importance of each factor. The hope is that our findings motivate future studies into important gene regulation factors in estrogen driven endometrial cancer.

Methods

Classification and analysis of bound and regulatory EREs

EREs were classified as ER bound using BEDTools (Quinlan and Hall 2010) applied to ER ChIP-seq data (Gertz et al. 2013) and were classified as regulatory using RNA-seq data (Gertz et al. 2013; Ginley-Hidinger et al. 2024). Features associated with bound and regulatory EREs were evaluated using the R packages Boruta (Kursa and Rudnicki 2010) and caret (Kuhn et al. 2020). For full details on ERE classification and analysis, see the Supplemental Methods.

Cell culture

Ishikawa, T-47D, and HCI-EC-23 cells were grown in full media: RPMI 1640 medium (Thermo Fisher Scientific 11875135) with 10% fetal bovine serum (Thermo Fisher Scientific 26140079) and 1% penicillin–streptomycin (Thermo Fisher Scientific 15070063). Hormone-depleted media were also used in preparation for hormone treatment and consisted of phenol red-free RPMI (Life Technologies 11835055) with 10% charcoal-dextran-stripped fetal bovine serum (Sigma-Aldrich F6765-100 mL) and 1% penicillin–streptomycin (Thermo Fisher Scientific 15070063). Cells were incubated with 5% CO2 at 37°C.

ChIP-seq

Cells were grown in hormone-depleted media for 5 days prior to either a 1 h DMSO (Fisher Scientific BP231; vehicle) treatment (SOX17), a 1 h 10 nM 17β-estradiol (E2; Sigma-Aldrich, E2758-1G) treatment (ER), or an 8 h DMSO treatment (H3K4me1) and were harvested for ChIP-seq (Johnson et al. 2007) as described previously (Reddy et al. 2009). Cells were fixed using 1% formaldehyde (Sigma-Aldrich 252549) for 10 min at room temperature, then quenched using glycine (Fisher Scientific BP381), and added to achieve a final concentration of 125 mM. Fixed cells were washed with cold PBS and harvested via scraping in PBS. Cells were pelleted at 500 RCF for 5 min, and PBS was removed. The cell pellet was resuspended in 1 mL of Farnham lysis buffer (5 mM PIPES at pH 8 [Fisher Scientific 50-155-760], 85 mM KCl [Sigma-Aldrich P9541], and 0.5% NP-40 [VWR M158-100ML]) supplemented with protease inhibitors (Thermo Scientific A32965) to prepare crude nuclei. Nuclei were pelleted at 500 RCF for 5 min, and the Farnham lysis buffer was removed. Nuclei were resuspended in RIPA lysis buffer (1× PBS [Thermo Scientific AM9624], 1% NP-40 [VWR M158-100ML], 0.5% sodium deoxycholate [Bioworld 40430018-3], and 0.1% SDS [Thermo Scientific AM9820]) supplemented with protease inhibitors (Thermo Scientific A32965) and sonicated for eight cycles (30 sec on, 30 sec off) at 4°C at 40% amplitude using a EpiShear probe sonicator (Active Motif). Chromatin immunoprecipitation was performed using 5 µg of antibody. Antibodies used for this study were SOX17 (R&D Systems AF1924), ER (Abcam AB3575), and H3K4me1 (Cell Signaling 5326S). ChIP-seq libraries were sequenced on an Illumina NovaSeq 6000. Reads were aligned to the GRCh37 (hg19) build of the human genome using Bowtie (Langmead et al. 2009): -m1 -t -best -q -S -l 32 -e 80 -n 2. The GRCh37 build of the human genome was used to align with previously published data sets in our analyses. We do not expect that realigning our data to the GRCh38 genome build would affect the results of this work, as we focused our analyses on uniquely alignable regions of the genome. MACS2 (Zhang et al. 2008) was used to call peaks with a P-value cutoff of 1 × 10−10 and mfold parameter between 15 and 100.

ATAC-seq

HCI-EC-23 cells were grown in hormone-depleted media for 5 days prior to a 1 h DMSO treatment and harvest for ATAC-seq (Buenrostro et al. 2013). Cell pellets were treated and lysed for 5 min as described previously (Lewis et al. 2025). The transposition reaction, DNA purification, and amplification were also performed as described previously (Lewis et al. 2025). Libraries were sequenced on an Illumina NovaSeq X at a depth of about 50 million reads/sample, and reads were trimmed to 50 bp. Bowtie (Langmead et al. 2009) was then used to align the trimmed reads to the hg19 build of the human genome using the following parameters: -m 1 -t ‐‐best -q -S −l 32 -e 80 -n 2. BAM files were created from SAM files and sorted using SAMtools (Li et al. 2009). Regions of accessible chromatin, termed peaks, were called using a cutoff of <0.05 for adjusted P-value and an mfold parameter between 15 and 100 using MACS2 (Zhang et al. 2008). Reads that aligned to the peaks were quantified across all samples using featureCounts (Liao et al. 2014).

CRISPR KO screen

Reporter cell lines containing a 3 × ERE-driven GFP construct were created from Ishikawa cells. GFP-reporter cells were then transduced with viral particles containing the Human Epigenetic Knockdown Library, a gift from Kivanc Birsoy (Addgene 162256) (Williams et al. 2020). Successfully transduced cells were selected using 1 µg/m puromycin (Thermo Scientific A1113803), grown on hormone-depleted media for 96 h, and treated with 10 nM E2 for 72 h. Cells were then sorted via flow cytometry based on GFP levels, genomic DNA was extracted and sequenced, and enrichment analysis was performed using MAGeCK (Li et al. 2014). For details on GFP reporter cell line creation, creation of viral particles, transductions, DNA extractions, sequencing, and enrichment analysis, see the Supplemental Methods and Supplemental Table S6.

Creation of ESR1 KO, SOX17 KO, and control gRNA lines

Ishikawa Cas9 expressing cells, described in the Supplemental Methods, were transduced with 5 µg/mL polybrene and lentiviral particles containing the LentiGuide Puro plasmids with ESR1 targeting, SOX17 targeting, and control guide RNAs at MOIs of 10–200 (see Supplemental Methods and Supplemental Table S6). Cells were treated and maintained with 1 µg/m puromycin in full media until validation. Western blotting was used to confirm the absence of ER and SOX17 in their respective KO lines after puromycin selection (see Supplemental Methods). After western blot confirmation, cells were plated at 1000 cells per 10 cm plate to allow single cells to form distinct colonies for isolation. Colonies were grown up to about 256 cells on full media, selected, and moved to individual wells of a 96-well plate before further passaging. Individual clones of the KO lines were evaluated for ER and SOX17 expression using western blotting.

ERE luciferase reporter assay

Ishikawa SOX17 KO, ESR1 KO, and control cells were grown in hormone-depleted media for 3 days prior to plating at 15,000 cells per well in a 96-well plate in hormone-depleted media. Then, the ERE luciferase reporter assay was performed as described previously using 10 nM E2 treatments (Rush et al. 2022).

RNA-seq

Ishikawa SOX17 KO and control cells were treated with 10 nM E2 or DMSO (vehicle) for 8 h. Following treatment, cells were lysed with buffer RLT Plus (Qiagen) containing 1% beta-mercaptoethanol (Sigma-Aldrich). RNA was extracted and purified using a Quick-RNA mini prep kit (Zymo Research). NEBNext Ultra II directional RNA library prep kit with poly(A) mRNA isolation was used to construct RNA-seq libraries according to the manufacturer's instructions (NEB). Sequencing reads were aligned to hg19 build of the human genome using HISAT2 (Kim et al. 2019). SAMtools (Li et al. 2009) was used to convert SAM files to BAM files. Genes were defined by the University of California Santa Cruz (UCSC) Known Genes (Kent et al. 2002), and reads that mapped to known genes were assigned with featureCounts (Liao et al. 2014). Read counts were normalized and analyzed for differential expression via DESeq2 (Love et al. 2014).

Data access

All raw and processed sequencing data generated in this study have been submitted to the NCBI Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/) under accession numbers GSE312322, GSE312323, GSE333956, GSE334097, and GSE334100.

Competing interest statement

The authors declare no competing interests.

Acknowledgments

Funding for this work came from the National Institutes of Health (NIH)/National Cancer Institute (NCI) grant R21 CA270580 to J.G. and from the Huntsman Cancer Institute. Research reported in this publication utilized the high-throughput genomics shared resource at the University of Utah and was supported by NIH/NCI award P30 CA042014. The flow cytometry shared resource and the drug discovery shared resource at the University of Utah were also utilized during this study. We thank Matthew Ginley-Hidinger for his help with coding and figure making, Wilbert Zwart and Sebastian Gregoricchio for their helpful suggestions, Jacob Polaski for his guidance in designing the ATAC-seq protocol, and other Gertz laboratory members for their input on the study and manuscript.

Author contributions: Conceptualization was by A.R. and J.G. Methodology was by A.R., C.M.R., N.K., H.A., B.K.D., and J.G. Investigation was by A.R., N.K., C.M.R., H.A., K.S.O., and A.J.B. Formal analysis was by A.R., J.M.V., and J.G. Writing of the original draft was by A.R. and J.G. Writing, reviewing, and editing were by all authors. Supervision was by J.R. and J.G. Funding acquisition was by J.G.

Footnotes

[1] Supplementary material [Supplemental material is available for this article.]

[2] Article published online before print. Article, supplemental material, and publication date are at https://www.genome.org/cgi/doi/10.1101/gr.281792.125.

[3] Freely available online through the Genome Research Open Access option.

References

  1. ↵
    Abewe H, Richey A, Vahrenkamp JM, Ginley-Hidinger M, Rush CM, Kitchen N, Zhang X, Gertz J. 2025. Estrogen-induced chromatin looping changes identify a subset of functional regulatory elements. Genome Res 35: 393–403. 10.1101/gr.279699.124
    Google ScholarCrossRefAbstract/Full TextPubMed
  2. ↵
    An S, Yeo KJ, Jeon YH, Song J-J. 2011. Crystal structure of the human histone methyltransferase ASH1L catalytic domain and its implications for the regulatory mechanism. J Biol Chem 286: 8369–8374. 10.1074/jbc.M110.203380
    Google ScholarCrossRefPubMed
  3. ↵
    Arafeh R, Shibue T, Dempster JM, Hahn WC, Vazquez F. 2025. The present and future of the cancer dependency Map. Nat Rev Cancer 25: 59–73. 10.1038/s41568-024-00763-x
    Google ScholarCrossRefPubMed
  4. ↵
    Buenrostro JD, Giresi PG, Zaba LC, Chang HY, Greenleaf WJ. 2013. Transposition of native chromatin for fast and sensitive epigenomic profiling of open chromatin, DNA-binding proteins and nucleosome position. Nat Methods 10: 1213–1218. 10.1038/nmeth.2688
    Google ScholarCrossRefPubMedWeb of Science
  5. ↵
    Carleton JB, Berrett KC, Gertz J. 2017. Multiplex enhancer interference reveals collaborative control of gene regulation by estrogen receptor α-bound enhancers. Cell Syst 5: 333–344.e5. 10.1016/j.cels.2017.08.011
    Google ScholarCrossRefPubMed
  6. ↵
    Carleton JB, Ginley-Hidinger M, Berrett KC, Layer RM, Quinlan AR, Gertz J. 2020. Regulatory sharing between estrogen receptor alpha bound enhancers. Nucleic Acids Res 48: 6597–6610. 10.1093/nar/gkaa454
    Google ScholarCrossRefPubMed
  7. ↵
    Ciocia A, Mestre-Farràs N, Vicent-Nacht I, Guitart T, Gebauer F. 2024. CSDE1: a versatile regulator of gene expression in cancer. NAR Cancer 6: zcae014. 10.1093/narcan/zcae014
    Google ScholarCrossRef
  8. ↵
    Doyon Y, Selleck W, Lane WS, Tan S, Côté J. 2004. Structural and functional conservation of the NuA4 histone acetyltransferase complex from yeast to humans. Mol Cell Biol 24: 1884–1896. 10.1128/MCB.24.5.1884-1896.2004
    Google ScholarCrossRefPubMedWeb of Science
  9. ↵
    Droog M, Nevedomskaya E, Kim Y, Severson T, Flach KD, Opdam M, Schuurman K, Gradowska P, Hauptmann M, Dackus G, 2016. Comparative cistromics reveals genomic cross-talk between FOXA1 and ERα in tamoxifen-associated endometrial carcinomas. Cancer Res 76: 3773–3784. 10.1158/0008-5472.CAN-14-1813
    Google ScholarCrossRefPubMed
  10. ↵
    Field A, Adelman K. 2020. Evaluating enhancer function and transcription. Annu Rev Biochem 89: 213–234. 10.1146/annurev-biochem-011420-095916
    Google ScholarCrossRefPubMed
  11. ↵
    Fullwood MJ, Liu MH, Pan YF, Liu J, Xu H, Mohamed YB, Orlov YL, Velkov S, Ho A, Mei PH, 2009. An oestrogen-receptor-α-bound human chromatin interactome. Nature 462: 58–64. 10.1038/nature08497
    Google ScholarCrossRefPubMedWeb of Science
  12. ↵
    Gertz J, Savic D, Varley KE, Partridge EC, Safi A, Jain P, Cooper GM, Reddy TE, Crawford GE, Myers RM. 2013. Distinct properties of cell-type-specific and shared transcription factor binding sites. Mol Cell 52: 25–36. 10.1016/j.molcel.2013.08.037
    Google ScholarCrossRefPubMedWeb of Science
  13. ↵
    Ginley-Hidinger M, Carleton JB, Rodriguez AC, Berrett KC, Gertz J. 2019. Sufficiency analysis of estrogen responsive enhancers using synthetic activators. Life Sci Alliance 2: e201900497. 10.26508/lsa.201900497
    Google ScholarCrossRefAbstract/Full TextPubMed
  14. ↵
    Ginley-Hidinger M, Abewe H, Osborne K, Richey A, Kitchen N, Mortenson KL, Wissink EM, Lis J, Zhang X, Gertz J. 2024. Cis-regulatory control of transcriptional timing and noise in response to estrogen. Cell Genom 4: 100542. 10.1016/j.xgen.2024.100542
    Google ScholarCrossRefPubMed
  15. ↵
    Gregoricchio S, Kojic A, Hoogstraat M, Schuurman K, Stelloo S, Severson TM, Mara O, Droog TA, Singh M, Glubb AA, 2025. Endometrial tumorigenesis involves epigenetic plasticity demarcating non-coding somatic mutations and 3D-genome alterations. Genome Biol 26: 124. 10.1186/s13059-025-03596-5
    Google ScholarCrossRefPubMed
  16. ↵
    Guimaraes-Young A, Neff T, Dupuy A, Goodheart MJ. 2015. The role of Sox17 in endometrial gland formation and carcinogenesis. Gynecol Oncol 137: 135. 10.1016/j.ygyno.2015.01.336
    Google ScholarCrossRef
  17. ↵
    Guo P, Dong X-Y, Zhao K-W, Sun X, Li Q, Dong J-T. 2010. Estrogen-induced interaction between KLF5 and estrogen receptor (ER) suppresses the function of ER in ER-positive breast cancer cells. Int J Cancer 126: 81–89. 10.1002/ijc.24696
    Google ScholarCrossRefPubMedWeb of Science
  18. ↵
    Heinz S, Benner C, Spann N, Bertolino E, Lin YC, Laslo P, Cheng JX, Murre C, Singh H, Glass CK. 2010. Simple combinations of lineage-determining transcription factors prime cis-regulatory elements required for macrophage and B cell identities. Mol Cell 38: 576–589. 10.1016/j.molcel.2010.05.004
    Google ScholarCrossRefPubMedWeb of Science
  19. ↵
    Hu R, Hilakivi-Clarke L, Clarke R. 2015. Molecular mechanisms of tamoxifen-associated endometrial cancer. Oncol Lett 9: 1495–1501. 10.3892/ol.2015.2962
    Google ScholarCrossRefPubMed
  20. ↵
    Huang F, Abmayr SM, Workman JL. 2016. Regulation of KAT6 acetyltransferases and their roles in cell cycle progression, stem cell maintenance, and human disease. Mol Cell Biol 36: 1900–1907. 10.1128/MCB.00055-16
    Google ScholarCrossRefPubMed
  21. ↵
    Hurtado A, Holmes KA, Ross-Innes CS, Schmidt D, Carroll JS. 2011. FOXA1 is a key determinant of estrogen receptor function and endocrine response. Nat Genet 43: 27–33. 10.1038/ng.730
    Google ScholarCrossRefPubMed
  22. ↵
    The ICGC/TCGA Pan-Cancer Analysis of Whole Genomes Consortium. 2020. Pan-cancer analysis of whole genomes. Nature 578: 82–93. 10.1038/s41586-020-1969-6
    Google ScholarCrossRefPubMed
  23. ↵
    Jiang G, Wang X, Sheng D, Zhou L, Liu Y, Xu C, Liu S, Zhang J. 2019. Cooperativity of co-factor NR2F2 with pioneer factors GATA3, FOXA1 in promoting ERα function. Theranostics 9: 6501–6516. 10.7150/thno.34874
    Google ScholarCrossRefPubMed
  24. ↵
    Johnson DS, Mortazavi A, Myers RM, Wold B. 2007. Genome-wide mapping of in vivo protein-DNA interactions. Science 316: 1497–1502. 10.1126/science.1141319
    Google ScholarCrossRefPubMedWeb of Science
  25. ↵
    Jordan VC. 2006. Tamoxifen (ICI46,474) as a targeted therapy to treat and prevent breast cancer. Br J Pharmacol 147: S269–S276. 10.1038/sj.bjp.0706399
    Google ScholarCrossRefPubMedWeb of Science
  26. ↵
    Jordan VC, Brodie AMH. 2007. Development and evolution of therapies targeted to the estrogen receptor for the treatment and prevention of breast cancer. Steroids 72: 7–25. 10.1016/j.steroids.2006.10.009
    Google ScholarCrossRefPubMedWeb of Science
  27. ↵
    Kent WJ, Sugnet CW, Furey TS, Roskin KM, Pringle TH, Zahler AM, Haussler D. 2002. The human genome browser at UCSC. Genome Res 12: 996–1006. 10.1101/gr.229102
    Google ScholarCrossRefAbstract/Full TextPubMedWeb of Science
  28. ↵
    Kim D, Paggi JM, Park C, Bennett C, Salzberg SL. 2019. Graph-based genome alignment and genotyping with HISAT2 and HISAT-genotype. Nat Biotechnol 37: 907–915. 10.1038/s41587-019-0201-4
    Google ScholarCrossRefPubMed
  29. ↵
    Kinnear S, Salamonsen LA, Francois M, Harley V, Evans J. 2019. Uterine SOX17: a key player in human endometrial receptivity and embryo implantation. Sci Rep 9: 15495. 10.1038/s41598-019-51751-3
    Google ScholarCrossRefPubMed
  30. ↵
    Klein-Hitpass L, Ryffel GU, Heitlinger E, Cato ACB. 1988. A 13 bp palindrome is a functional estrogen responsive element and interacts specifically with estrogen receptor. Nucleic Acids Res 16: 647–663. 10.1093/nar/16.2.647
    Google ScholarCrossRefPubMedWeb of Science
  31. ↵
    Kong SL, Li G, Loh SL, Sung WK, Liu ET. 2011. Cellular reprogramming by the conjoint action of ERα, FOXA1, and GATA3 to a ligand-inducible growth state. Mol Syst Biol 7: 526. 10.1038/msb.2011.59
    Google ScholarCrossRefPubMed
  32. ↵
    Kuhn M, Wing J, Weston S, Williams A, Keefer C, Engelhardt A, Cooper T, Mayer Z, Kenkel B, Team RC. 2020. Package ‘caret’. R J 223: 48.
    Google Scholar
  33. ↵
    Kursa MB, Rudnicki WR. 2010. Feature selection with the Boruta package. J Stat Softw 36: 1–13. 10.18637/jss.v036.i11
    Google ScholarCrossRefPubMed
  34. ↵
    Langmead B, Trapnell C, Pop M, Salzberg SL. 2009. Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol 10: R25. 10.1186/gb-2009-10-3-r25
    Google ScholarCrossRefPubMed
  35. ↵
    Lau WM, Doucet M, Huang D, Weber KL, Kominsky SL. 2013. CITED2 modulates estrogen receptor transcriptional activity in breast cancer cells. Biochem Biophys Res Commun 437: 261–266. 10.1016/j.bbrc.2013.06.063
    Google ScholarCrossRefPubMed
  36. ↵
    Lee CI, Goodwin A, Wilcken N. 2017. Fulvestrant for hormone-sensitive metastatic breast cancer. Cochrane Database Syst Rev 2017: CD011093. 10.1002/14651858.CD011093.pub2
    Google ScholarCrossRef
  37. ↵
    Levine DA. 2013. Integrated genomic characterization of endometrial carcinoma. Nature 497: 67–73. 10.1038/nature12113
    Google ScholarCrossRefPubMedWeb of Science
  38. ↵
    Lewis C, Levis H, Holbrook J, Polaski JT, Jacobsen TD, Mizels J, Gullbrand SE, Diekman B, Iatridis JC, Gertz J, 2025. ZNF865 (BLST) regulates back pain via cell senescence and DNA damage mechanisms. bioRxiv 10.1101/2025.06.13.659603
    Google ScholarCrossRefAbstract/Full Text
  39. ↵
    Li H, Handsaker B, Wysoker A, Fennell T, Ruan J, Homer N, Marth G, Abecasis G, Durbin R, 1000 Genome Project Data Processing Subgroup. 2009. The Sequence Alignment/Map format and SAMtools. Bioinformatics 25: 2078–2079. 10.1093/bioinformatics/btp352
    Google ScholarCrossRefPubMedWeb of Science
  40. ↵
    Li W, Xu H, Xiao T, Cong L, Love MI, Zhang F, Irizarry RA, Liu JS, Brown M, Liu XS. 2014. MAGeCK enables robust identification of essential genes from genome-scale CRISPR/Cas9 knockout screens. Genome Biol 15: 554. 10.1186/s13059-014-0554-4
    Google ScholarCrossRefPubMed
  41. ↵
    Liao Y, Smyth GK, Shi W. 2014. featureCounts: an efficient general purpose program for assigning sequence reads to genomic features. Bioinformatics 30: 923–930. 10.1093/bioinformatics/btt656
    Google ScholarCrossRefPubMedWeb of Science
  42. ↵
    Liu Z-H, Chen M-L, Zhang Q, Zhang Y, An X, Luo Y-L, Liu X-M, Liu S-X, Liu Q, Yang T, 2020. ZIC2 is downregulated and represses tumor growth via the regulation of STAT3 in breast cancer. Int J Cancer 147: 505–518. 10.1002/ijc.32922
    Google ScholarCrossRefPubMed
  43. ↵
    Love MI, Huber W, Anders S. 2014. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol 15: 550. 10.1186/s13059-014-0550-8
    Google ScholarCrossRefPubMed
  44. ↵
    Mhawech-Fauceglia P, Wang D, Samrao D, Godoy H, Pejovic T, Liu S, Lele S. 2012. Pair-Box (PAX8) protein-positive expression is associated with poor disease outcome in women with endometrial cancer. Br J Cancer 107: 370–374. 10.1038/bjc.2012.241
    Google ScholarCrossRefPubMed
  45. ↵
    Nakamura T, Mori T, Tada S, Krajewski W, Rozovskaia T, Wassell R, Dubois G, Mazo A, Croce CM, Canaani E. 2002. ALL-1 is a histone methyltransferase that assembles a supercomplex of proteins involved in transcriptional regulation. Mol Cell 10: 1119–1128. 10.1016/S1097-2765(02)00740-2
    Google ScholarCrossRefPubMedWeb of Science
  46. ↵
    Ng HH, Bird A. 2000. Histone deacetylases: silencers for hire. Trends Biochem Sci 25: 121–126. 10.1016/S0968-0004(00)01551-6
    Google ScholarCrossRefPubMedWeb of Science
  47. ↵
    Panigrahi A, O'Malley BW. 2021. Mechanisms of enhancer action: the known and the unknown. Genome Biol 22: 108. 10.1186/s13059-021-02322-1
    Google ScholarCrossRefPubMed
  48. ↵
    Pasini D, Di Croce L. 2016. Emerging roles for Polycomb proteins in cancer. Curr Opin Genet Dev 36: 50–58. 10.1016/j.gde.2016.03.013
    Google ScholarCrossRefPubMed
  49. ↵
    Paterni I, Granchi C, Katzenellenbogen JA, Minutolo F. 2014. Estrogen receptors alpha (ERα) and beta (ERβ): subtype-selective ligands and clinical potential. Steroids 90: 13–29. 10.1016/j.steroids.2014.06.012
    Google ScholarCrossRefPubMed
  50. ↵
    Qiu Y, Zhao Y, Becker M, John S, Parekh BS, Huang S, Hendarwanto A, Martinez ED, Chen Y, Lu H, 2006. HDAC1 acetylation is linked to progressive modulation of steroid receptor-induced gene transcription. Mol Cell 22: 669–679. 10.1016/j.molcel.2006.04.019
    Google ScholarCrossRefPubMedWeb of Science
  51. ↵
    Quinlan AR, Hall IM. 2010. BEDTools: a flexible suite of utilities for comparing genomic features. Bioinformatics 26: 841–842. 10.1093/bioinformatics/btq033
    Google ScholarCrossRefPubMedWeb of Science
  52. ↵
    Ragunthar T, Selvakumar S. 2019. Classification of gene expression data with optimized feature selection. Int J Recent Technol Eng 8: 4763–4769. 10.35940/ijrte.B1845.078219
    Google ScholarCrossRef
  53. ↵
    Rami F, Baradaran A, Kahnamooi MM, Salehi M. 2016. Alteration of GLIS3 gene expression pattern in patients with breast cancer. Adv Biomed Res 5: 44. 10.4103/2277-9175.178803
    Google ScholarCrossRefPubMed
  54. ↵
    Reddy TE, Pauli F, Sprouse RO, Neff NF, Newberry KM, Garabedian MJ, Myers RM. 2009. Genomic determination of the glucocorticoid response reveals unexpected mechanisms of gene regulation. Genome Res 19: 2163–2171. 10.1101/gr.097022.109
    Google ScholarCrossRefAbstract/Full TextPubMedWeb of Science
  55. ↵
    Rodriguez AC, Blanchard Z, Maurer KA, Gertz J. 2019. Estrogen signaling in endometrial cancer: a key oncogenic pathway with several open questions. Horm Cancer 10: 51–63. 10.1007/s12672-019-0358-9
    Google ScholarCrossRefPubMed
  56. ↵
    Rodriguez AC, Vahrenkamp JM, Berrett KC, Clark KA, Guillen KP, Scherer SD, Yang CH, Welm BE, Janat-Amsbury MM, Graves BJ, 2020. ETV4 is necessary for estrogen signaling and growth in endometrial cancer cells. Cancer Res 80: 1234–1245. 10.1158/0008-5472.CAN-19-1382
    Google ScholarCrossRefPubMed
  57. ↵
    Ross-Innes CS, Stark R, Teschendorff AE, Holmes KA, Ali HR, Dunning MJ, Brown GD, Gojis O, Ellis IO, Green AR, 2012. Differential oestrogen receptor binding is associated with clinical outcome in breast cancer. Nature 481: 389–393. 10.1038/nature10730
    Google ScholarCrossRefPubMedWeb of Science
  58. ↵
    Rush CM, Blanchard Z, Polaski JT, Osborne KS, Osby K, Vahrenkamp JM, Yang C-H, Lum DH, Hagan CR, Leslie KK, 2022. Characterization of HCI-EC-23 a novel estrogen- and progesterone-responsive endometrial cancer cell line. Sci Rep 12: 19731. 10.1038/s41598-022-24211-8
    Google ScholarCrossRefPubMed
  59. ↵
    Salvati A, Giurato G, Lamberti J, Terenzi I, Crescenzo L, Melone V, Palo L, Giordano A, Sabbatino F, Roscigno G, 2024. Essential gene screening identifies the bromodomain-containing protein BRPF1 as a new actionable target for endocrine therapy-resistant breast cancers. Mol Cancer 23: 160. 10.1186/s12943-024-02071-2
    Google ScholarCrossRefPubMed
  60. ↵
    Samuels M, Besta S, Betrán AL, Nia RS, Xie X, Gu X, Shu Q, Giamas G. 2025. CRISPR screening approaches in breast cancer research. Cancer Metastasis Rev 44: 59. 10.1007/s10555-025-10275-1
    Google ScholarCrossRefPubMed
  61. ↵
    Scelfo A, Piunti A, Pasini D. 2015. The controversial role of the Polycomb group proteins in transcription and cancer: how much do we not understand Polycomb proteins? FEBS J 282: 1703–1722. 10.1111/febs.13112
    Google ScholarCrossRefPubMed
  62. ↵
    Shaker N, Chen W, Sinclair W, Parwani AV, Li Z. 2023. Identifying SOX17 as a sensitive and specific marker for ovarian and endometrial carcinomas. Mod Pathol 36: 100038. 10.1016/j.modpat.2022.100038
    Google ScholarCrossRefPubMed
  63. ↵
    Siegel RL, Kratzer TB, Giaquinto AN, Sung H, Jemal A. 2025. Cancer statistics, 2025. CA Cancer J Clin 75: 10–45. 10.3322/caac.21871
    Google ScholarCrossRefPubMed
  64. ↵
    Simmen FA, Su Y, Xiao R, Zeng Z, Simmen RC. 2008. The Krüppel-like factor 9 (KLF9) network in HEC-1-A endometrial carcinoma cells suggests the carcinogenic potential of dys-regulated KLF9 expression. Reprod Biol Endocrinol 6: 41. 10.1186/1477-7827-6-41
    Google ScholarCrossRefPubMed
  65. ↵
    Theodorou V, Stark R, Menon S, Carroll JS. 2013. GATA3 acts upstream of FOXA1 in mediating ESR1 binding by shaping enhancer accessibility. Genome Res 23: 12–22. 10.1101/gr.139469.112
    Google ScholarCrossRefAbstract/Full TextPubMedWeb of Science
  66. ↵
    Van Den Brand T, Collier MD, Flach KD, Gregoricchio S, Mayayo-Peralta I, Dauyey Z, Schuurman K, Teunissen H, Zwart W, De Wit E. 2025. A 3D genome compendium of breast cancer progression. iScience 28: 113268. 10.1016/j.isci.2025.113268
    Google ScholarCrossRefPubMed
  67. ↵
    Walker CJ, O'Hern MJ, Serna VA, Kurita T, Miranda MA, Sapp CE, Mutch DG, Cohn DE, Goodfellow PJ. 2017. Novel SOX17 frameshift mutations in endometrial cancer are functionally distinct from recurrent missense mutations. Oncotarget 8: 68758–68768. 10.18632/oncotarget.20213
    Google ScholarCrossRefPubMed
  68. ↵
    Wang X, Li X, Wang T, Wu S-P, Jeong J-W, Kim TH, Young SL, Lessey BA, Lanz RB, Lydon JP, 2018. SOX17 regulates uterine epithelial–stromal cross-talk acting via a distal enhancer upstream of Ihh. Nat Commun 9: 4421. 10.1038/s41467-018-06652-w
    Google ScholarCrossRefPubMed
  69. ↵
    Williams RT, Guarecuco R, Gates LA, Barrows D, Passarelli MC, Carey B, Baudrier L, Jeewajee S, La K, Prizer B, 2020. ZBTB1 regulates asparagine synthesis and leukemia cell response to L-asparaginase. Cell Metab 31: 852–861.e6. 10.1016/j.cmet.2020.03.008
    Google ScholarCrossRefPubMed
  70. ↵
    Xiao T, Li W, Wang X, Xu H, Yang J, Wu Q, Huang Y, Geradts J, Jiang P, Fei T, 2018. Estrogen-regulated feedback loop limits the efficacy of estrogen receptor–targeted breast cancer therapy. Proc Natl Acad Sci 115: 7869–7878. 10.1073/pnas.1722617115
    Google ScholarCrossRefPubMed
  71. ↵
    Zhang Y, Liu T, Meyer CA, Eeckhoute J, Johnson DS, Bernstein BE, Nusbaum C, Myers RM, Brown M, Li W, 2008. Model-based Analysis of ChIP-seq (MACS). Genome Biol 9: R137. 10.1186/gb-2008-9-9-r137
    Google ScholarCrossRefPubMed
  72. ↵
    Zhang Y, Bao W, Wang K, Lu W, Wang H, Tong H, Wan X. 2016. SOX17 is a tumor suppressor in endometrial cancer. Oncotarget 7: 76036–76046. 10.18632/oncotarget.12582
    Google ScholarCrossRefPubMed
  73. ↵
    Zhou Y, Gerrard DL, Wang J, Li T, Yang Y, Fritz AJ, Rajendran M, Fu X, Stein G, Schiff R, 2019. Temporal dynamic reorganization of 3D chromatin architecture in hormone-induced breast cancer and endocrine resistance. Nat Commun 10: 1522. 10.1038/s41467-019-09320-9
    Google ScholarCrossRefPubMed
Back to top
Article contents

Announcement(s)