Multi-omics dissection of FOXA1 reveals cancer-type-specific transcriptional networks and functional risk variants

Summary

FOXA1 is a pioneer transcription factor that shapes lineage-specific regulatory programs in hormone-associated cancers. Here, we integrated FOXA1 cistromes, transcriptomes, GWAS loci, expression-associated quantitative trait loci (eQTL) datasets, clinical cohorts, and functional assays to examine how FOXA1-directed transcriptional networks contribute to prostate and breast cancer biology. We identified high-confidence FOXA1 direct target genes that converge on shared oncogenic pathways while retaining cancer-type-specific regulatory modules. These FOXA1-associated transcriptional programs generated prognostic signatures that stratified patient outcomes across independent cohorts. In prostate cancer, cancer-risk variants were enriched within FOXA1 binding regions, and selected functional variants modulated FOXA1 occupancy and downstream key effector gene (TLE4, USP39, and CPNE1) expression. Functional validation supported roles for these genetically regulated targets in prostate cancer cell proliferation and migration. Together, our findings connect inherited noncoding variation, FOXA1 chromatin binding, transcriptional regulation, and tumor-associated phenotypes, providing a framework for understanding FOXA1-centered regulatory mechanisms in hormone-associated cancers.

Graphical abstract

Highlights

• •FOXA1 target genes define cancer-type-specific transcriptional programs
• •FOXA1-derived signatures stratify prostate and breast cancer outcomes
• •Cancer risk-associated eQTL SNPs are enriched within FOXA1 binding regions
• •Functional SNPs modulate FOXA1 occupancy and effector gene expression

Molecular biology; Cell biology; Systems biology

Introduction

Transcription factors (TFs) that establish and maintain lineage-specific regulatory landscapes are central to development and disease.^1,2,3 Among them, FOXA1 is a paradigmatic pioneer factor that shapes chromatin accessibility and orchestrates transcriptional programs across diverse tissues. Its activity is especially prominent in hormone-driven malignancies including prostate and breast, where it dictates cellular identity, tumor initiation, and progression.^4,5,6 Despite FOXA1’s broad regulatory involvement, its transcriptional outputs display striking cell-type specificity, reflecting the complex interplay between pioneer activity, epigenomic context, lineage-determining cofactors, and non-coding genetic variation.^4,7,8,9,10 Elucidating this interplay is critical for understanding how a single TF can underlie both shared oncogenic mechanisms and tissue-specific cancer phenotypes.

Over the past decade, genome-scale chromatin and transcriptomic profiling has provided important insights into FOXA1 function. These studies have identified a conserved core of FOXA1-bound regions across cancer types, yet also revealed that downstream transcriptional consequences are frequently rewired in a tissue-dependent manner, highlighting FOXA1’s context-dependent regulatory logic. Mechanistic dissection has expanded this view: FOXA1 can drive widespread alternative splicing in prostate cancer (PCa),¹¹ remodel estrogen receptor (ER) cistromes to fuel metastasis in breast cancer (BCa),¹² or scan chromatin in ways distinct from other pioneer factors.¹³ Additional studies underscore how early oncogenic events, such as SPOP mutations, reconfigure FOXA1-mediated androgen responses.¹⁴ Together, these findings establish FOXA1 as a mechanistic hub in cancer biology. Although FOXA1 has been extensively studied as a lineage-defining pioneer factor, it remains unclear whether FOXA1-centered regulatory networks can be systematically linked to inherited susceptibility loci, direct transcriptional targets (DTGs), and clinically informative prognostic programs within a unified framework.

This knowledge gap is particularly pressing in PCa, where FOXA1 is indispensable for tumor development and progression. Intriguingly, GWAS have revealed that FOXA1 binding sites are disproportionately enriched for PCa risk SNPs compared with breast or liver cancer, suggesting that germline variation may reprogram FOXA1’s regulatory activity in a cancer-type-specific manner. Prior functional studies have shown that risk variants such as rs684232^15,16 and rs1859961 can alter FOXA1 and other TFs’ occupancy and transcriptional control of oncogenes.^17,18 However, a systematic map of FOXA1-bound risk loci, their downstream expression-associated quantitative trait loci (eQTL) genes (eGenes, whose expression levels are affected by genetic variants), and their phenotypic consequences in PCa has yet to be established.

Here, we address this challenge through an integrative, multi-omics framework. We combine large-scale FOXA1 cistromes, transcriptomic profiles, GWAS loci, and patient-derived datasets across prostate and BCas to construct a unified, functionally relevant FOXA1 target gene signature. Focusing on PCa, we identify three functional SNPs—rs1417078,¹⁹ rs4911493, and rs5832649—that display allele-specific FOXA1 binding and regulate expression of TLE4, CPNE1, and USP39, respectively. Among these, we establish TLE4 as a bona fide FOXA1 target at the 9q21.31 risk locus, where rs1417078 modulates gene expression and influences PCa cell phenotypes. These findings provide the first direct evidence that germline enhancer variants can redirect FOXA1 binding to shape target gene expression and clinical trajectories.

By unifying genetic, epigenomic, and functional evidence, our study reveals how FOXA1 acts as a conduit through which inherited noncoding variants drive lineage-specific oncogenesis. This work not only advances mechanistic understanding of FOXA1 regulation but also highlights potential opportunities for risk stratification, prognostic assessment, and therapeutic intervention in FOXA1-driven cancers.

Results

Systematic dissection of FOXA1 direct transcriptional targets reveals cancer-type-specific regulatory programs

While several studies have characterized individual FOXA1 target genes involved in tumor progression,^7,20,21,22 a comprehensive delineation of FOXA1’s DTGs across cancers has remained elusive. Prior work demonstrated that FOXA1 represses TGFB3 via an upstream enhancer, activating SMAD2 phosphorylation and promoting epithelial-mesenchymal transition (EMT), thereby enhancing tumor invasiveness.²³ Moreover, transcriptomic analyses in hormone-driven cancers such as prostate and breast have suggested that FOXA1 modulates thousands of genes.^24,25,26 However, the absence of an integrated gene signature that combines cistromic and transcriptomic data has hindered a mechanistic understanding of FOXA1’s oncogenic programs and their tissue specificity.

To define high-confidence FOXA1 direct-target programs in canonical AR-positive prostate and luminal BCa models, we performed integrative analyses of FOXA1 cistromes and transcriptomes from FOXA1 knockdown models across two tumor types: prostate (LNCaP; ENA: PRJNA678636 and PRJNA490188)^23,27 and breast (T47D; ENA: PRJNA844574, PRJNA732359).^12,28 We deliberately focused our analysis on promoter-proximal FOXA1 binding events as a stringent strategy to define high-confidence direct target genes (DTGs), rather than as a comprehensive model of FOXA1’s overall regulatory activity. Specifically, we defined high-confidence DTGs as those with transcription start sites (TSSs) located within −1 kb to +100 bp of FOXA1 ChIP-seq peaks, and that also exhibit significantly altered expression following FOXA1 silencing. This rigorous integration yielded 3,042 DTGs in PCa (1,803 downregulated, 1,239 upregulated) and 2,458 DTGs in BCa (1,497 down, 961 up) (Figures 1A and 1B). To assess whether FOXA1 binding regions align with chromatin accessibility, we integrated ATAC-seq data into our DTG analysis in LNCaP and T47D cells. In line with the strong FOXA1 ChIP-seq signals at DTG promoters, these loci showed marked ATAC-seq enrichment, indicating that most FOXA1-bound DTG promoters reside within open chromatin (Figures S1A and S1B). We next investigated the biological pathways governed by these FOXA1 DTGs. Pathway enrichment analyses, using curated gene sets from MSigDB²⁹ including Bioplanet,³⁰ Hallmark,³¹ KEGG,³² Reactome,³³ and WikiPathways,³⁴ revealed that FOXA1 DTGs converge on conserved oncogenic programs such as angiogenesis,³⁵ cell cycle regulation,³⁶ hormone signaling,¹² hypoxia response,²⁰ immune modulation,²⁵ lipid metabolism,³⁷ and TGF-β signaling²³ (Figure 1C). Importantly, despite these shared themes, we uncovered striking cancer-type-specific transcriptional modules. In PCa, FOXA1 directly modulated genes involved in ferroptosis, SUMOylation, ubiquitin-mediated proteolysis, and SEMA4D-mediated cell migration, suggesting a multifaceted role in redox control and post-translational regulation (Figures 1D and S1C). In BCa, FOXA1 governed pathways related to E2F targets, DNA replication, and DNA damage repair, highlighting its influence on proliferative and genomic integrity mechanisms (Figures 1E and S1D).

Figure 1.

Cell-type-specific genomic signatures of FOXA1 DTGs across different cancer types

(A and B) Heatmaps display RNA-seq and ChIP-seq data for FOXA1 DTGs in PCa (LNCaP) (A) and BCa (T47D) (B). Upregulated genes are shown in red and downregulated genes in blue, with expression levels from control samples subtracted so that regions corresponding to these DEGs appear white. The right image shows ChIP-seq signals for FOXA1, active enhancers (H3K27ac and H3K4me1), promoter-associated histone modifications (H3K4me3), and POLR2A within ±5 kb of the TSSs of DTGs. Scale bars indicating gene expression levels and ChIP-seq signal intensity are shown beneath each figure.

(C) Pathway enrichment analysis of FOXA1 DTGs across prostate and breast cancers highlights shared involvement in key biological processes, including angiogenesis, cell cycle regulation, growth and development, hormone signaling, hypoxia response, immune response, lipid metabolism, and TGFβ signaling. Analysis was performed using MSigDB collections (Bioplanet, Hallmark, KEGG, Reactome, and WikiPathways). Enrichment analysis was performed using Enrichr. p values were calculated using Fisher's exact test, and multiple testing correction was performed using the Benjamini–Hochberg procedure to generate adjusted p values.

(D and E) FOXA1 DTGs exhibit cancer-type-specific pathway enrichment: (D) PCa (LNCaP): ferroptosis, SMAD4-induced cell migration, ubiquitination, and associated pathways. (E) BCa (T47D): E2F target activation, DNA repair, DNA replication, and related processes. Pathway enrichment analysis was performed using the Enrichr R package. Enrichment p values were calculated by Enrichr using Fisher’s exact test.

Cross-cancer comparison uncovered a conserved transcriptional core, with 433 genes (138 up and 295 down) consistently altered by FOXA1 depletion (Figures S1E and S1F). Taken together, these data unveil a complex and layered transcriptional landscape orchestrated by FOXA1 across distinct malignancies, establishing a robust, context-informed FOXA1 target gene signature and revealing the transcriptional circuits through which FOXA1 governs both shared and tissue-specific tumor phenotypes. Notably, this DTG set represents a conservative promoter-proximal subset of FOXA1 targets. Given FOXA1’s well-established functions at distal enhancers, enhancer-mediated regulation likely contributes substantially to its broader transcriptional impact and remains an important subject for future investigation.

FOXA1-directed transcriptional programs define prognostic gene signatures across multiple malignancies

Tumor progression is driven by complex genetic and epigenetic alterations that rewire transcriptional circuits. Increasingly, mRNA-based gene signatures have proven valuable in stratifying patient risk and guiding therapeutic decisions.^38,39 Given the critical role of FOXA1 as a lineage-defining pioneer factor in multiple epithelial cancers, we hypothesized that FOXA1-directed transcriptional programs may yield clinically relevant prognostic markers across cancer types. To this end, we performed a comprehensive multi-cohort analysis to identify and validate FOXA1-regulated prognostic gene signatures in PCa and BCa, thus revealing shared and context-specific transcriptional vulnerabilities with translational implications.

We developed a robust analytical pipeline to pinpoint FOXA1 DTGs with prognostic relevance (Figure S2A). Starting with differential gene expression analyses using The Cancer Genome Atlas (TCGA) and multiple independent datasets,^40,41,42,43 we filtered FOXA1 DTGs based on stringent criteria (adjusted p < 0.01, |log₂FC| > 1). This yielded 96 and 286 DTGs in PCa and BCa, respectively (Figures 2A and 2B), all consistently dysregulated across multiple datasets—highlighting their conserved regulatory roles downstream of FOXA1.

Figure 2.

The FOXA1-directed target gene signature predicts prostate, breast, and liver cancer prognosis in clinical cohorts

(A and B) Venn diagrams displaying the overlap of DEGs (upregulated or downregulated) across multiple independent cohorts: (A) PCa: TCGA, CPGEA, and GSE133626 datasets. (B) BCa: TCGA, GSE225846, and GSE58135 datasets.

(C and D) mRNA expression levels of seven key genes in PCa tumors versus normal prostate tissues from the TCGA (n = 550) (C) and CPGEA (n = 272) (D) cohorts. Statistical significance was assessed by the Mann-Whitney U test. Boxplots show the median as the center line, the interquartile range as the box bounds, and whiskers indicating data dispersion. Individual points represent samples.

(E and F) mRNA expression levels of ten genes were evaluated in BCa tumors versus normal breast tissues from the TCGA (n = 1212) (E) and GSE225846 (n = 143) (F) cohorts. Statistical significance was assessed using the Mann-Whitney U test. Boxplots show the median as the center line, the interquartile range as the box bounds, and whiskers indicating data dispersion. Individual points represent samples.

(G and H) Prognostic evaluation of the seven-gene signature in PCa: (G) Training cohort (TCGA, n = 498); (H) Validation cohort (MSKCC, n = 140). Analyses include time-dependent ROC curves, scatterplots of risk scores correlated with patient survival status, heatmaps of mRNA expression, and Kaplan-Meier survival curves. p values were calculated using the log rank test.

(I and J) Prognostic evaluation of the ten-gene signature in BCa: (I) training cohort (TCGA, n = 1100); (J) validation cohort (METABRIC, n = 1980). Analyses include time-dependent ROC curves, scatterplots of risk scores correlated with patient survival status, heatmaps of mRNA expression, and Kaplan-Meier survival curves. p values were calculated using the log rank test.

To evaluate their prognostic potential, we first performed univariate Cox regression to identify genes significantly associated with recurrence-free survival (RFS). Subsequent Lasso-penalized Cox regression reduced these candidates to compact, high-performing prognostic signatures—a 7-gene panel for PCa LNCaP cells and a 10-gene panel for BCa T47D cells (Figures S2B and S2C, and Table S1). Given the context dependence of FOXA1 regulatory activity, especially in hormone-driven cancers, we assessed whether the DTGs identified in LNCaP and T47D were reproducible in additional models. In PCa, six of seven LNCaP-derived FOXA1 DTGs showed concordant directional expression changes after FOXA1 depletion in the independent AR-positive VCaP model, with TMEM100 as the only exception due to its extremely low expression in VCaP cells (Figure S2D). FOXA1 promoter/TSS-proximal occupancy was also observed for all seven DTGs in both LNCaP and VCaP, with binding at some loci additionally detected in 22Rv1 and C4-2B (Figures S2E–S2K). In BCa, a subset of the 10 T47D-derived DTGs showed concordant expression changes following FOXA1 perturbation in MCF-7, ZR-75-1, and MDA-MB-453, whereas others displayed discordant responses (Figure S2L). However, FOXA1 promoter/TSS occupancy remained broadly detectable across BCa models, being present for all 10 DTGs in ZR-75-1 and MDA-MB-453 and for all except STAT5A in MCF-7 (Figures S2M–S2P and S3A–S3F). These findings support broad conservation of FOXA1 occupancy at DTG loci but indicate that downstream transcriptional responses are more context-dependent, particularly in BCa.

We next evaluated these DTGs in human tumors and observed consistent dysregulation in primary tumors compared with normal tissues across multiple independent PCa and BCa cohorts, further underscoring their biological and clinical relevance. These genes showed consistent dysregulation in primary tumors compared with normal tissues across multiple independent prostate and breast cohorts, reinforcing their biological and clinical relevance (Figures 2C–2F, S3G, and S3J).

To examine the performance of the derived gene signatures, we performed receiver operating characteristic (ROC) curve analysis to assess their prognostic discrimination. In PCa, time-dependent ROC analysis demonstrated strong prognostic accuracy, with AUCs of 0.784, 0.722, and 0.668 for 1-, 3-, and 5-year RFS (Figure 2G, left). High-risk patients displayed distinct gene expression patterns—downregulation of TMEM100 and ALDH1A2, and upregulation of MYBL2, BUB1, and others—alongside markedly poorer RFS (Figure 2G, middle and right; Figure S3I). These findings were externally validated in the MSKCC and DKFZ cohorts, which showed comparable predictive power (AUCs: 0.757–0.838) and robust survival stratification (Figures 2H and S3H). Additionally, high FOXA1 risk scores in patients with TCGA PCa correlated with increased progression risk and decreased disease-specific survival (Figure S3I).

Parallel analyses in BCa (Figures 2I, 2J, and S3K–S3M) confirmed the robust prognostic capacity of the FOXA1-derived gene signature. In each case, FOXA1-regulated transcriptional programs effectively stratified patient outcomes, underscoring their value as context-dependent, prognostically informative biomarkers. These findings position FOXA1-regulated gene signatures as clinically informative, context-specific prognostic biomarkers in prostate and BCas. By integrating TF-directed regulatory networks with clinical outcome modeling, our analysis establishes a powerful framework for uncovering actionable targets for risk stratification and therapy.

FOXA1-driven transcriptional targets define a prognostic program across cancer types

To evaluate the clinical significance of FOXA1-regulated genes across cancer types, we conducted univariate Cox regression analysis. Higher FOXA1-DTG risk scores were significantly associated with reduced biochemical RFS in both PCa and BCa (Figures S4A–S4D). Multivariate Cox models, adjusting for key clinicopathological factors, confirmed the FOXA1 DTG risk score as an independent prognostic marker across both cancer types (Figures 3A, 3B, S4E, and S4F), further strengthening the robustness of the FOXA1 DTG signatures in clinic applications.

Figure 3.

The clinical impact of FOXA1 DTG on pan-cancer pathogenesis

(A and B) Multivariate Cox regression analysis in the TCGA datasets for PCa (n = 411) (A) and BCa (n = 994) (B) revealed that the risk score was an independent prognostic marker, showing the highest hazard ratio among clinical features. p values were calculated using the Wald tests. CI, confidence interval.

(C and D) Gene signatures were found to be upregulated during PCa (n = 550) (C) and BCa (n = 1212) (D) development in the TCGA cohorts. p values were assessed using the Mann-Whitney U test. Boxplots show the median as the center line, the interquartile range as the box bounds, and whiskers indicating data dispersion. Individual points represent samples.

(E) The upregulation of the gene signature was associated with increased severity of PCa, as indicated by tumor stage, metastasis stage, lymph node stage, Gleason score, and PSA levels. p values were calculated using the Kruskal-Wallis H test for comparisons across three or more groups and the Mann-Whitney U test for comparisons between two groups. The numbers of samples are indicated in the relevant figures. Boxplots show the median as the center line, the interquartile range as the box bounds, and whiskers indicating data dispersion. Individual points represent samples.

(F) The upregulation of the gene signature was associated with increased severity of BCa, as indicated by tumor stage, PR status, ER status, and HER2 levels. p values were calculated using the Kruskal-Wallis H test for comparisons among three or more groups and the Mann-Whitney U test for comparisons between two groups. The numbers of samples are indicated in the relevant figures. Boxplots show the median as the center line, the interquartile range as the box bounds, and whiskers indicating data dispersion. Individual points represent samples.

(G) DTG gene signatures showed a positive correlation with the CCP score in independent PCa and BCa cohorts. p values were calculated using the two-sided Pearson’s product-moment correlation test. The numbers of samples are indicated in the relevant figures.

Remarkably, FOXA1 DTG risk scores were significantly elevated in tumor tissues compared to normal controls, both in primary and metastatic lesions across both cancer types. In PCa, the risk scores were markedly higher in tumor specimens compared to normal tissues (Figures 3C and S4G–S4J), a pattern also observed in BCa (Figures 3D and S4K–S4N). Moreover, In PCa, higher FOXA1 risk scores were strongly associated with advanced tumor stage, nodal involvement, metastatic burden, elevated Gleason scores, and increased prostate-specific antigen (PSA) levels (Figures 3E and S5A). Similarly, in BCa, FOXA1 risk scores correlated with tumor stage, progesterone receptor (PR) and ER status, HER2 amplification, and metastatic phenotype (Figures 3F, S5B, and S5C). These findings underscore the prognostic significance of the FOXA1 signature in cancer detection, differential diagnosis, and cancer progression. In PCa and BCa, the FOXA1 DTG signature also showed a strong positive correlation with the cell cycle progression (CCP) score,⁴⁴ a proliferation-associated prognostic marker (Figures 3G and S5D), further linking FOXA1 activity with proliferative aggressiveness.

To assess the specificity and robustness of the FOXA1 DTG signatures, we compared them with the CCP signature, as DTG and CCP scores showed a strong positive correlation. Despite this association, the DTG signatures remained informative beyond the proliferation-associated CCP signal, indicating that they capture prognostic information beyond generic proliferative activity (Table S2). We further benchmarked the DTG signatures against standard clinicopathologic models across independent prostate and BCa cohorts. In all four cohorts, incorporation of the DTG signature significantly improved the clinical model and generally outperformed CCP, supporting its added prognostic value beyond established clinicopathologic variables (Table S2).

To explore the molecular underpinnings of high-risk states, we stratified patients based on FOXA1 risk scores and performed gene set enrichment analysis (GSEA). In high-risk PCa, upregulated genes such as IQGAP3, BUB1, MYBL2, HOXC4, and SLC22A10 were identified, while TMEM100 and ALDH1A2 were significantly downregulated (Figures 4A, S5E, and S5F). In BCa, high-risk tumors displayed upregulation of mitotic regulators (PLK1, SHCBP1, PBK) and stress response genes (ULBP1, IER5L), whereas STAT5A, TIPARP, and APOD were repressed in low-risk tumors (Figures 4B, S5G, and S5H).

Figure 4.

The transcriptional signature of FOXA1 links risk stratification to cell cycle and EMT activation

(A and B) Signature genes were top-ranked for PCa (n = 498) (A) and BCa (n = 1100) (B) in the TCGA cohorts. Median values of gene signatures were used as cutoffs to distinguish high- and low-risk score groups. Genes were prioritized based on signal-to-noise ratio calculations.

(C–F) GSEA results revealed significant upregulation of E2F targets and the G2M checkpoint pathways in PCa (C and D) and BCa (E and F) within the high-risk score group compared to the low-risk score group. FDR and NES were reported for each pathway. NES, normalized enrichment score.

(G and H) GSEA pathways associated with cell cycle, tumor proliferation, and EMT were identified as activated or depleted, corresponding to enrichment in upregulated (G) or downregulated (H) genes, respectively, when comparing high-risk score groups to low-risk score groups in PCa and BCa. Pathways enriched in PCa or BCa were denoted with “P” or “B,” respectively. Prostate: TCGA n = 498, CPGEA n = 136, DKFZ n = 118; breast: TCGA n = 1100, Metabric n = 1980, GSE2034 n = 286. The FDR q-values were estimated from permutation-based null distributions as implemented in GSEA.

GSEA consistently revealed enrichment of E2F targets and G2M checkpoint pathways in high-risk groups across both cancer types (Figures 4C–4F and S5I–S5M). These findings suggest that FOXA1-driven programs contribute to cell cycle dysregulation and tumor progression. Additional functional enrichment analysis identified coordinated activation of oncogenic pathways such as proliferation, DNA replication, and the EMT process, along with repression of tumor-suppressive programs in high-risk patients (Figures 4G, 4H, S6A, and S6B). Collectively, these integrative analyses position the FOXA1 DTG signature as a PCa and BCa biomarker with significant clinical utility. It not only provides prognostic insights but also mechanistically links FOXA1 activity to aggressive tumor behavior. By offering a unified molecular axis for risk stratification, this signature opens avenues for targeted therapeutic strategies across multiple malignancies.

FOXA1 orchestrates cancer-specific transcriptional programs through integration of inherited variation and chromatin binding

We next sought to elucidate how inherited genetic variation interacts with lineage-specific transcriptional regulation mediated by FOXA1, a pioneering TF with well-established roles in hormone-responsive cancers.^45,46 We hypothesized that FOXA1 integrates chromatin-level regulatory cues with germline variation to shape cancer-type-specific transcriptional programs. Given prior evidence that noncoding SNPs can disrupt FOXA1-chromatin interactions,^15,47 we investigated whether FOXA1 binding sites are preferentially enriched for germline variants associated with cancer risk.

To this end, we performed integrative analyses of FOXA1 ChIP-seq peaks and GWAS-identified SNPs across PCa, BCa, and liver cancer, with liver cancer serving as a positive control (Figure S7). In both PCa and BCa, cancer-specific GWAS SNPs and their LD proxies exhibited strong, cancer type-specific enrichment within FOXA1 binding sites. In contrast, no such enrichment was observed in liver cancer, likely due to the limited availability of FOXA1 ChIP-seq and GWAS data for this cancer type (Figures 5A, S8A, and S8B). These findings highlight a lineage-restricted regulatory landscape orchestrated by FOXA1 in hormonally driven tumors. Supporting this, FOXA1 common binding regions were preferentially enriched for eQTL SNPs compared to non-eQTL SNPs across all three cancer types (Figure 5B and Table S3), suggesting that germline variants within FOXA1-occupied cis-regulatory elements broadly contribute to gene expression regulation and tumorigenesis.

Figure 5.

The intersection of FOXA1-binding SNPs and eGenes and FOXA1 DTGs highlights that FOXA1 forms cell-specific genomic signatures and differentially regulates gene expression

(A) Cancer type-specific enrichment of GWAS LD SNPs in FOXA1 chromatin binding sites in PCa, BCa, and liver cancer. All peaks: the entire FOXA1 chromatin binding sites in each cancer type. Unique peaks: FOXA1 peaks uniquely identified in each type of cancer. Common peaks: common FOXA1 chromatin binding sites in all three cancer types. Statistical significance was assessed using the binomial test.

(B) eQTL SNPs are highly enriched in the common DNA binding regions of FOXA1 in prostate, breast, and liver cancers in comparison with non-eQTL SNPs. A total of 1042 and 847 eQTL SNPs were identified in PCa and BCa, respectively. Statistical significance was assessed using the binomial test.

(C) DisGeNET-enriched pathways of eGenes associated with prostate and breast cancer, highlighting their relevance to disease-specific pathways. p values for DisGeNET enrichment in Metascape were calculated using the hypergeometric test.

(D and E) Venn diagrams show the overlap between FOXA1 DTGs and FOXA1 eGenes across PCa (D) and BCa (E).

(F) Heatmap illustrates the co-expression of FOXA1 with FOXA1 DTG-eGenes identified in PCa and BCa across a broad range of cancers using TCGA datasets. The “X” indicates an insignificant expression correlation. The color gradient ranges from blue (negative correlation) to red (positive correlation), with intensity reflecting the magnitude of the correlation. Statistical significance was determined using the two-sided Pearson’s product-moment correlation test.

(G) Bar plots summarize the counts of significant expression correlations for each cancer type. FOXA1 exhibited the highest co-expression with its DTG-eGenes in BCa and PCa compared to other cancer types. p values were determined using the two-sided Pearson’s product-moment correlation test.

To elucidate the downstream biological significance of FOXA1 activity, we performed pathway enrichment analyses using DisGeNET. FOXA1-regulated eGenes exhibited cancer-type-specific disease associations: In PCa, they were enriched in pathways related to hereditary and familial PCa. In BCa, they were associated with mammary carcinoma and neoplasm-related pathways (Figure 5C). These observations underscore the context-dependent transcriptional outputs of FOXA1-regulated eGenes, which reflect the underlying tumor etiologies.

Notably, we identified a modest overlap between FOXA1 DTGs and eGenes—44 genes in PCa and 36 in BCa (Figures 5D, 5E, S8C, and S8D). This convergence suggests the existence of a core gene set of genes that function both as DTGs and as genetically regulated transcripts, potentially serving as key effectors of FOXA1-driven oncogenic programs. To explore their coordinated regulation, we assessed the co-expression patterns of these genes and FOXA1 across TCGA pan-cancer datasets. This analysis revealed significantly stronger DTG-eGenes and FOXA1 co-expression correlations in PCa and BCa relative to other cancer types (Figures 5F and 5G), supporting a model in which FOXA1 operates as a lineage-specific master regulator. By integrating inherited variation with chromatin occupancy, FOXA1 orchestrates disease-relevant transcriptional programs in a cancer-type-dependent manner. Collectively, in 22Rv1 cells, FOXA1 knockdown or deletion of the rs1417078-containing element increased TLE4 expression, suggesting that FOXA1 occupancy at this locus may exert a repressive or constraining effect in this model; this contrasts with the clinical association between the rs1417078 A allele and elevated TLE4 expression, indicating that the net regulatory output of this locus is likely allele-dependent and shaped by tumor-specific chromatin and cofactor contexts. These findings position FOXA1 as a central transcriptional hub that integrates genetic variation, epigenomic context, and transcriptional output to shape lineage-specific oncogenic trajectories.

FOXA1-bound SNPs link genetic variation to BCa progression

To directly assess whether FOXA1-binding SNPs link germline genetic variation to BCa progression, we first examined BCa eQTL genes associated with FOXA1-bound SNPs and found that they are strongly enriched for BCa-related traits and diseases (Figure S9A). We next performed SNP-linked gene enrichment analysis using Reactome pathways, and the top enriched pathways included DNA repair and cell cycle regulation (Figure S9B), two processes strongly implicated in BCa tumorigenesis and progression. These results suggest that FOXA1-associated regulatory variants have the potential to influence BCa biology through key oncogenic pathways. We then evaluated the clinical relevance of these FOXA1-binding eQTL SNPs by assessing their ability to predict progression and overall survival in patients with BCa. Approximately 20 SNPs showed significant associations with patient prognosis (Figure S9C), indicating that BCa eQTL SNPs could exert effects on BCa progression at the clinical level. Among these clinically relevant SNPs, we identified two FOXA1-binding eQTL SNPs, rs7907981 and rs7907988, whose GWAS-reported lead SNP rs7918232 was previously implicated in BCa risk.⁴⁸ The risk alleles of rs7907981 and rs7907988 were associated with significant upregulation of their eQTL gene MASTL (Figure S9D). In TCGA Breast tumors, MASTL expression is markedly elevated compared with normal breast tissues (Figure S9E). Furthermore, high MASTL expression is strongly associated with poorer progression-free and overall survival across multiple independent BCa cohorts (Figure S9F). Importantly, MASTL has been previously reported to have oncogenic functions in BCa, promoting tumor cell proliferation, mitotic progression, and survival.^49,50 Taken together, our findings provide functional support for the role of FOXA1-binding regulatory SNPs in modulating target gene expression and influencing BCa progression.

FOXA1 and its eGene orchestrate genetic susceptibility and govern clinical aggressiveness in PCa

Having observed that GWAS SNPs and eQTL signals were significantly enriched at FOXA1 binding sites and with strong co-expression between FOXA1 and DTG-eGenes in PCa and BCa (Figures 5A, 5B, 5F, and 5G), we next explored the biological mechanisms underlying these genetic loci and their clinical implications in disease predisposition and aggressiveness. Given that noncoding risk variants often reside in regulatory regions and can perturb TF binding affinity, we hypothesized that SNPs within FOXA1-binding sites may reprogram transcriptional outputs, thereby contributing to PCa susceptibility and progression.

Our integrative genomic analysis revealed striking enrichment of PCa-associated GWAS SNPs within the cistromes of key lineage-defining TFs, most notably FOXA1, alongside androgen receptor (AR), ERG, HOXB13, and MYC (Figure 6A). These results suggest that FOXA1 occupies a central position in the genetic regulatory landscape of PCa, acting as a pivotal transcriptional hub.

Figure 6.

PCa susceptibility loci are markedly enriched in FOXA1 binding sites as well as its eQTL genes are associated with PCa progression

(A) FOXA1 is top-ranked for the enrichment of GWAS SNPs among 10 important TFs in PCa, including AR, ERG, HOXB13, and MYC. Statistical significance was assessed using the binomial test.

(B) Genome-wide loss-of-function screen in 22Rv1, NCI-H660, VCaP, LNCaP, and PC3 identified essential genes including FOXA1 for cell survival. Lower scores indicate higher dependency on the gene for cell viability. MYC, AR, and GATA2 are well-known genes driving PCa cell proliferation and survival, whereas tumor suppressor TP53 does not favor PCa cell growth and survival.

(C) FOXA1 mRNA expression gradually decreased in a complete and stepwise prostate carcinogenesis model. EPT1 is mesenchymal cells derived from EP156T cells, EPT2 is cloned from foci formed in EPT1 cells, and EPT3-PT1 and EPT3-M1 are derived from a primary tumor in mouse prostate and a pool of abdominal metastases, respectively. Increasing malignant phenotypes were found in the stepwise 5-cell-line carcinogenesis model.

(D) FOXA1 expression was significantly higher in primary prostate tumors compared to normal prostate glands but decreased in metastatic tumors (n = 122). Statistical significance was assessed using the Kruskal-Wallis test. Boxplots show the median as the center line, the interquartile range as the box bounds, and whiskers indicating data dispersion. Individual points represent samples.

(E and F) Kaplan-Meier curves depict the association between FOXA1 expression levels and overall survival (n = 31) (E) and biochemical recurrence (n = 27) (F) in patients with PCa; log rank test. p values were calculated by the log rank test.

(G) Depletion of FOXA1 in 22RV1 through lentivirus-mediated shRNA interference. n = 3; technical replicates, error bars, mean ± SD; ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001; ns, non-significant; p values were evaluated using the two-tailed Student’s t tests. Source data are provided in Source Data file.

(H) FOXA1 knockdown reduces PCa cell proliferation. n = 3 technical replicates, error bars, mean ± SD, ∗∗p < 0.01 and ∗∗∗p < 0.001, p values were evaluated using the two-tailed Student’s t tests.

(I) Wound healing assay in the 22Rv1 cells infected with lentiviruses expressing shRNAs targeting FOXA1. n = 3 technical replicates, error bars, mean ± SD, ∗∗p < 0.01 and ∗∗∗p < 0.001, p values were evaluated using the two-tailed Student’s t tests.

(J) Circos visualization of PCa risk loci enriched in FOXA1 binding sites with corresponding eQTL genes.

(K–Q) Increased expression of the FOXA1 eGene signature is associated with tumor development and progression to metastasis (K, n = 550; L, n = 407; M = 150), higher Gleason score (N, n = 498), advanced tumor stages (O, n = 486); pathological metastasis (P, n = 82); and lymph node involvement (Q, n = 420). Statistical significance was assessed by two-sided Mann-Whitney U test or Kruskal-Wallis test. Boxplots show the median as the center line, the interquartile range as the box bounds, and whiskers indicating data dispersion. Individual points represent samples.

(R–U) Kaplan-Meier curves depict the association between the FOXA1 eGene signature expression levels and prognosis in patients with PCa: biochemical recurrence (R, n = 492; S, n = 120), metastasis (T, n = 493), and overall survival (U, n = 81). Statistical significance was assessed using the log rank test.

To functionally interrogate the role of FOXA1, we analyzed genome-wide CRISPR-Cas9 loss-of-function screen datasets across multiple PCa cell lines (22Rv1, NCI-H660, VCaP, LNCaP, and PC3). FOXA1 emerged as a critical dependency for cell survival, underscoring its essentiality in PCa biology (Figure 6B). Notably, FOXA1 expression exhibited a stage-dependent pattern: it was upregulated in primary tumors compared to normal tissue but significantly reduced in metastatic lesions across multiple independent transcriptomic datasets (Figures 6C, 6D, and S10A–S10L). These findings suggest that FOXA1 has context-dependent functions across PCa progression. While FOXA1 supports tumor-cell fitness in AR-driven disease, reduced FOXA1 expression in metastatic tumors is associated with poorer clinical outcomes and may reflect loss of a differentiated AR/FOXA1-driven luminal lineage program and emergence of more plastic, aggressive disease states.^10,51

Clinically, reduced FOXA1 expression correlated with poor overall survival (Figure 6E), higher risk of biochemical recurrence (Figure 6F), and diminished response to first chemotherapy and hormone therapy (Figures S10M and S10N), implicating FOXA1 loss in disease aggressiveness and therapeutic resistance. To validate these associations, lentiviral shRNA-mediated knockdown of FOXA1 in 22Rv1 cells significantly impaired proliferation (Figures 6G and 6H) and migration (Figure 6I), confirming its functional relevance in PCa pathogenesis.

We then investigated the genetic architecture underpinning FOXA1-mediated regulation. By intersecting FOXA1 ChIP-seq peaks with GWAS tag and LD SNPs, we identified 1,042 PCa risk-associated SNPs residing within the FOXA1 cistrome (Figure S10O and Table S3). Given that SNPs within TF motifs can directly alter binding specificity,^5,15 we performed a motif-SNP interaction test using the Enhancer Element Locator (EEL) algorithm⁵² and identified 14 SNPs that potentially affect FOXA1 binding. Notably, several of these SNPs reside within genomic regions linked to PCa susceptibility in large-scale GWAS, though their functional relevance has remained uncharacterized.^53,54 To functionally identify potential regulatory associations, we performed comprehensive cis-eQTL analyses and identified 23 eGenes across multiple independent PCa datasets, including TCGA, Stockholm, and Camcap cohorts, as well as GTEx,⁵⁵ PancanQTL,⁵⁶ and ncRNA-eQTL⁵⁷ databases, forming a robust FOXA1 eGene signature (Figure 6J).

Strikingly, this FOXA1 eGene signature was strongly associated with adverse clinicopathological features across multiple patient cohorts, including tumor development (Figures 6K and 6L), metastatic progression, high Gleason scores, advanced tumor stages, pathological metastasis, and lymph node involvement (Figures 6M–6Q and S10P–S10X). Higher FOXA1 eGene signature expression also predicted increased risk of biochemical relapse, metastasis, and shorter overall survival (Figures 6R–6U), further solidifying its clinical prognostic value. Together, these results reveal that FOXA1 and its genetically regulated target genes serve as critical determinants of PCa susceptibility, tumor progression, and patient outcomes. Through a comprehensive integration of GWAS, eQTL, CRISPR functional genomics, and clinical data, our study uncovers a FOXA1-centric regulatory axis that links noncoding germline variation to transcriptional rewiring and aggressive disease phenotypes.

FOXA1-bound eQTL SNPs define candidate functional regulatory variants contributing to PCa susceptibility

We next sought to elucidate the role of non-coding genetic variation in PCa risk, focusing on the regulatory potential of SNPs located within FOXA1 binding sites. Building on our previous work demonstrating that SNPs can disrupt TF binding motifs and chromatin interactions (e.g., for HOXB13, HOXA2, TMPRSS2-ERG, and HNF1B),^16,58,59 we evaluated FOXA1 occupancy at 14 PCa risk-associated SNP loci. To this end, we performed ChIP-qPCR in 22Rv1 and V16A PCa cell lines using a FOXA1 antibody, with IgG as a negative control and the TGFβ3 enhancer as a positive control.²³ FOXA1 binding was robustly detected at 10 of the 14 tested SNPs, supporting their candidacy as functional regulatory elements (Figures 7A and S11A, and Table S4). The remaining four SNPs (rs11890255, rs146710035, rs1978060, and rs9806221) were excluded from further analysis due to insufficient enrichment in ChIP-qPCR assays.

Figure 7.

Multiple eQTL SNPs at FOXA1 binding sites are potential candidate regulatory variants and are involved in modulating PCa progression

(A) ChIP-qPCR for FOXA1 chromatin binding at the FOXA1-binding SNP-containing region in 22Rv1 cell. n = 10 SNPs, mean ± SD of three technical replicates. ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001 compared to IgG control; ns, non-significant; p values were evaluated using the two-tailed Student’s t tests.

(B) RT-qPCR analysis to determine the mRNA expression levels of FOXA1 eGenes in 22Rv1 cells with partial knockout (KO) of each SNP region via CRISPR/Cas9 genome editing technology. n = 17 genes, mean ± SD of three technical replicates. ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001 compared to Beta-actin control; ns, non-significant; p values were evaluated using the two-tailed Student’s t tests.

(C) RT-qPCR analysis to determine the mRNA expression levels of FOXA1 eGenes in 22Rv1 cells with FOXA1 knockdown via siRNA. n = 11 genes, mean ± SD of three technical replicates. ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001 compared to Beta-actin control; ns, non-significant; p values were evaluated using the two-tailed Student’s t tests.

(D–F) Chromatin-binding of FOXA1, active histone marks (H3K27ac, H3K4me1, and H3K4me3), and POLR2A at representative genes TLE4 (D), CPNE1 (E), and USP39 (F).

(G–P) The PCa risk allele A at rs1417078 is associated with increased expression of TLE4 in TCGA (G, n = 405), Camcap (H, n = 119) and GTEx (I, n = 218) cohorts; risk allele G at rs4911493 is associated with increased expression of CPNE1 in the TCGA (J, n = 415), CPGEA (K–L, n = 134), GTEx (M, n = 218), and Camcap (N, n = 119) cohorts; risk allele ATTT at rs5832649 is associated with increased expression of USP39 in the TCGA (O, n = 405) and GTEx cohorts (P, n = 219). p values were assessed using a linear model in Matrix eQTL. Boxplots show the median as the center line, the interquartile range as the box bounds, and whiskers indicating data dispersion. Individual points represent samples.

(Q) eQTL analysis of the rs1417078 locus for genes within a 10-Mbp region, using the TCGA cohort of prostate tumors (n = 405). p values were assessed using a linear model in Matrix eQTL.

Next, we examined whether these 10 FOXA1-bound SNPs could modulate the expression of their associated target eGenes. Using genotyping and Sanger sequencing, we identified heterozygosity for seven of these SNPs across PCa cell lines: rs67026445 in LNCaP; rs4911493, rs6735074, rs6740722, rs11568818 in 22Rv1; and rs4911493 and rs11568818 in VCaP (Table S5). To assess the candidate regulatory effect of these SNPs, we conducted CRISPR/Cas9-mediated deletions⁶⁰ of their flanking genomic regions in 22Rv1 cells. While deletion of rs1887414, rs13069553, and rs7153397 had no significant impact on the expression of their candidate target genes, several SNPs showed promising regulatory effects (Figure 7B).

Given the established paradigm that TFs like FOXA1 often co-regulate the expression of their eQTL-linked targets,^15,58,59 we assessed co-expression between FOXA1 and its putative downstream genes in 22Rv1 and V16A cells (Figures 7C and S11B). Based on these co-expression patterns, together with FOXA1 chromatin binding and CRISPR evidence, we prioritized three FOXA1-bound SNPs—rs1417078, rs4911493, and rs5832649—and their respective eQTL targets (TLE4, CPNE1, and USP39) for in-depth mechanistic analysis. FOXA1 ChIP-seq data revealed strong co-occupancy of these SNP loci with active histone modifications (H3K27ac, H3K4me1, H3K4me3) and RNA polymerase II (POLR2A) around the TSSs of these target genes (Figures 7D–7F), supporting their roles as active enhancers.

Strikingly, risk allele A at rs1417078 was strongly correlated with elevated TLE4 expression across multiple distinct eQTL datasets (Figures 7G–7I and S11C). Similarly, the G risk allele at rs4911493 was associated with increased CPNE1 expression (Figures 7J–7N and S11D), while the ATTT/TTAATTTA/TTTA/TTTATTTA risk haplotype at rs5832649 was linked to higher USP39 levels (Figures 7O, 7P, and S10E–S10H). In-depth cis-eQTL analysis revealed that rs1417078 exhibited the most significant association with TLE4 within a 10 Mb genomic window across multiple cohorts (Figures 7Q, S11I, and S11J), and both rs4911493 and rs5832649 showed localized regulatory specificity for CPNE1 and USP39, respectively, within a 1 Mb interval in the TCGA dataset (Figures S11K and S11L). Collectively, these results define a set of functionally validated eQTL SNPs residing within FOXA1 binding sites as likely candidate regulatory non-coding variants that modulate the transcription of key PCa susceptibility genes. This analysis provides compelling evidence that lineage-specific TFs like FOXA1 interpret genetic risk by engaging functional SNPs within regulatory elements, thereby influencing PCa progression.

FOXA1-associated eQTL genes TLE4, CPNE1, and USP39 function as oncogenic drivers and prognostic biomarkers in PCa

Given the critical role of FOXA1 as a lineage-defining pioneer TF in PCa, we postulated that its eQTL-associated target genes may exert previously uncharacterized oncogenic functions. We therefore systematically evaluated the biological and clinical relevance of TLE4, CPNE1, and USP39 in PCa to clarify their contributions to PCa.

Using shRNA-mediated gene silencing in multiple PCa cell models, we found that knockdown of TLE4 significantly impaired the proliferative capacity of both 22Rv1 and PC3 cells (Figures 8A, 8B, S12A, and S12B), implicating TLE4 as an oncogenic driver in both AR-positive and AR-independent contexts. Similarly, targeted suppression of CPNE1 and USP39 notably inhibited cell proliferation and migration in 22Rv1 cells (Figures 8C–8H, S12C, and S12D), suggesting that these eQTL-associated genes contribute broadly to the promotion of malignant phenotypes. In line with experimental findings, complementary transcriptomic analyses also demonstrated that TLE4, CPNE1, and USP39 were significantly upregulated in prostate tumors compared to normal prostate glands (Figures 8I–8K and S12E–S12I), with expression further elevated in higher-grade tumors, lymph node metastases, and specimens with elevated Gleason scores (Figures 8L–8T and S12J–S12M). Importantly, high expression levels of TLE4 and CPNE1 were consistently associated with poor clinical outcomes, including reduced overall survival, diminished response to first hormone therapy, and increased risks of biochemical recurrence and metastasis (Figures 8U–8W and S12N–S12Q). Collectively, these results establish TLE4, CPNE1, and USP39 as downstream effectors of FOXA1-associated genetic risk variants and functionally validate their contributions to PCa aggressiveness. By integrating genetic, transcriptomic, functional, and clinical evidence, our study delineates a FOXA1-regulated oncogenic axis in PCa and suggests that eQTL-nominated genes may inform future studies of PCa progression, prognosis, and therapeutic relevance.

Figure 8.

eQTL genes TLE4, CPNE1 and USP39 as oncogenes with prognostic potential in PCa

(A and B) TLE4 knockdown reduces PCa cell proliferation in 22Rv1 (A) and PC3 (B) cell lines. n = 3 technical replicates, error bars, mean ± SD, ∗∗p < 0.01 and ∗∗∗p < 0.001, p values were evaluated using the two-tailed Student’s t tests.

(C–E) CPNE-1 knockdown reduces PCa cell proliferation (C) and migration (D and E) in the 22Rv1 cell line. n = 3 technical replicates; error bars, mean ± SD; ∗∗p < 0.01 and ∗∗∗p < 0.001; p values were evaluated using the two-tailed Student’s t tests.

(F–H) USP39 knockdown reduces PCa cell proliferation (F) and migration (G and H) in the 22Rv1 cell line. n = 3 technical replicates, error bars, mean ± SD, ∗∗p < 0.01 and ∗∗∗p < 0.001, p values were evaluated using the two-tailed Student’s t tests.

(I–T) TLE4, CPNE-1, and USP39 are highly expressed in prostate tumors compared to normal tissues (I, n = 406; J, n = 407; K, n = 272). Their elevated expression is also associated with advanced tumor stage (L = 491; M = 486; N, n = 491), lymph node metastasis (O, n = 425; P, n = 420; Q, n = 425), and higher Gleason score (R, n = 168; S, n = 498; T, n = 498). Statistical significance was assessed by two-sided Mann-Whitney U test or Kruskal-Wallis test. Boxplots show the median as the center line, the interquartile range as the box bounds, and whiskers indicating data dispersion. Individual points represent samples.

(U–W) Kaplan-Meier curves show the associations between gene expression levels of TLE4, CPNE1, and USP39, and PCa patient prognosis, as indicated by overall survival (U, n = 363; V, n = 363), and biochemical recurrence (W, n = 492). p values were assessed using the log rank test.

FOXA1-binding SNPs functionally link genetic variation to PCa progression

To uncover the clinical relevance of FOXA1-binding SNPs, we conducted a phenome-wide association study (PheWAS) leveraging the FinnGen cohort (n = 412,181) across 2,405 disease endpoints. This large-scale analysis identified two variants, rs1417078 and rs5832649, that displayed robust associations with malignant neoplasm of the prostate (Figure 9A). Although rs4911493 showed a comparatively weaker statistical association, it was still significantly linked to PCa, reinforcing its potential functional role across diverse populations. These findings highlight the cross-ethnic relevance of these regulatory SNPs in PCa susceptibility.

Figure 9.

Association of FOXA1 binding SNPs with clinical outcomes in PCa and their functional implications

(A) PheWAS of the associations between rs1417078, rs4911493, and rs5832649 and 2,405 disease endpoints in the FinnGen study (n = 412,181). p values were obtained from the FinnGen PheWAS summary statistics.

(B–E) Survival analyses indicate that patients carrying the rs1417078 AA or AG genotypes are more likely to experience earlier biochemical recurrence (B, n = 400) or metastasis (C, n = 399). Similarly, patients with PCa with the rs5832649 ATTT, TTAATTTA, TTTA, or TTTATTTA genotypes show an increased likelihood of earlier biochemical recurrence (D, n = 208) or metastasis (E, n = 207). p values were assessed using the log rank test.

(F) FOXA1 consistently demonstrated preferential binding to the A allele of rs4911493 over the G allele under both hormonal conditions. n = 3 technical replicates, mean ± SD of three technical replicates. ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001, p values were evaluated using two-tailed Student’s t tests.

(G and H) Genome browser representation of ChIP-seq signals for active histone marks (H3K27ac, H3K4me1 and H3K4me3), POLR2A, and TFs (AR, GATA2, and FOXA1) at the rs1417078 locus in PCa cells (G). The rs1417078 variant is located within a FOXA1 DNA-binding motif. Fold enrichment of sequences surrounding rs1417078 harboring either the A or G allele under FOXA1 ChIP-seq peaks (H).

(I) 4C contact profiles in LNCaP cells using the SNP rs1417078 as the viewpoint.

To assess the direct prognostic significance of these variants in patients with PCa, we performed Kaplan-Meier survival analyses in the TCGA cohort. Patients carrying the risk A allele at the rs1417078 exhibited significantly shorter biochemical recurrence-free and metastasis-free survival compared to those with the GG genotype (Figures 9B and 9C). Similarly, individuals with the rs5832649 ATTT/TTAATTTA/TTTA/TTTATTTA genotypes showed a markedly higher risk of recurrence and metastasis (Figures 9D and 9E), suggesting that these SNPs influence not only PCa onset but also its clinical trajectory.

Due to the limited availability of PCa-relevant cell lines heterozygous for these SNPs, we identified that rs4911493 is heterozygous in 22Rv1, while rs1417078 and rs5832649 were not. To test whether FOXA1 binding is allele-specific, we performed ChIP followed by allele-specific qPCR (AS-qPCR) in 22Rv1 cells under both ethanol (ETH) and dihydrotestosterone (DHT) treatments. FOXA1 consistently demonstrated preferential binding to the A allele of rs4911493 over the G allele under both hormonal conditions, providing direct evidence of allele-specific TF recruitment in vivo (Figure 9F), while the single-base contribution of rs1417078 and rs4911493 remains unresolved in the absence of site-directed mutagenesis, or allelic replacement experiments in an endogenous genomic context. These findings, combined with the epidemiological results, suggest that FOXA1-associated SNPs serve as functional and prognostic biomarkers for PCa progression.

Importantly, the discrepancy often observed between TF binding events and large-scale disease phenotypes, such as elevated cancer risk, highlights the complexity of regulatory SNP function. While histological and biochemical assays offer mechanistic insights, population-level associations, such as those reported here, provide a broader clinical context. Notably, the mechanistic basis of rs5832649 remains to be fully elucidated, warranting further investigation into its downstream regulatory consequences.

Next, we sought to characterize the cis-regulatory mechanism by which rs1417078 modulates expression of its eQTL target gene TLE4, a known FOXA1 DTG and prognostic factor in PCa. Integrative analysis of ChIP-seq data from PCa tissues and cell lines revealed that the rs1417078 locus is situated within a putative enhancer marked by H3K27ac and H3K4me1/3, and co-occupied by key PCa TFs, including AR, HOXB13, and GATA2 (Figure 9G). Motif enrichment analysis⁶¹ further demonstrated that the A allele sequence (TCTTTAAGTA) is significantly overrepresented within FOXA1 ChIP-seq peaks compared to the G allele sequence (TCTTTAAATA), suggesting enhanced FOXA1 binding potential driven by the A allele (Figure 9H).

To explore whether this enhancer physically interacts with TLE4, we constructed virtual 4C maps from Hi-C data in LNCaP cells.⁶² We observed preferential chromatin looping between rs1417078 and TLE4, but not other eQTL genes (Figure 9I), implicating allele-dependent long-range regulatory interactions that facilitate TLE4 transcription in the presence of the A allele. rs1417078 is a candidate regulatory SNP located within a FOXA1-bound regulatory element whose functional impact is supported by convergent evidence from chromatin occupancy, eQTL association, CRISPR perturbation, and reporter assays, but whose precise allele-specific effect remains to be established.

Collectively, these findings delineate a functionally coherent regulatory axis linking FOXA1, disease-associated enhancer SNPs, and PCa progression. Our integrative analyses suggest that the rs1417078 A allele is associated with enhanced FOXA1 binding potential and regulatory activity at this enhancer, promoting TLE4 upregulation and correlating with adverse patient prognosis. The combination of rs1407078 genotype and TLE4 expression further refines patient stratification and prognostic modeling. This work supports FOXA1-binding SNPs as mechanistic contributors to inherited PCa risk and suggests that these variants may have potential relevance for future studies of PCa risk stratification and disease management.

Discussion

FOXA1 has emerged as a quintessential lineage-defining TF that integrates chromatin dynamics, genetic variation, and oncogenic signaling to orchestrate context-specific transcriptional programs across diverse malignancies.^54,63,64,65 By integrating cistromic, transcriptomic, and patient-derived datasets, our study provides a comprehensive, cross-cancer dissection of FOXA1-driven regulatory networks in PCa and BCa. This framework not only delineates shared oncogenic circuits, including angiogenesis,³⁵ immune regulation,²⁵ and cell cycle control,³⁶ but also reveals striking lineage-specific pathways, such as ferroptosis regulation in PCa and DNA repair in BCa. These findings underscore the pleiotropic yet highly context-dependent nature of FOXA1’s transcriptional control.

A central innovation of our work is the explicit integration of germline genetics with FOXA1 cistromes, transcriptomic responses to FOXA1 perturbation, and clinical endpoints. Building on this framework, we developed FOXA1 DTG signatures using a stringent promoter-proximal strategy that required FOXA1 occupancy near gene promoters together with FOXA1-dependent expression changes. Importantly, FOXA1 expression, FOXA1 activity, and the derived FOXA1 gene signatures are related but not equivalent. Whereas FOXA1 expression reflects FOXA1 mRNA abundance, FOXA1 activity refers to the downstream transcriptional output associated with FOXA1 in a given biological context, which is shaped by chromatin state, lineage context, disease stage, and cofactor interactions. Accordingly, FOXA1 expression alone is not expected to uniformly predict FOXA1 regulatory activity across tumors, and our FOXA1 DTG signatures should therefore be interpreted as a context-informed readout of high-confidence direct FOXA1 target gene output rather than simple surrogates for FOXA1 expression. This distinction is especially relevant in light of FOXA1 heterogeneity across disease stages and subtypes, which is likely an important determinant of signature behavior. This integrative promoter-proximal DTG framework enabled the derivation and cross-cancer comparison of high-confidence direct FOXA1 target genes, revealing pathway-level convergence alongside gene-level lineage specificity. Consequently, our FOXA1-DTG signatures differ from previously reported FOXA1-, AR-, or ER-associated transcriptional programs, which are often based on co-expression, receptor-associated signatures, or binding data alone. Although these signatures show partial overlap with established receptor-associated programs, they are not equivalent to them and instead capture a context-dependent FOXA1-linked transcriptional component that is not fully represented by canonical AR- or ER-centered programs.

To strengthen the robustness and interpretability of these DTG signatures, we benchmarked them against standard clinicopathologic variables and the CCP proliferation signature. These analyses showed that the prognostic association of the DTG signatures could not be explained solely by generic proliferation-related transcriptional activity. The DTG signatures improved clinicopathologic models, indicating added prognostic value beyond conventional clinical factors, and also enhanced CCP-based models. By contrast, CCP added only limited prognostic information beyond the DTG signatures in some cohorts. Together, these findings indicate that the DTG signatures capture biologically and clinically meaningful information beyond standard clinical variables and generic proliferation-associated programs.

Importantly, the DTG signatures should be interpreted as context-informed FOXA1 regulatory programs rather than universal FOXA1 programs across all tumors. Because these regulatory maps are derived primarily from two canonical cell line models—LNCaP and T47D—their generalizability warrants careful consideration. In PCa, FOXA1 activity is strongly influenced by AR signaling and disease state, such that the LNCaP-derived program may reflect an AR-driven, model-specific transcriptional state rather than a prostate-wide FOXA1 program. Likewise, in BCa, FOXA1 targets defined in T47D likely represent a luminal, cell-line-specific regulatory landscape rather than a universal FOXA1 core program. Extending this framework to additional PCa states, including castration-resistant, double-negative, and neuroendocrine models, as well as additional BCa subtypes, will be important for distinguishing broadly conserved components of FOXA1 regulation from those that are context-restricted. Validation across such models would further strengthen the conclusions.

Although recent studies have established FOXA1’s pivotal roles in splicing regulation, hormone receptor reprogramming, pioneer factor dynamics, and mitotic bookmarking,^11,12,13,66 none have addressed whether germline variants within FOXA1 cistromes mediate functional enhancer regulation or contribute to patient risk stratification. Here, we demonstrate that cancer GWAS loci are significantly enriched in FOXA1 chromatin-binding regions, particularly in prostate and BCas, thereby defining FOXA1 cistromes as hotspots of functional noncoding variation.⁶⁷ By coupling eQTL analyses with FOXA1-driven transcriptional target genes, we identify convergent sets of regulatory SNPs and eGenes, positioning FOXA1 as a central hub linking inherited susceptibility with transcriptional dysregulation. In this framework, the subset of shared DTGs-eGenes may represent a conserved core FOXA1 program with potential relevance to cross-lineage mechanisms and FOXA1-dependent therapeutic vulnerabilities.^68,69 This germline-to-phenotype framework expands the FOXA1 field beyond prior mechanistic studies of hormone receptor reprogramming, pioneer factor dynamics, or splicing regulation, providing a unique bridge from inherited variation to clinical outcomes.

Our PCa analyses further exemplify this concept by identifying three functional noncoding variants—rs1417078, rs4911493, and rs5832649—that directly modulate FOXA1 binding and alter enhancer activity. Through integrative molecular and functional validation, we show that these SNPs regulate oncogenic effectors (TLE4, CPNE1, and USP39) that promote proliferation, migration, and poor clinical outcomes. Notably, the risk A allele at rs1417078 enhances FOXA1 binding, associates with elevated TLE4 expression, and stratifies patients by biochemical-recurrence and metastatic risk, thereby establishing a direct mechanistic link between germline risk alleles, enhancer regulation, and aggressive disease phenotypes. These findings illustrate how inherited variation can be functionally decoded through the lens of lineage-defining TFs to uncover oncogenic effectors and candidate markers associated with disease progression. Direct allele-specific FOXA1 binding was only validated for rs4911493 but not for rs1417078 or rs5832649 because suitable heterozygous PCa models were unavailable; definitive confirmation will require future experiments like luciferase reporter assays or electrophoretic mobility shift assays (EMSAs) in heterozygous or isogenic models.

Our study also addresses the paradoxical dynamics of FOXA1 expression during PCa progression. Elevated in primary tumors yet diminished in metastatic disease.^{10,23,51,70,71} Although FOXA1 remains functionally important in advanced PCa, lower FOXA1 expression in metastatic tumors may indicate erosion of the AR/FOXA1-associated luminal lineage state, which is generally linked to more differentiated behavior. In contrast, tumors with reduced FOXA1 and AR expression may acquire lineage plasticity, dedifferentiation, and therapy resistance. Such biphasic behavior, reminiscent of other lineage-determining TFs, highlights the context-dependent duality of FOXA1 and underscores the need for future work in therapy-resistant states to clarify its stage-specific functions.

Future integrative studies incorporating chromatin conformation mapping will be essential to define the broader enhancer-mediated FOXA1 regulatory network. rs1417078 is a candidate regulatory SNP located within a FOXA1-bound regulatory element whose functional impact is supported by convergent evidence from chromatin occupancy, eQTL association and CRISPR perturbation, but whose precise allele-specific effect remains to be established. Future studies should therefore employ site-directed mutagenesis-based luciferase assays and EMSA or other allele-specific binding approaches to clarify the mechanistic effect of rs1417078 on FOXA1 occupancy and enhancer activity.

Collectively, our findings establish FOXA1 as a master regulator that integrates germline variation with lineage-specific transcriptional programs to drive cancer susceptibility and progression. By identifying candidate regulatory SNPs and functionally validating downstream effectors, we provide mechanistic insights that bridge noncoding variation to clinical outcomes. This work expands biological understanding of FOXA1 in oncogenesis and highlights FOXA1-regulated networks and their genetic determinants as candidates for future studies of biomarker development and therapeutic relevance. Future studies incorporating regulatory genomics into patient stratification may build on these findings to refine risk prediction and better understand FOXA1-associated disease trajectories in PCa and other FOXA1-driven malignancies.

Limitations of the study

This study has several limitations. First, FOXA1 DTGs were defined using a conservative promoter-proximal strategy integrated with FOXA1 perturbation transcriptomes, which increased confidence in target assignment but did not capture the full spectrum of distal enhancer-mediated FOXA1 regulation. Second, the DTG signatures were mainly derived from LNCaP and T47D models and validated in public cohorts; therefore, further evaluation in additional prostate and BCa states, as well as prospective uniformly treated cohorts, is needed to assess their generalizability and clinical utility. Third, allele-specific FOXA1 binding was experimentally tested only for selected variants because of model availability, and additional candidate variants should be validated in genotype-matched or engineered isogenic systems. Finally, although multiple independent datasets were integrated and normalized, residual platform-related batch effects and cohort-specific biases cannot be fully excluded.

Resource availability

Lead contact

Requests for further information and resources should be directed to Prof. Gong-Hong Wei (gonghong_wei@fudan.edu.cn).

Materials availability

This study did not generate new unique materials.

Data and code availability

• •This paper analyzes existing, publicly available data. RNA-seq and microarray datasets, including cBioPortal: TCGA, CPGEA,⁴³ DKFZ,⁷² Metabric,⁷³ ENA: PRJNA678636,²⁷ PRJNA490188,²³ PRJNA844574,²⁸ PRJNA732359,¹² PRJNA1023495,⁷⁴ GEO: GSE133626,⁴² GSE58135,⁴¹ GSE225846,⁴⁰ GSE214846,⁷⁵ GSE77509,⁷⁶ GSE15654,⁷⁷ and GSE2034,⁷⁸ were retrieved from cBioPortal for Cancer Genomics^79,80 ENA or the GEO database.^81,82 ChIP-seq profiling data for FOXA1, AR, E2F1, ERG, NKX3.1, MYC, HOXB13, GATA2, TEAD1, and CTCF, as well as ATAC-seq data, were obtained from the Cistrome Data Browser.⁸³
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This work was funded by the Noncommunicable Chronic Diseases – the National Science and Technology Major Project (2026ZD0553304), the Shenzhen Medical Research Fund (SMRF) (C2503001), the Shanghai Interactional Collaborative Project (23410713300), the National Natural Science Foundation of China (82311530050; 82073082), the China Scholarship Council (CSC No. 201807040052), Key Research and Development Program of Anhui Province (2022i01020023), Jane ja Aatos Erkon säätiö, Sigrid Juséliuksen Säätiö, Syöpäjärjestöt, the Oulu University Scholarship Fund (grant no. 20230047), University of Oulu and Research Council of Finland Profi8 funding (decision no. 365202), Finnish Cultural Foundation (grant no. 00250994), and the Ida Montinin Säätiö (grant no. 20240007).

We gratefully acknowledge the use of Linux High-Performance Computing resources provided by CSC – IT Center for Science, the Medical Research Data Center in Shanghai Medical College of Fudan University, and the high-performance Computing Platform of Suzhou Institute of Systems Medicine, Chinese Academy of Medical Sciences and Peking Union Medical College.

Author contributions

G.-H.W. designed and conceptualized the work. B.L. performed most of the experiments and analyzed the data with significant assistance from X.Y. and Q.Z. carried out and interpreted all bioinformatics analyses. B.L., X.Y., Z.T., L.Q., Y.Y., R.J., and N.G. were involved in methodology; B.L., X.Y., and Q.Z. performed validation and formal analysis; Q.Z., A.M., and G.-H.W. performed investigation; B.L., Q.Z., and G.-H.W. drafted the original manuscript. B.L., Q.Z., and G.-H.W. reviewed and edited the manuscript with inputs from all authors; A.M. and G.-H.W. were involved in supervision; G.-H.W. provided resources and performed funding acquisition. All authors reviewed the results and approved the final version of the manuscript.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-FOXA1 antibody	Abcam	Cat# ab23738; RRID:AB_2104842
Rabbit IgG	Santa Cruz Biotechnology	Cat#sc-2027X; RRID:AB_737197
Bacterial and virus strains
lentiCas9-Blast	Addgene	Cat#52962
lentiGuide-Puro	Addgene	Cat#52963
pLKO.1-puro	Addgene	Cat#10879
pLKO.1-FOXA1-shRNA1-puro	Functional Genomics Unit (University of Helsinki)	TRCN0000020845
pLKO.1-FOXA1-shRNA2-puro	Functional Genomics Unit (University of Helsinki)	TRCN0000020846
pLKO.1-USP39-shRNA1-puro	Functional Genomics Unit (University of Helsinki)	TRCN0000017420
pLKO.1-USP39-shRNA2-puro	Functional Genomics Unit (University of Helsinki)	TRCN0000017422
pLKO.1-TLE4-shRNA1-puro	Functional Genomics Unit (University of Helsinki)	TRCN0000158555
pLKO.1-TLE4-shRNA2-puro	Functional Genomics Unit (University of Helsinki)	TRCN0000161885
pLKO.1-CPNE1-shRNA1-puro	Functional Genomics Unit (University of Helsinki)	TRCN0000007029
pLKO.1-CPNE1-shRNA2-puro	Functional Genomics Unit (University of Helsinki)	TRCN0000007032
Chemicals, peptides, and recombinant proteins
DMEM	Invitrogen	Cat#31966021
Low glucose DMEM	Invitrogen	Cat#21885025
RPMI1640	Merck	Cat#R8758
F-12K Medium	ATCC	ATCC 30-2004
Fetal bovine serum (FBS)	Gibco	10270-106
Dihydrotestosterone	Merck	Cat#D-073-1ML
SYBR Select Master Mix	Applied Biosystems	Cat#4472908
Lipofectamine 2000	Thermo Fisher Scientific	Cat#11668030
cOmplete™, Mini, EDTA-free Protease Inhibitor Cocktail	Roche	Cat#04693159001
Dynabead protein G	Invitrogen	Cat#10004D
Polybrene	Merck	Cat#H9268
Puromycin	Merck	Cat#P9620
Penicillin–streptomycin	Life Technologies	15140130
Triton X-100	Sigma	X100-500
Trypsin-EDTA	Gibco	25200-56
Critical commercial assays
MinElute PCR Purification Kit	QIAGEN	Cat#28006
RNeasy Mini Kit	QIAGEN	Cat#74106
High-Capacity cDNA Reverse Transcription Kit	Applied Biosystems	Cat#4368814
Cell Proliferation Kit II	Roche	Cat#11465015001
Deposited data
Homo sapiens reference genome (GRCh38)	Genome Reference Consortium	https://www.ncbi.nlm.nih.gov/grc/human
TCGA	Firehose Legacy	https://www.cbioportal.org/
CPGEA	Li et al.43	https://www.nature.com/articles/s41586-020-2135-x
DKFZ	Gerhauser et al.72	https://www.cbioportal.org/
Metabric	Curtis et al.73	https://www.cbioportal.org/
PRJNA678636	Park et al.27	https://www.ebi.ac.uk/ena/browser/view/PRJNA678636
PRJNA490188	Song et al.23	https://www.ebi.ac.uk/ena/browser/view/PRJNA490188
PRJNA844574	Anstine et al.28	https://www.ebi.ac.uk/ena/browser/view/PRJNA844574
PRJNA732359	Fu et al.12	https://www.ebi.ac.uk/ena/browser/view/PRJNA732359
PRJNA1023495	Nakagawa et al.74	https://www.ebi.ac.uk/ena/browser/view/PRJNA1023495
GSE133626	Kumar et al.42	https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE133626
GSE58135	Varley et al.41	https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE58135
GSE225846	Tang et al.40	https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE225846
GSE214846	Long et al.75	https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE214846
GSE77509	Yang et al.76	https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE77509
GSE15654	Hoshida et al.77	https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE15654
GSE2034	Wang et al.78	https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE2034
Cistrome Data Browser	Zheng et al.83	https://cistrome.org/db/#/
Experimental models: Cell lines
293T Cells	ATCC	Cat#CRL-11268; RRID:CVCL_1926
LNCaP	ATCC	Cat#CRL-1740; RRID:CVCL_1379
22Rv1	ATCC	Cat#CRL-2505; RRID:CVCL_1045
PC3	ATCC	CRL-1435; RRID:CVCL_0035
V16A	Derived from LNCaP	N/A
Oligonucleotides
Primers for RT-qPCR, see Table S7	This paper	N/A
Primers for allele specific ChIP-qPCR, see Table S8	This paper	N/A
Primers for ChIP-qPCR, see Table S9	This paper	N/A
Oligos for CRISPR sgRNAs, see Table S10	This paper	N/A
Software and algorithms
BEDTools version 2.30.0	Quinlan et al.84	https://bedtools.readthedocs.io/
Bowtie2 version 2.2.5	Langmead et al.85	https://bowtie-bio.sourceforge.net/bowtie2
ChIPseeker version 1.33.1	Yu et al.86	https://bioconductor.org/packages/release/bioc/html/ChIPseeker.html
Circos version 0.69-6	Krzywinski et al.87	https://circos.ca/
ComplexHeatmap version 2.6.2	Gu et al.88	https://bioconductor.org/packages/release/bioc/html/ComplexHeatmap.html
deepTools version 3.5.2	Ramírez et al.89	https://deeptools.readthedocs.io/
EEL version 1.1	Hallikas et al.52	https://www.cs.helsinki.fi/u/kpalin/EEL/
FastQC version 0.11.9	Bioinformatics Group at the Babraham Institute	https://www.bioinformatics.babraham.ac.uk/projects/fastqc/
ggplot2 version 3.4.4	CRAN	https://cran.r-project.org/web/packages/ggplot2
HOMER version 4.11.1	Heinz et al.90	http://homer.ucsd.edu/homer/
IGV version 2.12.3	Thorvaldsdóttir et al.91	https://igv.org/
MACS2 version 2.1.4	Zhang et al.92	https://pypi.org/project/MACS2/
MultiQC version 1.13.dev0	Ewels et al.93	https://multiqc.info/
R version 4.0.5	R	https://www.r-project.org/
SAMtools version 1.9	Danecek et al.94	https://www.htslib.org/
survival version 3.2.10	CRAN	https://cran.r-project.org/web/packages/survival
survivalROC version 1.0.3.1	CRAN	https://cran.r-project.org/web/packages/survivalROC/index.html
GSEA version 4.3.2	Subramanian et al.95	https://www.gsea-msigdb.org/gsea/index.jsp
Other
Bio-Spin® Chromatography Columns	Bio-Rad	732-6008
C18 macrospin columns	Nest Group	Cat# SMM SS18V
cBioPortal	Cerami et al.79	https://www.cbioportal.org/
CisBP	Weirauch et al.96	https://cisbp.ccbr.utoronto.ca/
Countess cell counting chamber slides	Thermo	Cat#C10283
Database for Annotation, Visualization and Integrated Discovery (DAVID)	Sherman et al.97	https://david.ncifcrf.gov/
Evosep One	Evosep	EV-1000
GDC Data Portal	The Cancer Genome Atlas Program	https://portal.gdc.cancer.gov/
GTEx	GTEx Consortium	https://gtexportal.org/home/
GWAS Catalog	Sollis et al.98	https://www.ebi.ac.uk/gwas/
Hybrid trapped ion mobility quadrupole TOF mass spectrometer	Bruker	TimsTOF Pro
IntAct Molecular Interaction Database	Del Toro et al.99	https://www.ebi.ac.uk/intact/home
JASPAR	Castro-Mondragon et al.100	https://jaspar.elixir.no/
Metascape	Zhou et al.101	https://metascape.org
Network Data Exchange	Pillich et al.102	https://home.ndexbio.org/index/
PancanQTL	Gong et al.56	http://gong_lab.hzau.edu.cn/PancanQTL/
ProHits-viz	Liu et al.103	https://prohits-viz.org
SCENIC+	Bravo González-Blas et al.104	https://resources.aertslab.org/cistarget/motif_collections/
Spacing pipeline (TF pairs calling)	Shen et al.105	https://github.com/zeyang-shen/spacing_pipeline
UCSC Toil RNAseq Recompute Compendium	Vivian et al.106	https://xena.ucsc.edu/

Experimental model and study participant details

Human participants

This study did not recruit new human participants or collect new human specimens. All human transcriptomic and clinical data analyzed in this study were obtained from previously published, de-identified, publicly available datasets, including TCGA, cBioPortal, GEO, GTEx, and other publicly accessible cohorts described in the original publications. Ethical approval and informed consent were obtained by the investigators of the original studies according to their institutional and national regulations; therefore, no additional ethical approval or informed consent was required for this secondary analysis.

Information regarding age, developmental stage, sex, gender, ancestry, race, ethnicity, and other demographic or clinical variables was used as reported in the original datasets, when available.

Sample sizes for each cohort are reported in the corresponding results sections, figure legends, and original publications. No new allocation of participants into experimental groups was performed in this study. Instead, samples were stratified computationally according to predefined analytical criteria, including tumor versus normal tissue, clinical characteristics, genotype, or median gene-signature risk score, as described in the corresponding STAR Methods sections.

This study investigated prostate cancer and breast cancer independently using publicly available datasets. The prostate cancer cohorts consisted of male participants, whereas the breast cancer cohorts consisted of female participants, reflecting the underlying disease populations. Therefore, sex was not included as an experimental variable or covariate in the analyses. The conclusions of this study are specific to each cancer type and should not be interpreted as comparisons between sexes.

Cell lines

The study used the 22RV1, LNCaP, VCaP, PC3, V16A and 293T cell lines, all cells are purchased from ATCC company, and the culture methods are followed as described by ATCC. Cell lines obtained from ATCC were authenticated by the supplier; no additional in-house STR authentication was performed after receipt. All cell lines used in this study were verified negative for mycoplasma contamination routinely using EZ-PCR Mycoplasma Detection Kit (20-700-20, Biological Industries). “One Shot™ Stbl3™ chemically competent Escherichia coli was used for plasmid propagation. Bacteria were maintained and expanded under standard laboratory conditions at 37°C using appropriate antibiotic selection when required. No animal models, primary cell cultures, plant models, or prospectively collected human biospecimens were used in this study.

Method details

Cell culture

The cell lines used in our project—22Rv1 (CRL-2505, ATCC), LNCaP (CRL-1740, ATCC), VCaP (CRL-2876, ATCC), PC3 (CRL-1435, ATCC), V16A, and 293T (CRL-11268, ATCC) were originally purchased from the American Type Culture Collection (ATCC). LNCaP and 22Rv1 cells were cultured in RPMI 1640 (R8758, Sigma) with 10% fetal bovine serum (F7524, Sigma), supplemented with 1% penicillin and streptomycin antibiotics (15140122, Gibco). The cells were grown at 37°C with 5% CO2.

To induce AR activity, 1 nM of the synthetic androgen methyltrienolone (R1881; dissolved in ethanol, from O.A. Jänne, University of Helsinki) or 100 nM dihydrotestosterone (DHT; dissolved in ethanol, from O.A. Jänne) was applied to 22Rv1 cells for 24 hours, as indicated in the related experiments.

Lentiviral constructs, lentivirus production and infection

Two sets of shRNA constructs in the pLKO.1-puro vector targeting FOXA1 and USP39 were purchased from the Functional Genomics Unit of the University of Helsinki for knockdown assays. shRNA against TLE4 was designed using BLOCK-iT RNAi Designer (Thermo Fisher Scientific) and cloned into the pLKO.1 vector (Addgene).

All shRNA constructs in the pLKO.1-puro vector targeting FOXA1, USP39, TLE4, and CPN-1 from the TRC1 shRNA library were also purchased from the Functional Genomics Unit of the University of Helsinki. The two most efficient shRNAs were selected for further testing from a set of five shRNAs against each gene (TRCN0000020845 and TRCN0000020846 against FOXA1; TRCN0000017420 and TRCN0000017422 against USP39).

HEK 293T cells (ATCC, CRL-11268) were seeded the previous day into 6-well plates at 70%–80% confluency. Lentiviral constructs were produced using the third-generation packaging system (1:1:1:3 ratio in a total amount of 10 μg, consisting of pVSVG-envelope plasmid, pMDLg/pRRE-packaging plasmid, pRSV-Rev-packaging plasmid, and lentiviral transfer vector) and Lipofectamine 2000 (11668019, Thermo Fisher Scientific) diluted in Opti-MEM according to the manufacturer's instructions. Two milliliters of fresh low-glucose DMEM GlutaMAX (21885025, Thermo Fisher Scientific) containing 10% fetal bovine serum (F7524, Sigma) and 0.1% penicillin and streptomycin (15140122, Gibco) were added. Over the next three days, virus-containing medium was collected every 24 hours, centrifuged for five minutes at 95 g, and then filtered using a 0.45 μm syringe filter unit. Samples were collected, frozen in liquid nitrogen, and stored at -80°C.

Target cells were seeded in a 6-well plate 16–18 hours before infection and grown to 70%–80% confluency. After removing the cell culture medium, the cells were transduced with lentivirus and polybrene at a final concentration of 8 μg/ml. Following a 24-hour incubation at 37°C and 5% CO2, the lentivirus-containing medium was replaced with pre-warmed fresh medium. After 48 hours of transduction, fresh medium containing puromycin was added at a final concentration of 2 μg/ml. Non-infected cells were used as control cells, and selection status was determined by the death of all control cells. Surviving cells were split and maintained at the same puromycin concentration. After three days, cells were collected for RNA or cell lysate preparation, followed by RT-qPCR or protein blotting to confirm shRNA knockdown efficiency. The two most efficient shRNA lentivirus vectors were selected for further analysis, including ChIP-qPCR, proliferation, and invasion assays.

Plasmids and gene cloning

TLE4 shRNA Oligos for pLKO.1 was designed by Sigma-Aldrich Mission shRNA (https://www.sigmaaldrich.com/FI/en/semiconfigurators/shrna?activeLink=productSearch). Primer sequences, cloning methods, and enzymes are shown in Table S6.

Transient transfections

We tested the individual set of four siRNAs (Qiagen) against FOXA1 by qRT-PCR. The two most effective single siRNAs (SI04154010 and SI04311888, Qiagen) against FOXA1 was chosen for further experiments. For siRNA transfection in 6-well plates, 3x10⁵ 22Rv1 cells per well were subjected to reverse transfection with a final concentration of 50nM FOXA1 or negative-control siRNA (Ctrl_Allstars_1, Qiagen) with HiPerFect Transfection Reagent (301705, Qiagen) according to the manufacturer's instructions.

Quantitative RT-PCR

A High-fidelity cDNA Reverse Transcription Kit (Life Technologies) was used to reverse transcribe total RNA from cultured cell lines into cDNA using the RNeasy Mini kit (Qiagen) according to the manufacturer's instructions. In triplicate or with further replications, genomic DNA and cDNA samples from ChIP or FAIRE analyses were quantified by RT-PCR and were normalized against an endogenous ACTB (β-actin) control. An Applied Biosystems Stratagene MX3005P machine was used to conduct quantitative RT-PCR using Power SYBR Green PCR Master. Mix or SYBR Select Master Mix. A minimum of three pairs of primers were designed for each target and tested for specificity and efficiency. RT-PCR quantification of the target was only achieved with primer pairs that were highly specific and amplification efficient. Primers are listed in Table S7.

Allele-specific quantitative RT-PCR

AS-qPCR was performed as previous instruction.¹⁸ In brief, the primers for allele-specific amplification of different alleles of FOXA1 SNPs (rs1417078, rs1887414, rs4911493, rs5832649, rs6735074, rs6740722, rs7153397, rs11568818, rs13069553, rs67026445) in the DNA samples from ChIP were designed as listed in Table S8.

Cell viability and proliferation assays

22Rv1 cells (2 x10⁴ per well) that were infected with lentiviral particles encoding FOXA1, USP39, CPNE1 or TLE4 shRNA were included in the cell proliferation assays. The experiments were performed on the 96-well plates with each well containing 100 μl medium of cells. The viability and proliferation of cells were assessed using XTT assays (Roche) at the specified time points by measuring the absorbance at 450 nm in accordance with the manufacturer's instructions. A total of three replicate wells were used for each treatment and time point. Results are representative of three independent experiments and differences. Significances were calculated by two-tailed t test.

Wound healing assays

96well-ImageLock plates were seeded with cells in appropriate culture medium to allow growth near 100% confluence. Following that, homogenous scratch wounds were created using WoundMaker and cells were washed twice with PBS. IncuCyte Live-Cell Imaging System from Essen BioScience was used to image the wounds of each well every 2 hours for a maximum of 180 hours. With ImageJ, the wound areas of each well were analyzed.

Chromatin immunoprecipitation (ChIP)

ChIP assay was performed for examining the endogenous interaction of FOXA1 in 22Rv1 with DHT treatment. The reaction was quenched with 125 mM glycine after 10 min of cross-linking with 1% formaldehyde at room temperature with slight shake. Hypotonic lysis buffer (20 mM Tris-HCl, pH 8.0, with 10 mM KCl, 10% glycerol, 2 mM DTT and complete protease inhibitor cocktail (Sigma)) was used to resuspend collected cells after being washed twice with cold PBS. Rotate in high speed for 30-50 min in cold room and spin to pellet cell nuclei 5 min at 2000g at 4°C. After washing with cold PBS, the isolated pellet of nuclei was resuspended in SDS lysis buffer (50 mM Tris-HCl, pH 8.1, 0.5% SDS, 10 mM EDTA, and Complete Protease Inhibitor (Roche)). The chromatin lysate has been cleared by centrifugation and diluted with 10 volumes of ChIP dilution buffer (16.7 mM Tris-HCl, pH 8.1, with 0.01% SDS, 1.1% Triton X-100, 1.2 mM EDTA, 167 mM NaCl and Complete Protease Inhibitor). The Q800R sonicator (QSonica) was used to sonicate nuclear extracts to generate chromatin fragments with an average size of 0.25-0.5 kb. Centrifuge for 30 min at 4°C and keep the supernatant, 70 μl protein-G Magnetic Beads were washed twice with blocking buffer (0.5% BSA in IP buffer) were used to incubate the supernatant overnight at 4°C. 5 μg of antibody (5 μg: rabbit monoclonal IgG (normal rabbit IgG, sc-2027, Santa Cruz Biotechnology) and rabbit monoclonal anti-FOXA1(ab23738, abcam)) was subjected to the supernatant with incubation at 4°C overnight. Afterwards, the beads-antibody complex was washed one time with RIPA Washing buffer (50 mM HEPES, pH 7.6, 1 mM EDTA, 0.7% (wt/vol) sodium deoxycholate, 1% (vol/vol) NP-40, 0.5M LiCl) 5 times, followed by two times of washing with AMBIC (100mM ammonium hydrogen carbonate solution) 2 times. The DNA-protein complex was precipitated in Dynabead protein G (10009D, Thermo Fisher Science), and was eluted beads in 50 μl of 2 x in extraction buffer (10ul 10%SDS+90 ulTE), and then was reversed cross-linked with Proteinase K (AM2548, Thermo Fisher Scientific) with final concentration 1 mg/ml and NaCl with final concentration 0.3 M for 16 h at 65°C in 1000 rpm. The DNA was purified using MinElute PCR Purification Kit (28006, QIAGEN), followed by ChIP-qPCR using primers that target DNA binding sequences in the genome. To prepare a ChIP library, a manufacturer's protocol was followed (Illumina TruSeq Sample Preparation Best Practices and Troubleshooting Guide). The samples were then sequenced and analyzed. Primers for ChIP-qPCR are listed in Table S9.

CRISPR/Cas9-mediated genome editing analysis

CRISPR design tool (https://benchling.com/ or http://chopchop.cbu.uib.no/) was applied to design two pairs of the oligos (sgRNA-top and sgRNA-bottom) knocking out FOXA1 SNPs using the dual vector lentiviral GeCKO strategy from the Zhang laboratory (Sanjana et al., 2014). lentiCas9-Blast (#52962, addgene) was used to produce lentiviral medium with the third-generation packaging system and further infect 22Rv1 cells as described before. The 22Rv1 cell had been transduced with lentiCas9-Blast lentiviral medium and selected for 6 days with 5 ug/ml blasticidin (Life Technologies). Cas9-expressing 22Rv1 stable cell lines were constructed.

At 37°C for 30 min and 95°C for 5 minutes in a thermocycler, phosphorylated oligos were annealed, then ramped down to 25°C at 5°C/min until 25°C was reached. Then, the annealed oligos were inserted into lentiGuide-Puro (#52963, addgene) plasmids, and the constructed plasmids were subjected to produce lentiviral medium with the third-generation packaging system, followed by infecting 70%-80% confluency 22Rv1 Cas9-expressing cells and further for puromycin screening for 5 days with 2 ug/ml puromycin (Life Technologies). After non-transfected cells died, dilution or FACS were used to isolate the successfully transfected cells. Genotyping and RT-qPCR were used to determine TLE4 gene expression in the positive clones after 2-3 weeks of seeding. The SNPs knocking out mixed clones were used to detect their eQTL genes expression level with RT-qPCR. CRISPR sgRNA oligos are listed in Table S10.

ChIP-seq analysis

All ChIP-seq data utilized in this study was obtained from publicly available resources. Raw sequence reads were first assessed with FastQC, followed by adapter and quality trimming using Trimmomatic¹⁰⁷; trimmed reads were re-evaluated with FastQC. Cleaned reads were aligned to the human hg38 genome using Bowtie2.⁸⁵ PCR duplicates were removed using sambamba markdup, and uniquely mapped reads were retained by filtering for primary alignments with mapping quality ≥30. Peaks were then called with MACS2⁹² with narrow peaks for FOXA1 and broad peaks for histone marks. Peak annotation was performed using the Bioconductor package ChIPseeker (v.1.34.1).⁸⁶ DeepTools (v.3.5.1)⁸⁹ was employed to generate heatmaps of ChIP-seq signal intensity within a ±3 kb window centered on the TSSs of DTGs.

RNA-sequencing and differential gene expression analysis

RNA-seq raw sequence data of FOXA1 utilized in this study were downloaded from public resources. Reads were initially evaluated using FastQC to assess read quality, followed by Trimmomatic for trimming or removing low-quality reads and adapters.¹⁰⁷ The resulting high-quality reads were aligned to the hg38 reference genome using STAR¹⁰⁸ with default settings. Read quantification was performed using HTSeq,¹⁰⁹ along with the primary assembly annotation file v39 from ENCODE. Differential expression analysis was conducted using the Bioconductor package DESeq2.¹¹⁰ Prior to analysis, genes with very low expression (cumulative read count < 2 across samples) were excluded. Significantly differentially expressed genes were identified with a false discovery rate (FDR) threshold of <0.05. The Variance Stabilizing Transformation (VST) method from DESeq2 was utilized to normalize gene expression values. RNA-seq heatmaps were generated using the R package “pheatmap” v.1.0.12.

Identification of eQTL eGenes

To identify eQTL genes relevant to PCa and BCa, we implemented a multistep analytical workflow. GWAS-identified lead SNPs associated with PCa and BCa risk were first compiled and expanded using linkage disequilibrium (LD) information from the 1000 Genomes Project. SNPs in substantial LD with the GWAS lead variants (r² ≥ 0.5) were retained as LD proxy SNPs. These LD-expanded SNP sets were then queried across multiple established eQTL resources, including GTEx,⁵⁵ PancanQTL,⁵⁶ and ncRNA-eQTL,⁵⁷ in addition to in-house patient cohorts containing paired genotype and transcriptomic data. SNPs exhibiting statistically significant associations with gene expression were classified as eQTL SNPs, and their corresponding genes were designated as eGenes; variants lacking significant expression associations were categorized as non-eQTL SNPs. Finally, we assessed whether eQTL SNPs were enriched within FOXA1 binding peaks and identified FOXA1-associated eQTL SNPs and their corresponding eGenes.

Development of the FOXA1 DTG signature

To develop the FOXA1 DTG signature, we integrated transcriptomic and epigenomic datasets from three cancer types: liver, prostate, and breast. Publicly available RNA-seq datasets were obtained, as specified in the manuscript. Corresponding FOXA1 ChIP-seq datasets for these cell lines were used to assess genomic FOXA1 binding. DTGs were identified by intersecting DEGs with regions of FOXA1 ChIP-seq signal coverage. A gene was classified as a direct FOXA1 target if a FOXA1 binding peak was located within a window extending from −1000 bp upstream to +100 bp downstream of the TSS.

Construction of the FOXA1 prognostic gene signature

To construct the FOXA1 prognostic gene signature, we first retained FOXA1-regulated genes meeting the criteria of differential expression (FDR < 0.05, |log₂FC| > 2) and exhibiting FOXA1 ChIP-seq binding signals within a window of −1000 bp to +100 bp relative to the TSS. The expression profiles of these candidate genes were then evaluated in tumor and normal tissues across three independent cohorts representing prostate, breast, and liver cancers. Genes that were consistently upregulated or downregulated across tumor types, using a threshold of p < 0.01 and |log₂FC| > 1, were retained for prognostic modeling. Patients with at least one month of follow-up were included in survival analyses. Prognostic relevance of candidate genes was first assessed using univariate Cox proportional hazards regression (p < 0.05) across both TCGA and independent clinical cohorts. The TCGA cohorts for PCa, BCa, and liver cancer served as training datasets. LASSO-penalized Cox regression was employed to perform feature selection and reduce model overfitting, identifying a subset of genes with the strongest predictive value for biochemical recurrence. A risk score model was constructed using the formula:

$$\text{Risk}\text{Score}=\mathrm{\Sigma }(\mathrm{\beta }\text{of}\text{gene}\_\mathrm{i}\times \text{Expression}\text{of}\text{gene}\_\mathrm{i})$$

where β coefficients were derived from the LASSO model, and expression values for genes with multiple probes were averaged. The resulting gene signature was subsequently validated in additional independent cohorts.

ROC curve and prognostic impact of the gene signature

Within the respective cohorts of PCa, BCa and liver cancer, patient-specific risk scores were computed based on the validated FOXA1 gene signature. Prognostic evaluation included time-dependent ROC curve analysis, performed using the survivalROC package (v1.0.3.1), to assess the model’s diagnostic accuracy. Kaplan–Meier survival analysis, coupled with log-rank testing (p < 0.05), was used to evaluate the effectiveness of risk stratification. In accordance with previous methodology, patients were dichotomized into high- and low-risk subgroups using the median risk score as the threshold.

Independent prognostic role of the gene signature

Univariate Cox proportional hazards modeling was first employed to evaluate the independent prognostic utility of the gene signature, with patients dichotomized into high- and low-risk subgroups based on median risk score thresholds. This initial analysis incorporated both molecular (gene signature) and clinicopathological confounders, including demographic variables (race, gender, age), tumor histology (grade, TNM staging), and metastatic status, to quantify their individual associations with survival outcomes. Covariates demonstrating statistical significance (p < 0.05) in univariate screening were subsequently integrated into a multivariate Cox regression framework, enabling the identification of robust prognostic factors while adjusting for competing variables. The hierarchical analytical design prioritized both biological relevance and statistical rigor, systematically isolating the signature's predictive capacity from established clinical determinants.

Differentiating performance of the prognostic signature

Nonparametric comparative analyses were systematically performed to assess distributional differences in risk scores. Boxplots, in conjunction with the Mann–Whitney U test, were used to evaluate baseline differences between normal tissues and neoplastic lesions. For multigroup comparisons involving tumor progression parameters, such as T stage, histologic grade, N stage, and M stage, the Kruskal–Wallis H test was applied to account for ordinal or categorical variables with three or more subgroups. A P-value of less than 0.05 was considered indicative of statistical significance in all analyses.

Gene set enrichment analysis

GSEA v.4.3.2⁹⁵ was implemented to mechanistically interrogate the biological pathways underpinning the prognostic gene signature’s functional relevance. Utilizing the MSigDB²⁹ as the annotation framework, a pre-ranked analytical strategy was executed where genes were ordered by their differential expression magnitude (signal-to-noise ratio, descending). Gene set permutations were performed under default parameters, including collapse mode for probe-to-gene symbol conversion and weighted enrichment statistic calculations. A p value of less than 0.05 was considered statistically significant for the analysis.

Survival analysis

Comprehensive survival modeling was implemented to quantify the prognostic influence of FOXA1 DTGs composite signature, FOXA1, and FOXA1 eQTL genes across multiple independent cohorts, specifically in pan-cancer and PCa cohorts. The analyses were performed using the R package “Survival” (v.3.5-7), and the results were illustrated through Kaplan-Meier curves. Participants were stratified based on the median levels of gene expression or the risk scores derived from the gene signatures. Initially, the “Surv” function was employed to construct survival models by providing time-to-event and event status data from clinical datasets. Subsequently, gene expression levels or risk scores were introduced into these models via the “survfit” function. In the final step, the Cox proportional-hazards framework was utilized to compute hazard ratios, assessing the relationship between patient survival durations and either gene expression levels or risk scores.

Genome-wide prediction of mammalian enhancers based on analysis of transcription-factor binding affinity

To determine whether identified SNPs intersect with TF binding sites relevant to PCa, we assembled cistromes for key TFs implicated in PCa biology, including FOXA1, AR, HOXB13, MYC, ERG, GATA2, NKX3-1, E2F1, TEAD1, and CTCF. Publicly available ChIP-seq datasets for these TFs were downloaded from Cistrome Data Browser.⁸³ To create unified cistromes, ChIP-seq peak regions (BED files) from various experiments for each TF were merged using Bedtools (v2.27.1).⁸⁴

eQTL analysis

The associations between SNP genotypes and the expression levels of USP39, TLE4, and CPNE1 were assessed through eQTL analysis using the ‘MatrixEQTL’ R package across TCGA, CPGEA, GTEx, Stockholm and TCGA cohorts. The eQTL analysis was applied by fitting a linear regression model between the expression and the genotype data, other parameters were left as default (pvOutputThreshold = 0.05, errorCovariance = numeric ()”). The transcriptional profiling in TCGA cohort was assessed by RNA-Seq. The TCGA cohort was genotyped on Affymetrix SNP array 6.

Comparative prognostic benchmarking of FOXA1 DTG signatures

The prognostic performance of the FOXA1 DTG signature was evaluated in TCGA prostate cancer, MSKCC prostate cancer, TCGA breast cancer, and METABRIC breast cancer cohorts. Samples with missing outcome information, non-positive follow-up time, or missing FOXA1 DTG signature score were excluded from the analysis. Cox proportional hazards regression models were used to compare the FOXA1 DTG signature with the CCP score and cohort-specific clinical variables. In PCa cohorts, clinical variables included Gleason score, tumor stage, lymph node stage, and/or PSA, depending on data availability. In BCa cohorts, clinical variables included tumor stage, lymph node stage, HER2 status, ER status, PR status, and/or the number of positive lymph nodes, depending on data availability. Incremental prognostic value was assessed using likelihood ratio tests between nested Cox models. Specifically, FOXA1 DTG signature-only and CCP-only models were compared with the combined FOXA1 DTG signature plus CCP model. Clinical-only models were compared with models adding either the FOXA1 DTG signature or CCP to evaluate whether each molecular signature improved prognostic performance beyond clinical covariates. Model discrimination was evaluated using Harrell’s concordance index. All analyses were performed in R (v.4.4.1) using the survival (v.3.8-3), dplyr (v.1.1.4), rms (v.8.1-0), riskRegression (v.2025.09.17), pec (v.2025.06.24), and tidyr (v.1.3.2) packages.

Quantification and statistical analysis

Statistical analyses were performed using RStudio v. 1.4.1106 and R version 4.1.0, unless otherwise noted. Statistical tests applied across normal tissues, primary and metastatic tumors were examined by the Mann–Whitney U test for comparing gene expression levels between two groups, and the Kruskal–Wallis H test for comparisons involving three or more groups. All survival analyses were conducted using the R package 'Survival', and Kaplan-Meier curves were evaluated using the log-rank test. For microarray-based expression profiling, gene probes were selected based on the lowest P-values. Circos plots were created using the Circos software (v.0.67).⁸⁷ ChIP-Seq heatmaps were generated using deepTools. Statistical thresholds followed oncology convention: ∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001, with all inferences two-tailed unless cohort size necessitated exact permutation tests.