Mapping the FOXA1 Interactome in ER+ Breast Cancer Cells using Proximity Labeling Reveals Novel Interactions with the Orphan Nuclear Receptor NR2C2

Abstract

FOXA1 is a pioneer transcription factor essential for chromatin accessibility and transcriptional regulation in hormone-driven cancers. In breast cancer, FOXA1 plays a central role in facilitating nuclear receptor binding, reprogramming enhancer landscapes, and promoting transcriptional changes associated with therapy resistance. While FOXA1’s function has been primarily studied in the context of estrogen receptor-α (ER), its broader protein interaction network remains incompletely defined. Here, we systematically map FOXA1-interacting proteins in ER-positive breast cancer cells using proximity-dependent biotin labeling (miniTurbo) combined with quantitative LC-MS/MS proteomics. We engineered MCF-7 cell lines stably expressing miniTurbo-tagged FOXA1 at either the N-terminus or C-terminus to ensure comprehensive coverage of interaction interfaces. This approach recovered known FOXA1 partners, including AR, MLL3, YAP1, and GATA3, and identified 157 previously unreported FOXA1 interactors. Notably, 42 of these novel partners, including NR2C2, were significantly associated with poor relapse-free survival in ER+ breast cancer patients. To demonstrate the utility of this resource, we characterized the FOXA1-NR2C2 interaction in depth. Integrating ChIP-seq and RNA-seq, we show that FOXA1 and NR2C2 co-occupy a subset of genomic regions and drive co-regulated transcriptional programs involved in tumor progression. Our study reveals an expanded FOXA1 interactome and new insights into its functional network in breast cancer, providing candidate proteins for further exploration as biomarkers or therapeutic targets.

INTRODUCTION

Breast cancer is the second leading cause of cancer mortality in women with nearly 75% of all breast tumors expressing the estrogen receptor alpha (ER) at the time of diagnosis ^1–3. Despite the therapeutic successes of endocrine therapies designed to target ER function, about 30–40% of tumors become resistant to endocrine therapy, either through de novo or acquired resistance, and still retain the expression of ER ^2–10. Notably, the majority of metastatic breast tumors arise from ER+ disease, underscoring the need to better understand the transcriptional and epigenetic mechanisms that support hormone signaling and therapy resistance ^8,11.

The pioneer transcription factor FOXA1 plays a pivotal role in shaping the chromatin landscape and facilitating DNA binding of nuclear receptors. In breast cancer, FOXA1 is essential for the majority of ER genomic interactions and its upregulation is associated with enhancer reprogramming and endocrine resistance ^12,13. Pioneer factors, including FOXA1, can bind condensed chromatin independently of other factors, displacing linker histones and enabling access for other transcriptional regulators ^14–16. FOXA1 is prototypical pioneer factor whose protein structure possess features of both linker histones and conventional transcription factors allowing it to displace linker histones in compacted nucleosomes and allow access to other transcription factors ^14,15. In hormone-dependent cancers, such as prostate and breast cancer, FOXA1 is critical for androgen and estrogen receptor action and indispensable for cell survival and proliferation ^15,17. Furthermore, FOXA1 is implicated in treatment resistance, as its upregulation and associated transcriptome are hallmarks of endocrine therapy resistance ^18,19.

Given its central regulatory role, FOXA1 represents a promising candidate for therapeutic targeting. However, the lack of well-defined ligand-binding pockets limits direct pharmacologic inhibition. Recent studies suggest that FOXA1 function is strongly influenced by its interactions with other chromatin and transcriptional regulators, including MLL3, YAP1, and GATA3 ^20–22. Thus, systematically defining FOXA1’s protein interactome may reveal novel co-regulators and therapeutic entry points in hormone-resistant breast cancer.

To map the FOXA1 interactome in ER-positive breast cancer, we employed proximity-dependent biotinylation using the miniTurbo biotin ligase ²³. Derived from the promiscuous biotin ligase mutant BirA-R118S, the 28-kD miniTurbo is missing the wild-type BirA N-terminal domain and has 13 novel mutations that lend to its smaller size and significantly faster biotinylation activity ²⁴. Unlike wild-type BirA found in E. coli, which creates biotinoyl-5’-AMP (bioAMP) and holds it in its active site until it can react with a specific biotin acceptor tag sequence, promiscuous BirA mutants release the bioAMP prematurely, creating a local pool of this highly reactive species ²⁵. This bioAMP then covalently bonds to nearby nucleophilic residues, such as lysine, within a labeling radius of ~10 nm ²⁶. The covalent bond between biotin and lysine residues on the labeled proteins, coupled with the strong affinity between streptavidin and biotin, allows for stringent washing steps and cleaner purification of interacting proteins for mass spectrometry analysis ^27,28. This method allows detection of both stable and transient protein-protein interactions without requiring crosslinking or fixation, offering an alternative to other proximity labeling assays.

Herein, we generated two MCF-7 breast cancer cell lines stably expressing miniTurbo fused to either the N- or C-terminus of FOXA1. Following biotin labeling and streptavidin-based enrichment, we identified 157 previously unreported FOXA1-interacting proteins by LC-MS/MS. Of these, 31 were significantly associated with poor relapse-free survival in ER+ breast cancer patients, including the orphan nuclear receptor NR2C2. To exemplify the utility of this proteomic resource, we investigated the FOXA1-NR2C2 interaction in depth. Through integrative ChIP-seq and RNA-seq analyses, we show that FOXA1 and NR2C2 co-occupy a subset of genomic loci to regulate transcriptional programs implicated in tumor progression. Together, our study expands the known FOXA1 interactome and provides mechanistic insight into its role in transcriptional regulation in breast cancer. These findings highlight potential targets for therapeutic intervention and contribute to a deeper understanding of pioneer factor-driven transcription programs in hormone-responsive tumors.

MATERIALS AND METHODS

Cell culture and reagents

MCF-7 (RRID:CVCL_0031) and T47D (RRID:CVCL_0553) breast cancer cells were acquired from the American Type Culture Collection (ATCC) were sequence-validated and maintained in Dulbecco’s Modified Eagle Medium (DMEM) (Gibco 11995065) + 10% FBS (SERADIGM 97068–085) + 1% penicillin/streptomycin (Corning MT3002CI) at 37°C under 5% CO₂ unless otherwise stated. Cells were tested and confirmed negative for mycoplasma every 6 months using the Universal Mycoplasma Detection Kit (ATCC 30–1012K). The cells were not passaged more than 20 times; new vials were thawed.

siRNA knockdowns

For FOXA1 knockdown experiments, the siRNA constructs used were: siNT (Dharmacon D-001206-14-05; siGENOME Non-Targeting Pool #2), siGAPD (Dharmacon D-001140-01-05; siGENOME GAPD Control siRNA), and siFOXA1 (sequence: GUGUAGACAUCCUCCGUAU[dT][dT]). MCF-7 cells were transfected with 40 nM siRNA using lipofectamine RNAiMAX (Invitrogen 13778030) for 48 hours following the manufacturer’s protocol. For NR2C2 knockdown experiments we used two siRNA oligos that target NR2C2 ordered from Sigma (SASI_Hs01_00148350) with the following sequences GCAUUUGAUACCUUAGCUA[dT][dT] and UAGCUAAGGUAUCAAAUGC[dT][dT]. The siRNA scrambled sequence is AUGUAUUGGCCUGUAUUAG.

Plasmid construction

The N-terminal construct (3X FLAG-miniTurbo-GGSGG-FOXA1) was prepared using FOXA1 DNA that was PCR-amplified from MCF-7 cells to contain HindIII and BsrGI cut sites at the 5’- and 3’-ends, respectively (primer sequences in 5’-3’ orientation: forward primer CTCAAGCTTATGTTAGGAACTGTGAAG; reverse primer CTCTGTACAGGAAGTGTTTAGGACG), and cloned with 3xFLAG-tagged miniTurbo into a Lenti-viral backbone containing blasticidin resistance. The C-terminal construct was synthesized by GeneWiz using the FOXA1 coding sequence (CDS) reference (NM_004496.5, minus the stop codon) followed by 3xHA-tagged miniTurbo at the C-terminus and cloned into the multiple cloning site of a lentiviral backbone using XbaI and BamHI restriction enzymes. Plasmids were verified using Sanger sequencing. Plasmid maps are found in Supplemental Figures. Plasmid sequences are available in Supplemental Files 1 & 2.

Viral packaging and MCF-7 cell line transduction

HEK293T (ATCC, CRL-3216) (RRID:CVCL_0063) cells were grown in Dulbecco’s Modified Eagle’s Medium (DMEM) (ATCC 30–2002) supplemented with 10% Fetal Bovine Serum (heat inactivated) (ATCC 30–2020) and 2mM L-glutamine (ATCC 30–2214) at 37°C and 5% CO₂. These HEK293T cells were used to package the N-terminal and C-terminal lentiviral backbone plasmids through use of the VSV-G envelope-expressing plasmid PMD2.G (Addgene plasmid #12259, RRID: Addgene_12259) and the lentiviral packaging plasmid psPAX2 (Addgene plasmid #12260, RRID:Addgene_12260). These plasmids were transfected into cells at a 4:2:3 ratio using lipid-based transfection Fugene 6 (Promega E269A) according to manufacturer’s guidelines. After 24 and 48 hours, the supernatant was carefully collected without disturbing the HEK293T cells, and filtered through a 0.45uM syringe filter. Virus was concentrated using the Lenti-X Concentrator (Takara Bio Cat # 631231) reagent per manufacturer’s instructions and resuspended in 1mL of target cell media.

The concentrated virus preparations were diluted 1:5 and 1:10 with DMEM, 10%FBS, and 10 μg /mL polybrene (Millipore TR1003G) and added to MCF-7 cells. Transduction efficiency was not assessed and instead a titration of cell densities (50% - 80% confluent) were used across various wells of a 6 well plate. Transduced cells were incubated for 72 hours followed by bulk selection using 5 μg/mL blasticidin until mock transduced control cells (cells transduced without virus) were dead (about 2 weeks changing selection media every 2–3 days). Surviving transduced MCF-7 cells were subsequently maintained and expanded in DMEM 10%FBS media supplemented with 1 μg /mL blasticidin. The efficiency of transduction in stable cell lines was approximated by comparing the relative amount of wild-type FOXA1protein vs modified FOXA1 in the transduced cell lines via western blots (Supp. Fig. 1 and 2).

Biotinylation assays and whole-cell lysate collection

FOXA1-miniTurbo biotinylation assays were performed according to Cho et al. (2020) Nat Protoc. ²³ with some modifications. Briefly, biological triplicates of un-transduced MCF-7 cells (controls) or MCF-7 cells stably expressing N-terminal or C-terminal FOXA1-miniTurbo were incubated with (controls + constructs) or without (constructs only) 0.5 mM biotin (Sigma B4501) for six hours at 37°C in 15-cm dishes. The biotin labeling reaction was stopped by removing each plate of cells from the 37°C incubator, placing on ice, and immediately rinsing/replacing all media with ice cold 1x PBS a total of five times prior to being collected in 1 mL cold 1X PBS supplemented with protease inhibitors [1X protease inhibitor cocktail (PIC, Sigma 11697498001) + 10 mM sodium butyrate (Sigma B5887) + 0.002 mM trichostatin-A (TSA, Sigma T1952–200UL) + 1X PhosSTOP (Roche 4906845001) + 1X phosphatase inhibitor cocktail (Sigma P5726)]. Cells were pelleted at 5000 x RPM at 4 °C for 5 minutes.

After pelleting cells, whole-cell lysates were collected using high-salt lysis buffer (50 mM Tris-HCl pH 8.0 + 500 mM sodium chloride (Sigma S7653) + 1% NP-40 + 0.5% sodium deoxycholate (Sigma 30970) + 0.1% SDS (Fisher BP1311–1)) supplemented with protease inhibitors as before with the addition of 50 U/sample benzonase (Sigma E8263). Lysates were incubated on ice for 30 minutes with frequent vortexing and then centrifuged at 17,000 x g for 10 minutes at 4°C to pellet cellular debris. Supernatants were collected and diluted 1:1 with dilution buffer [20 mM Tris-HCl pH 8.0 + 150 mM sodium chloride + 2 mM EDTA (Millipore 324506) + 0.5% Triton X-100 (Sigma T8787) + 1 mM dithiothreitol (DTT, Roche 10708984001)] supplemented with PIC, PhosSTOP, and TSA as before.

Streptavidin pulldowns

Pulldown reactions were performed according to Cho et al. (2020) Nat Protoc. ²³. Briefly, we combined 3 mg whole-cell lysate with 100 μL Pierce^™ Streptavidin Magnetic Beads slurry (Thermo Scientific^™ 88817) for an overnight incubation at 4°C on a rotator. Pulldown samples were washed two times with radio-immunoprecipitation assay (RIPA) buffer (50 mM Tris-HCl pH 8.0 + 150 mM sodium chloride + 0.1% SDS + 0.5% sodium deoxycholate + 1% Triton X-100 + 1x PIC), once each with 1 M potassium chloride (Sigma P9333), 0.1 M sodium bicarbonate (Sigma S5761), and 2 M urea (Sigma U5128) (in 10 mM Tris-HCl pH 8.0), then three times again with RIPA buffer followed by five times with 50 mM ammonium bicarbonate (ABC) (Fisher, 600–27-01).

MS sample preparation

Pulldown protein samples were subjected to on-bead trypsin digestion as previously described ²⁹. Briefly, after the last wash step of the pulldown, beads were resuspended in 50 μL of 50 mM ammonium bicarbonate, pH 8. On-bead digestion was performed by adding 1 μg trypsin and incubated with shaking overnight at 37°C. The following day, 1 ug trypsin was added to each sample and incubated with shaking at 37°C for three hours. Beads were pelleted and supernatants were transferred to fresh tubes. The beads were washed twice with 100 μL LC-MS grade water, and washes were combined with the original supernatants. Samples were acidified by adding trifluoroacetic acid to a final concentration of 2%, to pH ~2. Peptides were desalted using peptide desalting spin columns (Thermo), lyophilized, and stored at −80°C until further analysis.

LC-MS/MS

Samples were analyzed by LC-MS/MS using an Easy nLC 1200 coupled to a QExactive HF mass spectrometer (Thermo) using manufacturer’s guidelines. Samples were injected onto an Easy Spray PepMap C18 column (75 μm id × 25 cm, 2 μm particle size; Thermo) and separated over a two-hour method. The gradient for separation consisted of 5–48% mobile phase B at a 250 nL/min flow rate, where mobile phase A was 0.1% formic acid in water and mobile phase B consisted of 0.1% formic acid in 80% acetonitrile. The QExactive HF was operated in data-dependent mode where the 15 most intense precursors were selected for subsequent fragmentation. Resolution for the precursor scan (m/z 350–1700) was set to 60,000, while MS/MS scans resolution was set to 15,000. The normalized collision energy was set to 27% for HCD. Peptide match was set to preferred, and precursors with unknown charge or a charge state of 1 and ≥ 7 were excluded.

MS data analysis

Raw data files were queried against the reviewed human database (containing 20,396 entries, downloaded March 2021), appended with a contaminants database, using Andromeda within Enzyme specificity was set to trypsin, up to two missed cleavage sites were allowed, and methionine oxidation and N-terminus acetylation were set as variable modifications. A 1% false discovery rate (FDR) was used to filter all data. Match between runs was enabled (five-minute match time window, 20-minute alignment window), and a minimum of two unique peptides was required for label-free quantitation using the label-free quantification (LFQ) intensities.

Perseus was used for further processing ³⁰. Only proteins with >1 unique+razor peptide were used for LFQ analysis. Proteins with 50% missing values in experimental conditions were removed and missing values were imputed from normal distribution within Perseus. Log2 fold change (FC) ratios were calculated using the averaged Log2 LFQ intensities of each construct (+) biotin vs. negative controls (wild-type cells or construct (∡)biotin)), and Students t-tests were performed for each pairwise comparison, with p-values calculated. Proteins with significant p-values (<0.05) and Log2FC >1 were considered biological interactors.

Data visualization

Unless otherwise specified, plots in this manuscript were generated using R v4.2.3 ³¹. Cartoon and schematic diagrams were generated in Adobe Illustrator v28.0.

Determining unique MS hits

After filtering out known contaminants, the UNIPROT IDs of the proteins enriched in the N-terminal and C-terminal constructs following biotin labeling were compared to negative control samples to identify 185 commonly enriched (shared) protein hits. The UNIPROT IDs of these 185 hits were compared to those of all FOXA1 interactors found on the BioGRID database from theBioGRID.org (v4.4.217, accession date 19 Jan 2023) ³² to determine whether proteins were previously associated with FOXA1 or unique to this study.

Protein classification and pathway analyses

Protein classifications were assigned using the PANTHER Classification System (v17.0, accession date 25 Jan 2023) ³³. The lowest order (most specific) protein class annotation available was used for each protein, else higher order classifications were assigned for plotting. Hallmark pathway enrichment analyses were performed using the R packages msigdbr v7.5.1 ³⁴ and clusterProfiler v4.6.2 ^35,36 with Benjamini-Hochberg p-value adjustments and a cutoff of p<0.05.

METABRIC cohort data accession

The mRNA expression z-scores relative to all samples (log microarray) of the gene names associated with our 185 shared MS hits were downloaded for the METABRIC patients with mRNA data (cohort of 1505 ER⁺ BC patients with relapse-free survival information) from cBioPortal.org using a z-score threshold of ± 1.96 (accession date 02 Feb 2023) ^37–42. Two of the 185 hits, RPRD1B and YLPM1, did not have mRNA expression z-scores and were omitted from METABRIC-related analyses.

Survival analyses

Cox proportional hazards ratios (HR) for each gene (relative mRNA expression z-scores) were calculated for relapse-free survival of ER⁺ BC patients among the METABRIC cohort using the R package survival v3.5–5 ^43,44.

Western blots and antibodies

For standard Western blots, whole-cell lysates were loaded in 25 μg quantities. For biotinylation Western blots, aliquots of the pulldown samples in 50 mM ABC were eluted using 3X protein loading buffer supplemented with 2 mM biotin and 20 mM DTT as previously described ²³. Eluates were loaded into NuPAGE 4–12% Bis-Tris gels (Invitrogen NP0322BOX) for SDS-PAGE alongside 1% input and 1% flow-through protein samples. Proteins were then transferred to nitrocellulose membranes, stained with Ponceau S, imaged, and de-stained. Membranes were then blocked with 5% milk in 1X TBS supplemented with 0.05% Tween-20 (Sigma P9416) (TBST) for ~30 minutes at room temperature, washed three times with 1X TBST, incubated with 0.3 μg/mL streptavidin-horse radish peroxidase (HRP) in 3% bovine serum albumin for 30 minutes at room temperature, and finally washed three times with 1X TBST before visualizing using a 1:1 mixture of SuperSignal^™ West Pico PLUS Chemiluminescent Substrate (Thermo Scientific 34580) on a BioRad Chemi-Doc station. Antibodies used for Western blots include FOXA1 (Abcam ab23738, RRID: AB_2104842, Supp. Fig. 3; or ab170933, RRID: AB_3065104, Supp. Fig. 1), glyceraldehyde-3-phosphate dehydrogenase (GAPDH) (Abcam ab181602, RRID: AB_2630358), secondary goat anti-rabbit (GE NA934, RRID: AB_772206), and streptavidin-HRP (Thermo Scientific N100).

Co-immunoprecipitation(Co-IP)

Co-immunoprecipitation was performed using MCF-7 and T47D cells seeded in a 10cm plate and lysed with 1mL ice-cold lysis buffer (25 mM Tris-HCl (pH = 7.4) (Roche, 10812846001), 150 mM NaCl (Sigma, S7653–1KG), 1 mM EDTA (Sigma, E5134–250G), 1% IGEPAL CA-630 (Sigma, I8896–100ML)) supplemented with 1x protease (Roche, 11836145001) and phosphatase (Roche, 04 906 837 001) inhibitors, as well as 100 units of benzonase nuclease (Sigma, E8263–5KU) and were maintained at constant agitation for 30 minutes at 4°C. Samples were then centrifuged at 15,000 rpm for 10 minutes at 4°C, 50μL was collected of each sample for input (whole cell lysate). The rest of the supernatant was incubated with either 5μg of FOXA1 antibody (Abcam, Ab5089, RRID: AB_304744) or 5μg Goat IgG antibody (Sigma, NI02–100UG). The samples were then incubated overnight at 4°C with constant rotation. The following day the samples were incubated with 50μL of prewashed PureProteome Protein A/G Mix Magnetic Beads (Millipore, LSKMAGAG10) for 3 hours at 4°C with constant rotation. The beads in combination with any of the targets were isolated by magnet using the DynaMag^™-2 (Invitrogen, 12321D) and washed 3 times with cold PBS + 0.01% Tween 20 (Sigma, P9416–100ml). After the final wash, IP samples were eluted with 60μL 1x NuPAGE LDS Sample Buffer (Invitrogen, NP0007) + 1% 2-mercaptoethanol (BME) (Sigma, M6250–100ML), then they were heated at 70°C for 10 minutes. The samples were then transferred by isolating the beads using the magnet. The input samples were boiled with 12.5μL 4x NuPAGE LDS Sample Buffer (Invitrogen, NP0007) + 1% BME (Sigma, M6250–100ML) at 95°C for 5 minutes. Proceeded with western blotting. The antibodies for the western blot related to the coimmunoprecipitation are FOXA1 (0.3μg/mL) (Abcam, Ab5089) and NR2C2 (1:500 dilution) (Santa Cruz Biotechonology, sc-365895, RRID: AB_10844760). The secondary antibodies are: anti-mouse IgG (H+L) secondary antibody, HRP (1:10000) (Invitrogen, PA1–28748, RRID:AB_10982166) and anti-goat HRP (1:7500) (Abcam, A15999, RRID:AB_2534673).

Cell proliferation assays

Cell proliferation assays were performed as previously described in ⁴⁵. Briefly, MCF-7 and T47D cells were plated at 300,000 cells per well in a 6-well plate in antibiotic free DMEM media supplemented with 15% FBS. After 24 hours cells were transfected with siRNA targeting NR2C2 or Scramble for control per individual well. Biological replicates were seeded and transfected on separate days. Individual wells were transfected with 30pmol siRNA using Lipofectamine RNA iMax reagent (Invitrogen, 13778150). Cells were fixed with 4% Formaldehyde (Sigma, F8775–500ML) 72 hours after the transfection for 10 minutes at room temperature, then stained with 0.1% crystal violet (Sigma, V5265) in 10% methanol (VWR, 6101826396) for 30 minutes at room temperature. Cells were destained with 10% glacial acetic acid (Sigma, 695092–500ML) for 15 minutes at room temperature. The destain was transferred to a 96 well plate and the absorbance was read at 595nm to identify changes in cell proliferation rates. The representative images were taken before the destain. The relative cell growth was calculated by normalizing the absorbance reads to the control (siScramble) for each cell line. Statistical tests were performed using Welch t-test.

NR2C2 ChIP-seq samples and analysis

Biological duplicates of MCF-7 cells that were either untreated (control) or subjected to 48-hour 40 nM siRNA targeting of FOXA1 (siFOXA1) were fixed with 11% formaldehyde and submitted to Active Motif for Factor-Path ChIP-seq using the α-NR2C2 antibody (Abcam ab109301). Active Motif prepared the ChIP-seq libraries, performed 75-nt single-end sequencing using 40 μg chromatin per sample plus one pooled input from all four samples, and then aligned the Illumina chastity-filtered reads with phred+33 quality scores to the human genome (hg38) using BWA v0.7.12 (“bwa aln/samse”) with default settings.

Aligned reads were filtered to remove reads that were unmapped, not primary alignments, failed platform quality control, or had MAPQ < 30 using samtools ⁴⁶. PCR duplicates were marked by picard ⁴⁷ and removed using samtools. Alignments were then filtered against the ENCODE hg38 blacklist v2 (accession 14 June 2023) before replicates were pooled to generate consensus peaks using MACS v2.2.6 peak calling with the input sample used as background ⁴⁸. Finally, the R package DiffBind v3.8.4 was used to remove greylist regions, determine read densities per sample per peak, and calculate relative differential binding ^2,49. All DiffBind results are represented as relative change in NR2C2 binding in siFOXA1 samples compared to control samples.

ChIP-seq peak annotations

NR2C2 ChIP-seq peaks were annotated for features and nearest genes using the R package ChIPseeker v1.34.1 ^50,51. Enriched KEGG terms were discovered using clusterProfiler v4.6.2 ^35,36 with Benjamini-Hochberg p-value adjustments and a cutoff of p<0.05. Motifs enriched at the NR2C2 peaks were discovered using MEME-AME (analysis for motif enrichment) ⁵² using the motifs found in “JASPAR (non-redundant) DNA; JASPAR CORE (2022)” and the sequence scoring method used was average odds score with Fisher’s exact test and e-value ≤ 0.05. The R packages JASPAR2022 ⁵³ and TFBSTools v1.36.0 ⁵⁴ were used to pull classifications of identified motif enrichments. Venn diagrams in were generated using the motif IDs. Motif classifications refer to the motif family, motif class, or motif name, depending on available annotation data, where the family was preferentially used to group motifs by the lowest order classification possible.

NR2C2 versus FOXA1 ChIP-seq analysis

FOXA1 ChIP-seq was performed in our lab using the same parental MCF-7 cell line as for the NR2C2 ChIP-seq experiment. The vehicle-treated control samples (in biological duplicates) from the FOXA1 ChIP-seq experiment were analyzed using the same alignment and filtering pipeline as for the NR2C2 ChIP-seq samples to allow for parallel comparisons. NR2C2 binding profiles were then compared to those of FOXA1 using `bedtools intersect` from bedtools v2.30.0 ⁵⁵. The distances between NR2C2 and FOXA1 peaks were calculated using `bedtools closest`. Browser tracks were generated using the R package plotgardener v1.4.2 ⁵⁶. To overcome sequenced fragment length limitations, an "overlap" here is defined as NR2C2 peaks within 223 bp (half of the mean NR2C2 peak width) of the nearest FOXA1 peak, while "not overlapping" NR2C2 peaks were ≥ 223 bp away from the nearest FOXA1 peak.

siRNA NR2C2 RNA-seq

MCF-7 and T47D cells were seeded in 6-well plates at 300,000 cell per well in antibiotic free DMEM media supplemented with 15% FBS. After 24 hours cells were transfected with siRNA targeting NR2C2 or Scramble for control per individual well. Biological replicates were seeded and transfected on separate days. Individual wells were transfected with 30pmol siRNA using Lipofectamine RNA iMax reagent (Invitrogen, 13778150) per manufacturer’s instructions. The cells were lysed 72 hours post transfection and RNA was extracted using Zymo Quick-RNA^™ miniprep kit (Zymo Research, R1055), with on column DNaseI treatment. RNA libraries were prepared by the Puerto Rico OMIC Center (PROMIC) using the Illumina mRNA stranded kit and 1μg of RNA per sample. PROMIC performed paired end 75 bp sequencing on an Illumina NextSeq 550. Sequencing reads were trimmed and filtered for quality score using fastp v0.24.0 ⁵⁷. They were aligned to the human genome hg38 genome using STAR v2.7.11b ⁵⁸ and quantification was performed using gene counts parameter in STAR v2.7.11b. Differential gene expression analysis was performed with the DESeq2 v3.21 package in R ⁵⁹ using an adjusted pvalue < 0.05. Volcano plots of the RNA-seq results were generated using R package EnchancedVolcano v1.26.0. Heatmaps were generated using pheatmap package v1.0.13 from R. Hallmark pathway analysis was performed using R package fgsea v1.34.0 in conjunction with hallmark pathway signature from MsigDB ⁶⁰. Venn diagrams for the RNA-seq were generated using R packages: ggvenn v0.1.10 and VennDiagram v1.7.3.

Statistical Analysis

Statistical details of each experiment are detailed throughout the methods section. Fos the MS data analysis log2 fold change (FC) ratios were calculated using the averaged Log2 LFQ intensities of each construct (+) biotin vs. negative controls (wild-type cells or construct (−)biotin)), and students t-test performed for each pairwise comparison, with p-values calculated. Proteins with significant p-values (<0.05) and Log2FC >1 were considered biological interactors. For protein classification and pathway analysis, hallmark pathway enrichment analyses were performed using the R packages msigdbr v7.5.1 ³⁴ and clusterProfiler v4.6.2 ^35,36 with Benjamini-Hochberg p-value adjustments and a cutoff of p<0.05. For ChIP-seq annotations, enriched KEGG terms were discovered using clusterProfiler v4.6.2 ^35,36 with Benjamini-Hochberg p-value adjustments and a cutoff of p<0.05. Motifs enriched at the NR2C2 peaks were discovered using MEME-AME (analysis for motif enrichment) ⁵² using the motifs found in “JASPAR (non-redundant) DNA; JASPAR CORE (2022)” and the sequence scoring method used was average odds score with Fisher’s exact test and e-value ≤ 0.05.

DATA AVAILABILITY

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE ⁷⁹ partner repository with the dataset identifier PXD046652. ChIP-seq data generated in this study are publicly available in the Gene Expression Omnibus (https://www.ncbi.nlm.nih.gov/geo/) under the accession number GSE287088 and the RNA-seq data is found at the accession number GSE299760. No original code was used. Software and analysis pipelines are described in the methods section of the manuscript.

RESULTS

FOXA1-miniTurbo proximity labeling reveals 185 high-confidence interacting proteins

To characterize the FOXA1 interactome in estrogen receptor-positive breast cancer cells, we utilized proximity-dependent biotinylation via the miniTurbo biotin ligase ²³. Derived from the E. coli BirA enzyme, miniTurbo catalyzes the conversion of biotin and ATP into biotinoyl-5’-AMP (bioAMP), which facilitates covalent labeling of proteins in close proximity (~10 nm) to the fusion protein ^25,26. The covalent attachment of biotin to lysine residues, combined with the high-affinity streptavidin-biotin interaction, allows for stringent washing and efficient purification of labeled proteins for downstream mass spectrometry (MS) analysis ^27,28. When miniTurbo is fused to either the N- or C-terminus of FOXA1, it enables the capture of both stable and transient protein interactions in situ, without the need for crosslinking, fixation, or a recognition sequence on interacting partners (Fig. 1A, Supp. Fig. 4 and 5).

Figure 1. Proximity labeling delineates >185 FOXA1 binding partners.

A) Cartoon showing labeling of nearby proteins with biotin by the FOXA1-miniTurbo construct.

B) Experimental design for biotinylation Western blots and LC-MS/MS using biological triplicates of three cell lines (wild-type control MCF-7 and two transduced MCF-7 cell lines expressing either N-terminal or C-terminal FOXA1-miniTurbo fusions) either untreated (−) or treated (+) with 0.5 mM biotin for 6 hours.

C) Western blot showing expression of FOXA1 in MCF-7 control cells as well as the N-terminal (N-term) and C-terminal (C-term) FOXA1-miniTurbo-expressing MCF-7 cells. Each SDS-PAGE lane was loaded with 25 μg whole-cell lysate (WCL) before transfer to nitrocellulose membrane. FOXA1 was detected using anti-FOXA1 ab170933. *, delineates endogenous FOXA1 (~55 kDa); #, delineates exogenous FOXA1 construct (N-term, ~80 kDa; C-term, ~78 kDa).

D) Western blot against streptavidin-HRP showing enrichment of proteins pulled down from N-term and C-term FOXA1 constructs using streptavidin beads after treatment with 0.5 mM biotin for 6 hours. Control cells reveal endogenously biotinylated background protein bands at ~75 and 150 kDa.

E) Volcano plot representing the enrichment of each protein quantified by LC-MS/MS, shown as the log2FC (x-axis) by the -log10 p-values (y-axis) calculated using Studen’ts t-tests relative to the control group indicated on each panel. Values for the FOXA1, ESR1, and GATA3 proteins are indicated for reference. Left panels, N-terminal construct + biotin vs controls; right panels, C-terminal construct + biotin vs controls.

F) Venn diagram showing intersections of the N- and C-terminal enriched hits (p < 0.05 and Log2FC > 1) compared to both sets of controls: control cells with biotin (top) or FOXA1-miniTurbo cells without biotin (bottom). The central intersection represents 185 proteins enriched in both constructs compared to all control groups.

G) HALLMARK analyses of genes associated with the 185 shared hits from F are plotted as the number of genes found in each pathway and colored according to q-value.

H) Protein classifications of the 185 shared hits from F, as annotated by the PANTHER database. The 57 proteins without annotated classes were not plotted.

To implement this approach, MCF-7 cells were stably transduced with either N-terminal or C-terminal FOXA1-miniTurbo constructs. Cells were incubated with or without biotin, alongside biotin-treated untransduced controls, all in biological triplicates (Fig. 1B). Biotinylated proteins were purified from whole-cell lysates via streptavidin pulldown and either eluted for Western blot validation or digested for LC-MS/MS analysis (Fig. 1B). Expression of FOXA1-miniTurbo constructs (Fig. 1C; Supp. Fig. 1) and their biotinylation activity (Fig. 1D; Supp. Fig. 2) were confirmed by immunoblotting.

Mass spectrometry analysis identified 2,067 proteins across all experimental conditions (Supp. Table 1). Differential enrichment was assessed using Student’s t-tests comparing each FOXA1-miniTurbo construct with appropriate controls (transduced cells without biotin and untransduced cells with biotin). Proteins with a log₂ fold-change >1 and p < 0.05 were considered significantly enriched. The N-terminal FOXA1 construct yielded 746 enriched proteins, while the C-terminal construct identified 189 (Fig. 1E–F). The lower yield from the C-terminal construct may reflect reduced pulldown efficiency, potentially due to inefficient streptavidin binding and loss of biotinylated proteins in the flow-through (Supp. Fig. 2). However, both ER (ESR1) and GATA3, canonical FOXA1 interactors, were significantly enriched in both datasets, validating the approach.

To focus on the most confident interactions, we selected the 185 proteins enriched by both constructs for downstream analysis (Fig. 1F). Gene set enrichment analysis of these 185 shared proteins revealed significant overlap with HALLMARK pathways, including TNF-α signaling via NFκB, estrogen response early, and TGF-β signaling (Fig. 1G), consistent with FOXA1’s known roles. Furthermore, the interactome was enriched for transcription factors, cofactors, and chromatin-modifying proteins (Fig. 1H), aligning with FOXA1’s function as a chromatin-bound pioneer factor. Collectively, these data demonstrate that miniTurbo-tagged FOXA1 effectively labels expected nuclear partners in ER⁺ breast cancer cells and define a high-confidence set of 185 FOXA1-associated proteins for further mechanistic investigation.

Proximity labeling reveals 157 previously unreported FOXA1 interactions in ER⁺ breast cancer

To assess the novelty of the FOXA1 interactome identified via miniTurbo labeling, we compared the 185 high-confidence interactors to known FOXA1-binding proteins curated in the BioGRID database ³². The list of known FOXA1 interactors from BioGRID includes proteins identified using a variety of different methods: pulldowns from SFB-tagged FOXA1 expressed in HEK293T cells ⁶¹, crosslinking-based FOXA1 affinity pulldowns ⁶², and a high-throughput chemical screen using a luciferase-based reporter for FOXA1-DNA binding activity ⁶³. This comparison confirmed several established interactions, including MLL3 and YAP1, and revealed 157 previously unreported FOXA1-associated proteins in MCF-7 cells (Fig. 2A). To explore the potential clinical significance of these FOXA1 interactors, we analyzed relapse-free survival (RFS) associated with the expression of their corresponding genes using ER⁺ breast cancer patient data from the METABRIC cohort ^38,41,42. Of the 185 candidate genes, 57 were significantly associated with clinical outcome, with 42 of these (approximately 70%) correlating with poor prognosis (Fig. 2B). These findings emphasize FOXA1’s role in regulating transcriptional networks relevant to breast cancer progression.

Figure 2. Orphan nuclear receptor NR2C2 identified as a novel FOXA1 interactor with implications in clinical outcome for ER⁺ BC patients.

A) Venn diagram showing an overlap of 28 protein hits with previously identified FOXA1 interactors from BioGRID and the identification of 157 novel FOXA1 interactors.

B) Relapse-free survival hazard ratios (RFS HR) among an ER⁺ BC patient cohort from the METABRIC dataset are plotted for the 59 genes whose relative mRNA expression z-scores are associated with p < 0.05. The selected candidate, NR2C2, is indicated with a purple arrow. Purple, proteins not previously shown to interact with FOXA1; green, proteins previously associated with FOXA1 in the BioGRID database.

C) Kaplan-Meier plot of ER⁺ BC patient RFS based on NR2C2 relative mRNA expression z-scores from the METABRIC dataset. Purple, higher NR2C2 expression; green, lower NR2C2 expression.

D) Violin boxplot of ER⁺ BC patient tumor histologic grade based on NR2C2 relative mRNA expression z-scores from the METABRIC dataset

E) Co-Immunoprecipation western blots confirming the interaction between FOXA1 and NR2C2 in MCF-7 Cells (left) and T47D breast cancer cells (right).

To illustrate the utility of this proteomic dataset, we focused on novel FOXA1 interactors associated with the highest hazard ratios, those most predictive of unfavorable outcomes. Among the top five candidates was the orphan nuclear receptor NR2C2, which has been implicated in cancer biology but remains understudied in the context of breast cancer. Given its classification as a nuclear receptor, similar to ER and AR which are well-known FOXA1 partners, NR2C2 may play analogous roles in transcriptional regulation. Although its endogenous ligand is unknown, the presence of a ligand-binding pocket suggests potential druggability. We therefore prioritized NR2C2 for further analysis. Expression analysis in the METABRIC cohort revealed that elevated NR2C2 mRNA levels are associated with significantly worse relapse-free survival (Fig. 2C and Supp. Fig 6) and correlate with increasing histologic tumor grade (Fig. 2D). To confirm the interaction between FOXA1 and NR2C2, co-immunoprecipation assays (Co-IP) were performed in two ER+ breast cancer cell lines, MCF-7 and T47D cells, validating the mass spectrometry results (Fig. 2E). Given that FOXA1 functions as a pioneer factor for multiple nuclear receptors and NR2C2 has been shown to modulate transcriptional activity of nuclear receptors, we hypothesize that FOXA1 either facilitates chromatin access for NR2C2 or cooperates with it at clinically relevant genomic loci.

Comprehensive annotation of the chromatin binding profile of NR2C2 in breast cancer cells

To characterize the genome-wide binding profile of NR2C2 and assess whether FOXA1 functions as a pioneer factor for NR2C2, we performed NR2C2 ChIP-seq in MCF-7 cells under basal conditions and following siRNA-mediated FOXA1 knockdown. Because the NR2C2 cistrome had not been previously annotated, this experiment not only identified NR2C2 binding sites across the genome but also enabled the assessment of their dependence on FOXA1 (Fig. 3A). We performed a 48-hour siRNA knockdown of FOXA1, confirmed by Western blot (Supp. Fig. 3), and subsequently conducted NR2C2 ChIP-seq using a specific anti-NR2C2 antibody (see Methods). Differential binding analysis identified 13,270 consensus NR2C2 binding sites shared between control and siFOXA1 conditions (Supp. Table 2). Applying a stringent cutoff (FDR < 0.1 and log2 fold change > +/−0.5), we found 191 NR2C2 peaks to be differentially bound upon FOXA1 depletion. Strikingly, the majority of these sites (170) exhibited decreased NR2C2 binding upon FOXA1 depletion (Fig. 3B). KEGG pathway enrichment analysis of these regions revealed that loss of NR2C2 binding was associated with genes involved in critical cellular pathways including endocytosis, cell cycle regulation, and cellular senescence (Fig. 3C).

Figure 3. NR2C2 DNA binding (ChIP-seq) is altered by FOXA1 knockdown in MCF-7 cells.

A) NR2C2 ChIP-seq pipeline illustrating sample setup and computational analyses for n=2 biological replicates of wild-type MCF-7 cells that were untreated (control) or treated with 48-hour 40 nM siRNA against FOXA1 (siFOXA1).

B) Volcano plot showing differential NR2C2 binding affinity in siFOXA1 samples relative to controls. Pink, gain of binding affinity (21 sites); blue, loss of binding affinity (170 sites); gray, no significant change.

C) KEGG pathway enrichment analyses of genes nearest to the NR2C2 binding sites are shown for both the control and siFOXA1 samples.

D) Heatmaps showing per-sample NR2C2 binding affinity across the DiffBind universe of 13270 peaks (left) or the 191 differentially bound peaks (right). Each row represents a single peak colored by Z-score (purple, lower; yellow, higher).

E) Distribution of peak features for NR2C2 binding across the universe of peaks (left) or differentially bound peaks (right) for both the control and siFOXA1 samples.

F) Motifs enriched at NR2C2 peaks are represented as Venn diagrams (left) for the universe of NR2C2 peaks (top) and the differentially bound peaks (bottom), with the unique motifs from each diagram classified in the dot plot (right). Any motif classification that contained only a single motif lists that motif by name; classifications with > 1 motif are represented as dots, where size corresponds to the number of unique motifs in that class.

To globally assess NR2C2 binding under control and FOXA1-depleted conditions, we compared binding intensity across all consensus peaks and the subset of differentially bound peaks (Fig. 3D). In the global set of peaks, NR2C2 binding distributions were largely similar between conditions (Fig. 3E, left panel). The differentially bound NR2C2 peaks, however, occurred at diverse genomic features. Motif analysis revealed 940 shared motifs across both conditions (including the NR2C2 motif itself), while 107 motifs were unique to control samples and 17 motifs were specific to the FOXA1-depleted condition (Fig. 3F, Supp. Table 3). All motifs found in the differentially bound peaks were a subset of those shared in the full set of peaks. Notably, NFIL3 was the only motif uniquely enriched in NR2C2 gain-of-binding peaks upon FOXA1 depletion; the remaining motifs were either shared or enriched specifically in loss-of-binding peaks (Fig. 3F, bottom Venn).

A detailed examination of condition-specific motif enrichment, both in the universe of peaks and among differentially bound regions, revealed distinct motif classes (Fig. 3F, right). For instance, NR2C2 peaks under control conditions were enriched for motifs absent in the FOXA1 knockdown condition, including those for FOXH1 and ER. Furthermore, the NR2C2 loss-of-binding sites were enriched for 37 FOX family member motifs not found in gain-of-binding sites, highlighting a likely dependency of NR2C2 binding on FOXA1 at these loci. Collectively, these data suggest that FOXA1 may act as a pioneer factor for NR2C2 at a subset of binding sites. Importantly, since the ligand for NR2C2 remains unknown, it is plausible that additional FOXA1-dependent NR2C2 sites exist but remain unbound under the unliganded conditions of this experimental setup. Nevertheless, this study represents the first comprehensive annotation of NR2C2 binding sites and lays the foundation for further mechanistic studies.

FOXA1 depletion reduces NR2C2 binding at key loci such as the ESR1 promoter.

To investigate which genomic loci may contribute to the proximity-based interaction between FOXA1 and NR2C2, we compared NR2C2 ChIP-seq data from control and FOXA1-depleted MCF-7 cells with previously generated FOXA1 ChIP-seq profiles from the same parental cell line ⁶⁴. We defined co-bound sites as NR2C2 and FOXA1 peaks overlapping within 223 bp (half the mean NR2C2 peak width). Under control conditions, 19.3% of NR2C2 binding sites overlapped with FOXA1 peaks, compared to 16.6% in FOXA1-depleted cells (Fig. 4A), suggesting that NR2C2 binding is partly dependent on FOXA1 at specific loci. The distribution of distances between NR2C2 and FOXA1 peaks remained broadly similar across conditions. However, in the FOXA1-depleted state, NR2C2 binding sites shifted slightly upstream of FOXA1 peaks, as indicated by an increased frequency of upstream sites and decreased frequency of downstream sites compared to control cells (Fig. 4B).

Figure 4. Comparison of NR2C2 and FOXA1 ChIP-seq binding patterns suggests distinct subtypes of NR2C2 dependency on FOXA1 in MCF-7 cells.

A) Venn diagram representing the overlapping intervals between FOXA1 (gray; from vehicle-treated MCF-7 cells in different study) and NR2C2 ChIP-seq from controls (purple) or siFOXA1-treated (green) MCF-7 cells.

B) The distance in bp between NR2C2 and FOXA1 peaks are shown as an estimator of the cumulative distribution function (ECDF) plot, which has been zoomed into the relative ± 5-kb region around FOXA1 sites. Purple solid line, NR2C2 binding in controls; green dashed line, NR2C2 binding in siFOXA1-treated samples.

C) Cartoon illustrating six distinct binding mechanisms of NR2C2 as related to both FOXA1 expression (presence/absence of FOXA1) and binding (overlapping sites). Purple, NR2C2 binding in control MCF-7 cells; green, NR2C2 binding in siFOXA1-treated MCF-7 cells; gray, FOXA1 binding in vehicle-treated MCF-7 cells (different experiment). Below the cartoon are the numbers of distinct peaks and associated genes that fit into each category.

D) Browser tracks showing binding of NR2C2 in control MCF-7 cells (purple), NR2C2 in siFOXA1-treated MCF-7 cells (green), and FOXA1 in vehicle-treated MCF-7 cells (gray). Example of NR2C2 loss-of-binding upon siFOXA1 at the ESR1 locus (DiffBind (DB peaks 29–31).

E) Browser tracks showing binding of NR2C2 in control MCF-7 cells (purple), NR2C2 in siFOXA1-treated MCF-7 cells (green), and FOXA1 in vehicle-treated MCF-7 cells (gray). Example of the NR2C2 gain-of-binding site at LINC02957 (DB peak 42).

Closer examination of NR2C2 and FOXA1 co-bound sites revealed six distinct patterns of NR2C2 binding based on FOXA1 expression (present or depleted) and peak overlap with FOXA1 (Fig. 4C; Supp. Table 2). The most prevalent mode was FOXA1-independent binding, in which NR2C2 binding remained unchanged regardless of FOXA1 expression or binding. In this category, 3,841 NR2C2 peaks (corresponding to 3,089 nearest gene annotations) overlapped with FOXA1, while 9,236 NR2C2 peaks (7,322 genes) did not.

Differentially bound NR2C2 peaks were classified into four additional categories based on whether they were lost or gained following FOXA1 depletion. NR2C2 peaks that decreased in binding (i.e., FOXA1-dependent sites) occurred at both FOXA1-overlapping (105 peaks, 101 genes) and non-overlapping sites (65 peaks, 65 genes). Notably, FOXA1 depletion reduced NR2C2 binding at the ESR1 promoter (Fig. 4D), suggesting a potential role for NR2C2 in regulating estrogen receptor expression. Among the NR2C2 peaks that gained binding following FOXA1 depletion, only one overlapped with FOXA1 (at the LINC02957 locus), while the remaining 20 were located at non-overlapping sites (Fig. 4E). We note that our overlap definition was stringent, expanding the window we used to define an overlap could reveal additional shared binding sites. Altogether, these results validate a chromatin-based interaction between FOXA1 and NR2C2 and support the idea that FOXA1 regulates NR2C2 recruitment at specific genomic loci. These findings underscore the need for further investigation into the transcriptional consequences of NR2C2 binding and its functional interplay with FOXA1.

NR2C2 regulates conserved transcriptional programs enriched in cancer pathways and dependent on FOXA1 chromatin binding

To comprehensively define the transcriptional programs regulated by NR2C2 and identify genes co-regulated by NR2C2 and FOXA1, we performed RNA-sequencing in MCF-7 and T47D breast cancer cells following siRNA-mediated knockdown of NR2C2. Cells were treated with either a scrambled non-targeting siRNA or siRNA specifically targeting NR2C2 (Fig. 5). After 72 hours, NR2C2 emerged as the most downregulated gene in MCF-7 cells and the fourth most downregulated gene in T47D cells, confirming efficient knockdown across three biological replicates (Fig. 5D and E). Transcriptome analysis revealed a consistent and conserved gene expression program regulated by NR2C2 in both cell lines, with enrichment in hallmark cancer pathways including estrogen response, UV response, hypoxia, p53 signaling, and epithelial-to-mesenchymal transition (EMT) (Fig. 5F). Although the endogenous ligand for NR2C2 remains unknown, these data define a robust set of differentially expressed genes downstream of NR2C2 (Supplemental Tables 4 and 5), supporting its role as a regulator of key oncogenic pathways in ER+ breast cancer.

Figure 5. NR2C2 regulates conserved transcriptional programs enriched in cancer pathways and cell proliferation.

A) NR2C2 RNA-seq pipeline illustrating sample setup and computational analyses for n =3 biological replicates of wild-type MCF-7 and T47D cells that were treated with 30pmol siRNA against NR2C2(siNR2C2) or Scramble (siScramble) for 72 hours.

B) Heatmap showing per-sample top 50 differentially expressed genes (adjusted < 0.05) in MCF-7 cells

C) Heatmap showing per-sample top 50 differentially expressed genes (adjusted < 0.05) in T47D cells

D) Volano plot showing enriched upregulated and downregulated genes in MCF-7 cells. In dark gray NS = not significant, in light gray Log (2) FC = fold change < 1 and > -1, in purple Adj. pvalue < 0.5, and in green both adjusted pvalue < 0.05 and fold change > 1 and < -1.

E) Volano plot showing enriched upregulated and downregulated genes in T47D cells. In dark gray NS = not significant, in light gray Log (2) FC = fold change < 1 and > -1, in purple Adj. pvalue < 0.5, and in green both adjusted pvalue < 0.05 and fold change > 1 and < -1.

F) Hallmark gene signatures associated to differentially expressed genes in common between MCF-7 and T47D cells. In orange pathways enriched in siNR2C2 group and in red pathways enriched in siScramble group. From 3 biological replicates per cell line with pvalue < 0.05.

G) Venn diagram showing the overlap of commonly down regulated genes upon NR2C2 knockdown in MCF-7 and T47D cells.

H) Browser tracks showing binding of NR2C2 in control MCF-7 cells (purple), NR2C2 in siFOXA1-treated MCF-7 cells (green), and FOXA1 in vehicle-treated MCF-7 cells (gray) near the promoter of EXO1.

I) Boxplot for normalized counts of EXO1 in MCF-7 and T47D from RNA-seq upon treatment with 30pmol siRNA against NR2C2(siNR2C2) or Scramble (siScramble) for 72 hours. From 3 biological replicates per cell line with adjusted pvalue shown. Mean and standard error is presented.

J) Kaplan-Meier plot of ER+ BC patient overall survival based on EXO1 relative mRNA expression z scores from the METABRIC dataset. Purple, higher EXO1 expression; green, lower EXO1 expression.

K) Bar graph for proliferation assays in MCF-7 and T47D treated with 30pmol siRNA against NR2C2(siNR2C2) or Scramble (siScramble) for 72 hours. All value were normalized to siScramble group. Representative images shown below. From 3 biological replicates per cell line and pvalues were calculated using t-test. Mean and standard error is presented.

To integrate transcriptomic and chromatin binding data, we analyzed 105 genomic loci co-bound by NR2C2 and FOXA1 that showed FOXA1-dependent NR2C2 occupancy, as determined by ChIP-seq (Fig. 4C). We mapped these peaks to their nearest genes and intersected this list with genes significantly downregulated upon NR2C2 knockdown, thereby identifying targets co-dependent on both FOXA1 chromatin binding and NR2C2 expression. Among these targets, we identified several genes associated with poor prognosis in ER+ breast cancer patients, including EXO1, which encodes a 5′ to 3′ exonuclease critical for mismatch repair and genome stability ⁶⁵. Our data demonstrate that EXO1 expression is driven by promoter-proximal NR2C2 binding that requires FOXA1’s pioneering activity (Fig. 5H). Expression of EXO1 is significantly reduced upon NR2C2 knockdown (Fig. 5I) and high EXO1 expression correlates with poor clinical outcomes in ER+ breast cancer (Fig. 5J). Like EXO1, we identified other examples of FOXA1/NR2C2 co-dependent genes such as CCNA2 and RUVBL2 (Supp. Fig. 7).

Given the broad transcriptional impact of NR2C2 on cancer-relevant genes, we next assessed its functional role in proliferation. NR2C2 knockdown significantly reduced proliferation in both MCF-7 and T47D cells (Fig. 5K and L), linking its gene regulatory activity to tumor cell growth. Collectively, these findings define a transcriptional program governed by NR2C2, highlight genes co-regulated by FOXA1 and NR2C2, and nominate NR2C2 as a potential therapeutic target in ER+ breast cancer.

DISCUSSION

By fusing miniTurbo to FOXA1 in two separate constructs and performing proximity labeling followed by LC-MS/MS analysis, we identified 157 previously unreported protein candidates within the FOXA1 interactome (Fig. 1 and 2). Our results indicate that fusing miniTurbo to either the N- or C-terminus of FOXA1 does not significantly impair its chromatin localization, enabling the identification of high-confidence interactors common to both constructs. This is supported by the detection of well-established FOXA1 interactors (Fig. 2A), proteins involved in known FOXA1-regulated pathways, and known chromatin-associated proteins including transcription factors, histone modifiers, and chromatin-binding proteins. Notably, among these novel interactors, 42 are predictive of poor relapse-free survival in ER+ breast cancer, underscoring the relevance of FOXA1’s interactions with factors central to disease progression. This proteomic platform could be further leveraged to identify FOXA1 interactors enriched under specific conditions, such as estrogen stimulation or treatment with endocrine therapies like tamoxifen. Overall, this study reveals an expanded FOXA1 interactome and provides a valuable resource for uncovering new mechanistic insights and potential therapeutic targets.

To demonstrate the utility of this resource, we explored the FOXA1-NR2C2 interaction in depth. This represents the first study to perform ChIP-seq of endogenous NR2C2 in cancer cells and to integrate these data with RNA-seq (both in MCF-7 and T47D cells) to systematically link chromatin binding with gene expression changes (Fig 3 and 4). We identified multiple putative NR2C2 binding mechanisms in relation to FOXA1 protein levels and chromatin occupancy, including potential co-regulation at the promoter of the estrogen receptor gene, ESR1 (Fig 4). Through integrative ChIP-seq and RNA-seq analyses, we show that FOXA1 and NR2C2 co-occupy a subset of genomic loci that regulate transcriptional programs implicated in tumor progression (Fig. 5). Upon FOXA1 depletion, NR2C2 binding was lost at 170 loci, suggesting FOXA1-dependent recruitment of NR2C2 at these sites. Conversely, we also observed loci where NR2C2 binding was enhanced upon FOXA1 depletion, implying that FOXA1 may inhibit NR2C2 binding at specific sites. This dual regulatory behavior is reminiscent of FOXA1’s interaction with the androgen receptor (AR) in prostate cancer, where FOXA1 has been shown to both facilitate and restrict AR binding, depending on genomic context ⁶⁶.

Although most studies of NR2C2, also known as testicular receptor 4 (TR4), have been conducted in the context of prostate cancer metastasis ^67–70, several lines of evidence support a potential role for NR2C2 in ER+ breast cancer. NR2C2 is expressed in mammary tissue at both the transcript and protein levels ^71,72, and like AR, it has been shown to suppress ER-mediated transcription in MCF-7 cells via direct interaction between its ligand-binding domain and ER, which inhibits ER homodimerization and DNA binding ⁷³. Furthermore, NR2C2 binding sites in MCF-7 cells expressing GFP-tagged NR2C2 are enriched for Forkhead motifs, suggesting a functional link with FOXA1 ⁷⁴.

As FOXA1 is a known pioneer factor for ER and AR and plays a central role in hormone receptor-driven breast and prostate cancers ⁷⁵, the identification of NR2C2 within the FOXA1 interactome is particularly noteworthy. Although NR2C2 is considered an “orphan” receptor due to the absence of known ligands, its structure is characteristic of nuclear receptors, with distinct nuclear localization, DNA-binding, ligand-binding, transactivation, and dimerization domains ⁷⁶. Based on its DNA-binding domain, NR2C2 belongs to the steroid/thyroid receptor subfamily, like ER. Functionally, it binds DNA as a homodimer at direct repeats of the AGGTCA motif to regulate transcription ⁷⁷. NR2C2 also modulates the activity of other nuclear receptors, including RAR and RXR, by competitively binding hormone response elements or through direct protein-protein interactions. In prostate cancer, NR2C2 can heterodimerize with AR, repressing transcription of both AR and NR2C2 target genes, thereby revealing a complex network of steroid receptor signaling ⁷⁷.

Given NR2C2’s apparent role in repressing ER activity through protein-protein interactions, it will be interesting to determine whether its function could be therapeutically leveraged to suppress ER signaling in breast cancer. This may be achievable by identifying NR2C2’s elusive ligand(s) or by targeting post-translational modifications (PTMs) of NR2C2. A recent large-scale proteogenomic study in ER+ breast cancer patients identified multiple NR2C2 PTMs, including phosphorylation and acetylation ⁷⁸. If any of these PTMs are cancer-associated, or if modifying these residues enhances NR2C2’s repressive activity on ER, such mechanisms could provide an epigenetic strategy to inhibit ER signaling that is distinct from current endocrine therapies. While NR2C2 was the focus of our functional follow-up, we also identified 42 additional FOXA1 interactors associated with poor relapse-free survival in ER+ breast cancer. Collectively, our study demonstrates that defining the FOXA1 interactome in breast cancer cells and integrating these data with patient outcomes offers a powerful approach to uncover potential therapeutic targets or prognostic biomarkers in breast cancer.

IMPLICATIONS STATEMENT.

These findings expand the FOXA1 interactome in breast cancer and uncover new candidate proteins with potential as biomarkers and therapeutic targets in hormone-driven tumors.