Integrated Multi-omic Profiling Identifies BRD8/EP400 as a Pivotal Chromatin Module Mediating Anti-HER2 Response in HR+/HER2+ Breast Cancer.

Introduction
Hormone receptor (HR)+/HER2+ breast cancers, characterized by expression of estrogen receptor (ER) and/or progesterone receptor (PR) and HER2, represent ~13% of all breast cancers1 and exhibit more aggressive features than HR+/HER2− tumors2, necessitating intensive treatment approaches3. Despite standard care combining chemotherapy with dual anti-HER2 therapy4, chemotherapy-free alternatives have shown limited efficacy1, with neoadjuvant trials of antiestrogen and HER2-targeted therapies achieving pathological complete response in only ~20% of cases5,6.
This modest response reflects the complex biology of HR+/HER2+ tumors, including receptor heterogeneity and ER-HER2 signaling crosstalk7. While studies demonstrate that each pathway can compensate when the other is blocked8,9, and dual HER2 blockade can induce a luminal A-like phenotype10, the molecular mechanisms remain unclear. The heterogeneous expression of receptors and limitations of bulk RNA-sequencing approaches have impeded the identification of anti-HER2 responsive genes that mediate ER activation, leaving major signaling and proliferation pathways regulated by anti-HER2 agents poorly understood.
Among the 46 proteins in the human proteome harboring bromodomains (BRD), the acetyl-lysine binding domain, BRD8 is among the least studied11. BRD8 is a subunit of the p400/Tip60 chromatin remodeler/Histone Acetyl Transferase (HAT) complex12. BRD8 activates ER target gene TFF1 by incorporating H2A.Z at its promoter13. Moreover, BRD8 promotes cell proliferation in various cancer types including colorectal cancer, hepatocellular carcinoma, and glioblastoma14–16. Knockdown of BRD8 in colorectal cancer cell lines impairs cell growth and induces cell death14,17, and renders cells sensitive to spindle poisons and proteasome inhibitors14. Recently, BRD8 was reported to repress wild type (WT) p53 function by incorporating H2A.Z to p53 target gene loci thus sustaining proliferation in TP53WT glioblastoma (GBM). BRD8 has been proposed as a therapeutic target for patients with TP53wt GBM because ablation of BRD8 enhances chromatin accessibility, leading to p53 transactivation and cell cycle arrest 16. However, the functions of BRD8 in regulating breast cancer growth and treatment resistance have not yet been reported.
To understand how ER/HER2 crosstalk mediates therapeutic resistance to dual ER/HER2 blockade, we focused on one direction of the crosstalk, i.e., how anti-HER2 agents activate ER signaling, using BT474 cells in which HER2 signaling is dominant over ER signaling. Our scRNA-sequencing analyses identified earlier neratinib-induced genes in a highly responsive cell population, in which BRD8 was found to drive massive ER target gene activation. Knockout of BRD8 abolished the activation of the ER signaling pathway induced by neratinib treatment, sensitizing cells to anti-HER2 agents. By performing snRNA and ATAC-sequencing of a PDX model after long-term neratinib treatment, we confirmed co-induction of BRD8 with ER and ER target genes by neratinib and that neratinib-induced genes shared open chromatin regions enriched in ER, FOX, and ETS family transcription factor (TF) binding motifs. Moreover, the chromatin decompaction is mediated by p400/Tip60 mediated deposition of H2AZac, and BRD8 regulates ER-dependent and -independent growth and survival in HR+/HER2+ cells. In line with these findings, BRD8 ablation re-sensitizes neratinib-resistant HR+/HER2+ cells to neratinib, suggesting that combinatory targeting of BRD8 and HER2 will be effective in treating HR+/HER2+ BC that developed resistance to dual ER/HER2 blockade.

Materials and Methods

Cell lines and cell culture
HEK293T (ATCC, Cat# CRL-11268, RRID: CVCL_1926), BT474 (ATCC, Cat# HTB-20, RRID: CVCL_0179) and MDA-MB-361 (ATCC, Cat# HTB-27, RRID: CVCL_0620) cell lines were purchased from ATCC. HEK293T and MDA-MB-361 were maintained in Dulbecco’s Modified Eagle Medium (DMEM) (Gibco) supplemented with 10% fetal bovine serum (FBS) (VWR) and 1% Penicillin-Streptomycin (P/S) (Gibco). BT474 cell lines were maintained in RPMI 1640 medium (Gibco) supplemented with 10% FBS (VWR) and 1% Penicillin-Streptomycin (P/S) (Gibco). All cell lines were cultured at 37 °C under standard conditions (5% CO2), used at low passage numbers, and periodically tested for Mycoplasma contamination. Cell line authentication was performed using short tandem repeat (STR) profiling at the Translational Research Initiatives in Pathology Laboratory, University of Wisconsin–Madison.

Generation of BT474 and MDA-MB-361 BRD8 knockout cell lines
Guide RNAs (gRNAs) targeting exon 2 of BRD8 were produced by mixing CRISPR RNA (crRNA) and transactivating CRISPR RNA (tracrRNA) at equimolar concentrations in a microcentrifuge tube for a final duplex concentration of 100 μM. The resulting mixture was heated at 95°C for 5 min and allowed to cool to room temperature. Ribonucleoprotein particles (RNPs) were produced by mixing Cas9 enzyme and gRNAs targeting exon 2 of BRD8 and incubating at room temperature for 10 min. 1×106 BT474 and MDA-MB-361 cells were transfected with RNPs using the Lonza 4D-Nucleofector system and seeded in 6 well plates. Five days later, single cells were sorted into 96-well plates. Following single colony formation, protein was extracted from single clones and the parental cells. Western blot was performed to confirm that BRD8 was knocked out in BRD8 KO clones. The sequences of the two sgRNAs are listed below:
sgRNA-1: AGCTTCTCTCGGATGGACCA, sgRNA-2: CGGATGGACCATGGCTCTGT

Organoid culture
Organoids were cultured as previously described18. Briefly, organoids were embedded in 80 μl Matrigel domes in a 24-well plate or 10 μl Matrigel domes in the 96-well plate. Plates were flipped and incubated for 30 min at 37 °C to allow Matrigel to solidify. Following incubation, subtype-specific medium was added after the matrigel domes solidified. For all breast cancer subtypes, 10 μM Y-27632 was freshly added to the PDxO base medium (Advanced DMEM/F12 with 5% FBS, 10 mM HEPES, 1× Glutamax, 1 μg/ml hydrocortisone, 50 μg/ml gentamicin and 10 ng/ml human EGF). For ER+/HER2+ breast cancer, 10 nM heregulin-β1 (HER2 positive specific), 100 ng/ml FGF2 and 1 mM NAC (ER positive specific) were added before use. Medium was exchanged every 3 to 4 days. Mature organoids were passaged using cell recovery solution (Corning, Cat# 76332–050) according to the manufacturer’s instructions, and then dissociated into single cells in TrypLE for 10–15 min at 37 °C.

Virus packaging and stable cell line or orgaoids generation

Lentiviral Packaging and Transduction
Lentiviral particles were produced in HEK293T cells cultured in 10-cm dishes. Cells were transfected with 2 μg pME-VSVG (Addgene #98286), 4 μg psPAX2 (Addgene #12260), and 4 μg of either the lentiviral shBRD8 expression vector or the corresponding control vector. Viral supernatants were collected 48 h post-transfection and used immediately for transduction. To generate stable BRD8 knockdown organoids, dissociated organoids were mixed with 1 mL viral supernatant in 1 mL complete medium supplemented with polybrene (8 μg/mL). The mixture was centrifuged at 1,000 rpm for 2 h at 10°C (spin infection). Following centrifugation, the supernatant was removed, and organoids were resuspended in Matrigel and plated in 24-well plates. Transduced organoids were selected with puromycin (2 μg/mL) for at least 2 weeks. The sequences of the BRD8 shRNAs were as follows: shBRD8–1: AGATGTTATTGTTCGGAAATT, shBRD8–2: GCCGAAATAGTAGCTGGAGTT

Retroviral Packaging and Stable Cell Line Generation
Retroviral particles were generated in HEK293T cells cultured in 10-cm dishes. Cells were transfected with 2 μg VSVG, 4 μg PHIT60, and 4 μg of either pLNCX-WT-BRD8, pLNCX-ΔBD-BRD8 (bromodomain-deficient mutant), or empty control vector. Viral supernatants were harvested 48 h after transfection for infection. To establish stable BRD8 rescue cell lines, 5 × 105 BRD8 knockout (KO) cells were seeded into 6-well plates. The following day, cells were infected with 1 mL retroviral supernatant mixed with 1 mL fresh medium containing polybrene (8 μg/mL). After infection, cells were selected with G418 (400 μg/mL) for at least 2 weeks to generate stable cell lines expressing either wild-type or bromodomain-deficient BRD8.

Cell viability and cell proliferation assays
An MTT assay was used for measuring cell viability. Briefly, each well of a 96 well plate was seeded with 1 × 104 cells in 100 μl medium. The next day after cells adhered, either fulvestrant (100 nM), neratinib (200 nM), or a combination of both drugs was added to each well. Drugs were refreshed every two days. After 6 days of treatment, medium was removed and precipitate crystals were dissolved in 100 uL of DMSO, and then plate was read at 570 nm using a microplate reader.
A sulforhodamine B (SRB) assay was used for measuring cell proliferation19. Briefly, each well of a 96 well plate was seeded with 1 × 104 cells in 100 μl medium. Following treatment, cells were then fixed by adding 25 μl of 50% trichloroacetic acid (TCA) in each well and incubated at 4 °C for 1 h, after staining with SRB (Sigma-Aldrich, Cat#S1402) solution, the excess dye was removed by washing repeatedly with 1% acetic acid. Allow the plate to air-dry at room temperature. The protein-bound dye was dissolved in 100 uL 10 mM Tris base solution and plate was read at OD 510 nm using a microplate reader.

Annexin V and PI staining
Annexin V and PI staining was performed using the Annexin V Apoptosis Detection Kit following the manufacturer’s instructions (eBioscience, Invitrogen, Cat# 88–8005-7). Briefly, BT474 parental and BRD8 KO cells were treated with or without neratinib (200 nM) for 4h, then cells were harvested and washed twice with PBS. Subsequently, they were stained with propidium iodide (PI) and Annexin V. Following 15 min of incubation in dark at room temperature, cell apoptosis was measured by flow cytometry.

Cell Titer 96 AQueous One Solution Cell Proliferation Assay (MTS) for organoids
Organoids were first digested into single cells, and 2,000 cells were embedded in 10 μl Matrigel domes in a 96 well plate. Three days post-seeding, organoids were treated with fulvestrant (100 nM), neratinib (200 nM), or a combination of both drugs. Treatments were refreshed every 3 days for 2 to 3 weeks. Organoid viability was tested by MTS according to the manufacturer’s instructions (Promega, Cat#G3581).

Western blot
Western blot was performed as previously described20. Briefly, cells were collected and resuspended in lysis buffer (50 mM Tris (pH 7.4), 150 mM NaCl, 1 mM EDTA, 1 mM EGTA, 1 mM NaF, 1 mM Na3VO4, 1% Triton X-100, 10% glycerol, 0.25% deoxycholate, and 0.1% SDS). Approximately 30 μg of protein was resolved by SDS-PAGE. Proteins were then transferred to nitrocellulose membranes using a BioRad Turbo Blot. Blots were blocked with 5% nonfat milk or 5% BSA for 1–2 hours, then incubated with primary antibodies diluted in blocking buffer (Table S1) at 4°C overnight on a rotator. Blots were then incubated with HRP- conjugated goat anti-rabbit or mouse IgG secondary antibodies for 1 h at room temperature. Membranes were incubated with SuperSignal West Pico ECL (Thermo Fisher Scientific, Waltham, MA) followed by exposure using a BioRad ChemiDoc.

Co-immunoprecipitation
For Co-IP from cell cultures, BT474 cells in 15 cm dishes were washed with cold PBS and harvested in immunoprecipitation buffer [50 mmol/L Tris–HCl, pH 8.0, 150 mmol/L NaCl, 1% IGEPAL CA-630, 10% Glycerol, and Complete Protease Inhibitor Mixture (Roche)]. The lysate was then rotated at 4°C for 1 hour, followed by centrifugation at 14,000 rpm for 20 minutes. The supernatants were then combined with 100 μL of protein G Sepharose preincubated with antibodies against 3 μg ER (Santa Cruz Biotechnology, Cat# sc-8002) or 3 μg BRD8 (Thermo Fisher, Cat# 702872), followed by rotating at 4°C overnight on a rotator. The protein G Sepharose was pelleted and washed three times using immunoprecipitation buffer. The precipitates were resolved on SDS–PAGE gel and subjected to immunoblot analysis.

RNA extraction and real-time quantitative PCR (RT-qPCR)
Total RNA was extracted from cell lines and organoids using the E.Z.N.A Total RNA kit (Omega Bio-tek). 2 μg of RNA was reverse transcribed using the iScript cDNA Synthesis kit (Bio-Rad, Cat#1708891) according to the manufacturer’s instructions. RT-qPCR was performed using 2X SYBR Green Master Mix (Roche Scientific, Basel, Switzerland) according to the manufacturer’s instructions (including cycling parameters), and Bio-Rad the BioRad CFX96 Touch Real-Time PCR Detection System. Primer sequences are listed in Table S2.

Single cell RNA-seq data analysis
scRNA- seq was performed by the UW- Madison Biotechnology Center Gene Expression Center. For the single cell preparation, BT474 cells were treated with DMSO, fulvestrant (100 nM), neratinib (200nM) or a combination of fulvestrant and neratinib for 4h. Cells were digested with trypsin into single cells and resuspended with 1x PBS (Ca++/Mg++-free) + 0.04% BSA (w/v non-acetylated) + 1U/ml RNase Inhibitor for cell number counting and cell viability measurement using the Countess® II Automated Cell Counter. The scRNA-seq libraries were constructed using the Chromium Next GEM Single Cell 3’ v3 Reagent Kit according to the manufacturer’s guidelines (10x Genomics). cDNA libraries were uniquely sample indexed and pooled for sequencing. Libraries were sequenced on Illumina NovaSeq 6000 instruments using 50 bp paired-end sequencing targeting 3000 cells per sample with a depth of 60000 reads per cell according to the manufacturer’s recommendations (10x Genomics).
Filtered gene count matrices of each treatment condition were generated by the 10X Genomics Cell Ranger software package for downstream analysis using Seurat v421. Quality control (QC) metrics are shown in Figure S1A. We excluded cells with high mitochondrial content (>=25%) or less than 200 genes expressed. Single-cell doublets were filtered using DoubletFinder v322 with the estimated ratio recommended by 10X Genomics. After pre-processing using the Seurat pipeline with default parameters, including log normalization (NormalizeData function), identifying top variable genes, and principal component analysis, four datasets were integrated using the Harmony method23 based on fuzzy-clustering to reduce unwanted technical variation and batch effects. The integrated dataset was then clustered using community-based clustering at resolutions from 0.1 to 2 through the Seurat FindCluster function to determine the optimal resolution with a minimum of 5% of cells in each cluster at >=2 samples. UMAP dimensionality reduction was performed with the RunUMAP function in Seurat, using 25 PCs corrected by Harmony. Gene signatures for each cluster and condition were identified using differential gene expression analysis employing MAST24 which is implemented using the FindMarkers function. Neratinib-induced gene signatures were identified by differentially expressed (DE) genes (average_log2FC cutoff 0.5 and FDR-adjusted p-value < 0.01) in pairwise, comparing the neratinib/combination treatments with the DMSO/fulvestrant treatments. To identify ER target genes expressed in the neratinib treatment condition compared to DMSO, the log2FC of the % of expressing cells in the neratinib enriched (1,3,7) and neratinib excluded (0,2,5) clusters was calculated. Cells from the combination and fulvestrant treatments were not included, and only ER target genes expressed in at least 5 cells in each treatment were evaluated. Data visualization was performed with Seurat functions Dim Plot, Vin Plot, and Heatmap, as well as with functions from the data visualization libraries ggplot2 and ComplexHeatmap25.

Multi-omics single nucleus RNA-seq and ATAC-seq
Nuclear dissociation was performed based on the 10X Genomics’ Nuclei Isolation Guide (CG000366). HCI032 tumors were cut into <2 mm pieces, homogenized using a Dounce homogenizer in 1ml of ice-cold lysis buffer until the solution was uniform, and incubated on ice for 5 minutes with an additional 0.5 ml of lysis buffer. The suspension was filtered through a 70 μm Flowmi Cell Strainer, followed by a 40 μm strainer and centrifuged at 500 × g for 5 min at 4°C. The pellet was washed twice with 1 ml of cold wash buffer and twice with 1ml of cold dilution buffer, with a 5-minute incubation, and centrifugation at 500 × g for 5 min at 4°C after each wash. The pellet was then resuspended with 1ml of sort buffer (2.5% BSA in DPBS with DTT and RNAse inhibitor) and 5 μl 7-ADD stain solution and incubated at room temperature for 10 minutes, followed by a 20-minute incubation on ice. Following centrifugation, the nuclei were washed twice with 500 μl sort buffer and filtered through a 40 μm Flowmi Cell Strainer after each wash. Nuclei were then centrifuged at 500 × g for 5 min at 4°C and resuspend in diluted Nuclei Buffer (10× Genomics, PN-2000153, add 0.1% RNase inhibitor). FACS sorted and counted using a Countess® II Automated Cell Counter.
Library preparation. The multi-omics snRNA-seq and snATAC-seq libraries were constructed using the Chromium Next GEM Single Cell Multiome ATAC + Gene Expression Reagent Bundle Kit (Cat# 1000283, 10x Genomics) according to the manufacturer’s guidelines. Libraries were sequenced on Illumina NovaSeq 6000 instruments using 50 bp paired-end sequencing targeting 6000 cells per sample with a depth of 60000 reads per cell according to the manufacturer’s recommendations (10x Genomics).

Multi-omics single nucleus RNA-seq and ATAC-seq analysis
Filtered gene count and peak count matrices for each treatment condition were generated using the 10X Cell Ranger ARC software package version 2.0.1. Quality control (QC) metrics are shown in Figure S1B–C. Downstream analysis was performed using the R libraries Seurat v4 (RNA) and Signac v1 (ATAC)26.
Nuclei with fewer than 200 genes expressed, and greater than 5000 genes expressed were excluded from the snRNA-seq analysis. After pre-processing, the Seurat pipeline was used to perform log normalization, identify highly variable genes, and perform principal component analysis23. The integrated dataset was clustered using community-based clustering at resolutions from 0.1 to 1 through the FindCluster function. UMAP dimensionality reduction was performed with the RunUMAP function using 30 PCs corrected by Harmony. Differentially expressed genes were identified using MAST24 from FindMarkers function in Seurat. Neratinib/Combination and DMSO/Fulvestrant signatures were identified by differentially expressed (DE) genes (average_log2FC ≥ 0.25 and FDR-adjusted p-value < 0.01) in a pair-wise comparison between clusters enriched for nuclei from the neratinib and combination treatments (clusters 0,2,5) against clusters enriched for cells in the DMSO and Fulvestrant treatments (clusters 1,4,6). Pearson correlations of BRD8 and ESR1, GREB1, and TFF1 were calculated using the R cor.test function based on the average expression of the genes in each cluster.
A joint snATAC-seq Signac object across all treatments was constructed by constructing a Seurat object from the filtered peak count matrix, retaining peaks that with observed counts in at least 10 cells per sample and cells with at least 200 peaks detected. We then merged the individual objects into a single one. For consistent analysis and downstream multiomic analyses, all cells used for snRNA-seq analysis were retained for snATAC-seq analysis. The resulting nuclei by peak matrix was normalized using the TF-IDF transform and dimensionally reduced using the singular value decomposition for Latent Semantic Indexing (LSI). Harmony was employed on the 2nd through 50th LSI components to remove unwanted variation using Treatment as the batch variable. Motif activity was calculated using chromVAR27 in Signac package. Pseudo-bulked motif activity differences were used to identify motifs induced by neratinib as the difference of motif activity means in DMSO-treated nuclei from the mean motif activity in neratinib-treated ones. ATAC fragments in the gene body and 2kbp upstream were counted using the GeneActivity function in Signac to calculate gene activity for a given gene. For global association analysis of neratinib-specific differentially expressed genes and differentially accessible peaks, we first found the differentially expressed genes between cells treated with neratinib against cells treated with DMSO using the MAST24 method implemented in Seurat via the FindMarkers function. We then found differentially accessible peaks by using the logistic regression framework implemented in Seurat via FindMarkers with peak counts set as a latent variable. Cells were separated by treatment (DMSO, neratinib) and the Signac LinkPeaks function was then applied to each set of cells. The resulting dataframes were filtered to genes enriched in neratinib with an adjusted p-value less than 0.1 and average log2FC greater than 0.5 and filtered to differentially accessible peaks for each treatment.

BC-GenExMiner-based analysis of gene expression and correlation
We analyzed the correlation between BRD8, ESR1, PGR, GREB1 and TFF1 expression levels using the Breast Cancer Gene-Expression Miner v4.8 (bc-GenExMiner). For this analysis, we utilized the RNA-seq dataset within the “Correlation” module of bc-GenExMiner. We specifically selected the “Targeted correlation” option and entered BRD8 and ESR1 or other genes as genes of interest. The analysis was performed across all breast cancer patient samples available in the database. The results provided Pearson’s correlation, coefficient values and associated p-values indicating the statistical significance of the correlation.
We analyzed the BRD8 gene expression using the Breast Cancer Gene-Expression Miner v4.8 (bc-GenExMiner). For this analysis, we utilized the RNA-seq dataset within the “Expression” module of bc-GenExMiner. We specifically selected the “Targeted “ option and entered BRD8 gene as gene of interest. The analysis was performed across ER positive and ER negative breast cancer patient samples available in the database. The results provided p-values indicating the statistical significance of the expression across two different subtypes. This analysis was performed according to previously described methods28,29.

RNA-seq data analysis
BT474 parental and BRD8 KO cells were treated with neratinib (200 nM) for 4h, and total RNA was extracted using the E.Z.N.A total RNA kit (Omega Bio-tek). RNA sequencing libraries were prepared using the Illumina TruSeq RNA Library Prep Kit v2 following the manufacturer’s instructions. Each library was sequenced in single read mode, 1 × 50 bp, using the HiSeq4000 platform.
RNA-seq samples (3x replicates each) were mapped using the STAR method30 with default parameters, outputting aligned and sorted BAM files with 28–40 Mio uniquely mapped reads. Raw gene counts were normalized using log counts per million (logCPM) and transformed using limma-voom method31 from the limma package32. Contrasts were constructed between BRD8 KO, and the BT474 parental samples treated with neratinib or DMSO and limma linear models were employed to find differentially expressed genes based on the contrasts. The TREAT33 method was employed to identify differentially expressed genes greater than a specified fold-change threshold by extending the empirical Bayes moderated t-statistic presented in34. Differentially expressed genes were then found using the topTreat function in limma. Genes with an FDR-corrected p-value less than 0.05 and an absolute log2FC greater than 1 were called up/down-regulated genes in each comparison condition.
Gene Set Enrichment Analysis (GSEA)35 was performed using GenePattern36 using the gene expression matrix for all genes across all samples and replicates.

Chromatin immunoprecipitation-seq (ChIP-seq) and ChIP-RT-qPCR
Parental or BRD8 KO cells in 15 cm dishes treated with or without neratinib (200 nM for 4h) were cross-linked with 1% formaldehyde for 14 min and quenched with 125 mM glycine for 5 min. Cells were washed with ice-cold PBS twice and scraped into cold PBS and collected by centrifugation (1500 rpm, 5 min). Cross-linked cells were then lysed with lysis buffer 1 (10 mM HEPES pH 7.0, 10 mM EDTA, 0.5 mM EGTA, 0.25% Triton X-100, supplemented with 0.5 mM PMSF before use) for 10 min at 4°C. The crude nuclear pellets were collected by centrifugation at 1500 rpm for 4 min at 4°C, followed by washing with buffer 2 (10 mM HEPES, pH 7.0, 200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, supplemented with 0.5 mM PMSF before use) for 10 min at 4°C. Nuclear pellets were collected by centrifugation (1500 rpm for 4 min at 4°C) and resuspended in nuclear lysis buffer (50 mM Tris-HCl pH 8.1, 10 mM EDTA, 1% SDS, supplemented with 1 mM PMSF and 1× protease inhibitor cocktail (Sigma-Aldrich) and incubated on ice for 10 min. Chromatin was sheared by sonication in an ice-water bath at 4°C using a Branson Sonifier 450 with a microtip (40% amplitude, 3 s on, 10 s off, 5 min of total pulse time). Chromatin was then centrifuged at 15,000 rpm for 15 min at 10 °C. The concentration of nuclear proteins was measured by using the BioRad Protein Assay (BioRad) and incubated with the indicated antibodies overnight. The following day, immune complexes were incubated with Protein A/G Dynabeads for 2h at 4°C, following by washing with low salt wash buffer (20 mM Tris-HCl, pH 8.1, 150 mM NaCl, 2 mM EDTA, 0.1% SDS, 1% Triton X-100) once, high salt wash buffer (20 mM Tris-HCl, pH 8.1, 500 mM NaCl, 2 mM EDTA, 0.1% SDS, 1% Triton X-100) once, LiCl wash buffer (10 mM Tris-HCl, pH 8.1, 0.25 M LiCl, 1 mM EDTA, 1% NP-40, 1% deoxycholate) once, and TE wash buffer (10 mM Tris-HCl, pH 8.0, 1 mM EDTA pH 8.0) twice. Each wash step was carried out on a tube rotator at 4°C for 5 min. The immunoprecipitate was eluted twice in freshly prepared elution buffer (1% SDS, 0.1M NaHCO3). The elution and input were digested with proteinase K (200 μg/ml) at 55 °C for 2 h, and then incubated at 65 °C overnight. DNA was purified using the Qiagen PCR Purification Kit according to the manufacture’s protocol. ChIP-qPCR was then performed. (Primer sequences are listed in Table S3)

ChIP-seq library construction
ChIP-seq libraries were constructed using the Ovation Ultralow System V2 1–16 Kit (NuGEN Technologies) according to the manufacturer’s protocol. Briefly, 2ng of ChIP DNA was end-repaired and ligated to specific adaptors. The adapted DNA was purified using Agencourt® Ampure XP Beads (Beckman Coulter, Cat. #A63881), followed by PCR amplification (13 cell cycle for 2ng DNA input). The amplified DNA was purified again with Ampure XP Beads. The quality of the library, including the library size distribution, purity, and concentration was assessed at the Sequencing Facility Center of the University of Wisconsin-Madison. Qualified libraries were sequenced on an Illumina Novaseq X Plus platform with paired end reads of 150 bases.

ChIP-seq data analysis
ENCODE ChIP-seq pipeline version 2.2.1 (https://www.encodeproject.org/chip-seq/transcription_factor/) was used to align sequencing reads to human genome assembly hg38, call peaks, and calculate fold-change signals of target over input. Library size normalization was also done by ENCODE ChIP-seq pipeline version 2.2.1, This pipeline is public available on GitHub (https://github.com/ENCODE-DCC/chip-seq-pipeline2). Only peaks from chromosomes 1 to 22 and X were considered. Protein-coding transcripts and long non-coding RNAs from GENCODE basic annotation (version 38) were used to define exons and introns. Proximal promoter is defined as 1kb upstream of a gene and distal promoter is defined as 4kb upstream of a proximal promoter. Remaining regions are defined as intergenic regions. The genomic locations of ChIP-seq peaks are determined by peaks’ overlap with the five types of genomic regions. Moreover, the bigWig files that were used to make peak signals have been deposited to GEO as described in manuscript’s ‘Data availability’ section. TF motifs enriched in peaks were computed using CentriMo from the MEME Suite (version 5.3.3) using the HOCOMOCO Human TF Mononucleotide Model (version 11) on the 500 bp flanking each peak’s center. Gene set over-representation analysis was performed using the Bioconductor package limma (version 3.60.5)’s goana function for GO terms or the kegga function for the Hallmark gene sets from the Molecular Signature Database (version 2023.1).

Mouse xenograft experiments
All animal work was performed in accordance with protocols approved by the Research Animal Resource Center of UW-Madison, and the study was compliant with ethical regulations regarding animal research. Three days prior to tumor engraftment, estrogen pellets in beeswaxA (prepared in house) were implanted into the dorsal side of the neck of 6–8-week-old female NSG mice. For PDXs HCI032 and HCI007 18, fresh human breast tumor fragments were implanted into the cleared inguinal mammary fat pads. When tumors reached ~100 mm3, mice were randomized into four groups, and treated with either vehicle, fulvestrant (5 mg/mouse/week s.c.), neratinib (40 mg/kg P.O.) or a combination of fulvestrant and neratinib for four to five weeks. For cell line xenografts, 5×106 MDA-MB-361 parental or BRD8 knockout cells were suspended in 100 μL of PBS/Matrigel (1:1), injected into the mammary fat pads. After tumors reached about 100 mm3, mice were randomized into two groups and treated with either vehicle or neratinib (40mg/kg/d, orally) treatment for two weeks. Mice were euthanized, and tumors were collected. Tumor volume was measured every 3–4 days using a Vernier caliper, and tumor volume was estimated using the following formula: V = (length × width × height × 0.5) mm3.

Immunohistochemistry (IHC) staining
IHC staining was performed as previously described37. Briefly, tumors were fixed in 3.7% formaldehyde overnight and processed for paraffin embedding by the UWCCC Experimental Pathology Lab. Slides cut from the paraffin-embedded tissue were incubated at 63 °C for 15 min, followed by three changes of xylene solution for six minutes each and four graded ethanol solutions (100%, 95%, 85% and 70%) for 6 minutes each. Antigen retrieval was performed by microwaving the slides in citric acid solution (PH 6.0), for 5 min in a microwave, and the slides were allowed to be incubated in the resulting heated solution for 23 min. After slides had cooled to room temperature, tissue was outlined using a pap pen, and peroxidases were inactivated by incubating slides in hydrogen peroxide solution (3%) (Biocare Medical, Cat#PX968 M) for 10 min. Slides were washed with distilled water and rinsed with TBS. Slides were blocked with blocking buffer (5% goat serum+2% BSA) for 1h. Slides were rinsed with Tris-buffered saline with 0.1% Tween® 20 Detergent (TBST), following by incubation with avidin (Biocare Medical, Cat#AB972L) for 15 min. Slides were rinsed with TBST and then incubated with biotin (Biocare Medical, Cat#AB972L) for 15 min. Slides were stained with primary antibody (diluted in antibody dilution buffer, CST, Cat#12378S) (Table S1) overnight and washed with TBST, and then stained with second antibody (Biocare Medical) for 15 min. Slides were rinsed with TBST and stained with streptavidin HRP label (Biocare Medical, Cat#HP604L) for 15 min. Finally, slides were stained with DAB using the DAB detection kit (Biocare Medical, Cat#BDB2004L) to visualize expression signal nuclei were counterstained with hematoxylin (Sigma, Cat# HHS16–500ML).

Statistical analysis
Statistical significance was evaluated using P values from unpaired two-tailed Student t-tests to compare two independent groups. For experiments involving two-group comparisons (e.g., control vs. treatment, wild-type vs. knockout), we used a two-tailed Student’s t-test as indicated in the figure legends. For experiments involving multiple group comparisons, we used one-way or two-way ANOVA followed by appropriate post-hoc tests (Tukey’s or Sidak’s multiple comparisons test) as now specified in the legends. For correlation analyses, we used the Pearson correlation coefficient. Data are presented as the mean ± s.e.m. or mean ± s.d as labelled in the figure legends. Significance was set to a P value of <0.05. The sample size (n) is indicated in the figure legends. Details for sequence data analyses and statistical significance are described in the specific Materials and Methods section.

Data availability
The scRNA-seq, ChIP-seq and bulk RNA-seq data generated in this study is available at the GEO database (https://www.ncbi.nlm.nih.gov/geo/) with accession number GSE230328 (scRNA-seq, bulk RNA-seq), GSE286038 (multi-omics snRNA/snATAC-seq) and GSE283036 (ChIP-seq). RNA-seq data analyzed in this study was obtained from the bc-GenExMiner database (https://bcgenex.ico.unicancer.fr). All other raw data are available upon request from the corresponding author.

Code availability
The code for analyses performed in this paper has been uploaded to Github (https://github.com/pkhatri94/Gaoetal_Omics_Analysis). Due to file size constraints at Github, the single-cell data file is saved at the Zenodo link in the readme on the front page of the github link.

Results

Characterization of HR+/HER2+ cell lines exhibiting ER/HER2 signaling crosstalk
Crosstalk between ER and HER2 signaling has been observed preclinical models38, where inhibition of one pathway activates the other as an escape mechanism39,40. To investigate the mechanism of resistance to dual ER/HER2 blockade, we first validated the bidirectional crosstalk between ER and HER2 in two ER+/HER2+ cell lines, BT474 and MDA-MB-361, by treating with ER antagonist (fulvestrant), HER2 tyrosine kinase inhibitors (TKIs) lapatinib or neratinib, or their combination41. In both cell lines, HER2 inhibition-induced ER target gene expression was reversed by fulvestrant treatment, confirming functional crosstalk (Figure 1A–B, S2A–B). Protein analyses confirmed activation of HER2 signaling upon ER inhibition (Figure S2C).
To demonstrate the functional output of signaling crosstalk, we measured cell proliferation in response to drug treatment. BT474 cells were highly responsive to HER2 inhibition but partially resistant to fulvestrant, with minimal additive effects when both pathways are targeted (Figure 1C). In contrast, MDA-MB-361 cells showed partial resistance to both ER and HER2 inhibition, with greater suppression from dual blockade (Figure 1D). Notably, BT474 cells expressed higher levels of HER2 and its downstream effectors, despite comparable ER expression with MDA-MB-361 cells (Figure 1E). These baseline differences in HER2 pathway activation explain why BT474 cells depend primarily on HER2 signaling, whereas MDA-MB-361 cells retain greater ER pathway activity and display divergent ESR1/PGR response following lapatinib treatment. Collectively, these findings demonstrate that while both ER and HER2 pathways drive MDA-MB-361 growth, HER2 signaling is the dominant proliferative driver in BT474 cells. We therefore selected BT474 cells for subsequent single-cell transcriptome analyses to investigate ER pathway activation in response to neratinib treatment.
To assess whether ER/HER2 crosstalk is conserved in vivo, we treated two HR+/HER2+ patient-derived xenograft (PDX)42 models, HCI032 and HCI007, with vehicle, fulvestrant, neratinib, or their combination (Figure 1F, S2D). IHC confirmed co-expression of ER and HER2, with variable expression across tumor cells (Figure 1G, S2E). HCI032 tumors were more sensitive to neratinib, while HCI007 showed partial responses to either treatment (Figure 1H–J, S2F–G). Dual blockade resulted in greater tumor growth inhibition and reduced Ki67 staining in both models (Figure 1H, 1K; S2F, S2H). Notably, neratinib upregulated ER and PR, whereas fulvestrant increased p-ERK (Figure 1K–L, S2H–I). Despite enhanced growth suppression, tumors remained proliferative, suggesting that while ER/HER2 crosstalk is conserved in vivo, dual blockade partially overcomes the functional crosstalk between ER and HER2, leading to stronger growth inhibitory effects.

scRNA-seq analysis of BT474 cells identifies BRD8 as a neratinib-responsive gene, which co-segregates with high expression of ESR1 and PGR
The lack of complete response to ER and HER2 inhibitors in vivo resembles the partial response to dual ER/HER2 blockade observed in HR+/HER2+ patients, implying that ER and HER2 signaling crosstalk accounts for drug resistance. Although BT474 cells express HER2 at high levels and are highly sensitive to anti-HER2 treatment (Figure 1C and 1E), bulk RNA-seq hinders the discovery of anti-HER2 responsive genes responsible for ER activation because HR+/HER2+ cells display transcriptional heterogeneity and plasticity43. To identify acute neratinib-responsive genes that are linked to ER activation, we performed single-cell RNA sequencing (scRNA-seq) after BT474 cells were treated with DMSO, fulvestrant, neratinib, or combination for 4 hours to identify the earlier responsive genes (Figure 2A). An average of 3,287 cells per treatment were analyzed after QC filtering (Figure S1A, 3A), scRNA-seq data were integrated and clustered into 10 groups, then visualized by Uniform Manifold Approximation and Projection (UMAP) (Figure 2A). Neratinib and fulvestrant induced distinct clusters: clusters 0, 2, 5 with DMSO/fulvestrant, clusters 1, 3, 7 with neratinib/combination, and clusters 4, 6, 8, 9 present across all treatments (Figure 2B–C). The top ten genes from each cell cluster are presented in the heatmap (Figure S3B). We analyzed genes up-regulated by neratinib or neratinib plus fulvestrant versus DMSO/fulvestrant and identified BRD8 among the top hits (Figure 2D). BRD8, a component of the p400/Tip60 complex, has been shown to enhance ER target gene expression13. Although oncogenic roles of BRD8 are reported in colorectal, glioblastoma, and lung cancers14,16,44, its specific role in mediating response to anti-HER2 therapy remains unexplored. The dual involvement of BRD8 in both ER signaling regulation and broader cancer progression highlights it as a compelling candidate for mediating resistance to HER2-targeted therapies. GSEA of neratinib-induced clusters showed heterogeneous cell cycle responses: cluster 1 was linked to reduced division, while clusters 3 and 7, with BRD8 upregulation, were enriched for mitotic and E2F target pathways (Figure S3C–E), supporting BRD8’s role in cell cycle progression and potential adaptive resistance to HER2-targeted therapy. However, analyses of bulk RNA-seq data from bc-GenExMiner database revealed a significant positive correlation between BRD8 and ESR1 (Figure 2E), as well as between BRD8 and ER target genes (PGR, GREB1, TFF1) in breast tumors (Figure S3F). Moreover, BRD8 expression was elevated in ER+ compared to ER− breast cancer patients, a pattern observed both across general breast cancer population (Figure S3G) and specifically within the HER2-positive subgroup (Figure S3H). These findings suggest a consistent relationship between BRD8 and estrogen receptor signaling across clinical breast cancer samples.
At the single-cell level, BRD8 was strongly induced by neratinib or combination in clusters 0, 5, and 7 (Figure 2F), alongside ESR1, PGR, and GREB1, especially in cluster 7 (Figure 2G–I). To simplify this analysis, we aggregated clusters into DMSO/fulvestrant-enriched (clusters 0, 2, 5), neratinib/combination-enriched (clusters 1, 3, 7), and mixed clusters (clusters 4, 6, 8, 9), which recapitulated the trend of coordinated induction of BRD8 and ER target genes by neratinib (Figure S4A). To examine if neratinib treatment leads to an activation of a broad array of ER target genes, we used a list of >400 putative direct ER target genes identified in an ER ChIP-seq experiment of MCF-7 cells treated with 17β-estradiol. 227 genes were detectable in our scRNA-seq data, with 88 upregulated in neratinib-enriched clusters, including ESR1, GREB1, and PGR. The top 10 are labeled (Figure S4B). Those ER target genes were enriched in neratinib as well as neratinib + fulvestrant treatment groups (Figure S4C), suggesting that fulvestrant cannot ablate the activation of ER target genes induced by neratinib. Together, these data show that neratinib treatment not only induces BRD8 expression in BT474 but also induces the expression of several ER-target genes tested at the single-cell level.

BRD8 is required for neratinib-induced ER signaling activation
Among the acute neratinib-induced genes, BRD8 was selected for further study due to its positive correlation with ESR1 in breast tumors (Figure 2E) and its role in regulating ER target genes13. We posit that BRD8 expression induced by neratinib is a prerequisite for the activation of other ER target genes. To test this, we compared BRD8 induction with other ER-interacting TFs in BT474 and MDA-MB-361 cells. RT-qPCR showed BRD8 and ER target genes, but not ER-interacting TFs (FOS, JUN, GATA3, FOXO3A), were induced at 4 h by neratinib, while at 48 h, both ER-interacting TFs and target genes were induced (Figure 3A–B, Figure S5A–F). These results confirmed BRD8 as an early neratinib-responsive gene in ER+/HER2+ cells. Neratinib-induced BRD8 expression appeared to sustain over time at both mRNA and protein levels (Figure 3C–D and Figure S5G–H). Furthermore, BRD8 was induced by multiple anti-HER2 agents, including neratinib, lapatinib, pertuzumab, trastuzumab, and T-DM1 (Figure 3E and Figure S5I), along with ESR1 and PGR (Figure 3F). Notably, neratinib also increased BRD8 expression in HR+/HER2+ HCI032 and HCI007 PDX models (Figure 3G–J). Collectively, our data supports that BRD8 is an early neratinib-responsive gene and can be induced by various types of anti-HER2 agents.
To determine whether BRD8 induction is required for ER target gene activation, we generated BRD8 knocked out (KO) BT474 and MDA-MB-361 cells using CRISPR/Cas9 (Figure 3K and Figure S5J) and treated them with neratinib. RT-qPCR results revealed that BRD8 KO significantly impaired neratinib-induced upregulation of ESR1, PGR, and GREB1 (Figure 3L and Figure S5K), indicating that BRD8 serves as a critical node linking the ER and HER2 proliferative pathways. Notably, neither ER inhibition nor activation of ER affected BRD8 expression (Figure S5L–M), suggesting that BRD8 is not a direct ER target gene but rather function as a regulator of ER activity. Next, we tested whether disturbing this positive feedback loop sensitizes HR+/HER2+ cells to anti-HER2 agents, we compared neratinib sensitivity between BRD8 KO and parental cells. BRD8 KO significantly enhanced neratinib sensitivity in both BT474 and MDA-MB-361 cells (Figure 3M and Figure S5N).
To determine whether BRD8 is also required for neratinib-induced ER activation in tumors, we employed three well-characterized HR+/HER2+ organoids derived from PDXs (PDxOs): HCI032, HCI005 and HCI00718. In HCI032 PDxOs, BRD8 was successfully knocked down (KD) using shRNA (Figure 3N). RT-qPCR analysis confirmed neratinib-induced upregulation of BRD8, ESR1 and ER target genes in control PDxOs (Figure 3O). However, BRD8 KD significantly reduced neratinib-induced expression of ESR1, PGR and GREB1 (Figure 3O). Consistent with the cell line data, BRD8 KD enhanced neratinib sensitivity in all three PDxOs (Figure 3P), while neratinib treatment decreased p-HER2 levels regardless of BRD8 status (Figure 3Q). Similar results were observed in HCI005 and HCI007 PDxOs (Figure S6A–H).
To determine whether the increased sensitivity to anti-HER2 therapy is not specific to neratinib, we treated BT474 parental and BRD8-knockout cells, as well as HCI032 control and BRD8-knockdown organoids, with multiple anti-HER2 agents. BRD8 depletion significantly decreased the IC50 values of lapatinib and T-DM1 (Figure S6I–L). Together, these data suggest that neratinib-induced BRD8 enables ER activation in cell lines and PDxO models.
Given that single-cell sequencing revealed that neratinib induces both ER-dependent and ER-independent signaling pathways, we next investigated whether BRD8 also functions in ER−/HER2+ cells. We knocked down BRD8 in SKBR3 cells (Figure S7A) and treated both control and BRD8-depleted cells with multiple anti-HER2 agents. BRD8 knockdown significantly reduced IC50 values across all anti-HER2 treatments compared with control cells (Figure S7B–D), indicating enhanced drug sensitivity. These findings suggest that BRD8 contributes to anti-HER2 therapy resistance through both ER-dependent and ER-independent mechanisms.

Multi-omics single-nucleus RNA sequencing identified coordinated induction of BRD8, ER, and ER target genes by neratinib in a patient-derived xenograft (PDX)
Having identified BRD8 as an early neratinib-responsive gene in BT474 cells with expression patterns mirroring ER and its targets, we next investigated whether BRD8 and ER are co-induced by neratinib in the same cells with a more heterogeneous tumor context. We performed multi-omics single-nucleus RNA sequencing and ATAC sequencing (snRNA/sn-ATAC seq) on HCI032 PDX tumors following long-term drug treatment (Figure 4A). We selected this model because neratinib treatment strongly inhibited HCI032 xenograft growth while inducing ER and PR activation (Figure 1H and 1L).
An average of 4624 nuclei were analyzed for each treatment group after QC filtering (Figure S1B–C, 8A). snRNA-seq data were classified into 11 clusters based on transcription profiles and visualized using UMAP (Figure 4B). As in BT474 scRNA-seq (Figure 2), neratinib and fulvestrant induced distinct clusters: 1, 4, 6 in vehicle/fulvestrant; 0, 2, 5, 10 in neratinib/combination; and 3, 7, 8, 9 shared across all treatments (Figure 4C–D). The top ten marker defining each cluster are presented in the heatmap (Figure S8B).
GSEA pathway analysis of neratinib-induced clusters revealed distinct transcriptional programs across neratinib-responsive cell populations (Figure S8C–F). Clusters 0 and 2 exhibited strong enrichment of estrogen response signatures, consistent with compensatory ER pathway activation. Clusters 0 and 5 showed downregulation of negative regulators of the ERK1/ERK2 cascade alongside activation of alternative growth-promoting pathways, suggesting MAPK-dependent bypass mechanisms. The smaller cluster 10 displayed enrichment of Notch signaling with concurrent suppression of EMT programs. These findings demonstrate that anti-HER2 treatment elicits heterogeneous pathway-level responses within the tumor cell population, with distinct subsets activating different survival and proliferative programs. This transcriptional diversity likely contributes to incomplete therapeutic response and provides a mechanistic framework for understanding how residual tumors persist under HER2-targeted therapy.
Differential expression gene (DEG) analyses identified known ER target genes, such as GREB1, PGR, and BCL2, among the top up-regulated genes in neratinib and combination treatment conditions (Figure 4E). While BRD8 and ESR1 were not among the most highly neratinib-induced genes, both showed significantly upregulation following neratinib or combination treatment across multiple clusters (Figure 4F). Critically, neratinib or neratinib plus fulvestrant treatment induced coordinated expression of BRD8 with ER and ER target genes, including GREB1, TFF1, and PGR, across multiple clusters (Figure S8G), with pronounced co-segregation observed in representative clusters 0 and 4 (Figure 4G). Correlation analysis confirmed statistically significant association between BRD8 and ESR1, GREB1, and TFF1 expression in neratinib treated cells (Figure S9A). Single-cell expression patterns of BRD8, ESR1, GREB1, and PGR visualized across UMAP clusters further demonstrated this coordinated response (Figure S9B). Together, these findings demonstrate that co-induction of BRD8 with ER and its target genes is conserved between BT474 cells and PDX tumors following neratinib treatment, highlighting a consistent role of BRD8 in mediating ER signaling activation upon anti-HER2 therapy.

Open chromatin regions are enriched in ER, FOX, and ETS family transcription factor binding motifs
To identify TF binding changes upon neratinib treatment, we performed motif enrichment analysis using the multi-omics snRNA/ATAC-Seq dataset. Figure 4H showed that the TF-associated motif activity induced by neratinib or combination treatment differed significantly from vehicle and fulvestrant treatments, consistently with the snRNA-seq data. This result indicates that a distinct set of TFs drives neratinib-responsive gene expression.
Moreover, differential motif activity analyses identified the top 30 enriched TF binding motifs at neratinib-induced genes, featuring ER, forkhead family (FOX) TFs, and ETS family TFs (Figure 4I). Several of these TFs have established roles in endocrine resistance. For example, ELF5, an ETS family TF, is a luminal progenitor marker in the mammary gland, exhibits high expression levels associated with endocrine resistance 45,46. FOXA1, a pioneer factor for ER, is overexpressed in endocrine-resistant cells and promotes transcriptional reprogramming that confer endocrine resistance47. In HER2-positive cell lines and patient tumors, FOXA1 activates alternative growth pathways in response to anti-HER2 therapies48.
Specifically, neratinib-increased gene activity (e.g., PGR, GREB1, and BCL2) coincided with the open chromatin regions enriched in ER, FOX, and ETS TF binding motifs (Figure 4J). Moreover, we analyzed the global relationship between differentially expressed genes and differentially accessible chromatin regions and found that neratinib treatment is associated with a statistically greater number of ATAC-seq peaks (p<2e-16) linked to neratinib-specific DEGs, indicating enhanced coordination between chromatin accessibility and transcriptional responses (Figure S9C). In contrast, we did not observe increased chromatin accessibility at the BRD8 locus following neratinib treatment. The BRD8 promoter maintained an accessible chromatin state under both untreated and neratinib-treated conditions (Figure S9D), indicating that enhanced chromatin accessibility is not the primary mechanism driving BRD8 transcriptional induction. Instead, this constitutively accessible chromatin state likely enables rapid transcriptional activation through alternative mechanisms, such as recruitment of specific TFs, changes in histone modifications, or activation of distal enhancer–promoter interactions.
Together, snRNA/sn-ATACseq analyses of the HCI032 PDX model validated BRD8 as a mediator of ER activation upon neratinib treatment and revealed that neratinib-induced chromatin remodeling creates an open chromatin landscape enriched for ER, FOX, and ETS transcription factor binding motifs.

BRD8 activates ER, FOX, and ETS target genes by modulating H2AZ acetylation and chromatin accessibility
To interrogate the mechanism by which BRD8 regulates ER activation, we examined whether BRD8 interacts with ER. First, co-immunoprecipitation (Co-IP) results showed that BRD8 was immunoprecipitated with an anti-ER antibody (Figure 5A). ChIP-qPCR analysis further demonstrated neratinib-induced recruitment of BRD8 to ER target genes (Figure 5B). Conversely, BRD8 knockout reduced ER recruitment to PGR and GREB1 loci following neratinib treatment (Figure 5C), indicating that BRD8 reciprocally regulates ER chromatin occupancy.
The EP400 complex is responsible for depositing histone variant H2AZ and regulates ER target gene expression13. TIP60 and BRD8 function as acetylase49 and reader protein50 within this complex, respectively. Given that BRD8 maintains H2AZ occupancy at p53 target loci through the EP400 complex16, we investigated whether BRD8 similarly activates ER target genes through H2AZ deposition and acetylation. Co-IP experiments confirmed interaction between BRD8 and H2AZ, acetylated H2AZ (H2AZac), and TIP60 in BT474 cells (Figure 5D).
To determine whether neratinib promotes the assembly of ER-BRD8-TIP60-H2AZ complex and whether BRD8 bridges ER-TIP60 interaction, we performed co-IP assays in parental and BRD8 KO BT474 cells treated with vehicle or neratinib. Neratinib treatment enhanced interactions among ER, BRD8, and TIP60, while BRD8 depletion abolished ER -TIP60 interaction (Figure 5E). Since H2AZ acetylation marks transcriptionally active chromatin 51, we examined whether BRD8 regulates ER activation by modulating H2AZac deposition upon neratinib treatment. Strikingly, BRD8 KO significantly reduced overall H2AZac levels in both BT474 and MDA-MB-361 cells (Figure 5F) and impaired neratinib-induced recruitment of H2AZac (Figure S10A) and TIP60 (Figure S10B) to ER target genes. Additionally, BRD8 KO markedly reduced H2AZ deposition at ER target genes (Figure S10C), indicating that BRD8 profoundly influences EP400 complex activity.
To assess whether BRD8’s bromodomain is required for chromatin engagement, we generated BRD8-KO BT474 cells rescued with either wild-type (WT) or bromodomain-deficient (BD-deficient) BRD8 (lacking amino acids 782–885, Figure S10D). H2AZac ChIP-qPCR at ER target genes (PGR, GREB1) showed that neratinib-induced H2AZac deposition was lost in BRD8 KO cells, restored by WT BRD8, but not by the BD-deficient mutant (Figure S10E), demonstrating the bromodomain is essential for EP400 recruitment and H2AZ acetylation. Moreover, TIP60 knockdown abolished neratinib-induced ER–BRD8 interaction (Figure S10F) and impaired recruitment of both ER and BRD8 to target genes (Figure S10G–H). These results indicate that TIP60 functions upstream to enable chromatin engagement by ER and BRD8.
Taken together, these data support the following mechanistic model: Upon neratinib treatment, TIP60-mediated histone acetylation creates a chromatin environment that recruits BRD8 via its bromodomain. BRD8 then scaffolds the ER-EP400 complex at ER target loci, stabilizing TIP60-mediated H2AZ acetylation and promoting ER-dependent transcription. The mutual dependency between BRD8 and TIP60 for ER target gene activation indicates they operate within a coordinated regulatory pathway rather than parallel compensatory mechanisms.
To assess whether BRD8 regulates TFs (i.e., ER, FOX, and ETS) via H2AZ and H2AZac, we performed ChIP-seq in parental and BRD8 KO BT474 cells treated with vehicle or neratinib. Both neratinib treatment and BRD8 KO reduced average H2AZ and H2AZac signals (Figure 5G–H), with neratinib causing more peaks lost than gained (Figure S11A). Neratinib induced a large gain of H2AZac peaks (14,533 unique), while total H2AZ occupancy remained relatively stable, highlighting H2AZac’s role in gene activation (Figure S11B, Table S4). Strikingly, ~80% of neratinib-induced H2AZ/H2AZac peaks were lost upon BRD8 KO (Figure 5I–K), establishing BRD8 as a key regulator of chromatin-associated H2AZac. GO and Hallmark analysis of BRD8-dependent H2AZac peaks revealed enrichment for estrogen response, cell cycle, and migration (Figure 5L). For instance, neratinib-induced H2AZac peaks on GREB1 (Figure 5M), oncogenic transmembrane protein SLITRK652 (Figure 5N) and cell migration regulator NET153 (Figure S11C) were abolished by BRD8 KO, whereas H2AZ signals at these genes remained largely unaffected by neratinib.
To further characterize pathways directly affected by BRD8, we analyzed H2AZ and H2AZac peaks lost upon BRD8 KO. Hallmark and GO analysis revealed effects on estrogen response, cell cycle, migration, and MAPK pathways (Figure S11D–E). For example, BRD8 KO resulted in a marked decrease in H2AZ and H2AZac peaks at E2F154 (key regulators of cell cycle) and JUN, an AP-1 component driving breast cancer proliferation and survival in breast cancer55(Figure S11F–G). Motif enrichment of neratinib-induced H2AZac peaks identified FOX family TFs (FOXK1, FOXO3, FOXM1, FOXA1), GATA3, and stem cell regulators SOX556,57/PO5F158 (Figure 5O), with 120 of 128 motifs significantly diminished upon BRD8 depletion. FOX family TFs and GATA3 are well-established regulators of both ER-dependent and -independent growth-promoting pathways in breast cancer48,59–63. Consistently, forkhead motifs were highly enriched in HCI032 PDX snATAC-seq (Figure 4I), indicating FOX TFs regulate neratinib-induced genes in both cell lines and tumors.
Further analysis of TF binding motifs at H2AZ peaks (177 motifs) and at H2AZac peaks (135 motifs) lost upon BRD8 KO revealed that ETS family TF motifs were most prominently enriched (Figure S11H). This finding aligns with the enrichment of ETS family motifs in neratinib-treated HCI032 PDX tumors identified in snATAC-seq (Figure 4I).
Together, these findings demonstrate that BRD8, which is induced by anti-HER2 treatment, coordinates with ER, FOX, and ETS family TFs to activate multiple growth-promoting pathways. BRD8 functions through the EP400 complex to drive chromatin deposition of H2AZ and H2AZac at ER or other growth factor-related TF target genes. Therefore, BRD8 serves as an essential hub for orchestrating growth-promoting signaling crosstalk under anti-HER2 treatment conditions, and its targeting may block this signaling crosstalk to enhance therapeutic efficacy.

BRD8 regulates growth and survival pathways in ER+/HER2+ cells
To profile the BRD8-regulated transcriptome with or without functional HER2, we performed bulk RNA-sequencing of BT474 and BRD8 KO BT474 cells after a 4-hour vehicle or neratinib treatment. Compared to BT474, BRD8 KO cells showed downregulation of 183 genes, including ER targets (TFF1, PGR, GREB1) and cell cycle genes like MYC, and upregulation of 628 genes, including CDKN1A, a p53 target, aligning with previously reported functions of BRD8 in glioblastoma16(Figure 6A). GSEA analysis revealed increased apoptosis and decreased MYC and estrogen response signatures in BRD8 KO cells (Figure 6B–D). Upon neratinib treatment, BRD8 KO cells showed further repression of ER targets and increased pro-apoptotic genes (e.g. TIMP2) (Figure 6E). BRD8 KO also downregulated RBM24 and AFF3 (Figures 6A, 6E). Notably, the lower expression of RBM24 and AFF3 was within the gene signature to predict better response to neratinib or lapatinib64,65, suggesting that reducing BRD8 levels may enhance HER2 inhibitor sensitivity. GSEA of neratinib-treated BRD8 KO cells confirmed that neratinib augmented p53 and apoptosis pathways upon BRD8 depletion (Figure S12A–B) while further suppressing estrogen response and cell cycle gene signatures compared to neratinib-treated parental cells (Figure 6F and 6G), supporting the collaborative effects of BRD8 ablation and neratinib treatment.
We validated RNA-seq data using RT-qPCR. GREB1 and PGR levels were decreased in BRD8 KO BT474 and MDA-MB-361 cells (Figure S12C), along with loss of activation of GFRA1, an ER phosphorylation regulator66 (Figure S12D) and upregulation of CDKN1A. Whereas E2F1 and MYC were downregulated in BT474 BRD8 KO cells (Figure 6H). Moreover, combined neratinib treatment and BRD8 KO exerted stronger suppression of E2F1 than either neratinib or BRD8 KO alone (Figure 6I). Similar results were observed in MDA-MB-361 cells and HCI007 PDxO following BRD8 knockdown or knockout (Figure S12E–F). BRD8 KO significantly reduced the proliferation of BT474 (Figure 6J) and MDA-MB-361 (Figure S12G). Moreover, BRD8 re-expression in knockout cells significantly rescued the growth inhibition phenotype, while BRD8 overexpression in parental cells enhanced proliferation (Figure S12H–I). BRD8 overexpression also restored expression of BRD8-regulated target genes (Figure S12J–K), providing functional validation of BRD8’s role in supporting tumor cell growth.
Furthermore, BRD8 knockout induced apoptosis, an effect further enhanced by neratinib treatment, as evidenced by increased Annexin V+/PI− cells (Figure 6K) and elevated cleaved caspase-3 levels (Figure S12L). In conclusion, BRD8 is critical for growth and survival of ER+/HER2+ cells, and BRD8 ablation augments the growth inhibitory effects of anti-HER2 agents.

High BRD8 gene signature correlates with poor response to anti-HER2 agents in human patients and knock out of BRD8 sensitizes neratinib-resistant xenografts to neratinib in vivo
Resistance to anti-HER2 therapy remains a clinical challenge in HR+/HER2+ breast cancer treatment, underscoring the need for reliable biomarkers to predict therapeutic responses. Our investigation identified RBM24 and AFF3 as genes co-induced with BRD8 in RNA-seq data, with expression levels directly regulated by BRD8 in BT474 and MDA-MB-361 cells (Figure 7A–B). Notably, previous studies have demonstrated that lower expression of these genes correlates with improved responses to HER2-targeting agents including neratinib and lapatinib65,67. We therefore developed a 3-gene BRD8 signature (BRD8, RBM24, and AFF3) and evaluated the predictive potential in two independent clinical trials: PAMELA68,69 and CALGB-4060170,71 (Figure 7C). In PAMELA trial68,69, a chemotherapy-free neoadjuvant regimen for HER2-positive breast cancer, patients with a BRD8high/RBM24high/AFF3high gene expression signature exhibited significantly higher levels of residual disease (RD) (Figure 7D) and were six times less likely to achieve pathological complete response (pCR) (p < 0.001) (Table 1A) compared to those with lower expression. This association held across HER2+ subtypes, independent of hormone receptor (HR) status (Table 1B), consistent with BRD8’s broader role in regulating ER-dependent and ER-independent growth pathways.
Similarly, in CALGB-40601, where the current standard of care for high-risk HER2-positive breast cancer combines chemotherapy with dual anti-HER2 therapy4. The BRD8high/RBM24high/AFF3high signature was associated with increased RD across HER2+ patients, regardless of HR status (Figure 7E and Table 1C–D). These findings support the BRD8 signature as a robust predictor of poor response to anti-HER2 therapy operating through both ER-dependent and ER-independent mechanisms.
To validate the robustness of our three-gene BRD8 signature, we tested an expanded BRD8-regulated gene set, which performed worse in both clinical trials (Figure S13A–B). Similarly, BRD8 alone predicted only pCR in HR+ patients in the PAMELA trial (Figure S13C). Overall, the combined BRD8/RBM24/AFF3 signature showed superior predictive power across both HR+ and HR− HER2+ breast cancers.
To investigate BRD8 in anti-HER2 resistance, we employed neratinib-resistant MDA-MB-361 xenografts. While parental tumors were resistant to neratinib, BRD8 knockout slowed tumor growth and restored neratinib sensitivity, leading to tumor regression (Figure 7F–H). BRD8 KO was confirmed by IHC (Figure 7I–J). Neratinib induced BRD8, ER, and PR in parental xenografts, an effect abolished by BRD8 KO (Figure 7K–P). Consistently, BRD8 KO reduced c-Myc and Ki67 levels (Figure 7Q–T). Together, our data supports the strong growth-promoting effect of BRD8 in HR+/HER2+ tumor models and identify BRD8 as a potential target to improve the sensitivity of anti-HER2 agents (Figure S14).

Discussion
Patients with HR+/HER2+ breast cancer frequently exhibit suboptimal responses to current standard-of-care treatments4, and typically show lower pCR rates when treated with de-escalated, personalized dual ER/HER25,6. Despite these challenges, limited research has focused on understanding the resistance mechanisms or identifying actionable vulnerabilities that could enable more precise and effective treatment strategies. Through single-cell RNA sequencing of an HR+/HER2+ cell line and human tumors, we identified BRD8 as a gene consistently and rapidly induced across all tested anti-HER2 agents. BRD8 functions as a critical hub orchestrating crosstalk between ER and HER2 signaling pathways. Mechanistically, BRD8 scaffolds the ER-EP400/TIP60 complex at chromatin, driving H2AZ acetylation and ER-dependent gene expression upon anti-HER2 therapy. Beyond ER signaling, BRD8 regulates FOX and ETS family TFs that activate ER-independent growth and survival pathways, thereby conferring resistance to anti-HER2 therapies. Importantly, BRD8 gene signature successfully predicts clinical response to anti-HER2 therapies, establishing BRD8 as both a biomarker of resistance and a therapeutic vulnerability that could be targeted to enhance the efficacy of dual ER/HER2 blockade strategies.
Resistance to dual ER/HER2 blockade therapy is attributed to heterogeneous expression of ER and HER2 and extensive crosstalk between ER and HER2 signaling. Yet how anti-HER2 therapy activates ER signaling remains unclear. Our single-cell transcriptomic approach identifies BRD8, the least studied bromodomain protein in the EP400 histone acetylation complex, as co-induced with ER target genes following neratinib treatment in BT474 cells (Figure 2F–I) and an HR+/HER2+ PDX model (Figure 4F–G, Figure S9A).
Integrated snRNA-seq and ATAC-seq analyses showed that neratinib-responsive genes are associated with open chromatin enriched for ER, FOX, and ETS motifs (Figure 4I). Previous studies show FOX family members are induced by HER2 inhibition and drive resistance. For instance, lapatinib induces FOXO3a to enhance ER-dependent transcription60 and FOXO1/3 to regulate c-MYC, promoting lapatinib resistance63. FOXA1 similarly activates alternative growth pathways in HER2-positive breast cancer48. Importantly, H2AZac peak signals at FOX binding motifs were significantly reduced upon BRD8 depletion in our ChIP-seq experiments (Figure 5O), indicating that BRD8 functions cooperatively with FOX family TFs under anti-HER2 treatment conditions.
The coordinated induction of BRD8 with multiple TFs explains its potent growth-promoting activity. BRD8 KO impaired estrogen response and MYC target signatures and induced apoptosis (Figure 6B), consistent with its role in regulating growth and survival across cancers14–16. Given BRD8’s pivotal role in regulating both ER-dependent and -independent growth pathways, its depletion disrupts ER and HER2 signaling crosstalk and decreases growth of neratinib-resistant tumors in vivo (Figure 7F–7H), establishing BRD8 as a critical therapeutic vulnerability even in anti-HER2-resistant settings.
While our study demonstrates that BRD8 is transcriptionally upregulated in response to HER2 inhibition and mediates therapeutic resistance, the molecular mechanism linking HER2 inhibition to BRD8 transcriptional activation remains to be elucidated. Defining the specific transcription factors, signaling intermediates, and cis-regulatory elements responsible for this adaptive response represents an important direction for future investigation.
Regarding the mechanism by which BRD8 activates growth and survival genes upon neratinib treatment, our data indicate that BRD8 activates growth and survival genes upon neratinib treatment through coordinated action with multiple TFs and the EP400 complex. Neratinib enhances ER interaction with EP400, specifically involving TIP60 and the acetyl-lysine reader BRD8; depletion of either component disrupts ER–coactivator complex formation (Figure 5E, S10F). While BRD8 is not part of the EP400 core cryo-EM structure72, it is linked to H2AZ deposition in glioblastoma16. Importantly, our ChIP-seq data show that BRD8 regulates both H2AZ deposition and H2AZ acetylation, but H2AZ acetylation, rather than H2AZ itself, is the key driver of BRD8-mediated gene activation upon neratinib treatment. Mechanistically, BRD8’s bromodomain recruits or stabilizes EP400 at chromatin, while TIP60 is required for ER and BRD8 chromatin association, collectively promoting H2AZ acetylation and transcription factor engagement (Figure S10D–H). Neratinib-induced H2AZac peaks are enriched for forkhead TF motifs, which are sensitive to BRD8 depletion (Figures 4I, 5O). Together, these results support a model (Figure S14) in which anti-HER2 treatment induces BRD8, leading to EP400 recruitment to ER, FOX, and ETS TFs, driving growth-promoting programs and therapeutic resistance.
Patients with HR+/HER2+ breast cancer often show lower pCR rates to anti-HER2 therapy, but predictive biomarkers are limited. The 27-gene HER2DX signature65,73,74, developed by Dr. Charles M. Perou, predicts pCR and survival, and renders hormone receptor status non-predictive when applied to neoadjuvant trastuzumab-treated patients. Our three-gene BRD8 signature (BRD8, RBM24, AFF3), derived from BT474 transcriptomics, also predicts anti-HER2 response (Figures 6A, 6E, 7C–E, S13, Table 1) and was validated in PAMELA and CALGB 40601 trials, where BRD8high/RBM24high/AFF3high patients had higher residual disease. Unlike HER2DX, the BRD8 signature currently serves primarily as a predictive biomarker. High BRD8 signature scores consistently mark treatment resistance, supporting its utility to identify patients needing alternative or intensified therapies. Given that BRD8 depletion disrupts growth-promoting pathways that confer resistance to anti-HER2 therapy, targeting BRD8 in combination with HER2 inhibition represents a promising strategy to improve outcome of HER2+ breast cancer patients.

Supplementary Material
123456789101112131415161718