Comparative analysis of the AIB1 interactome in breast cancer reveals MTA2 as a repressive partner which silences E-Cadherin to promote EMT and associates with a pro-metastatic phenotype

Steroid regulated cancer cells use nuclear receptors and associated regulatory proteins to orchestrate transcriptional networks to drive disease progression. In primary breast cancer, the coactivator AIB1 promotes estrogen receptor (ER) transcriptional activity to enhance cell proliferation. The function of the coactivator in ER+ metastasis however is not established. Here we describe AIB1 as a survival factor, regulator of pro-metastatic transcriptional pathways and a promising actionable target. Genomic alterations and functional expression of AIB1 associated with reduced disease-free survival in patients and enhanced metastatic capacity in novel CDX and PDX ex-vivo models of ER+ metastatic disease. Comparative analysis of the AIB1 interactome with complementary RNAseq characterized AIB1 as a transcriptional repressor. Specifically, we report that AIB1 interacts with MTA2 to form a repressive complex, inhibiting CDH1 (encoding E-cadherin) to promote EMT and drive progression. We further report that pharmacological and genetic inhibition of AIB1 demonstrates significant anti-proliferative activity in patient-derived models establishing AIB1 as a viable strategy to target endocrine resistant metastasis. This work defines a novel role for AIB1 in the regulation of EMT through transcriptional repression in advanced cancer cells with a considerable implication for prognosis and therapeutic interventions.


Bioinformatic analysis
Gene expression Survival analysis. Previously published microarray datasets were downloaded from NCBI GEO and these include AI-treated primary breast cancer datasets (GSE55374, GSE20181, and GSE41994), tamoxifen treated datasets (GSE66532, GSE9195, GSE17705 and GSE12093) and AI-treated metastatic ER + breast cancer (GSE41994). In analysis containing multiple array platforms, data was summarized with Ensemble alternative CDF and normalized with Robust Multi-array Average (RMA), before integration using ComBat to remove dataset-specific bias. survivAll R package was used to generate visual and numerical comparisons of all possible points of separation, enabling quantitative expression data to be reliably evaluated (1).
RNA-sequencing data on patient-matched primary breast cancer and metastasis (2)(3)(4) was obtained and raw CPM counts for genes of interest were analysed for differential expression. cBioPortal (5) was utilized to obtain mutational data with respect to metastatic breast cancer cohort and the Metabric cohort. Oncoplot and tables were generated using the built-in features and standard parameters. Oncoplot and tables were generated using the built-in features and standard parameters. cBioPortal was utilized for overall survival analysis of AIB1 mRNA expression. Default settings were used with automatically generated survival probabilities. Patients with deaths due to other or unknown causes are censored at the date of death and all other patients are censored at the date of last contact.
Cistrome data browser (Cistrome DB) (6) was utilized to access ChIP-seq data (7) and visualise the relevant binding peaks using the genome browser feature. Cistrome data browser (Cistrome DB) (6) was utilized to access ChIP-seq data (7) and visualise the relevant binding peaks using the genome browser feature. Differential binding analysis of ChIP-Seq data Publicly available sequence alignment files from ChIP-Seq of AIB1,MTA2 and input control samples from MCF7 cell line were downloaded from the ENCODE portal (8) with the following identifiers: ENCFF722BQX, ENCFF976AUQ, ENCFF828SPJ, ENCFF043VGV, ENCFF583FAG, ENCFF049MPQ). Corresponding binding peak calls were downloaded from Cistrome DB (accession numbers: 101717, 101716, 63873, 63874). The DiffBind R package was used to identify a consensus peak set between AIB1 and MTA2 peak calls, differential binding analysis and visualization (9). The number of sequencing reads per ChIP sample overlapping each consensus peak were counted, and adjusted for the number of reads from input control sample. Differentially bound peaks were identified using DESeq2 and called statistically significant based on a false discovery rate (FDR) < 0.05. ChIPseeker R package was used to annotate both the consensus and differentially bound peak sets based on the hg19 reference human genome (10). ChIPseeker was further used to a) calculate and plot the percentage of peaks distributed in each gene feature and b) calculate and plot the distance of each peak relative to the transcription start site of the primary transcript for each gene. Written and informed consent was acquired prior to collection of all patient tumor tissue under RCSI Institutional Review Board approved protocol (#13/09; ICORG09/07). PDX samples were established by processing and implanting tumor fragments into the mammary fat pad of 6 week-old NOD-SCID mice supplemented with 17β-estradiol pellets implanted subcutaneously as described (12). Upon successful engraftment, the PDX model was propagated by serial transplantation and biobanked. PDX tumors T638 and T347 have been previously validated and described (13). T328 is an ER+ve endocrine resistant tumor. All samples were sampled for IDEXX testing and histological studies. HCI-05 and -11 PDX models have been previously described (14).

RIME
The rapid immunoprecipitation mass spectrometry of endogenous proteins (RIME) was utilized to purify endogenous AIB1 and identify interacting proteins. After a 72 hr steroid depletion, the cells were seeded at 70% confluency for 24 hr into steroid depleted media with no hormone stimulation. At this point the cells were cross-linked, chromatin extracted and immunoprecipitated with either AIB1 antibody (10 µg; Santa Cruz; sc-25742) or negative control IgG (10 µg; Diagenode; C15410206). Raw data was processed, analyzed and results visualized as previously described (15). Those GO terms that were called as statistically categories using gene symbol ids as input (16). The hypergeometric test was used with significant GO terms called based on a bonferroni adjusted p-value<0.05 and a q-value<0.01 to control for the false discovery rate. Significant GO terms were visualised using the dotplot function.

Chromatin Immunoprecipitation (ChIP)
Steroid starved cells were fixed, crosslinked and ChIP was carried out as described previously (17). Chromatin was sonicated using Bioruptor Pico (Diagenode) and chromatin was subjected to immunoprecipitation with the AIB1 antibody (10 µg; Santa Cruz; sc-25742), MTA2 (10 µg; Santa Cruz; sc-55566) or negative control IgG (10 µg; Diagenode; Rabbit C15410206) bound to protein A Dynabeads (ThermoScientific). The proteins were then removed from the DNA by reverse crosslinking overnight and DNA was extracted using phenol-chloroform-isoamyl alcohol. q-PCR was carried out on eluted ChIP DNA utilising SYBR Green PCR (Qiagen). Primers sequences were: CDH1, F: CGTCTATGCGAGGCCGGGT; R: AACTGACTTCCGCAAGCTCACA. The delta delta (∆∆) Ct comparative method was used for analysis and data are presented as relative fold enrichment over IgG control. ChIP-re-ChIP was performed as described above with the addition of elution step after first IP pull followed by the sequential ChIP pull with either antibody of interest or IgG (13). plasmid containing full-length AIB1 was used for overexpression studies and empty pcDNA3.1 was used as a control plasmid. SI-2 (SML1762, Sigma), new class of potent small-molecule inhibitors for AIB1, has been described previously (20). Fulvestrant (ICI-182780) and

Cell culture and Functional assays
Palbociclib (PD-0332991) were sourced from Selleckchem. Mammopshere assay. Cells were seeded in a 6 well ultra-low adherence plate (Corning, MA) (500 cells/well) and cultured in mammosphere media (DMEM F12 supplemented with EGF and B27 [Sigma]). Mammospheres were maintained and analysed as previously described (21). EMT phenotype profiling. LetR cells were steroid depleted for 72 hours and serum starved for 24 hours. The cells were plated into a 6-well plate and stimulated with 10% serum in the presence or absence of AIB1/MTA2.
After 48 hours the cells were imaged and whole cell lysate was isolated for EMT protein profiling. Migration and cell growth assays. Cellomics Cell Motility Kit (Thermo Scientific) was used to measure individual cell migration after 24 hours as per manufacturer's instructions as previously described (17). Steroid starved and siRNA transfected cells were grown in 12-well plates (2000 cells/well) and manually counted using a haemocytometer.

Patient-derived tumor explants (PDTEs)
Metastatic PDX tissue was cut and dissected into 2-4mm 3 tumor fragments and placed on 1cm 3 hemostatic gelatin dental sponges (Vetspon, Novartis) pre-soaked with human mammary epithelial media as described previously (22). The ex vivo patient-derived tumor explants (PDTEs) were cultured and tumor pieces randomized to treatment with either SI-2 (50nM), 4-OHT (10 -6 M) or DMSO for 72 hours after which they were paraffin embedded and IHC stained. IHC protein positivity levels of pAIB1, AIB1 and pER were quantified utilizing Aperio ImageScope software using the positive pixel algorithm. The investigator was not blinded during image analysis. The viability of the tumors was evaluated by utilizing ki67 as a proliferation marker to confirm viable, proliferating cells (22).

RNA-sequencing
RNA-seq was carried out in biological triplicate in MCF7 and LetR cells comparing siNT (Scramble siRNA) cells versus cells depleted of AIB1 using a SMARTpool ON-TARGETplus siRNA. RNA was extracted using Qiagen RNeasy Kit and subjected to 100bp PE sequencing using the BGISEQ 500 platform. Briefly, clean reads were mapped to reference genome using Bowtie (23) and gene expression was calculated with RSEM (24). Differentially expressed genes (DEGs) were detected with DEseq2 (25). With the KEGG pathway analysis, the pathway functional enrichment was performed using phyper, a function of R.

Protein studies
Cell nuclear fraction was lysed using NE-PER kit (Thermo-Fisher Scientific). EMT profiling experiments were carried out with whole cell lysates. In the case of PDX tumors, biological replicates were mixed, minced and lysed in ice cold RIPA buffer while sonicating with a Bioruptor Pico (Diagenode). For co-immunoprecipitation experiments, 400 µg of nuclear protein was immunoprecipitated (IP) with AIB1 (10 µg, SC-25742) or IgG (10 µg, C15410206) using Dynabeads Protein A (10002D, Life Technologies) following standard IP protocol.
Protein extracted for western blots or IPs was electrophoresed and immunoblotted with AIB1  (17). Antigen retrieval and primary antibody incubations were carried out as per manufacturer's instructions. Positivity scores were assessed and scored utilizing a modified Allred IHC scoring system and the H-score for nuclear protein expression.