BET Inhibition as a Rational Therapeutic Strategy for Invasive Lobular Breast Cancer

Invasive lobular carcinoma (ILC) is the second most common histologic subtype of breast cancer after invasive ductal carcinoma (IDC), accounting for approximately 10% to 15% of breast tumors (1). It is characterized by inactivation of E-cadherin and neoplastic cells that invade the stroma in a “single-file” pattern (2, 3). Due to their unique stealth-like growth pattern through the stroma, ILC tumors are usually diagnosed at a later stage than the more common IDC subtype (4, 5). Women with ILC are usually older, have used hormone replacement therapy, and are more likely to have hormone receptor–positive disease (∼90% are ER⁺; refs. 5, 6). ILC patients have an improved survival compared with IDC patients at an early time point; however, at a later time point of 10 years, IDC have a survival advantage over ILC patients (7). This cross-over in survival could be linked to clinical course and differences in metastasis: ILCs are more frequently bilateral and are also more likely to metastasize to the bone, peritoneum, and gastrointestinal tract (5). Several groups (8, 9), including our own (10), have also now defined distinct molecular subtypes of ILC. However, despite these clear biological and molecular differences compared with IDC, ILC are currently treated in the same manner as all other ER⁺breast cancers, with antiendocrine therapy as the first line of treatment. Like IDC, antiendocrine resistance has emerged as a significant problem in the management of ILC (6, 11). For IDC patients, adjuvant chemotherapy may offer additional benefit in aggressive cases. ILC is considered to be chemoresistant, whether aggressive cases of ILC may benefit from adjuvant chemotherapy is currently unclear (12). As such, there is a pressing need to develop tailored therapeutic options for ILC patients who relapse on endocrine therapy.

Deregulated transcription mediated by epigenetic events is a recurring theme in cancer, especially in the acquired resistance setting (13–15). The bromodomain and extraterminal domain (BET) family of proteins (BRD 2, 3, 4, and T) are a family of chromatin readers (16). BET proteins bind acetylated lysine residues on nucleosomal histones and recruit transcription factors and chromatin modifying enzymes to gene promoter and enhancers, thereby regulating transcription (17–19). Through inhibition of the BRD proteins, with drugs such as JQ1, we have gained understanding into the function of BRD proteins in both normal and cancerous cells (20). JQ1 has shown efficacy in a wide variety of cancers, including different subtypes of breast cancer, and is known to downregulate important oncogenes required for survival (21–24). Sensitivity to JQ1 is not completely understood; however, recent work has shown that maintained transcriptional repression is required for efficacy in acute myeloid leukemia (AML; ref. 25) and triple-negative breast cancer (TNBC; ref. 21). To date, there are no published reports on the sensitivity of ILC to BET inhibition, and we aimed to assess the transcriptional profile regulated by BRD proteins in ILC.

All the ILC cell lines tested were sensitive to growth inhibition following JQ1 treatment. However, two of the cell lines were intrinsically resistant to JQ1-induced apoptosis. We identified an increase dependence on BCL-2 antiapoptotic proteins following JQ1 treatment in ILC. We used RNA-sequencing (RNA-seq) to identify pathways that may lead to intrinsic JQ1 resistance and showed evidence of FGFR1 rewiring in the apoptotic-resistant ILC cell line. We show that combination treatment of JQ1 and FGFR1 is highly effective at inhibiting growth of ILC cell lines in 3D models. Next, we showed that 20% of ILC patients in the RATHER cohort had FGFR1 amplification. Interestingly, high expression of the BET gene BRD3 was associated with poor survival in the RATHER cohort and the association appeared specific for ILC, as identified using the METABRIC microarray data. In summary, we have identified that high expression of BRD3 mRNA is associated with poor survival in ILC and that BET inhibition, in combination with FGFR1 inhibition, may be an effective therapeutic combination for ILC.

The RATHER cohort consists of 61 primary ILC samples (Supplementary Table S1). Paired-end strand-specific RNA-seq was carried out on the 61 clinical samples derived from ILC tumors sourced from two biobanks (Cambridge and NKI). Read pairs were aligned to the GRCh37 genome using TopHat (version 2.0.10), and quantified against the Ensembl 75 annotation using featureCounts (version 1.4.6). DESeq2 (version 1.6.3) was used to apply a regularized log transformation, and Limma (version 3.22.7) was used to remove batch effects associated with biobank.

The METABRIC ILC cohort consists of 147 primary ILC samples, 129 of the samples were ER⁺and used in the analysis. The METABRIC ER⁺IDC cohort consists of 1,140 patient samples. The protocol for METABRIC microarray analysis has been published previously (26).

JQ1 was a kind gift from the laboratory of James Bradner at the Dana-Farber Cancer Institute. ABT-199, ABT-263, and PD173074 were purchased from Selleck Chemicals. WEHI was purchased from ChemScene. Tamoxifen and fulvestrant were purchased from Sigma-Aldrich.

For Western blotting, the antibodies used were: ER alpha (Leica Biosystems, NCL-ER-6F11), E-cadherin (Cell Signaling Technology (CST, 5296), PgR (CST, 3157), BCL-2 (Santa Cruz, sc-7382), BCL-XL (CST, 2764), MYC (Abcam, ab32072), PARP (CST, 9542), FGFR1 (CST, 9740), FGFR4 (CST, 8562), STAT5 (Santa Cruz Biotechnology, sc-74442), phoshoSTAT5 (CST, 4322), STAT3 (CST, 9139), phosphoSTAT3 (CST, 9145), α/β-tubulin (CST, 2148), β-actin (Sigma-Aldrich, A5316, Santa Cruz, sc-81178). For flow cytometry, the antibodies used were CD45-PE (BD Biosciences, 555483) and IgG-PE (BD Biosciences, 555749).

The SUM44-PE (hereon referred to as SUM44), MDA-MB-134VI (hereon referred to as MM134), OCUB-M, and CAMA-1 cell lines were obtained from the RATHER consortium and were validated at the start of the project by short tandem repeat profiling, carried out by American Type Culture Collection. All ILC cell lines and the T-47D IDC cell line were grown in RPMI-1640 supplemented with 10% FBS, 1% L-glutamine, and 1% penicillin/streptomycin. IDC cell lines ZR-751 and MCF-7 were cultured in DMEM supplemented with 10% FBS, 1% L-glutamine, 1% penicillin/streptomycin, and 10 nmol/L estradiol. IDC cell lines LCC-1 and LCC-9 were cultured in phenol-red free DMEM with 5% FBS, 1% penicillin/streptomycin. Mycoplasma testing was carried out routinely every 6 to 8 weeks.

Cell viability was assessed using the MTT (3-(4,5-dimethylthiazol-2-yl)-2,5diphenyltetrazolium bromide) assay. SUM44 and MM134 cells lines were seeded at a density of 5,000 cells/well in a 96-well plate in 200 μL media overnight at 37°C. The OCUB-M and CAMA-1 cell lines were seeded at a density of 6,000 cells/well. Cell lines were treated with JQ1 for 96 hours. MTT (1.25 mg/mL) was added to the cells and incubated at 37°C for 3 hours. The media/MTT was aspirated off, and the formazan crystals were dissolved in dimethyl sulfoxide (DMSO). Absorbance was measured at 570 nm on a SpectraMax M2 plate reader.

Apoptosis was assessed using annexin V–FITC/propidium iodide (PI) staining. SUM44 and MM134 cell lines were seeded at a density of 75,000 cells/well in a 24-well plate overnight at 37°C. The OCUB-M and CAMA-1 cell lines were seeded at 50,000 cells/well. Cell lines were treated with JQ1 for 96 hours. Cells were trypsinized and resuspended in annexin binding buffer (10 mmol/L Hepes pH 7.4, 140 mmol/L NaCl, 2.5 mmol/L CaCl₂) and stained with 0.0005 mg/mL of annexin V–FITC and 0.001 mg/mL of PI for 15 minutes and analyzed on the BD Accurri 6 plus (BD Biosciences).

Cells were seeded in 1 mL of media per well in a 24-well plate and incubated at 37°C for 24 hours. Cells were then treated with JQ1 (0.03, 0.1, 0.3, 1, 3, and 10 μmol/L) for 96 hours. Cells were then trypsinized and resuspended in 1 mL of PBS. The cell suspension was supplemented with ethanol to a final concentration of 70% with agitation and incubated on ice for 15 minutes to fix. The fixed cells were pelleted and resuspended in 500 μL of PI solution in PBS: 50 μg/mL propidium iodide, 0.1 mg/mL RNase A, and 0.05% Triton X-100 and incubated for 40 minutes at 37°C. 3 mL of PBS was added to the sample followed by centrifugation at 1,500 rpm for 5 minutes. The supernatant was removed and cells were resuspended in 500 μL of PBS for flow analysis on BD Accuri C6 and FCS Express software.

Cells were harvested and resuspended in 50 μL of RIPA lysis buffer (150 mmol/L sodium chloride, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate, 50 mmol/L Tris pH 8.0) supplemented with 1× protease inhibitors. The cells were vortexed in lysis buffer and incubated on ice for 30 to 40 minutes. Protein (15–25 μg) was resolved on 8% to 10% SDS-polyacrylamide gels and transferred to a PVDF membrane and blocked with 5% milk for 1 hour at room temperature before antibody probing.

RNA was extracted using TRI Reagent (Sigma-Aldrich) and quantified. cDNA was synthesized from 1 μg RNA using an Eppendorf Mastercycler Gradient Thermal Cycler (Eppendorf). qRT-PCR was performed in an Applied Biosystems 7500 fast real-time PCR system (Applied Biosystems) using GoTaq qRT-PCR Master Mix (Promega). qRT-PCRs were performed in technical duplicate. The primers were designed using PrimerBank (https://pga.mgh.harvard.edu/primerbank/). Primer sequences are listed in Supplementary Table S2.

The SUM44 cell line was seeded at a density of 4 million cells and the MM134 seeded at a density of 5 million cells in T75 flasks and treated for 48 hours with 1 μmol/L JQ1. RNA was extracted as previously described and cleaned using RNeasy Mini Kit (Qiagen). RNA was quantified using NANODROP 1000 (Mason Technology), and quality was validated using Bioanalyser. RNA (100 ng) was used for library preparation. Libraries were prepared as per manufacturing instructions using TruSeq Stranded mRNA Library Prep Kit for NeoPrep (Illumina). Paired-end RNA-seq was carried out using the NEXTseq 500 Sequencing System (Illumina). Paired-end data in the form of fastq files were downloaded from Illumina BaseSpace. Data were aligned to the human hg19/GRCh37 reference using STAR version 2.5.2a (27). Read counts were produced by the featureCounts tool from the SubRead package. These counts were then combined for all samples and used as input for differential gene-expression analysis. Differential expression analysis was conducted using the DESeq2 package (28) in the R statistical environment (29). A heat map and principal component analysis plot were produced for the full data set to determine the quality of the count data and similarity of samples from each condition. Fragments-per-kilobase per million reads were produced using the edgeR package (30) rpkm call. Heat maps of the top 200 DE genes were produced using Perseus software, and gene ontology analysis was carried out using the DAVID functional annotation tool (31).

Cells were transfected with nontargeting negative control (siNT3), BRD2, BRD3, BRD4, and FRS2α ON-TARGETplus siRNA SMARTpool (Dharmacon, GE Healthcare Life Sciences) using Lipofectamine 2000 (Invitrogen) according to the manufacturer's guidelines. siRNA was transfected at a concentration of 0.04 μmol/L for Western blotting analysis and at a concentration of 0.033 μmol/L for the MTT assay.

Two hundred microliters of Matrigel/well was placed in a 12-well plate and incubated at 37°C for 15 minutes. 30,000 cells/well in 2% Matrigel/RPMI of either SUM44 or MM134 cells were added on top, followed by a further incubation at 37°C for 20 minutes. Five hundred microliters of 2% Matrigel/RPMI was then added per well on top of the cells. Cells were incubated overnight at 37°C followed by drug treatment for 72 hours. Media were changed twice weekly, and each condition was imaged on days 1, 8, and 15. On Day 15, 4 μmol/L Calcein-AM in serum-free RPMI media was incubated at 37°C for 15 minutes. Media were then removed from the 3D cultures and 1 mL of Calcein-AM/RPMI mix was added/well for 30 minutes at 37°C prior to imaging.

Survival analysis was carried out using the R survival package (version 2.38-3). For the RATHER data set, Cox regression analysis on continuous expression was stratified by biobank. The tumors were grouped into low- and high-expression categories (split by median expression value) for the purposes of category-based analysis. Multivariate Cox regression analysis was carried out using covariates: gene expression, biobank, tumor size (cm), number of positive lymph nodes, age at diagnosis, and histologic grade. We reported the hazard ratio (HR), 95% confidence interval, and the P values to indicate the effect and significance of the covariates. For the METABRIC data set, a Cox model using the levels of expression (log-intensity z-scores) as a nonlinear variable was modeled with the hazard ratio plotted as a smooth function of the z-score. The χ²test was used to assess the expression of each gene. For the METABRIC survival curves, the tumors were grouped into either low- and high-expression categories (split by median expression value) or by normal expression and overexpression categories (split normal 90% vs. overexpressing 10%). We also tested the null hypothesis of having the same survival function using the log-rank test. Differential expression from the RNA-seq data following 1 μmol/L JQ1 was determined using DESeq2 that uses the Wald test to determine statistical significance. Nonlinear regression analysis was used to plot dose–response curves. T test was used to assess statistical significance following JQ1 treatment for qPCR. All data, unless stated otherwise, show ± standard error of the mean (SEM) from three independent experiments (n = 3). GraphPad prism was used to calculate statistics. *, P ≤ 0.05; **, P ≤ 0.01; ***, P ≤ 0.001; ****, P ≤ 0.0001.

Previous reports have shown that BETi is an effective treatment strategy for diverse breast cancer subtypes, in particular following acquired resistance to standard therapies (23, 24, 32). To date, no studies have assessed the sensitivity of ILC to BETi and it has been shown that ILC cell lines are de novo resistant to endocrine therapy, and therefore are an excellent model to identify new therapies for endocrine-resistant patients (33). One issue with ILC is there is an insufficiency of ER-positive ILC cell lines for use in preclinical studies (34). For this study, two published verified ER-positive ILC cell lines, SUM44 and MM134, and two “ILC-like” cell lines, OCUB-M and CAMA-1, were used. All demonstrate loss of E-cadherin expression, which is characteristic of ILC, and 3 of 4 are positive for ER (Supplementary Fig. S1A). We tested the sensitivity of ILC cell lines to JQ1, a BET family inhibitor. Importantly, JQ1 inhibited the growth of all ILC cell lines tested (Fig. 1A). JQ1 induced apoptosis in two cell lines, the SUM44 and OCUB-M cell lines, but two of the cell lines, MM134 and CAMA-1, were intrinsically resistant to JQ1-induced apoptosis (Fig. 1B; Supplementary Fig. S1B and S1C). In agreement, JQ1 induced PARP cleavage in the SUM44 and OCUB-M but not in the MM134 and CAMA-1 cell lines (Fig. 1C). We then assessed if JQ1 altered the expression of the BCL-2 family of proteins that regulate mitochondrial apoptosis. At the protein level, there was a downregulation of the antiapoptotic protein BCL-XL in all four cell lines but no change in BCL-2. There was a downregulation of the proapoptotic protein BAX in the two cell lines that were intrinsically resistant to JQ1 (Fig. 1D). We found that BCL-XL mRNA expression is maintained or increased in the two apoptotic-resistant cell lines (MM134 and CAMA1; Fig. 1E). Therefore, to determine how the pro- and antiapoptotic proteins were interacting in the cell lines intrinsically resistant to JQ1, we performed dynamic BH3 profiling following JQ1 treatment (35, 36). Interestingly, we observed an increase in the response to the HRK BH3 peptide (specifically binds BCL-XL) and the NOXA BH3 peptide (specifically binds MCL-1) indicating that prodeath signaling has occurred and is being kept in check by the antiapoptotic BCL-2 proteins that are keeping the cells alive (Fig. 1F). In summary, although JQ1 induces growth arrest in all four ILC cell lines, two of the cell lines (MM134 and CAMA-1) are intrinsically resistant to apoptosis and have an increased dependence on antiapoptotic BCL-2 family proteins following JQ1 treatment.

Based on dynamic BH3 profiling, we identified a potential therapeutic vulnerability in the intrinsically resistant ILC cells following JQ1 treatment: a dependence on antiapoptotic BCL-2 family proteins. Therefore, we assessed if inhibiting antiapoptotic proteins with the BH3 mimetic ABT-263 (inhibits BCL-2, BCL-XL, and BCL-w) was synergistic with JQ1. Interestingly, synergy was detected following combination treatment in three of four ILC cell lines, with enhanced growth inhibition also observed in the CAMA-1 cell line (Fig. 2A). Similarly, we observed an enhanced cell killing particularly in the two JQ1 apoptotic-resistant cell lines (MM134 and CAMA-1), following combination treatment of JQ1 and ABT-263 (Fig. 2B). As ABT-263 inhibits BCL-2, BCL-XL, and BCL-w, we also validated our results from the BH3 profiling, showing enhanced BCL-XL dependence, using the BCL-XL–specific inhibitor WEHI in the OCUB-M and CAMA-1 cell lines (Supplementary Fig. S2C).

BCL-2 was highly expressed at the protein level in the two JQ1 intrinsically resistant cell lines (MM134 and CAMA-1; Fig. 1D). Therefore, we also assessed if ABT-199, the FDA approved BCL-2–specific inhibitor, was synergistic with JQ1. Remarkably, given the high expression of BCL-2, we did not detect any additive effects following combination of ABT-199 with JQ1 (Supplementary Fig. S2A and S2B), indicating that inhibition of BCL-2 alone is not sufficient to induce apoptosis following JQ1 treatment in the ILC cell lines.

Next, we measured the efficacy of JQ1 and ABT-263 drug combination in long-term 3D cultures (34). Importantly, sensitivity to drugs in 3D cell culture has been reported to better represent the in vivo environment (35). We treated the JQ1-sensitive SUM44 and the JQ1-resistant MM134 cell lines with JQ1 and ABT-263 alone or in combination and imaged the cells over two weeks (Fig. 2C and E). 3D clusters of cells formed in both cell lines and were viable (Fig. 2C and E). Interestingly, in the SUM44 cells, it appeared that the cells invaded in a single-file pattern through the Matrigel. In both the SUM44 and the MM134 cell lines, JQ1 treatment significantly inhibited the number (Fig. 2D and F) and size of the cell clusters, although ABT-263 treatment alone only had a minor effect. Importantly, the combination of JQ1 and ABT-263 caused the greatest reduction in the size and number of the cell clusters in both the cell lines with a statistical difference compared with JQ1 treatment alone (SUM44 P = 0.0010; MM134 P = 0.0211). These data indicate that the combination of JQ1 and ABT-263 is more effective than JQ1 treatment alone at inhibiting the growth of ILC cell lines in 3D culture.

Recent studies (21, 25) report the continued transcriptional repression as evidence of JQ1 sensitivity. Here, we have evidence that MYC is downregulated at the protein level in all four cell lines; however, in the MM134 cells MYC expression increases again at 72 hours (Fig. 3A). This is confirmed at the mRNA level, MYC is downregulated in the two apoptotic-sensitive cell lines, although it is not significantly changed in the two apoptotic-resistant cell lines (Fig. 3B).

Next, we aimed to measure the transcriptome of both the apoptotic-sensitive (SUM44) and apoptotic-resistant (MM134) ILC cell lines following JQ1 treatment. We hypothesized that the apoptosis-resistant cells may utilize transcriptional rewiring to upregulate genes to maintain survival following JQ1 treatment. The cell lines, treatment conditions, and replicates clustered together in a heat map of differentially expressed genes (Fig. 3C). Principal component analysis demonstrates limited variance between the different replicates used for RNA-seq (Supplementary Fig. S3A). A volcano plot was generated for the differentially expressed genes for each cell line. Following JQ1 treatment, there were a large number of genes downregulated and upregulated in both cell lines (Fig. 3D). In order to identify pathways altered by JQ1, gene ontology analysis was performed using the DAVID functional annotation tool and KEGG pathway analysis (31, 37). The top 10 significantly downregulated and upregulated pathways in both the SUM44 and MM134 cell lines following JQ1 treatment are shown (Fig. 3E and F; Supplementary Fig. S3B and S3C). Common pathways downregulated in both cell lines include cell cycle, DNA replication, purine metabolism, and pyrimidine metabolism (Fig. 3E).

In order to identify factors contributing to JQ1-induced apoptotic resistance, we looked at pathways uniquely upregulated in the MM134 cell line following JQ1 treatment (Fig. 3F). Pathways altered included the MAPK signaling pathway, Wnt signaling pathway, and insulin resistance (Fig. 3F; Supplementary Fig. S3D). We validated a change in Wnt signaling following JQ1 treatment by measuring Wnt 9a, Wnt 11, and Wnt 4 in the ILC cell lines (Supplementary Fig. S3D). We found that Wnt 11 and Wnt 4 were increased in the apoptotic-resistant cell lines (MM134 and CAMA1), although they were not statistically changed in the apoptotic-sensitive cell lines (SUM44 and OCUB-M). We knocked down Wnt 11 by siRNA and detected a change at the mRNA level, but there was not a change at the protein level and it did not enhance JQ1-induced cell death (Supplementary Fig. S3E and S3F). We therefore focused our analysis on the genes in the MAPK pathway, as the fibroblast growth factor receptors (FGFR) were among the set of genes upregulated specifically in the apoptotic-resistant cell line (MM134). Previously, it was shown that approximately 12% of ILC patients have FGFR1 amplification (38). Next, we confirmed by qRT-PCR a significant increase in the expression of FGFR 1, 2, 3, and 4 in the MM134 following treatment with JQ1 at the mRNA level (Fig. 3G). Interesting, we also found a significant increase in FGFR1 at the mRNA level in the CAMA-1 cell line (the other apoptotic-resistant cell line) following JQ1 treatment, indicating that this may be a general mechanism of resistance to JQ1 in ILC.

Next, we investigated if rewiring of kinase signaling through the FGFR pathway could be maintaining the survival of the MM134 cells following JQ1 treatment. First, we knocked down FGFR 1–4 in the MM134 cells with siFRS2α (Supplementary Fig. S4A and S4B). We detected statistically significant enhanced cell killing in the siFRS2α compared with control treated with JQ1 (Supplementary Fig. S4C). At the protein level, we detected a time-dependent increase in FGFR1 with no obvious change in FGFR4, following JQ1 treatment (Fig. 4A). As FGFR1 inhibitors are in clinical trials and FGFR1 amplification has been indicated in endocrine resistance (39), we decided to test if FGFR1 rewiring following BET inhibition also occurs in IDC breast cancers. First, we assessed if the IDC cell lines were sensitive to BETi with JQ1. The dose–response curves show that the IDC cells that we tested were remarkably insensitive to JQ1 (Supplementary Fig. S4D). Similar to the JQ1-resistant ILC cell lines we detected an increase at the mRNA (Supplementary Fig. S4E) and at the protein level in FGFR1 in three of five IDC cell lines following JQ1 treatment (Supplementary Fig. S4F). These data suggest that the resistance phenotype of rewiring through receptor tyrosine kinases is generally applicable to both ILC and IDC cell lines following JQ1 treatment.

Next, we tested the combination of an FGFR1 inhibitor, PD173074, with JQ1 as this may be a more clinically relevant treatment combination for translation. We detected enhanced cell death following combing FGFRi with JQ1 in the ILC MM134 (Fig. 4B) and in the IDC cell lines (Supplementary Fig. S4G). JQ1 treatment alone caused an increase in FGFR1 downstream targets p-STAT3 and p-STAT5 in the MM134 cells. Inhibition of FGFR1 in the presence of JQ1 prevented downstream signaling, as is evidenced by reduced p-STAT3 and p-STAT5 (Fig. 4C). Next, we assessed if this combination treatment led to a long-term growth inhibition of ILC in 3D cultures (Fig. 4D). Interestingly, the FGFR1 inhibitor alone was relatively ineffective at reducing the growth of MM134 cells in 3D culture. However, combination of the FGFR1 inhibitor with JQ1 significantly inhibited growth (Fig. 4E and F). In addition, we tested the combination of JQ1 and FGFR1 inhibition in the CAMA1 cell line. Although neither of the inhibitors alone had any effect on the growth in 2D (Supplementary Fig. S5A) the combination treatment of JQ1 and FGFR1 inhibitor greatly reduced the size of the CAMA-1 spheroids in 3D with similar preliminary results of reduced number and size of spheroids in IDC cell lines (MCF7 and ZR-751; Supplementary Fig. S5B and S5C). These data indicate that JQ1 and FGFR1 inhibitor combination treatment is effective at inhibiting the growth of ILC in long-term 3D cultures.

Having confirmed the effective combination of FGFR1 and BET inhibition in vitro, next we wanted to determine if we could identify the poor responding ILC patients who may be sensitive to this combination. As part of the EU FP7 project RATHER, we obtained copy-number alterations and paired-end RNA-seq of 61 primary ER⁺ILC patients. This cohort of ILC patient tumors had accompanying clinical data with 6.8 years' median clinical follow-up (see Supplementary Table S1). We identified that 20% of patients in the RATHER cohort had an FGFR1 amplification (Fig. 5A). This is a higher number than was observed in the TCGA cohort, where only 12% of ILC patients had an FGFR1 amplification (Fig. 5A).

From the RNA-seq data, we investigated if BRD 2, 3, and 4 mRNA was associated with survival in ILC, as this had not been previously analyzed. The mRNA expression was grouped into low and high based on median value, and we found that high expression of the BRD3 transcript was significantly associated with poor outcome in ILC (Fig. 5B; P = 0.004). Similar results were found with a significant association of BRD3 mRNA with survival when using expression as a continuous variable (Supplementary Table S3). Multivariate Cox regression analysis also confirmed that BRD3 mRNA expression (Supplementary Table S4: HR 73.34; 95% CI, 1.44–3741.2; P = 0.032), age at diagnosis (HR 1.092; 95% CI, 1.01–1.19; P = 0.025) and nodal status (HR 1.207; 95% CI, 1.04–1.41; P = 0.015) were independent prognostic factors of reduced survival in ILC.

To assess if this association of BRD3 mRNA with survival was specific for ILC, we turned to the previously published microarray data from METABRIC (2,433 patient samples; ref. 26). We assessed the association of BRD mRNA expression and outcome in 129 ER⁺ILC patient samples compared with 1,140 ER⁺IDC patient samples. Using the median cutoff, high BRD4 mRNA was significantly associated with reduced survival only in ER⁺ILC (Fig. 5C; P = 0.029), there was a trend that high BRD3 mRNA was associated with poor outcome of ER⁺ILC patients but it was not significant (P = 0.121). Interestingly, using a Cox model, where the expression is considered as a continuous variable, we found that only BRD3 mRNA expression was significantly associated with poor response in the METABRIC ILC cohort (Supplementary Fig. S6A; P = 0.008). Based on the Cox model, it appeared that the cutoff for BRD3 may be higher than median in the microarray data set and hence we also assessed the data using a 90/10% cutoff and found that high BRD3 mRNA expression was associated with poor disease-specific survival in ER⁺ILC (Supplementary Fig. S6B; P = 0.001). However, we did not find an association of BRD 2, 3, and 4 mRNA with survival in ER⁺IDC assessed by median cutoff (Fig. 5D; BRD2 P = 0.289; BRD3 P = 0.077; BRD4 P = 0.116) or 90/10% cutoff (Supplementary Fig. S6B; BRD2 P = 0.29; BRD3 P = 0.17; BRD4 P = 0.27). The data from the two independent ILC cohorts of patients (Fig. 5B and C; Supplementary Fig. S6A and S6B) suggest that BRD 3/4 mRNA may play a significant role in tumor progression in ILC and that FGFR1 is commonly amplified in ILC. Combined, these data rationalize testing the therapeutic combination of BET inhibitors and FGFR1 inhibitors for the treatment of recurrent ILC patients.

ILC is an extremely understudied subtype of breast cancer, as it is routinely not separated from the more common IDC in molecular biology studies and clinical trials (1). Recently, there has been a growing interest in understanding the molecular mechanisms driving ILC biology using transcriptomic and proteomic analysis, with the aim of uncovering potential therapeutic targets (9, 10). Numerous reports in cancer (40, 41) and specifically in breast cancer (23, 24) have demonstrated the efficacy of BETi in the acquired resistance setting. Currently, there are limited treatment options for ILC patients once antiendocrine treatment fails. Indeed, the MM134 cell line has previously been shown to exhibit de novo tamoxifen resistance via recognition of tamoxifen as an agonist (33). To date, no report had assessed the sensitivity of ILC to BETi. We found that BETi with JQ1 inhibited the cell growth of all ILC cell lines tested and downregulated the important growth regulating gene MYC. This is in agreement with numerous reports showing a downregulation of MYC across different cancer types, potentially due to a preferential inhibition of superenhancers by BET inhibitors (41–43).

Although inhibition of cell growth (cytostatic agents) is an effective treatment strategy for reducing tumor size, the killing of tumor cells (cytotoxic agents) ensures that no cells are left alive to repopulate the tumor, enabling relapse. Although we observed that JQ1 was cytostatic in all cell lines, cytotoxicity was only induced in two of the cell lines (SUM44 and OCUB-M), although the other two cell lines were intrinsically resistant to JQ1-induced apoptosis (MM134 and CAMA-1). Using dynamic BH3 profiling (35, 36, 44) we determined that JQ1 induces a dependence on antiapoptotic proteins and we detected synergy with the BH3 mimetic ABT-263 (inhibits BCL-2, BCL-xl, and BCL-w) in both 2D and 3D cultures. We see a time-dependent downregulation of BCL-XL at the protein level, even in the apoptotic-resistant cell lines (MM134 and CAMA-1); however, this leads to an enhanced dependence on BCL-XL, as detected by dynamic BH3 profiling and sensitivity to the BH3 mimetic WEHI. These data are in contrast to those published in castration-resistant prostate cancer, where BCL-XL was downregulated in the JQ1 apoptotic-sensitive cells but not in the apoptotic-resistant cells (40). Similar results were found in TNBC where gain of a superenhancer was detected at the BCL-XL locus as a mechanism of maintaining BCL-XL expression and resistance to JQ1 (21). Downregulation of BCL-2 has been detected in JQ1-sensitive MLL-fusion leukemia (45), and high BCL-2 mRNA expression has been suggested as a biomarker of JQ1 sensitivity in AML (46). Interestingly, the two apoptosis-refractory cell lines had high expression of BCL-2 but this was unaltered by JQ1 treatment and we did not detect any synergy when we combined ABT-199 (BCL-2–specific inhibitor) with JQ1 in ILC. Therefore, although the apoptotic family of proteins has been implicated in sensitivity to JQ1 treatment in both hematologic and solid tumors, the mechanism of BCL-2 regulation following JQ1 treatment appears to be different in ILC.

Previous reports have shown that a compensatory transcriptional mechanism can lead to JQ1 resistance and that MYC mRNA expression was an indicator of transcriptional plasticity driving resistance (25). We found a durable reduction in MYC mRNA expression in the two apoptotic-sensitive ILC cell lines, although there was no significant reduction in the two apoptosis-resistant cell lines. To identify compensatory mechanisms that are restoring transcription in the apoptosis-resistant cell lines we performed RNA-seq following JQ1 treatment. We looked at pathways that were uniquely upregulated in the apoptosis-resistant MM134 cell line following JQ1 treatment. Enhanced signaling was identified in the Wnt pathway, which has previously been shown as a rewiring mechanism in leukemia (25, 47) and in the MAPK signaling pathway, which has also been previously identified as a resistance mechanism in ovarian cancer and colon cancer (48, 49).

Following JQ1 treatment we observed increased expression of FGFR 1, 2, 3, and 4 at the mRNA level and an increase in FGFR1 at the protein level in the ILC MM134 cells and in a series of JQ1-insensitive IDC cell lines, indicating a potential compensatory mechanism through tyrosine kinase signaling. Indeed, a recent report in neuroblastoma showed that acquired resistance to JQ1 is caused by enhancer remodeling leading to receptor tyrosine kinase activation and dependencies on PI3K signaling (50). In the list of top 20 genes with altered enhancers, both FGF1 and FGFR1 gained enhancers in the JQ1-resistant cells. Although we do not know the exact mechanism for the enhanced FGFR1 expression in the ILC and IDC cell lines following JQ1 treatment, we would hypothesize that altered rewiring of transcriptional programs may play a role. FGFR1 has previously been shown to drive endocrine resistance in luminal-type breast cancers (39) and FGFR1 amplification is deemed to have a significant role in ILC tumor cell survival (38). Indeed, in a recent study by Desmedt and colleagues comparing the genomic landscape of metastatic ILC compared with primary tumors, FGFR1 copy-number gains (together with MYC and CCND1) were found to be enriched in metastatic ILC lesions (personal communication). Interestingly, even though MM134 cells have FGFR1 amplification, treatment with an FGFR1 inhibitor alone did not cause a significant reduction in the growth of the cells in 2D or 3D culture. However, combination treatment with JQ1 and an FGFR1 inhibitor prevented downstream FGFR signaling, including phosphorylation of STAT3 and STAT5, and significantly reduced growth in 3D culture. We found that FGFR1 was amplified in approximately 20% of ILC patients in the RATHER cohort. Previously, it was shown that 15% of ER-positive breast cancer patients and approximately 5% of triple-negative breast cancers (10, 38, 51) have FGFR1 amplification, indicating that rewiring signaling through FGFR1 may be a potential targetable resistance mechanism in FGFR-amplified breast cancer patients in general.

Currently, it is not known if BRD2, BRD3, or BRD4 have independent functions in cells. Previously, it was shown that BRD3 interacts with GATA1 in erythroid cells (52). In tamoxifen-resistant ER-positive breast cancer there is also evidence that BRD3/BRD4 recruit WHSC1, a histone methyltransferase, to the ER promoter (24). Moreover, in ER-positive breast cancer, there is evidence that BRD4 is required for the recruitment of RNA polymerase II to estrogen response elements (ERE) elements (47, 48). Intriguingly, a BRD4 gene signature was identified as a predictor of survival in breast cancer (53). We show that high expression of BRD3 mRNA is associated with poor survival in ILC. We further confirmed our finding across a different platform, and again show that BRD3/BRD4 is associated with poor survival in a second ILC cohort (METABRIC), consisting of a total of 190 primary ILC samples. Whether BRD3 has a specific function in ILC is not yet known; however, high BRD3/BRD4 mRNA association with poor responding tumor highlights the rationale for specifically targeting these proteins with BET inhibitors in ILC.

Currently, there is a lack of ER-positive mouse models for investigating ILC (54). Although 3D cultures lack the complexity of the in vivo microenvironment, they are a significant improvement on 2D cultures and ultimately serve as a better physiologic model for predicting in vivo responses to treatment (55, 56). A recent study reported that ILC cell lines had varying 3D culture morphologies, further supporting drug validation in both 2D and 3D models (57). Here, for the first time, we show the utility of 3D cultures as a technological platform for assessing drug combinations in ILC. Interestingly, in the SUM44 cells we see what appears like a single-file pattern of invasion in 3D, emphasizing that this model system is more physiologically relevant in the study of ILC.

In summary, here we assess the efficacy of JQ1 as a potential therapeutic for ILC and further describe the transcriptomic landscape of BRD proteins in ILC using RNA-seq. Using pathway analysis, we show that transcriptional rewiring in apoptosis-resistant ILC is driven in part by tyrosine kinases including FGFR1. Importantly, FGFR1 is amplified in approximately 12% to 20% of ILC patients and is enriched in metastatic ILC, highlighting the clinical relevance of our findings. In conclusion, our data suggest combined iBET and FGFR1 inhibition as a rational therapeutic strategy for ILC patients.

B. Moran is an employee of OncoMark Limited. J.P. Crown is an employee of OncoMark Limited, reports receiving commercial research grants from Roche, Eisai, Boehringer Ingelheim, and Puma Biotechnologies, and speakers bureau honoraria from Eisai, Puma Biotechnologies, Boehringer Ingelheim, Pfizer, Roche, Seattle Genetics, Genomic Health, Pierre Fabre, AstraZeneca, and Merck Sharp & Dhome. C. Caldas reports receiving commercial research grants from AstraZeneca, Servier, Genentech, and Roche and is an unpaid consultant/advisory board member for AstraZeneca and Illumina. W.M. Gallagher is an employee of and holds ownership interest (including patents) in OncoMark Limited, and reports receiving commercial research grants from and is an unpaid consultant/advisory board member for Carrick Therapeutics. T. Ní Chonghaile reports receiving commercial research grants from AbbVie. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L. Walsh, R. Bernards, C. Caldas, W.M. Gallagher, D.P. O'Connor, T. Ní Chonghaile

Development of methodology: L. Walsh, K.E. Haley, S. Das, T. Ní Chonghaile

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.E. Haley, B. Mooney, F. Tarrant, A. Di Grande, S. Das, C.M. Dowling, D. Varešlija, S.-F. Chin, S. Linn, L.S. Young, K. Jirström, J.P. Crown, C. Caldas, W.M. Gallagher, D.P. O'Connor

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Walsh, K.E. Haley, B. Moran, F. Tarrant, S.F. Madden, Y. Fan, S. Das, O.M. Rueda, D. Varešlija, L.S. Young, C. Caldas, W.M. Gallagher, D.P. O'Connor, T. Ní Chonghaile

Writing, review, and/or revision of the manuscript: L. Walsh, K.E. Haley, B. Moran, S. Das, S.-F. Chin, S. Linn, J.P. Crown, C. Caldas, W.M. Gallagher, D.P. O'Connor, T. Ní Chonghaile

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Walsh, B. Moran, K. Jirström, C. Caldas

Study supervision: L.S. Young, R. Bernards, W.M. Gallagher, D.P. O'Connor, T. Ní Chonghaile

TNC is supported by Breast Cancer Now 2016NovPR849 and Wellcome Trust SEED award 202079/Z/16/Z. D.P. O'Connor is supported by the Susan G Komen Foundation under the Career Catalyst Research grant CCR18547488 and Science Foundation Ireland under the CDA award 15/CDA/3438. D.P. O'Connor and W.M. Gallagher are supported by the Irish Cancer Society Collaborative Cancer Research Centre BREAST-PREDICT Grant CCRC13GAL and the European Union Seventh Framework Programme under the RATHER project 258967. W.M. Gallagher is also supported by the Science Foundation Investigator Programme OPTi-PREDICT (15/IA/3104).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.