The multidrug resistance efflux transporter ATP-binding cassette subfamily G member 2 (ABCG2) is not only overexpressed in certain drug-resistant cancers but is also highly expressed in the mammary gland during lactation, carrying xenobiotics and nutrients into milk. We sought to investigate the molecular mechanisms involved in the upregulation of ABCG2 during lactation. Expression profiling of different mouse Abcg2 mRNA isoforms (E1a, E1b, and E1c) revealed that E1b is predominantly expressed and induced in the lactating mouse mammary gland. Despite this induction, analyses of CpG methylation status and published ChIP-seq datasets reveal that E1b promoter sequences in the virgin gland are already hypomethylated and marked with the open chromatin histone mark H3K4me2. Using a forced-weaning model to shut down lactation, we found that within 24 h there was a significant reduction in Abcg2 mRNA expression and a loss of signal transducer and activator of transcription-5 (STAT5) occupancy at the mouse Abcg2 gene. Luciferase reporter assays further showed that some of these STAT5-binding regions that contained interferon-γ-activated sequence (GAS) motifs function as an enhancer after prolactin treatment. We conclude that Abcg2 is already poised for expression in the virgin mammary gland and that STAT5 plays an important role in Abcg2 expression during lactation.
- drug transporter
the breast cancer resistance protein, referred to herein by its gene name ABCG2, is the second member of the G subfamily of the ATP-binding cassette transporter superfamily, which plays a role in cancer drug resistance through active drug efflux (9). ABCG2 is expressed not only in breast cancer but also in other solid tumors and in hematological malignancies (reviewed in Refs. 33, 42).
Besides playing a prominent role in cancer drug resistance, ABCG2 also serves as a toxin/drug efflux transporter at tissues with barrier function such as intestine, kidney, and the blood-brain barrier (40, 48). A role for ABCG2 as a toxin efflux transporter is also exemplified in the mammary gland, as ABCG2 protein is dramatically upregulated during lactation, excreting drugs and toxins into breast milk (20). This phenomenon, however, apparently puts the offspring at risk of unwarranted exposure to toxins and drugs. Whereas this seemingly paradoxical role has since been clarified in part by more recent findings that ABCG2 may function as a nutrient transporter (e.g., riboflavin) (16), the biological mechanism for the dramatic upregulation of ABCG2 during lactation is still largely unknown. The mechanisms by which ABCG2 is induced in the lactating mammary gland may inform us as to how ABCG2 is regulated in certain cancer cells and provide us with novel therapeutic strategies to combat ABCG2-associated cancer drug resistance.
In both humans and mice, transcription of ABCG2/Abcg2 can start from alternative promoters leading to at least three mRNA isoforms (E1a, E1b, and E1c) that differ in 5′-untranslated sequence but code for the same protein (32, 34, 61). Regulation of ABCG2 expression is complex. A number of mechanisms, such as regulation by estrogen receptor (10), progesterone receptor (49), aryl hydrocarbon receptor (43), and hypoxia-inducible factor 1α (24), have been explored, but none offers a clear explanation for lactation-associated upregulation of ABCG2 in the mammary gland.
Development of the mammary gland from virgin to lactation is highly dependent on hormones (36, 37). One hormone of particular importance is prolactin (PRL), which is vital for both induction and maintenance of lactation. The binding of PRL to its receptor (PRLR) on the surface of mammary epithelial cells activates a complex signaling network (5). Among the many pathways activated, the best-characterized is mediated by the tyrosine kinase Janus kinase-2 (JAK2), which phosphorylates signal transducer and activator of transcription 5 (STAT5). Phosphorylated STAT5 dimerizes and translocates to the nucleus, where it binds to DNA with interferon-γ-activated sequence (GAS) motif 5′-TTCNNNGAA-3′ to modulate gene expression (4). Milk protein genes such as β-casein (CSN2) (14) and whey acidic protein (WAP) (27) are regulated in this manner. The importance of PRLR and STAT5 in lactation is seen in knockout mice, whereby deletion of one of these proteins results in impaired mammary gland development and lactation (3, 30).
There is a growing body of evidence that the expression of milk protein genes may also be modulated by epigenetic mechanisms (39). DNA methylation, which is associated with repressed gene expression when present near important regulatory elements and promoters, has been proposed as a mechanism for tissue- and developmental stage-dependent expression of several milk protein genes (19, 38, 44, 47). Furthermore, compared with the liver, which does not express high levels of milk proteins during lactation, in mammary epithelial cells the casein gene cluster and Wap gene promoters are enriched with open chromatin histone marks such as dimethylated histone 3 on lysine 4 (H3K4me2) and acetylated histone 3 (H3Ac) during lactation (38).
We have previously shown that PRL induces ABCG2 by direct binding of STAT5 to a GAS motif near the ABCG2 gene promoter in T-47D human breast cancer cells grown in vitro (52). Here, we sought to understand the mechanism(s) regulating ABCG2 induction in the physiological context of the lactating mammary gland in vivo. To this end, we examined the expression of different Abcg2 mRNA isoforms, and investigated epigenetic and STAT5-mediated transcriptional regulation of Abcg2 in the mouse lactating mammary gland.
MATERIALS AND METHODS
Mouse HC11 mammary epithelial cells were a gift from Dr. Jason Matthews (Department of Pharmacology and Toxicology, University of Toronto). Mouse EpH4 mammary epithelial cells were generously provided by Dr. Senthil Muthuswamy (Princess Margaret Cancer Centre, University Health Network, Toronto, ON, Canada). Human T-47D breast cancer epithelial cells were purchased for the American Type Culture Collection (Manassas, VA). Recombinant human epidermal growth factor (EGF), and DMEM-F12 were purchased from Life Technologies (Burlington, ON, Canada). Human recombinant insulin, RPMI 1640 medium, and fetal bovine serum (FBS) were from Wisent (Montreal, QC, Canada). Growth factor reduced Matrigel was purchased from BD Bioscience (Mississauga, ON, Canada). Recombinant human and mouse PRL were obtained from Dr. A. F. Parlow at the National Institute of Diabetes and Digestive and Kidney Diseases National Hormone and Peptide Program, Harbor-UCLA Medical Center (Torrance, CA), stored as lyophilized powder, and dissolved in phosphate-buffered solution (PBS) to working concentrations. All other reagents, such as dexamethasone and hydrocortisone, and unless specified, were from Sigma-Aldrich Canada, (Oakville, ON, Canada).
FVB/N and C57BL/6 mice originally from Jackson Laboratory and Charles River Laboratories, respectively, were bred and housed at the Toronto Centre for Phenogenomics (TCP). All protocols performed were approved by the TCP Animal Care Committee and are in accordance with the Canadian Council on Animal Care guidelines. Female virgin and lactating (7–10 days) mice were between 12 and 14 wk old. All lactating mice were first-time mothers, and litter sizes were controlled to six pups immediately after parturition. For gene expression analyses comparing virgin and lactating mice, pups were removed 2–3 h prior to tissue collection. For forced-weaning experiments to trigger involution of the mammary gland, pups were either kept with the mother (control) or removed from the mothers at 24 and 48 h prior to tissue collection. FVB/N mice were used for all forced-weaning experiments. Tissues were collected between 9:00 AM and 12:00 noon and either snap-frozen in liquid nitrogen or immersed in RNAlater (Qiagen). Tissues were subsequently stored at −80°C.
Isolation of mammary epithelial cells.
Mouse mammary epithelial cells (MECs) were isolated using immunomagnetic negative selection as reported previously (53), with modifications. Briefly, the third and fourth pairs of mammary glands were excised from virgin FVB mice and dissociated at 37°C in Complete EpiCult-B medium (STEMCELL Technologies, San Diego, CA) supplemented with 5% FBS and collagenase-hyaluronidase (STEMCELL Technologies) for 5 h with agitation. Dissociated mammary glands were further processed as per manufacturer's instructions (STEMCELL Technologies) to obtain single cell suspensions. Cells were then incubated with antibodies against mouse CD31, CD45, Ter-119, and CD140a (eBioscience, San Diego, CA). Mouse MECs were separated by immunomagnetic negative selection using anti-rat IgG MACs MicroBeads and an autoMACS Pro Separator (Miltenyi Biotec, San Diego, CA).
Cell culture and PRL treatment.
HC11 cells were grown in RPMI 1640 medium supplemented with 10% FBS, 10 ng/ml EGF, 5 μg/ml insulin, 4.7 g/l HEPES, and 2.2 g/l sodium bicarbonate. For experiments to assess mRNA expression after PRL treatment, HC11 cells were seeded at 1.5 × 105 cells per well on 12-well plates and grown for 2–3 days to 100% confluence. Cells were then subsequently incubated in maintenance medium (RPMI 1640 supplemented with 8% FBS, 5 μg/ml insulin, 1 μM dexamethasone, 4.7 g/l HEPES, and 2.2 g/l sodium bicarbonate) as per Hovey et al. (17), containing 0.1% vol/vol PBS or induced with recombinant mouse PRL (0.1–1 μg/ml) for 72 h. EpH4 cells were grown and treated as previously described (54), with modifications. EpH4 cells were grown in DMEM-F12 supplemented with 2% FBS and 5 μg/ml insulin. For PRL treatment, cells were first serum starved for 24 h and then plated at 6 × 104 cells per well on poly-HEMA coated 24-well plates (Corning) in 0.5 ml of RPMI 1640 medium. After 24 h, 0.5 ml of treatment medium was added to each well such that EpH4 cells were treated in RPMI 1640 supplemented with 2% Matrigel, 10 μg/ml insulin, and 2 μg/ml hydrocortisone in the presence or absence of recombinant mouse PRL (1 μg/ml). Forty-eight hours after treatment, cells from three wells were sequentially pelleted by short centrifugation using a table top centrifuge, pooled, and used for gene expression analyses. T-47D cells were grown in RPMI 1640 supplemented with 10% FBS and treated with recombinant human PRL as reported previously (52). All cells were maintained in a humidified incubator at 37°C under a 5% CO2 atmosphere.
Genomic DNA isolation, bisulfite treatment, and pyrosequencing.
Genomic DNA (gDNA) was isolated from snap-frozen mammary glands (4–6 mice per group) or cultured mammary epithelial cells using the DNA QIAamp Mini Kit (Qiagen). Samples were treated with RNase A during the lysis step according to the manufacturer's instructions. Approximately 1 μg of gDNA was treated with sodium bisulfite using the EpiTect Bisulfite Kit (Qiagen) and eluted from the DNA purification column using 20 μl of EB buffer. Sequence-specific primers to the mouse Abcg2 CpG island (assayed −62/+18 region, primer sequence available upon request) and M13 biotin-labeled universal primers were used to amplify 1 μl of sodium bisulfite-treated DNA in a 25-μl reaction. The PCR product (20 μl) was subsequently pyrosequenced using PyroMark Q24 (Qiagen), and results are presented as mean %methylation ± SD from 5 CpG sites within the E1b promoter.
Gene expression analyses.
Total RNA was isolated from mouse mammary gland and liver tissues using Qiazol, treated with DNase I, and further purified using RNeasy mini columns (Qiagen). RNA was isolated from cells using the RNeasy Mini Kit coupled with on-column DNase treatment. RNA was quantitated using a Nanodrop 2000 spectrophotometer (Wilmington, DE) and integrity assessed by agarose gel electrophoresis. RNA was reverse transcribed to cDNA using oligo(dT)16 and SuperScript II (Invitrogen). Mouse Abcg2 alternative mRNA isoforms were detected by real-time reverse transcriptase-polymerase chain reaction (RT-PCR) using previously published primers (61) and Power SYBR green master mix (Life Technologies). PCR products were then inserted into pCR2.1-TOPO (Invitrogen) by TA-cloning. The resulting plasmids (pCR2.1-E1a, pCR2.1-E1b, and pCR2.1-E1c) were linearized using HindIII and serial diluted (10-fold) to generate standard curves for absolute quantitation of Abcg2 mRNA isoforms using the standard curve method. Plasmid copy number was determined using OligoCalc [http://basic.northwestern.edu/biotools/OligoCalc.html] (23). The TaqMan Universal PCR Master Mix - no AmpErase UNG and inventoried TaqMan probes and primer sets (Applied Biosystems, Life Technologies) were used to detect mouse β-casein (Csn2), whey acidic protein (Wap), cytokeratin-18 (Krt18), glyceraldehyde-3-phosphate dehydrogenase (Gapdh), and β-actin (Actb) mRNA. Relative mRNA expression was determined using the 2−ΔΔCT method. All real-time RT-PCR was performed using the ABI 7500 system (Applied Biosystems) set at 50°C for 2 min, 95°C for 10 min, and then 40 cycles of 95°C for 15 s and 60°C for 1 min.
Firefly luciferase reporter plasmids driven by different lengths of the E1b promoter and 5′-flanking region were constructed by amplifying C57BL/6 genomic DNA using Pfu Turbo (Stratagene/Agilent; Santa Clara, CA) and various forward primers and a common reverse primer. Forward primers contained a KpnI restriction site, and the common reverse primer contained a XhoI restriction site. The PCR product was treated with KpnI/XhoI, purified using a QIAquick spin column, and cloned into pGL4.10[luc2] (Promega). Firefly luciferase reporter plasmids driven by a minimum promoter under control of various STAT5 binding regions from the mouse Abcg2 gene were generated by PCR using regions of interest from mouse genomic DNA as template and by digesting PCR products with KpnI or XhoI before cloning into pGL4.23[luc2/minP] (Promega). The mAbcg2-GAS4mut1 and mAbcg2-GASmut2 plasmids, which contain a single mutation in the GAS4 motif, were constructed by digesting chemically synthesized double-stranded DNA (gBlocks Gene Fragments, Integrated DNA Technologies) with XhoI and inserting the purified digested fragment into pGL4.23. E1b promoter-driven Lucia luciferase reporter constructs were generated by amplifying the −377/+199 and −71/+199 E1b promoter region using forward and reverse primers that contain HindIII and SpeI restriction sites, respectively. The PCR product was treated with HindIII and SpeI and cloned into pCpGfree-basic-Lucia (InvivoGen), which is devoid of CpG sites throughout the entire backbone. All plasmids were sequenced and prepared for transfection using Qiagen Plasmid Midi and Maxi kits. Primer sequences are listed Table 1.
In vitro methylation of plasmid DNA.
Plasmid DNA (4 μg) was incubated with 20 U of CpG methyltransferase, M.SssI, and 640 μM S-adenylmethionine (SAM) in a 50-μl reaction containing 1× NEBuffer 2 (50 mM NaCl, 10 mM Tris·HCl, 10 mM MgCl2, and 1 mM dithiothreitol). Mock methylation reactions in which SssI methyltransferase was not added were performed in parallel. Reactions were first incubated for 4 h at 37°C followed by heat inactivation of the SssI methyltransferase by incubating for 20 min at 65°C. An additional round of methylation was performed by adding 10 U of SssI methyltransferase and 16 nmol SAM to the methylation reaction and incubating for 5 h at 37°C. Treated DNA were purified using QIAquick spin columns. Methylation status was assessed by agarose gel electrophoresis, testing for the absence of linearized plasmid after incubation with the CpG methylation-sensitive AatII that would otherwise cut a nonmethylated CpG site within the insert.
Transient transfection and luciferase reporter assay.
For all transfections, plasmids were first diluted in OPTIMEM and then incubated with 3:1 Fugene HD:DNA for 15 min at room temperature. Fugene HD-DNA complexes were then diluted with an appropriate amount of OPTIMEM and added dropwise to each well. HC11 cells were seeded at 9 × 104 cells per well on 24-well plates and grown for 24 h in growth medium prior to transfection. For deletion analysis, cells were transfected with 100 ng/well pGL4.10 and variations of this construct driven by different lengths of the E1b promoter region. Transfection efficiency was monitored by cotransfection with 1 ng/well pRL-TK, which constitutively expresses Renilla luciferase. Twenty-four hours after transfection, cells were lysed in 100 μl of phosphate lysis buffer (PLB), and 2.5 μl of lysate was used to measure luciferase activity with the Dual Luciferase Reporter Assay Kit (Promega) on a Sirius Luminometer (Berthold Detection Systems, Pforzheim, Germany). For promoter methylation experiments, cells were transfected with 50 ng/well mock or CpG methylated pCpG-free-basic-lucia reporter driven by −71/+199 or −377/+199 of the Abcg2 E1b promoter region. To control for transfection efficiency, cells were cotransfected with 2 ng/well pGL3-control (a gift from Dr. David S. Riddick, Department of Pharmacology and Toxicology, University of Toronto), which constitutively express firefly luciferase. After 24 h, medium containing secreted lucia luciferase was collected, and cells were lysed in 100 μl of PLB. Lucia luciferase activity was quantitated using 5 μl of medium and 50 μl of QUANTI-Luc (InvivoGen). Firefly luciferase activity was quantitated using 10 μl of lysate and 50 μl of luciferase assay reagent II (LARII; Promega). For functional assessment of enhancer activity, T-47D cells were transfected as previously described (52). Briefly, T-47D cells were seeded at 1.25 × 105 cells per well on 24-well plates and transfected with 300 ng/well pGL4.23 or variations of this constructs containing different STAT5 binding regions (or GAS4 mutation) of the mouse Abcg2 gene. Cells were cotransfected with 300 ng/well pcDNA3-STAT5A (a gift from Dr. Dwayne Barber, Ontario Cancer Institute, Princess Margaret Hospital, Toronto, ON, Canada) to overexpress mouse STAT5A. After transfection, cells were serum starved overnight and treated with 200 ng/ml recombinant human PRL for 24 h. Luciferase activity was measured using the Dual Luciferase Reporter Assay Kit (Promega).
Preparation of mammary gland tissue lysate.
Mammary gland tissue lysate was prepared as previously described for cells (52) but modified for tissues. Snap-frozen mammary gland tissue was cut into small pieces over dry ice and then homogenized in RIPA buffer using a Polytron PT 2100 (Kinematica AG) set at 15,000 rpm. The tissue lysate was incubated end over end at 4°C for 20 min, and insoluble debris was pelleted by centrifugation at 10,000 rpm for 10 min at 4°C. The supernatant (excluding the top fatty layer) was transferred to a new microfuge tube and centrifuged again. The resulting supernatant was aliquoted and stored at −80°C. Protein concentration was determined using the Bradford assay.
Immunodetection of proteins by SDS-PAGE/western blot.
Equal amount of protein (10 or 20 μg) was resolved in 4–12% Bis-Tris gradient gels and transferred to nitrocellulose membranes (Hybond C) using the Novex NuPAGE SDS-PAGE system (Invitrogen, Life Technologies). Membranes were blocked overnight at 4°C with 5% skim milk or 5% BSA in TBST when phosphorylated STAT5 (p-STAT5) was being detected. Blots were incubated with primary antibody for 1 h at room temperature or overnight at 4°C and then with the appropriate horseradish peroxidase-conjugated secondary antibody for 1 h at room temperature. Bands were detected using enhanced chemiluminescence and film exposure. For densitometric analysis of band intensity, images of films were captured using the Fluorchem 8000 Gel Documental System coupled to a Fisher Scientific CCD camera. Captured images were then analyzed using Image J [NIH (41)]. Antibodies used were as follows: ABCG2 (1:5,000, BXP-53, Abcam), β-actin (1:10,000, A5441, Sigma Aldrich), phospho-STAT5 (1:10,000, 71-6900, Invitrogen), STAT5A (1:10,000, sc-1081, Santa Cruz Biotechnology), and STAT5B (1:4,000, sc-1656, Santa Cruz Biotechnology).
Tissues were fixed overnight in 10% neutral buffered formalin, dehydrated, and paraffin embedded. Paraffin-embedded tissue sections mounted onto glass slides were rehydrated, and microwave/pressure cooker-mediated heat-induced antigen retrieval was performed using sodium citrate buffer (10 mM sodium citrate, 0.05% Tween 20, pH 6.0). Tissue sections were then blocked with 10% goat serum in incubation buffer (1% BSA, 0.3% Triton X-100, and 0.05% Tween 20) for 1 h. Tissues were incubated with different antibodies under conditions described below. Unless specified, all incubations with blocking buffer and antibodies were performed at room temperature in a humidified chamber. Tissue sections were incubated with secondary antibodies for 1 h. Slides were washed with PBST before incubation with the next antibody. Negative slides prepared from tissue sections of lactating mouse mammary gland incubated with secondary antibodies but without primary were processed in parallel. To immunodetect cytokeratin-18, tissue sections were blocked in 10% goat serum supplemented with 20 μg/ml F(ab′)2 goat anti-mouse (Jackson ImmunoResearch) and then incubated overnight with 1:50 anti-cytokeratin 18 (Abcam) followed by 1:100 Alexa 488-conjugated goat anti-mouse (Invitrogen). For detection of p-STAT5, slides were blocked with 10% donkey serum, incubated with 1:100 anti-p-STAT5 (Invitrogen) overnight at 4°C, and then with 1:200 Cy3-conjugated donkey anti-rabbit (Jackson ImmunoResearch). ABCG2 was detected with 1:100 anti-ABCG2 (BXP-53, Abcam) overnight at 4°C followed by 1:200 Cy3-conjugated anti-rat (Jackson ImmunoResearch). Cell nuclei were stained with 5 μM Draq5 (Cell Signaling) for 2 h. Slides were mounted with Fluorescence Mounting Medium (Dako, Agilent Technologies) and stored at 4°C in the dark. Spinning disk confocal images were acquired with a Zeiss AxioVert 200M microsope and analyzed using Volocity 6.3 (PerkinElmer).
Chromatin Immunoprecipitation (ChIP) was performed as previously described (52) but adapted for tissue. Snap-frozen mouse mammary glands were cut into small pieces (<3mm2) over dry ice. The tissue was fixed and homogenized in 1% formaldehyde in PBST (0.1% vol/vol Tween 20) at room temperature for 10 min. Fixation was quenched with 0.125 M glycine at room temperature for 5 min. Fixed tissue homogenate was centrifuged at 1,000 g for 5 min at 4°C. The tissue pellet was washed twice with ice-cold PBST containing protease inhibitor cocktail and then resuspended in 1 ml of ice-cold cell lysis buffer (5 mM PIPES, 85 mM KCl, 0.5% vol/vol NP-40 and 1× protease inhibitor cocktail). After 20 min on ice, nuclei were pelleted at 1,000 g for 5 min at 4°C. The resulting pellet was resuspended in 400 μl of TSEI buffer (20 mM Tris, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% SDS, pH 8.0) supplemented with protease inhibitor cocktail and incubated on ice for at least 20 min before sonication. Samples were sonicated 10 times at 20% amplitude, 0.5 s on, 1.0 s off, for 20 s to a total of sonication time of 200 s using a Digital Branson Sonifier 450. After sonication, the samples were centrifuged at 13,000 rpm for 10 min at 4°C, and the supernatant (soluble chromatin) was precleared with 40 μl of 50% protein A agarose beads. Precleared soluble chromatin (150 μl) was then incubated with 2 μg IgG or Stat5 antibody (N-20x, Santa Cruz Biotechnology) at 4°C overnight. Immune complexes were precipitated with 40 μl of 50% protein A agarose beads for 2 h on a rotating wheel at 4°C. Beads were pelleted and then washed three times with TSEI, once with TSEII (20 mM Tris, 500 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% SDS, pH8.0), once with LiCl buffer (20 mM Tris, 250 mM LiCl, 1 mM EDTA, 1% NP-40, 1% sodium deoxycholate, pH8.0), and twice with TE buffer. Immune complexes were eluted from the beads by incubating with 120 μl of 1%SDS in TE buffer on a rotating wheel at room temperature for 4 h. Eluted complexes and input DNA (4%) was incubated at 65°C overnight to reverse cross-links and then treated with 55 μg proteinase K (ThermoScientific) for 1 h at 52°C. DNA was purified using EZ-10 spin columns (BioBasic) and eluted twice with 50 μl of EB buffer (Qiagen). Recruitment was determined by real-time PCR using 2 μl of eluted ChIP DNA or diluted input DNA (1:4) in a 20-μl reaction containing Power SYBR green mix (Applied Biosystems) and 400 nM each of forward and reverse primers. Primer sequences are listed in Table 1.
Analyses of published ChIP-seq datasets.
Preprocessed ChIP-seq datasets from Rijnkels et al. (2013) (38), already aligned to the mm9 reference genome by the authors, were downloaded from the Gene Ontology Omnibus site (http://www.ncbi.nlm.nih.gov/geo; GEO Series GSE25105). Multiple replicates for MEC and liver samples were concatenated before processing. Data reads mapped to chromosome 6 were extracted and processed using MACS peak-finding algorithm version 1.4.2 (59), comparing histone mark data to the matching background control data. Data for lactating mammary cells were also compared directly with the matching data from virgin cells. The algorithm parameters were adjusted for tag size 36 bp, and for the effective-genome size reduced by the ratio 68.6% as recommended by MACS usage guidelines for the mm9 genome. Due to the input data comprising multiple replicates, all duplicate tags were retained. The paired-peaks model building was switched off, which is appropriate for potentially long histone mark regions. Other algorithm parameters were left as default values. Raw ChIP sequencing datasets corresponding to the Yamaji et al. (2013) study (55) (GEO data series GSE40930) were retrieved in Fastq format via the DDBJ resource (http://www.ddbj.nig.ac.jp). Data were first aligned to a reference mouse genome (mm9) using the Bowtie algorithm (25). Sequencing reads that were mapped to chromosome 6 were then processed using MACS, comparing the data to the matching background control sets. Data corresponding to the wild-type mammary cells were also compared with matching Stat5a knockouts. The tag size and effective-genome size were set as described above, and paired-peaks model building was switched off for H3K4me3 datasets. Other parameters were left at default values. Results generated by the MACS algorithm were visually mapped onto the genomic neighborhood of Abcg2 using the UCSC Genome Browser (22). Visualization included the locations of statistically significant peaks of DNA occupancy, as well as WIGGLE data representing sequencing-fragment pileup in the original ChIP treatment data (calculated by MACS at every 10 bps, with a further smoothing window of 5 pixels applied to the image in the browser). Whenever multiple replicates were present in input data, the WIGGLE data values were scaled down appropriately. Data analysis used the publicly available resources Galaxy (2, 12, 13) and Cistrome (29), as well as custom-made software scripts for data pre- and postprocessing. Nucleotide sequences extracted from ChIP-seq datasets were interrogated for STAT5 binding sites (matrix similarity score ≥0.8) using MatInspector (www.genomatrix.de).
Data from all animal experiments are presented as means ± SD, whereas in vitro results are expressed as means ± SE. Statistical significance was tested using Student's t-test, or analysis of variance (ANOVA) followed by post hoc pairwise multiple comparisons using Tukey's or Dunnett's test. A P value of <0.05 was considered significant. All statistical analyses were performed using SPSS Statistics v. 21 (International Business Machines).
The E1b isoform is the predominant isoform expressed in the mouse lactating mammary gland.
To gain insight into the Abcg2 mRNA isoform expression profile, we quantitated the expression of E1a, E1b, and E1c isoforms (Fig. 1A) in the mammary gland of virgin and lactating FVB and C57BL/6 mice (Fig. 1, B and C). The expression of all three Abcg2 mRNA isoforms was very low or undetectable in the virgin gland. In contrast, all three isoforms, particularly E1b, which was the main transcript expressed, were robustly detected in the lactating mammary gland. All three isoforms were robustly detected in lineage-depleted cell populations enriched for MECs isolated from the mammary glands of virgin FVB mice (Fig. 1B, top). Importantly, however, the level of RNA standardized expression was an order of magnitude lower than that observed in the epithelia-rich lactating mammary gland. This demonstrates that increased expression of Abcg2 mRNA during lactation was not only due to relative abundance of epithelial cell content during lactation but also due to Abcg2 mRNA increase within the epithelial cell population. The liver also showed E1b predominance, but, unlike in the mammary gland, expression of Abcg2 mRNA isoforms in liver did not change with lactation. These results suggest that regulatory mechanisms governing increased Abcg2 transcript expression during lactation are specific to the mammary gland.
The mouse Abcg2 gene has a CpG island in the E1b-specific promoter.
There are numerous reports that milk protein genes that are dramatically upregulated during lactation may be under epigenetic control (19, 38, 39, 44, 47). Given that Abcg2 is also highly upregulated during lactation, we hypothesized that a similar epigenetic mechanism might regulate its expression. Based on the UCSC Genome Browser (mouse mm9 assembly) default track setting for CpG islands, there is only one such island in the entire mouse Abcg2 gene. Interestingly, this CpG island (schematically shown in Fig. 2A) is located at −231/+178 relative to the transcription start site of the E1b isoform, which was the predominant isoform expressed and induced in the lactating mammary gland. To determine the functional significance of this CpG island, we first investigated E1b promoter activity in HC11 mouse mammary epithelial cells by using luciferase reporter assays. The E1b promoter region was able to drive luciferase reporter activity, and sequential deletion analysis demonstrated that a minimal promoter is localized within the −71/+199 region (Fig. 2A). To study the effect of CpG methylation on promoter activity, the −377/+199 and −71/+199 E1b promoter regions were subcloned into a novel Lucia luciferase reporter plasmid that was completely devoid of CpG sites. These constructs were then subjected to methylation of CpG sites that are present only in the E1b promoter inserts. Successful methylation was confirmed by agarose gel electrophoresis, testing for the absence of linearized plasmid after incubation with CpG methylation-sensitive AatII that would otherwise cut a nonmethylated CpG site within the insert (data not shown). Methylation of the E1b promoter region almost completely abolished reporter activity in HC11 cells (Fig. 2B). This suggests that methylation of the CpG island could potentially impact transcriptional activity of the E1b promoter. Therefore, a loss of CpG methylation at the E1b promoter in the mammary gland could potentially facilitate upregulated expression of Abcg2 during lactation.
The E1b promoter is hypomethylated in both virgin and lactating mouse mammary glands but accumulates H3K4me2 during lactation.
Given the above findings, we postulated that there is differential CpG methylation between virgin and lactating mammary glands at the E1b promoter. Unexpectedly, CpG sites within the E1b promoter region were already hypomethylated in the mammary gland of virgin mice (FVB: 3.1 ± 0.3% methylation; C57BL/6: 2.4 ± 0.2% methylation) and remained hypomethylated during lactation (FVB: 2.4 ± 0.3% methylation; C57BL/6: 2.3 ± 0.3% methylation). Moreover, these sites were hypomethylated in mammary epithelial cell fractions isolated from virgin FVB mice (1.9 ± 0.2% methylation). As expected, identical results were seen in the liver, which expressed similar levels of Abcg2 isoforms between virgin (FVB: 2.1 ± 0.6% methylation; C57BL/6: 2.2 ± 0.7% methylation) and lactation (FVB: 2.7 ± 1.0% methylation; C57BL/6: 2.2 ± 0.8% methylation). In addition to CpG methylation, the state of chromatin marked by histone modifications can also modulate gene expression. Therefore, we analyzed recently published H3K4 dimethylation (H3K4me2) ChIP-seq data for enrichment of this open chromatin histone mark at the mouse Abcg2 gene in virgin MECs and lactating mammary gland (38). H3K4me2 was present at the promoter region of E1a, E1b, and E1c in both virgin MECs and mammary gland of lactating mice (Fig. 3). However, a much more significant H3K4me2 enrichment, further characterized by a broader genomic coverage, was observed at E1b and E1c promoter regions in the lactating mammary gland compared with virgin MEC. This is consistent with our mRNA data that both isoforms are upregulated during lactation (Fig. 1). Together, these results suggest that Abcg2 is already poised for expression in the virgin mammary gland and that induction of E1b and E1c during lactation is likely mediated by transcription factors activated by lactation-associated hormones.
PRL does not induce Abcg2 mRNA in mouse MEC lines.
Previously, we demonstrated that PRL upregulated human ABCG2 mRNA in T-47D breast cancer cells, in part, through recruitment of STAT5 to a GAS motif within the ABCG2 gene (52). To determine whether PRL induces mouse Abcg2 mRNA in vitro, we treated HC11 and EpH4 mouse mammary epithelial cells with recombinant mouse PRL. Despite induction of STAT5-dependent PRL-responsive genes, Cish (cytokine-inducible SH2-containing protein) and Csn2 (β-casein), PRL did not induce Abcg2 mRNA or individual isoforms in either cell line (Fig. 4, A–C). Surprisingly, in contrast to the whole mammary gland or epithelial cells isolated directly from the mammary gland, which were hypomethylated at the E1b promoter, E1b promoter sequences were partially methylated in HC11 and EpH4 cells even prior to PRL induction (Fig. 4D). This suggested that the state of the E1b promoter is different between mammary gland cells grown in vitro and in vivo, and it may explain the lack of induction observed with PRL treatment in tissue culture.
Acute loss of STAT5 activity in the lactating mammary gland significantly reduces Abcg2 mRNA expression.
To examine the role of STAT5 in inducing mouse Abcg2 during lactation in vivo, we used a forced-weaning model to manipulate STAT5 activity in the lactating mouse mammary gland (26). In the forced-weaning model, pups are prematurely removed from the lactating mother to trigger cessation of lactation and to induce the mammary gland to undergo involution. Others have shown that changes to the mammary gland such as loss of activated STAT5 and increased STAT3 activation occurs during the first 48 h (26). There is still a high level of epithelial cells present that allow the gland to resume lactation if the suckling stimulus is restored. This model has been used by others to compare STAT5 binding between lactating and nonlactating mammary glands (at an early stage of involution) that have comparable epithelial content (7). As expected, and consistent with this model, there was a dramatic loss of activated (phosphorylated) STAT5 as early as 24 h after forced weaning (Fig. 5A). We also observed a reduction of total STAT5B (the minor isoform in the mammary gland) at 24 h and a reduction of both STAT5 isoforms at 48 h after forced weaning. This difference in STAT5 activity between lactating and nonlactating mammary glands was specific to luminal epithelial cells (Fig. 5B). The loss of activated STAT5 after forced weaning was accompanied by a significant reduction in Abcg2 mRNA expression and expression of milk protein genes (Csn2 and Wap) (Fig. 6A). This decline in total Abcg2 mRNA was reflected in a significant reduction of not only the predominant E1b isoform but also the other isoforms (Fig. 6B). Despite some changes to cell morphology, ABCG2 protein remained highly expressed in the mammary gland after forced weaning (Fig. 7A). In particular, luminal epithelial cells that express ABCG2 protein (Fig. 7B) remained largely intact even at 24 and 48 h after forced weaning. This supports the validity of the forced-weaning model as a means for obtaining mammary gland tissue with relatively intact epithelial integrity that is depleted of active STAT5.
STAT5 is recruited to the mouse Abcg2 gene in the lactating mammary gland.
To gain insight into whether STAT5 is bound to the mouse Abcg2 gene in the lactating mammary gland, we first analyzed recently published ChIP-seq data from Yamaji et al. (55). Those authors performed an elegant experiment whereby mammary fat pads from mice carrying a number of Stat5a and Stat5b gene copies were transplanted into the mammary fat pad of nude mice, which were then mated to induce pregnancy-related mammary gland development. Mammary glands were then collected at day 1 of lactation. Our analysis of this dataset showed that STAT5A was bound to nine regions along the Abcg2 gene in the donor mammary gland from wild-type mice (Fig. 8). Recruitment of STAT5A to these regions was significantly reduced in mammary glands from STAT5A knockout mice. The top five regions herein denoted as peaks 1 to 5 that showed particularly high affinity for STAT5A were selected for further analysis. With the exception of peak 2, which contains one-half of a consensus GAS motif, all other peaks contained at least one putative GAS motif (Table 2). Using ChIP, we analyzed STAT5 recruitment to these five regions in the lactating mouse mammary glands and compared it to STAT5 recruitment in nonlactating mammary glands from mice after forced weaning or from virgin mice. STAT5 was recruited to all five regions of the Abcg2 gene during lactation but not to other regions that contain a putative consensus GAS motif (GAS_AB, GAS_C, defined in silico) or a region that is completely devoid of GAS motif sequences (GAS_neg) (Fig. 9). As a whole, reduced STAT5 recruitment was noted at all five sites in Abcg2 and at known STAT5-binding GAS motifs in the Csn2 and Wap genes after forced weaning and in the virgin mammary gland. These results robustly demonstrate that STAT5 is recruited to Abcg2 in the mouse mammary gland during lactation.
Functional assessment of STAT5 binding regions for enhancer activity.
Many of the regions that bound STAT5 during lactation were located a significant distance away from all three Abcg2 gene promoters. Therefore, instead of using the native Abcg2 gene promoters, we used a luciferase reporter construct with a minimal promoter to assess the functional activity of GAS motifs (GAS 1 to 5) and surrounding regions, as well as STAT5 binding region peak 2. We note that this minimal promoter reporter system did not function in mouse HC11 cells. This result is supported by our observation that PRL failed to induce reporter activity of GAS6v2.4 (data not shown), which we had previously shown to be highly inducible by PRL in T-47D cells. For this reason, we used T-47D cells to assess functional activity of these GAS motifs/STAT5 binding regions. Among various regions examined for enhancer activity, only the reporter construct containing GAS4 was strongly induced by PRL treatment (Fig. 10A). A more modest but statistically significant induction was also observed for the reporter construct containing peak 2. We further examined the effect of overexpressing STAT5A on reporter activity. Ectopic expression of mouse STAT5a further induced reporter activity of the mAbcg2-GAS4 construct after PRL treatment (Fig. 10B). The reporter activity of mAbcg2-Peak2 and mAbcg2-GAS2 (which is present in peak 1) was also enhanced by overexpression of mouse STAT5A. This suggests that, in addition to GAS4, other GAS motifs/STAT5 binding regions, under certain defined conditions such as those in vivo, could also be functional. Detailed analysis of these GAS motifs is the subject of future work. Here, we show that single mutations in the GAS4 motif significantly attenuated reporter activity after PRL treatment (Fig. 10C). Together, these results demonstrate that some STAT5 binding regions may serve as functional enhancer elements during lactation.
Nearly one decade after Jonker et al. (2005) reported that expression of ABCG2 protein is upregulated in the lactating mammary gland (20), it remains to be determined how this occurs. Here, we show that the E1b promoter region, which regulates expression of the predominant Abcg2 isoform upregulated in the lactating mouse mammary gland, is already hypomethylated and marked with the open chromatin histone mark H3K4me2 in the virgin mammary gland. Furthermore, we show that STAT5 is bound to multiple regions along the Abcg2 gene in the mouse mammary gland during lactation.
The predominance of E1b transcript expression in the mammary gland is reminiscent of other tissues such as intestine (34) and liver (8). While most transcripts were E1b, E1c was also induced during lactation. It is interesting that this induction during lactation was tissue specific, since it was not seen in liver. This agrees with previous reports that ABCG2 protein expression does not change in mouse liver, kidney, and intestine during lactation (31), further supporting the idea that regulatory mechanisms governing Abcg2 expression during lactation are specific to the mammary gland.
To our knowledge, this is the first study to examine epigenetic mechanisms that may regulate Abcg2 in mice. In humans, CpG methylation at the ABCG2 promoter is inversely related to ABCG2 expression in myeloma cell lines and patient myeloma cells (46). Furthermore, CpG methylation at the ABCG2 promoter inhibits transcription of ABCG2 in UOK121 and UOK143 human renal carcinoma cells (45). We showed that CpG methylation dramatically reduced luciferase reporter activity driven by the E1b promoter, suggesting that methylation in this context can potentially modulate expression. However, this does not play a role in the regulation of Abcg2, since the gene was already hypomethylated in the mammary gland of virgin mice and remained hypomethylated at lactation. Similarly, we showed that promoter regions of E1b and E1c isoforms were already marked by H3K4me2 in the virgin mouse mammary gland, and that during lactation there is further enrichment for this histone mark at both promoters. It is not clear whether this enrichment of H3K4me2 is the cause or effect of enhanced E1b isoform expression during lactation. Taken together, these epigenetic marks indicate that mouse Abcg2, particularly the E1b promoter, is already poised for expression in the virgin gland and that it adopts a more open conformation during lactation, which is consistent with enhanced expression of the Abcg2 gene.
As discussed above, since Abcg2 is already poised for expression in the virgin mammary gland, external/environmental signals such as PRL, a key lactogenic hormone, which we previously showed induces ABCG2 in human T-47D cells, are likely to play a very important role in upregulation of Abcg2 during lactation (52). However, PRL failed to induce Abcg2 in PRL-responsive mouse HC11 and EpH4 mammary epithelial cells. It is well recognized that human/mouse ABCG2/Abcg2 may be regulated differently, as shown by the induction of human but not mouse ABCG2 in response to aryl hydrocarbon receptor agonists in vitro and in vivo (15, 43). Given the data presented herein and discussed below, however, species-specific regulation of Abcg2 seems to be an unlikely explanation. Jäger et al. showed that in vivo downregulation of AP-2 in the mouse mammary gland during lactation (57) was not seen in HC11 cells after differentiation with lactogenic hormones (18). In the present study, we further show that an Abcg2 CpG island is differentially methylated in mouse mammary tissues and isolated virgin MECs compared with HC11 and EpH4 cells. It is unclear whether this accounts for the lack of response to PRL at Abcg2 in HC11 and EpH4 cells, but it highlights inherent differences between MECs grown in vitro vs. those grown in vivo, where additional environmental signals can dramatically alter the biology of these cells.
We focused on studying the STAT5 pathway, since this transcription factor mediated induction of ABCG2 by PRL in human T-47D cells in vitro (52). Indeed, there is a growing body of evidence that STAT5 plays a role in Abcg2 expression in vivo. In the experimental system used by Yamaji et al., following parturition, Abcg2 transcript expression was dramatically reduced in donor mammary glands from Stat5a knockout mice compared with wild-type mice (55). In the liver, male Stat5b knockout mice similarly show reduced expression of Abcg2, compared with wild-type counterparts (6). Due to the crucial role of STAT5 in mammary gland development (30), there are limited ways to study the role of STAT5 in mammary gland gene expression during established lactation. In the present study, we used forced weaning to turn off STAT5 activity in the lactating mammary gland. This model allowed us to compare gene expression and STAT5 binding between mammary gland tissues with comparable tissue architecture. We acknowledge that during the early phase of involution, in addition to STAT5, the activities of a variety of other signaling cascades are altered (50). It remains to be determined whether these other mechanisms also contribute to the reduction of Abcg2 mRNA during early involution.
In the present study, we show that STAT5 was bound to multiple sites along the mouse Abcg2 gene during lactation. Most of these sites, with the exception of peak 2, contained classical GAS motifs. It has been reported that in mouse embryonic fibroblasts overexpressing STAT5 up to 40% of DNA regions bound by STAT5 do not contain a GAS motif (60). It has yet to be determined how STAT5 binds to regions that lack classical GAS motifs, but perhaps neighboring transcription factors can help stabilize the STAT5-DNA interaction.
Among the five regions of Abcg2 that bound STAT5, three were located more than 30 kb from the E1b promoter. This is interesting because in general gene expression level in the mammary gland immediately following parturition is correlated with STAT5 binding to the proximal region (∼+1 to −10 kb) of the gene promoter (55). This suggests that the more proximal GAS element (i.e., GAS4) may be more important. STAT5 binding to regions greater than 10 kb from the transcription start site can still have functional consequences, as with the human perforin gene, which is upregulated by interleukin-2 via STAT5 activation and recruitment to enhancer elements located 15 and 1 kb upstream of the promoter (58). It is not clear whether these STAT5 binding regions distal from the E1b promoter are an exception and to what degree, if any, they contribute to Abcg2 expression in the mammary gland at lactation. We note that based on our analysis of Yamaji et al. data (55), RNA polymerase was not only colocalized to STAT5 binding regions containing the more proximal GAS3 and GAS4 elements but also to the distal GAS5 site. Therefore, it is plausible that multiple STAT5 binding sites act together to induce Abcg2 during lactation. Additional experiments will be required to test this hypothesis.
Interestingly, despite the presence of a half-site that is sufficiently close in proximity to GAS3 to support binding of STAT5 tetramer (28), which would favor strong activation, under the conditions of our luciferase reporter assay we did not observe any enhancer activity with the reporter containing GAS3 and flanking regions that include the half-site. Instead, the STAT5 binding region that contains GAS4 in intron 1 of the E1b isoform showed the highest enhancer activity following PRL treatment. While upstream enhancer elements have traditionally received more attention, the presence of functional GAS motifs within intronic regions is a relatively well-documented phenomenon. For example, the first intron of the human neural cell adhesion molecule 2 (NCAM2) gene, which is a known target of STAT5, contains two functional GAS motifs (35). Additional examples of STAT5 target genes with a functional GAS motif in the intron region include the human and mouse FOXP3 gene (56, 62), rat Igf1 gene (51), and a variant of mouse Akt1 (7).
Interestingly, Abcg2 was not identified by Yamaji and colleagues as a gene bound by STAT5 during lactation (55). We speculate that in their original analysis the +1 to −50 kb region relative to a RefSeq (NM_011920.3) transcription start site that corresponds to the E1a promoter was analyzed, instead of E1b. For this reason, STAT5 recruitment to regions downstream from this site, which include the regulatory regions of E1b and E1c, the predominant isoforms in the mammary gland, were not accounted for. However, during preparation of this paper, the same group reported on the combined use of RNA-seq and ChIP-seq data for trimethylated H3K4 (an epigenetic mark that preferentially mark transcription start sites of active genes; see Refs. 1 and 11) and STAT5 to show that Abcg2 was transcribed from a novel promoter in mouse mammary gland samples obtained at pregnancy and at parturition (lactation day 1) (21). This novel promoter coincides with the E1b promoter region. Interestingly, STAT5 was bound to the Abcg2 gene at lactation but not during early pregnancy (day 6). Furthermore, there was an approximately twofold increase in trimethylated H3K4 at the E1b promoter at lactation (day 1) compared with early or midpregnancy (21). These findings further support our conclusions that E1b is the predominant isoform expressed during lactation and that STAT5 plays a key role in the upregulation of mouse Abcg2 in the lactating gland.
We show here for the first time that Abcg2 mRNA, in particular the E1b isoform, is induced in the mouse lactating mammary gland. This induction does not require alterations to epigenetic patterns at the CpG sites, since the E1b promoter is already hypomethylated and marked with H3K4me2 in the virgin gland. During lactation, activation of STAT5 leads to recruitment of STAT5 to multiple regions along the Abcg2 gene, which ultimately plays a role in the dramatic upregulation of this gene. Given that both the biological actions of PRL/STAT5 and the induction of ABCG2 expression in mammary tissues during lactation are well-conserved phenomena, we speculate that these findings may have broad applicability across species, including humans.
This work was supported by Grant MOP 13747 (to S. Ito) and a Canada Graduate Scholarship (to A. M. L. Wu) from the Canadian Institutes of Health Research.
No conflicts of interest, financial or otherwise, are declared by the author(s).
Author contributions: A.M.L.W., D.B., P.A.H., and S.I. conception and design of research; A.M.L.W., M.Y., P.D., and W.W. performed experiments; A.M.L.W., M.Y., P.D., and A.L.T. analyzed data; A.M.L.W., A.L.T., P.A.H., and S.I. interpreted results of experiments; A.M.L.W. and A.L.T. prepared figures; A.M.L.W. drafted manuscript; A.M.L.W., M.Y., P.D., A.L.T., W.W., D.B., S.E.E., R.W., P.A.H., and S.I. approved final version of manuscript; A.L.T., D.B., S.E.E., R.W., P.A.H., and S.I. edited and revised manuscript.
We thank Rijnkels and colleagues (38), and Yamaji and colleagues (55), for making the dataset available. We acknowledge Sarah Fodor (Toronto Centre for Phenogenomics) for helping with the animal work. Special thanks to Dr. Sanaa Choufani and Youliang Lou of the Pyrosequencing Service/Weksberg Lab at the Hospital for Sick Children for designing and performing the pryrosequencing assay. We also express gratitude to Drs. David Riddick and Jason Matthews (University of Toronto) for helpful discussion throughout the preparation of this manuscript.
- Copyright © 2014 the American Physiological Society