Septin 9 isoform expression, localization and epigenetic changes during human and mouse breast cancer progression

Fluorescence in situ hybridization (FISH) and immunofluorescence analysis using tissue microarrays (TMAs) have been described previously [22]. Flash-frozen normal breast tissue, breast tumors and matching adjacent tumor-free areas used for real-time qRT-PCR and DNA methylation analysis were either collected at Jacobi Medical Hospital, kindly provided by Dr. Adrian L. Harris (Weatherall Institute of Molecular Medicine, University of Oxford, Oxford, UK), or obtained from the Cooperative Human Tissue Network (Additional file 1, Supplementary Table 1). The samples obtained for this study are part of an approved protocol for the collection of breast tissue samples (CCI protocol 2007-433, approved by the Albert Einstein College of Medicine Committee on Clinical Investigations) and were generated from leftover biopsies. Patient consent forms were obtained in compliance with the Declaration of Helsinki.

Mouse primary tumors were dissected from polyoma virus middle T antigen (PyMT) mice [23] (gift from Dr. Jeffrey W. Pollard, AECOM, Bronx, NY, USA) at the indicated time points. Tumor-bearing glands were isolated and cut in half. One portion was processed for Western blot analysis and genetic studies, and the other was embedded in paraffin. Four-weeks-old C57B6 female mice serving as normal controls were killed, and the fourth inguinal MGs were dissected. The use of murine models for this study was approved by the Institutional Animal Care and Use Committee at AECOM (protocol 20090502).

The cell lines used for experiments are shown in Additional file 1, Supplementary Table 2. MCF10A cells were cultured as previously reported [24, 25], hTERT-HME1 and 184A1 cells were cultured in mammary epithelium basal medium supplemented with the MEGM BulletKit (CC-3150; Lonza Group Ltd, Basel, Swtizerland) and all other cell lines were grown in DMEM supplemented with 10% fetal bovine serum. Because of the relatively high level of SEPT9 mRNA in the MCF10A cells, we analyzed the same cell line obtained from two different sources, Dr. Rachel Hazan (Department of Pathology, AECOM, Bronx, NY, USA) and the American Type Culture Collection (Manassas, VA, USA) (Additional file 1, Supplementary Table 2). HCT116 is a colorectal cancer cell line that was included in the analysis as a control for our DNA methylation data, since its DNA is highly methylated [26] and because of the prognostic role of DNA methylation of SEPT9 shown in colon cancer [27].

Nine bacterial artificial chromosome (BAC) clones mapping to SEPT9, HER2 and CEP17 were obtained from the BACPAC Resources Center (Children's Hospital Oakland Research Institute, Oakland, CA, USA). Clones and their genetic mapping are shown in Additional file 2, Supplementary Figure S1A. BAC clones mapping to the mouse Sept9 gene were previously reported [3]. DNA isolation, labeling, hybridization and imaging were performed as described previously [22, 28]. Slide imaging was carried out with a Zeiss Axiovert 200 inverted microscope (Carl Zeiss MicroImaging, Inc., Thornwood, NY, USA) using SpectrumOrange (SO ™)-specific (Abbott Molecular, Des Plaines, IL, USA), Cy5-specific and 4', 6-diamidino-2-phenylindole-specific filters (Chroma Technologies, Bellows Falls, VT, USA). Twenty-five cells were analyzed for each cell line, and locus-specific signals for SO (SEPT9) and Cy5 (HER2) were manually counted and plotted for further analysis. Unsupervised hierarchical clustering and heat map generation representing the signals for each cell were performed using R version 2.8 statistical software [29]. On TMAs, signals within each cell were counted for a minimum of 20 cells and scored as follows: two copies for each locus were classified as normal; cells with two copies of CEP17 and three to four copies of SEPT9 were classified as low copy number gain (+); cells with loss of CEP17 and at least two copies of SEPT9 and cells with two copies of CEP17 and five or more copies of SEPT9 were classified as moderately high copy number gain (++); cells with more than five copies of SEPT9 were scored as high copy number gain (+++).

Fisher's exact test of independence was used to test the null hypothesis of no association between grade and copy number, and the Cochran-Mantel-Haenszel test (based on Pearson's ρ) was generated to test the null hypothesis of no linear association, given the ordinal nature of both grade and copy number categories.

Human TMAs (previously described), mouse mammary adenocarcinomas and age-matched normal tissues were baked at 60°C for 1 hour, deparaffinized in xylene (twice for 10 minutes each time) and rehydrated in a graded series of ethanol washes. Antigen retrieval was performed using Vector Antigen Unmasking Solution (H-3300; Vector Laboratories, Burlingame, CA, USA) for 20 minutes. Incubation with the following antibodies diluted in 3% goat serum/PBS was performed overnight at 37°C: anti-SEPT9 polyclonal antibody (gift from Dr. Koh-ichi Nagata, Institute for Developmental Research, Aichi, Japan [16]) detected with anti-rabbit Alexa Fluor 680 dye (Invitrogen, Carlsbad, CA, USA), monoclonal anti α-tubulin (DM1A clone; Sigma-Aldrich, St Louis, MO, USA) detected with anti-mouse Alexa Fluor 488 dye (Invitrogen); anti-nucleolin (gift of Dr. Thomas Meier, Department of Anatomy & Structural Biology, AECOM) detected with anti-mouse Alexa Fluor 488 dye (Invitrogen). All primary antibodies were used at a 1:1,000 dilution.

Stable MCF7 transfectants expressing GFP-tagged isoforms of SEPT9 were fixed in 3.7% formaldehyde for 10 minutes at room temperature. All images were acquired with an Olympus BX61 inverted microscope equipped with a UPlanSApo 63 × objective oil immersion lens (Olympus Imaging America, Inc., Melville, NY, USA), a mercury arc lamp for excitation, narrow band filters for all fluorescence emission, and a SensiCamQE CCD camera system (PCO AG, Kelheim, Germany) using a FISH view.

Total RNA was isolated from cancer cell lines, tissues and stably transfected GFP clones as reported previously [3]. cDNA was reverse-transcribed from 5 μg of total RNA using random primers and SuperScript II Reverse Transcriptase (Invitrogen). SEPT9 and GFP primers were designed with Primer3 software [30] (Additional file 1, Supplementary Table 3) along with glyceraldehyde 3-phosphate dehydrogenase (GAPDH) primers. SEPT9_v1 and SEPT9_v3 isoform primers were either as published [21] or designed by hand based on unique sequences at the 5' end. Real-Rime qRT-PCR was performed using Applied Biosystems Fast SYBR Green Master Mix and the StepOnePlus Real-Time PCR System (Life Technologies Corp., Carlsbad, CA, USA). Data normalization and analysis were performed as described previously [3]. P values for statistically significant differences in mRNA levels were calculated using a t-test.

Proteins (40 μg) were extracted from cell lines and tissues (Additional file 1, Supplementary Table 4) and separated on 7.5% acrylamide SDS-PAGE gels, then transferred to nitrocellulose membranes. Alternatively, NuPAGE Novex 4% to 12% Bis-Tris gradient gel and 3-(N-morpholino) propanesulfonic acid SDS running buffer (Invitrogen, Cergy-Pontoise, France) were used to achieve maximal resolution of SEPT9 bands. SEPT9 was detected by enhanced chemiluminescence using the anti-SEPT9 antibody (1:4,000 dilution) provided by Dr. Nagata [31], which recognizes human and mouse SEPT9_v1 through SEPT9_v4 isoforms and some uncharacterized high-molecular-weight isoforms; the antibody provided by Dr. Cossart (Institut Pasteur, Paris, France [32]) (1:1,000 dilution), which recognizes human SEPT9_v1 to SEPT9_v5 isoforms; or an antibody that recognizes only SEPT9_v3 [33] (provided by Dr. Tachibana, Osaka City University, Osaka, Japan) (Additional file 3, Supplementary Figure S2). Anti-α-tubulin DM1A antibody was used at a 1:2,000 dilution (Cell Signaling Technologies, Danvers, MA, USA), and anti-rabbit, anti-mouse and anti-rat secondary antibodies coupled to horseradish peroxidase were used at 1:2,000 dilutions (Jackson Immunoresearch Laboratories, Suffolk, UK). Quantification of bands was performed with the ImageJ software package (National Institutes of Health, Bethesda, MD, USA) and normalized to α-tubulin [34]. All positive bands between 37 and 75 kDa were counted as the SEPT9-related signal based on the expected molecular weight range defined by the five known isoforms (Additional file 1, Supplementary Table 5), and this signal was normalized to the α-tubulin signal at 50 kDa.

The SEPT9_v1, SEPT9_v2, SEPT9_v3, SEPT9_v4 and SEPT9_v5 isoforms [GenBank:AF189713, GenBank:AJ312319, GenBank:AF123052, GenBank:AJ312322 and GenBank:AJ312320] were cloned from normal breast cDNA (Clontech Laboratories, Inc., Mountain View, CA, USA) using an isoform-specific forward primer containing a XhoI cassette and a reverse primer common to all isoforms containing a HindIII cassette (Additional file 1, Supplementary Table 3). Two micrograms of isoform-specific plasmid were transfected into subconfluent MCF7 cells using FuGENE 6 Transfection Reagent (Roche Applied Science Indianapolis, IN, USA). Forty-eight hours after transfection cells were selected with 800 μg/mL G418 for 14 days. mRNA and protein levels were quantified for four clones per SEPT9-specific isoform (total of 20 clones) by real-time qRT-PCR and Western blot analysis (Additional file 4, Supplementary Figure S3).

The genomic region chr17:72,824,475-72,827,999 of the hg18 build of the human genome was selected based on mammalian conservation and the publicly available ENCODE Project integrated regulation tracks [35]. The fragments shown in Figure 4B were cloned into the pGL3-Basic Luciferase Reporter Vector (Promega, Madison, WI, USA) and transfected into MCF7 cells, where lysates were quantified with the Dual-Luciferase Reporter Assay System (Promega). Data from three technical and two biological replicates were plotted for a total of six data points for each fragment.

Primers within the SEPT9_v3 promoter region were designed to include and target 10 different CpG sites using the MethPrimer program (Li Lab, University of California San Francisco, San Francisco, CA, USA) [36]. Matrix-assisted laser desorption/ionization time-of-flight mass spectrometry was performed using the MassARRAY EpiTYPER assay platform (SEQUENOM Inc., San Diego, CA, USA) on bisulfite-converted DNA as previously described [37]. For bisulfite conversion of DNA, we used the Zymo DNA purification kit (Zymo Research Corp., Irvine, CA, USA). MassARRAY primers (SEQUENOM Inc.) were designed to cover the genomic region HSA17 using the hg17 build of the human genome (chr17:72,823,233-72,825,460) [38] (Additional file 1, Supplementary Table 3). For 5-aza-2'-deoxycytidine treatment, cell lines were plated in 100-mm Petri dishes, allowed to attach overnight and treated with 2 μM 5-aza-2'-deoxycytidine for three days. Fresh 5-aza-2'-deoxycytidine stock was made every day and replaced together with media. Statistical analysis of the differences in methylation percentages between paired tumor and adjacent normal tissue samples at each CpG site was performed using nonparametric Wilcoxon signed-rank tests. P values were adjusted using the Holm-Bonferroni step-down method to correct for multiple testing [39]. The average methylation percentage across all CpG sites was compared in tumors versus tumor-free adjacent tissues as well as normal tissues using a paired t-test. Regression analysis to determine the coefficient of correlation between SEPT9_v3 mRNA and DNA methylation levels was performed using Microsoft Excel version 11.6.3 software (Microsoft Corp., Redmond, WA, USA).

Transwell inserts (8 μM; Millipore, Billerica, MA, USA) were coated with 25 μg/mL type I rat tail collagen (BD Biosciences Franklin Lakes, NJ, USA). Cells (5 × 10⁴ in DMEM 0.3% BSA) were added in triplicate for each sample to the transwell inserts and allowed to migrate for 10 hours at 37°C. Nonmigratory cells were removed and filters were fixed in 3.7% formaldehyde/PBS for 15 minutes and stained with 0.2% crystal violet dye for 10 minutes. Migrated cells were counted and averaged from 10 fields of view per filter at 20 × magnification using an Axio Observer.A1 inverted microscope (Carl Zeiss MicroImaging, Inc.). Two independent biological replicate assays were analyzed. For cluster analysis, cells were seeded at 3.8 × 10³/cm² and allowed to grow for 5 days. The number of cells comprising clusters found within one-fourth of a 60-mm plate was counted for each cell line, again with the Axio Observer.A1 inverted microscope at 20 × magnification.

Using FISH to determine DNA copy number, a locus-specific probe for SEPT9 was cohybridized with a probe for HER2, a well-established oncogenetic biomarker that maps to the same chromosome arm as SEPT9 (Additional file 2, Supplementary Figure S1A). Gene amplification of both loci was observed in most cancer cell lines and human breast carcinoma samples (Figure 1A). The nontumorigenic epithelial cell line MCF10A exhibited the lowest copy number for both SEPT9 and HER2 oncogenes (2.6 and 2.28 average number of copies per cell, respectively) (Additional file 2, Supplementary Figure S1B). The nontumorigenic cell line HBL100, however, carries a higher average number of copies per cell for both genes (8.96 and 8.68), which is consistent with the knowledge that over time, cultured, immortalized cell lines derived from nontumorigenic breast tissue may acquire cytogenetic alterations similar to those of cancer cell lines [40]. Unsupervised clustering based on SEPT9 signals counted within each cell separated the 12 cell lines into low and high copy number clusters (Figure 1B). The low SEPT9 copy number group included the nontumorigenic HBL100 and MCF10A cell lines, while the high copy number group included the highly tumorigenic SKBR3 and MDA-MB-231 cell lines. Analysis of SEPT9 and HER2 copies for each cell line within a single cell revealed that the amplification of SEPT9 and HER2 represents two independent cytogenetic events (Additional file 2, Supplementary Figure S1C).

We aimed to determine whether genomic amplification of SEPT9 results in increased mRNA levels in human breast cancer cell lines and whether this change translates into higher protein levels. Quantification of total SEPT9 mRNA levels (Figure 1C) showed that the cell lines expressed variable SEPT9 mRNA levels, but we found no direct correlation between mRNA levels and DNA copy number, suggesting that a more complex mode of regulation might occur at this genomic locus.

Because cytogenetic alterations of cell lines in two-dimensional culture systems may not accurately reflect what occurs in vivo, we examined the differences in SEPT9 expression in normal and tumor human breast tissues at the mRNA and protein levels. SEPT9 mRNA expression was quantified in 93 breast tissue samples consisting of normal tissue from individuals who had undergone reduction mammoplasty and tumors and their corresponding adjacent tumor-free breast tissue from cancer patients. There was a significant increase in SEPT9 in tumor tissues compared to both normal and tumor-free adjacent tissues (Figure 2A). Additionally, we analyzed SEPT9 protein content in breast tissues and tumors and found a significant increase in SEPT9 protein expression in tumors compared to normal tissues (Figure 2B and Additional file 5, Figure Supplementary S4B). Notably, we found higher expression in adjacent tumor-free tissues compared to normal tissues. This result suggests that either (1) SEPT9 protein expression might be influenced by the tumor microenvironment, thus affecting the adjacent tumor-free area of the breast, or (2) SEPT9 expression is altered early in the premalignant stage.

The observed differences in SEPT9 gene amplification and mRNA and protein levels in breast tumors compared to normal breast tissues prompted us to test whether these changes are associated with tumor grade. We analyzed SEPT9 copy number in a tumor set including benign tissues, ductal carcinoma in situ (DCIS) and breast tumors classified as poorly or moderately differentiated (grade II or III, respectively). For well-differentiated tumors (grade I), only one sample was available, and although it was included in the analysis, it was not used to draw conclusions. Only approximately one-half of the DCIS samples analyzed presented normal SEPT9 copy numbers, suggesting that genomic amplification of this oncogene occurs during early stages of tumorigenesis (Table 1). The maximum copy number detected in the primary tumors indicated that there are lower levels of gene amplification than those observed in breast cancer cell lines. The most significant finding of this analysis was that SEPT9 copy number increased in grades II and III, suggesting that high SEPT9 amplification correlates with poor clinical outcome.

To further examine whether SEPT9 gene amplification and increased SEPT9 mRNA and protein levels occur during tumor progression, we used the mammary carcinoma PyMT mouse model [41] to analyze stage-specific Sept9 expression. We detected Sept9 amplification in MGs with premalignant adenomas and observed that Sept9 gain (five signals per cell and higher) increases in advanced stage IV adenocarcinomas (Figure 2C). Similarly to human primary tumors, murine MG adenocarcinomas also expressed higher Sept9 mRNA (Figure 2D) and protein levels (Figure 2E) than normal MGs. At the protein level, a differential pattern of Sept9 isoform expression was reproducibly observed in normal tissue, hyperplastic tissue, adenoma and early and late carcinomas. There was an increase in Sept9 protein expression compared to normal MGs as early as the hyperplasia stage (six weeks). The most striking difference was a prominent decrease in the expression of a 120-kDa band with a concomitant increase in the expression of a 90-kDa band beyond hyperplasia (Figure 2E). Because these bands did not match any known isoform of Sept9, they were excluded from the quantification of Sept9 protein expression levels.

The observed quantitative changes in SEPT9 expression between normal tissue and breast tumors and during tumor progression impelled us to perform immunofluorescence on the same TMAs used for FISH to detect whether significant changes in SEPT9 intracellular localization occurred. All cells expressed characteristic SEPT9 cytoplasmic staining, but, surprisingly, normal breast tissues bore a strong nuclear signal in the luminal layer of the epithelium that was lost in tumor cells (Figure 2F and Additional file 1, Supplementary Table 6). We also analyzed MGs isolated from four-week-old virgin C57B6 mice and found similar intracellular localization (Additional file 1, Supplementary Table 6). Additionally, when we analyzed PyMT tissues at various stages of tumorigenesis, we found that approximately 40% of cells (all luminal) exhibited nuclear staining in hyperplasia, whereas only about 2% of cells carried a nucleolar signal in advanced tumor stages. On the basis of the morphology and size of the nuclear staining, it appears that SEPT9 localizes to the nucleolus of normal breast cells, which was confirmed by its colocalization with nucleolin (Figure 2F). Collectively, these results suggest that in normal breast tissue, SEPT9 maintains both cytoplasmic and nucleolar localization, whereas during tumorigenesis only cytoplasmic expression is retained.

The lack of a direct correlation between SEPT9 copy number increases and mRNA levels prompted us to examine epigenetic modifications, specifically DNA methylation, at potential alternative transcription start sites (TSSs) in breast cancer. Alternative splicing of SEPT9 mRNA has previously been proposed as the primary mode of regulation of SEPT9 isoforms [19]. Recently, it has been suggested that DNA methylation of CpG sites near specific TSSs coding for individual isoform variants may be an alternative regulatory mechanism [18]. On the basis of our genomic and computational analyses of the mammalian SEPT9 gene, we found a highly conserved genomic region containing a CpG island that maps upstream to the SEPT9_v3 start codon, suggesting the presence of an alternative promoter (Figure 3A, top). We designed specific primers that targeted and quantified the methylation percentage at each CG dinucleotide within this region by using the highly sensitive MassARRAY technique on our panel of cell lines. Cell lines clustered into two groups, one group with the SEPT9_v3 TSS hypomethylated (average 18.4% methylation across all CpG sites) and one group with the SEPT9_v3 TSS hypermethylated (76.4%) (Figure 3A). To determine how DNA methylation alters SEPT9_v3 transcription, we quantified specific SEPT9_v1 and SEPT9_v3 mRNA levels in cell lines, using SEPT9_v1 for comparison since this specific isoform has been shown to increase in cancer (Figure 3B). As expected, SEPT9_v1 was expressed in most cell lines, whereas significant differences in mRNA levels were observed for the SEPT9_v3 isoform. A subset of cell lines (T47D, MDA-MB-435, MCF7 and BT474) expressed nearly undetectable levels of SEPT9_v3, whereas other cell lines (HS578T, HBL100, MDA-MB-231, MCF10A, SKBR3 and MDA-MB-468) expressed high levels. Interestingly, cell lines with hypermethylation at the SEPT9_v3 start site exhibited lower levels of SEPT9_v3 mRNA, whereas cell lines with hypomethylation at the SEPT9_v3 start site expressed higher SEPT9_v3 mRNA levels (R² = 0.678; significance deviation from zero: P = 0.0031) (Figures 3A and 3B). This finding also corresponded to repression of SEPT9_v3 expression at the protein level in cell lines that exhibited hypermethylation at the SEPT9_v3 start site (Figure 3C and Additional file 5, Supplementary Figure S4A).

To confirm the functional consequence of DNA methylation at the SEPT9_v3 start site on transcription, a subset of cell lines was treated with the DNA demethylating agent 5-aza-2'-deoxycytidine. As expected, treatment of cell lines displaying hypermethylation of the SEPT9_v3 start site resulted in increased SEPT9_v3 mRNA expression (Figure 4A, left graph), whereas it had no effect on cell lines exhibiting hypomethylation (Figure 4A, right graph). Taken together, these results strongly indicate that DNA methylation of an alternative promoter near the SEPT9_v3 TSS is a principal mechanism responsible for decreased SEPT9_v3 expression. The presence of an active alternative promoter (Figure 4B, green bar) was confirmed by performing a luciferase reporter assay in which the genomic region upstream of the SEPT9_v3 TSS was targeted (Figure 4B).

Next we analyzed the methylation status of the CpG sites proximal to the SEPT9_ v3 TSS in human breast tumor samples, their matching tumor-free adjacent areas and normal breast tissues from reduction mammoplasty patients (Figure 4C). MassARRAY analysis revealed an overall statistically significant difference in methylation levels between breast tumor tissues, the corresponding matching tumor-free area and normal breast tissues (52.9% versus 41.7% and 44.5%, respectively) (Figure 4C). Moreover, when DNA methylation was measured at each of the 10 individual CpG sites, we also found statistically significant differences in methylation levels between the tumors and the matching tumor-free tissue samples. These results suggest that SEPT9_v3 may frequently be silenced in breast tumors by methylation of its promoter region.

We sought to determine whether each SEPT9 isoform has similar intracellular distribution by generating stable MCF7 clones, each expressing one GFP-tagged SEPT9 isoform (Additional file 4, Supplementary Figure S3). All GFP-SEPT9 constructs were expressed in septin-like filamentous form, confirming functionality (Figure 5). Differences were observed in the organization of SEPT9 filaments between the nuclear region and the rest of the cytoplasm, thus the distribution of nuclear and cytoplasmic immunofluorescence was quantified for each of the isoform clones. SEPT9_v1 was mainly expressed within the nucleus, whereas SEPT9_v2 localization was sparse in that region. The other SEPT9 isoforms were relatively equally distributed between the nucleus and the rest of the cytoplasm.

Because of the presumed specific oncogenic nature of SEPT9_v1 [21] and the differences in interphase cellular distribution of SEPT9 isoforms, we tested the promigratory potential of each SEPT9 isoform in our stable MCF7 clones (Figure 6A). GFP-SEPT9_v1, GFP-SEPT9_v3, GFP-SEPT9_v4 and GFP-SEPT9_v5 showed significant increases in the number of migratory cells, whereas GFP-SEPT9_v2, despite a similar mRNA expression level of total SEPT9 in cells (Additional file 4, Supplementary Figure S3D), was not as promigratory as the parental cell line. On the basis of this observation, we sought to determine whether increased migration is the result of a loss of cellular interaction, where single-cell growth is prevalent, as opposed to clustering, a common characteristic of epithelial cells. We counted the average number of cells present in a cluster for each GFP clone after five days of growth (Figure 6B). The data revealed that parental MCF7 tended to grow in clusters containing at least 25 cells. The least migratory isoform (GFP-SEPT9_v2) had the lowest number of cells per cluster in the 1-5 range compared to those that were most migratory (GFP-SEPT9_v1, GFP-SEPT9_v3, GFP-SEPT9_v4 and GFP-SEPT9_v5).