PIAS1 Regulates Breast Tumorigenesis through Selective Epigenetic Gene Silencing

Epigenetic gene silencing by histone modifications and DNA methylation is essential for cancer development. The molecular mechanism that promotes selective epigenetic changes during tumorigenesis is not understood. We report here that the PIAS1 SUMO ligase is involved in the progression of breast tumorigenesis. Elevated PIAS1 expression was observed in breast tumor samples. PIAS1 knockdown in breast cancer cells reduced the subpopulation of tumor-initiating cells, and inhibited breast tumor growth in vivo. PIAS1 acts by delineating histone modifications and DNA methylation to silence the expression of a subset of clinically relevant genes, including breast cancer DNA methylation signature genes such as cyclin D2 and estrogen receptor, and breast tumor suppressor WNT5A. Our studies identify a novel epigenetic mechanism that regulates breast tumorigenesis through selective gene silencing.


Introduction
Both genetic and epigenetic alterations contribute to cancer development [1][2][3]. Tumor suppressors and epigenetic gatekeeper genes are frequently silenced by epigenetic mechanisms during tumor initiation and progression [3][4][5]. Extensive studies have been performed in the identification and characterization of altered DNA methylation in breast cancer development and progression. More than 100 genes have been reported to be aberrantly hypermethylated in breast tumors or breast cancer cell lines [1,6]. Many of these genes play important roles in the regulation of cell cycle, apoptosis, angiogenesis, metastasis and tumor initiation. It has been proposed that breast cancer-specific DNA methylation signatures can extend our ability to classify breast cancer and predict outcome beyond what is currently possible [6]. Epigenetic therapy holds a promising potential for the successful treatment of cancer since epigenetic changes are reversible as opposed to mutations [7]. The approval of DNA methylation and histone deacetylase (HDAC) inhibitors for cancer treatment offers new promise for epigenetic therapy. However, these drugs are rather nonspecific, and the development of more effective strategies for epigenetic therapy requires a thorough understanding of the molecular specificity involved in epigenetic gene silencing.
Most tumors are composed of a mixture of cancer cells, and the heterogeneity of tumors is the major obstacle to effective cancer therapy. It has been demonstrated that a sub-population of cancer cells, referred to as tumor-initiating cells (TICs) or cancer stem cells, is tumorigenic when transplanted into immunosuppressed nonobese diabetic/severe combined immunodeficiency (NOD/ SCID) mice [8]. TICs display some key properties of stem cells including self-renewal and multilineage differentiation [9]. In addition, TICs are found to be resistant to radiation and conventional chemotherapies [10][11][12][13]. Therefore, TICs may largely contribute to tumor cellular heterogeneity, tumor progression and tumor recurrence [14][15][16][17].
PIAS1 is a member of the PIAS (protein inhibitor of activated STAT) transcriptional regulator family that possesses SUMO (small ubiquitin-like modifier) E3 ligase activity [18]. Biochemical and genetic studies indicate that PIAS1 is a physiologically important transcriptional repressor of STAT1 and NF-kappaB [19][20][21]. PIAS1 is rapidly activated by phosphorylation on Ser90 residue in response to a variety of stimuli, including proinflammatory signals, TCR activation and growth factors. Activated PIAS1 is then recruited to gene promoters to repress transcription [22,23]. Recent studies indicate that PIAS1 mediates a novel epigenetic regulatory mechanism to control natural regulatory T cell (Treg) differentiation. PIAS1 binds to the Foxp3 promoter to maintain a repressive chromatin state through the recruitment of DNA methyltransferases (DNMTs) and HP1gamma [24]. These findings indicate that this newly identified PIAS1 epigenetic mechanism plays an important role in T cell differentiation.
In this paper, we report that PIAS1 is important for breast tumorigenesis. Elevated PIAS1 expression was observed in breast cancer patient samples. PIAS1 knockdown in breast cancer cells inhibited tumor growth in vivo. Most interestingly, mechanistic studies indicate that PIAS1 suppresses a number of-genes clinically relevant to breast tumorigenesis through epigenetic mechanisms. These studies suggest that targeting the PIAS1 epigenetic signaling pathway may represent a novel therapeutic strategy for cancer treatment.

Elevated PIAS1 expression in primary human breast cancer tissues
To test whether PIAS1 is involved in breast cancer progression, immunohistochemistry (IHC) tissue arrays were performed to examine the expression of PIAS1 protein in a panel of primary human breast tumor samples. PIAS1 is a nuclear protein, but it diffused to the cytoplasm under formalin fixation conditions ( Figure S1 in File S1), a phenomenon observed with other nuclear proteins [25]. IHC analyses indicate that PIAS1 is significantly upregulated in primary breast cancer samples at early stages of breast ductal carcinoma in situ (DCIS) as well as invasive ductal carcinoma (IDC) (Fig. 1a).

PIAS1 is important for breast tumorigenesis
To directly test whether PIAS1 plays a functional role in breast tumorigenesis, RNA interference approach was used to knockdown the expression of PIAS1 protein in MDA-MB231 cells. Stable cell lines expressing a scramble short hairpin RNA (control shRNA) or two independent PIAS1 shRNAs (shRNA1 and shRNA2) were obtained. Western blot analysis showed that PIAS1 expression was significantly suppressed by both PIAS1 shRNAs, although a more efficient inhibition by PIAS1 shRNA2 was observed (Fig. 1b). PIAS1 knockdown did not affect the growth of MDA-MB231 cells under the conventional serumcontaining conditions (DMEM) (Fig. 1c, left panel). In contrast, when these cells were cultured under serum-free growth factorenriched conditions (Stem Cell Media; SCM), which favor normal stem cells and more closely resemble primary tumors than the DMEM condition [26], PIAS1 knockdown significantly inhibited the survival of MDA-MB231 cells (Fig. 1c, right panel). To directly test the effect of PIAS1 knockdown on tumor growth in vivo, xenograft experiments were performed in SCID mice. PIAS1 knockdown significantly inhibited the tumor formation of MDA-MB231 cells in both the subcutaneous and the fat pad models (Fig. 1d), suggesting an important role of PIAS1 in the regulation of breast tumorigenesis.

PIAS1 regulates the self-renewal of breast tumor initiating cells (TICs)
The finding that PIAS1 knockdown affects breast cancer cell survival specifically under the conditions that favor stem cell growth suggests a possibility that PIAS1 may play a role in the regulation of breast cancer stem cells/tumor-initiating cells (TICs). Previous studies suggest that the ALDH + subpopulation of breast cancer cells is highly enriched in breast TICs [27]. ALDEFLUOR assays revealed that PIAS1 knockdown almost completely eliminated the ALDH + population (Fig. 2a), supporting the hypothesis that PIAS1 knockdown inhibits breast TICs. To further test whether PIAS1 regulates breast TICs, the control and PIAS1 knockdown MDA-MB231 cells were subjected to mammosphere assays [14,28,29]. PIAS1 knockdown significantly inhibited the formation of mammospheres (Fig. 2b), suggesting that PIAS1 regulates the self-renewal of breast TICs. PIAS1 Ser90 phosphorylation and SUMO ligase activity are required for PIAS1-mediated regulation of breast TICs Previous studies indicate that PIAS1 is activated by Ser90 phosphorylation to bind to chromatin and repress transcription of target genes in response to pro-inflammatory stimuli [22], a process that is dependent on the SUMO ligase activity of PIAS1. We explored whether PIAS1 can also be activated by growth factor signals. Western blot analysis revealed that PIAS1 became phosphorylated on Ser90 in response to EGF or Heregulin in various breast cancer cell lines, including MDA-MB231, BT-20, BT-474 and HCC-1954 (Fig. 2c). To test the importance of PIAS1 Ser90 phosphorylation and PIAS1 SUMO ligase activity in the regulation of breast TICs, PIAS1 shRNA1 knockdown MDA-MB231 cells were rescued with either an empty vector (Vec), wild type PIAS1 (WT), PIAS1 Ser90 mutant (S90A), or PIAS1 SUMO ligase defective mutant (W372A) through an shRNA escape approach, in which silent mutations were introduced into PIAS1 expression constructs to escape the inhibitory effect of PIAS1 shRNA. Western blot analysis indicated that the expression of WT or mutant PIAS1 proteins in the rescued cell lines was comparable to that of the MDA-MB231 control cells (Fig. 2d).
Consistent with the previous results (Fig. 1c), the introduction of either WT or S90A and W372A PIAS1 mutants did not affect cell growth under the conventional DMEM conditions (Fig. 2e, left panel). In contrast, when these cells were cultured under SCM conditions, only WT, but not the vector (Vec) or W372A mutant PIAS1 reconstituted cells, rescued cells from cell death (Fig. 2e, right panel). PIAS1 S90A mutant showed minor increase in cell survival, although the increase is not statistically significant (Fig. 2e, right panel). In addition, mammosphere assays were performed to examine the ability of WT or PIAS1 mutants to support the selfrenewal of TICs. The introduction of PIAS1 WT into PIAS1 knockdown cells promoted the formation of mammospheres (Fig. 2f). The introduction of PIAS1 S90 or W372 mutant resulted in minor increases in mammospheres, although the increases are not statistically significant (Fig. 2f). Consistently, ALDEFLUOR assays indicated that PIAS1 WT, but not S90 or W372 mutant, restored the population of ALDH + TICs (Fig. 2g). Taken together, these studies suggest that the observed inhibition of TICs in PIAS1 knockdown cells is due to the reduction of PIAS1 expression, and that both PIAS1 Ser90 phosphorylation and SUMO ligase activity are required for the maintenance of the breast TICs.

PIAS1 selectively represses a subset of genes clinically relevant to breast cancer
We explored the molecular mechanism of PIAS1-mediated regulation of breast TICs. Gene profiling studies were performed to identify PIAS1 downstream genes involved in tumorigenesis. Total RNAs from the control and PIAS1 knockdown MDA-MB231 cells cultured under DMEM or SCM conditions were subjected to microarray analysis. Since PIAS1 is a transcriptional repressor and PIAS1 knockdown inhibited self-renewal of breast TICs under SCM conditions, we focused on the genes that were preferentially upregulated in PIAS1 knockdown cells under SCM conditions (Table S1 in File S1). Interestingly, among the group of genes strongly induced by PIAS1 knockdown, several genes are known to be clinically relevant to breast cancer, including breast cancer DNA methylation signature genes Cyclin D2 (CCND2) and Estrogen receptor (ESR1), candidate tumor suppressor WNT5A, progestagen-associated endometrial protein (PAEP), as well as leucine zipper, downregulated in cancer 1 (LDOC1). WNT5A, which signals through a non-canonical WNT pathway, is a candidate tumor suppressor in breast cancers [30]. The loss of WNT5A is associated with early relapse in invasive ductal breast carcinomas (IDC) and short recurrence-free survival. PAEP (also known as GDA/PP14) is an epithelial differentiation-related gene. PAEP expression is associated with a more favorable prognosis in breast and ovarian cancers, and PAEP inhibits breast tumor growth in SCID mice [31,32]. CCND2 is frequently silenced in a variety of human cancers, including breast and ovarian cancers, through promoter hypermethylation [33,34]. LDOC1 has been reported to be downregulated in pancreatic and gastric cancer cells [35].
The induction of these genes identified by microarray was validated by quantitative real time PCR (Q-PCR) analysis in two independent PIAS1 knockdown MDA-MB231 cell lines ( Fig. 3a and Table S2 in File S1). As a control, PIAS1 knockdown did not show significant effect on the expression of WNT1 and CCND1, which show sequence homologies, but are functionally distinct in tumorigenesis, from WNT5A and CCND2, respectively. These results suggest that PIAS1 shows specificity in gene repression. Consistently, the transcription of WNT5A and CCND2, but not CCND1, was also elevated in PIAS1 knockdown xenograft tumor samples (Fig. 3b).

PIAS1 promotes self-renewal of breast TICs through WNT5A suppression
The WNT pathway is known to play a role in the regulation of self-renewal of stem cells [36][37][38]. Consistent with the gene expression results, higher levels of WNT5A protein were detected in PIAS1 knockdown MDA-MB231 cells (Fig. 4a). WNT5A shRNA was introduced into PIAS1 shRNA2 knockdown MDA-MB231 cells to inhibit WNT5A expression. While WNT5A shRNA1 efficiently inhibited the expression of WNT5A, WNT5A shRNA2 showed only a minor inhibition of WNT5A expression (Fig. 4b). Mammosphere assays showed that the suppression of WNT5A expression by WNT5A shRNA1 significantly enhanced the formation of mammospheres (Fig. 4c). Furthermore, the exogenous administration of recombinant WNT5A protein efficiently inhibited the mammosphere formation of parental MDA-MB231 cells (Fig. 4d). Consistently, the knockdown of WNT5A by shRNA1 significantly enhanced the tumor growth of PIAS1 knockdown MDA-MB231 cells in vivo (Fig. 4e). These studies support a role of WNT5A in PIAS1 knockdown-mediated inhibition of the self-renewal of breast TICs and breast tumorigenesis.

PIAS1 promotes epigenetic gene silencing in breast cancer cells
Our recent studies showed that PIAS1 restricts nTreg differentiation by recruiting DNMTs to the Foxp3 promoter to promote DNA methylation and epigenetic silencing [24]. We explored whether the PIAS1 epigenetic pathway also operates in breast cancer cells. Our previous results showed that PIAS1 inhibits the expression of CCND2, ESR1 and WNT5A; but not CCND1 (Fig. 3). Hypermethylation of the CCND2, ESR1 and WNT5A loci has been reported in various cancer types [39][40][41][42]. Therefore, chromatin immunoprecipitation (ChIP) assays were performed to test whether PIAS1 was associated with the genomic loci with close proximity to the reported methylation regions of the CCND2, ESR1 and WNT5A genes in MDA-MB231 cells. As shown in Fig. 5a, PIAS1 bound to the CCND2, ESR1 and WNT5A loci in the control shRNA cells, while the binding was reduced in PIAS1 shRNA2 cells. Furthermore, PIAS1 was not associated with the CCND1 promoter (Fig. 5a), consistent with the finding that PIAS1 does not affect the expression of CCND1 (Fig. 3). Interestingly, PIAS1 knockdown resulted in a substantial increase of the active histone mark histone H3 acetylation (AcH3) on the WNT5A gene (Fig. 5b). In contrast, the repressive modifications, such as histone H3 K27 trimethylation (H3K27me3) and histone H3 K9 trimethylation (H3K9me3), were considerably reduced in PIAS1 knockdown cells (Fig. 5b). Similar changes in AcH3 and H3K9me3 were observed in the CCND2 promoter (Fig. 5c). As a control, H3K9me3 was readily detectable in the centromeric satellite repeat, Satellite 2, a heterochromatin region [43] in MDA-MB231 cells (Fig. 5d). More importantly, the H3K9me3 level was not affected in PIAS1 knockdown cells, suggesting that PIAS1 does not affect global heterochromatin structure (Fig. 5d). In addition, while H3K9me3 and H3K27me3 were not detectable in the CCND1 promoter, the AcH3 level was not affected by PIAS1 knockdown (Fig. 5e), consistent with the finding that PIAS1 does not affect the CCND1 expression (Fig. 3). Taken together, these results suggest that PIAS1 regulates histone modifications of its target genes.
CCND2 and ESR1 are signature genes that are frequently methylated in breast cancer [1]. Bisulfite sequencing analysis indicated that the promoter of CCND2 was methylated in MDA-MB231 control cells, which was significantly reduced in PIAS1 knockdown cells (Fig. 6a). Similar reductions in DNA methylation were observed in ESR1 and WNT5A genes (Fig. 6a). Consistently, ChIP assays indicated that both DNMT1 and DNMT3A bind to the CCND2 promoter in the control MDA-MB231 cells, while the binding was compromised by PIAS1 knockdown (Fig. 6b). These studies suggest that PIAS1 recruits DNMTs to promote DNA methylation in breast cancer cells.

Discussion
Although extensive studies have been performed in the identification and characterization of altered DNA methylation and epigenetic modifications in breast cancer development and progression [1,[4][5][6], the molecular mechanism involved in this process has not been understood. Studies described in this manuscript have identified a novel epigenetic control mechanism in promoting selective epigenetic silencing in breast cancer. Our results suggest that the PIAS1 epigenetic pathway, which has been previously shown to function in regulatory T cell differentiation [24], is up-regulated in breast cancer and is involved in promoting DNA methylation and epigenetic silencing of breast cancer signature genes such as ESR1 and CCND2, as well as the breast tumor suppressor WNT5A.
Microarray analysis of PIAS1 knockdown breast cancer cells has uncovered an essential role of PIAS1 in the suppression of a group of genes previously known to be clinically relevant to breast cancer, such as WNT5A (Fig. 3). The WNT family of proteins can signal through the canonical beta-catenin-dependent or the noncanonical beta-catenin-independent pathway [36][37][38]. WNT5A belongs to the nontransforming class of the WNT gene family that  Table S1 and Table S2 in File S1. (b) Same as in a except that total RNA from fat pad tumor xenograft samples were used (n = 5). Error bars represent SEM. P values were determined by non-paired t-test. doi:10.1371/journal.pone.0089464.g003 PIAS1 Epigenetic Pathway in Breast Cancer PLOS ONE | www.plosone.org activates non-canonical signaling pathways. The biology of WNT5A is cell-type dependent, and it has been reported that WNT5A may signal through different WNT receptors to cause different cellular responses [44]. Gene targeting studies indicate that WNT5A is required for normal mammary gland development, and Wnt5a-null ammary tissue shows an accelerated developmental capacity [45]. In addition, WNT5A overexpression inhibits tumorigenesis of uroepithelial cell carcinoma and suppresses mammary cell migration [46,47]. The loss of WNT5A is associated with early relapse in invasive ductal breast carcinomas and short recurrence-free survival, supporting WNT5A as a candidate breast tumor suppressor [30]. The tumor suppressor function of WNT5A has also been suggested in other human cancers [48,49]. In this report, we showed that PIAS1-mediated regulation of the self-renewal of breast TICs is largely achieved through the transcriptional repression of WNT5A. The exogenous administration of recombinant WNT5A protein to MDA-MB231 breast cancer cells suppressed mammosphere. Consistently, WNT5A inhibition by shRNA rescued PIAS1 knockdownmediated suppression of mammosphere and tumor growth in vivo ( Fig. 4). Our results suggest that the PIAS1-WNT5A pathway regulates the self-renewal of breast TICs.
PIAS1 is a nuclear SUMO E3 ligase that functions as a transcriptional repressor. PIAS1 is activated by Ser90 phosphorylation in response to proinflammatory stimuli. Activated PIAS1 is then recruited to gene promoters to repress transcription [22,50]. In this report, we showed that PIAS1 is also phosphorylated/ activated in response to growth stimuli, and the ability of PIAS1 to regulate the self-renewal of breast TICs requires PIAS1 Ser90 phosphorylation as well as PIAS1 SUMO E3 ligase activity (Fig. 2). Our studies suggest that PIAS1 may act as a sensor protein in the nucleus that responds to growth and inflammatory stimuli in the tumor microenvironment to regulate the self-renewal of TICs through epigenetic gene regulation.
In conclusion, studies described in this paper suggest that PIAS1 plays an important role in promoting selective epigenetic silencing during breast tumorigenesis. It is possible that the PIAS1 epigenetic pathway may provide a link between inflammation and the development of breast cancer (Fig. 7). Targeting the PIAS1 epigenetic pathway may represent a novel therapeutic strategy for the treatment of breast cancer.

Mice, cell Lines and reagents
Tissue samples from breast cancer patients used for IHC analysis were purchased from commercial companies (Imgenex and Full Moon Biosystems). The work is exempt from Human Research since the data were analyzed anonymously. Human breast cancer cell lines MDA-MB231, BT-20, BT-474 and HCC-1954 were obtained from ATCC. MDA-MB231 cells were maintained in DMEM supplemented with 10% fetal bovine serum (FBS) and 1% Penicillin/Streptomycin. All other cells were maintained in RPMI supplemented with 10% FBS and 1% Penicillin/Streptomycin. Stem Cell Media (SCM) is composed of DMEM/F-12 (Cellgro) supplemented with 0.4% BSA, 1% Penicillin/Streptomycin, 2 mM Glutamine, 25 ng/ml human EGF (R&D), 25 ng/ml human basic FGF (R&D) and 5 ug/ml human insulin (Sigma). The following agents have also been used: Heregulin (Upstate), anti-pPIAS1 (Ser90-phosphorylated PIAS1) [22], polyclonal anti-PIAS1 [20], [51]; anti-Tubulin (Sigma), anti-WNT5A/B (Cell Signaling) and recombinant murine WNT5A protein (R&D). This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by The UCLA Institutional Animal Care and Use Committee (Protocol Number: 1999-015-43A).

shRNA knockdown and reconstitution
Oligonucleotides encoding a control small hairpin RNA (shRNA) or various shRNAs targeting PIAS1 or WNT5A were cloned under the control of the U6 promoter in the Lentiviral vector CS-CP for PIAS1 (containing a puromycin-resistant marker) or CS-CH for WNT5A (containing a hygromycinresistant marker), which was modified from the CS-CG Lentiviral vector [52]. Lentiviruses were generated by co-transfecting 293T cells with shRNA constructs together with helper plasmids pCMV-VSV-G and pHR'8.9DVPR using the calcium phosphate method. The wild type (WT), S90A and W372A mutant PIAS1 escape expression constructs were generated by insertion of WT or mutant PIAS1 cDNA fragments into the Lentiviral expression vector bearing a Hygromycin-resistant marker. These PIAS1 cDNAs carried 4 silent mutations that can escape the inhibition by shRNA without changing the codons of the protein (only the third nucleotide of each codon was altered). Lentiviruses were obtained as described above and target cells were infected with viral supernatant followed by Hygromycin selection.
In vitro mammosphere formation. In vitro mammosphere assays with MDA-MB231 cells were performed as described [12,27]. Briefly, cells were seeded under SCM conditions at indicated densities on 35 mm petri dish. Fresh human EGF (25 ng/ml), basic FGF (25 ng/ml) and Insulin (5 ug/ml) were supplemented every 2 days. Spheres were counted under a microscope after 5-7 days of culture.

Cell proliferation assay
For MDA-MB231 cell proliferation under DMEM conditions, cells were seeded in DMEM plus 10% FBS and 1% Penicillin/ Streptomycin at a density of 1610 5 cells per well in 6 well plate, and stained with trypan blue and counted for viable cells everyday for 5 days. Alternatively, cells were seeded at a density of 3,000 cells per well in 96 well tissue culture plate (08-772-3; Fisher), and cell growth was determined everyday for 4-5 days by CellTiter96 AQueous One Solution Cell Proliferation Assay as instructed by the manufacturer (Promega). For cell growth under SCM conditions, cells were seeded in SCM at a density of 3,000 cells per well in 96 well non-treated microplate (12-565-226; Fisher), and supplemented with fresh human EGF (25 ng/ml), basic FGF (25 ng/ml) and Insulin (5 ug/ml) every 2 days. Cell growth was determined at indicated time points by CellTiter96 AQueous One Solution Cell Proliferation Assay as instructed by the manufacturer (Promega). Triplicates were performed for each time point of the growth curve.

ALDEFLUOR assay
The ALDH + cell population was determined using an ALDEFLUOR assay kit as instructed (StemCell Technologies). Briefly, cells grown under SCM condition were incubated with the ALDH substrate BAAA in ALDEFLUOR assay buffer in the presence or absence of the specific ALDH inhibitor diethylaminobenzaldehyde (DEAB) at 37uC for 45 min, followed by flow cytometry. The ALDH + population of each sample was determined using its own negative control (DEAB containing sample) as a reference.

In vivo tumorigenesis
Exponentially growing cells were trypsinized and resuspended in serum free-DMEM or RPMI, mixed with equal volume of Matrigel (BD Biosciences), and injected into the fat pad, or subcutaneously into the flank of 6-10 week old SCID beige mice (UCLA Department of Radiation Oncology) in a volume of 150 ul per site. Tumors were measured weekly with a caliper and tumor volume was calculated as width x length x height x 0.526.

Tissue microarray analysis
Tissue microarray slides were obtained from Imgenex and Full Moon Biosystems, and immunohistochemistry (IHC) staining was performed using polyclonal anti-PIAS1 [20] as instructed by the manufacturers. ''Normal duct'' was defined as normal breast tissues from healthy individuals as well as histologically normal tissues adjacent to tumors. Semiquantitative assessment of PIAS1 staining was performed using a 0-2 scale (0 = negative; 1 = weak staining; 2 = strong staining) based on the average intensity per epithelial cell, and PIAS1 score was defined as the product of PIAS1 staining scale and the percentage of PIAS1 positive staining. A total of 3 independent tissue arrays containing 30-100 samples each were performed and the data were pooled. Statistic analysis was performed using non-parametric two-tailed Kruskal Wallis test with alpha level equals 0.05 for all tests, since both ''Normal duct'' and ''IDC'' populations are not normally distributed.

Microarray analysis
MDA-MB231 cells containing a control shRNA or PIAS1 shRNA2 were cultured in DMEM plus 10% FBS (DMEM) or SCM for 30 h. Total RNA was prepared and subjected to microarray analyses using the human genome U133A 2.0 array chip (Affymetrix) as described [19]. The microarray data is presented in Table S1 in File S1 and has been deposited to Gene Expression Omnibus (GEO) database (GSE44024).

Quantitative real time PCR (Q-PCR)
Quantitative real time PCR (Q-PCR) analyses were performed with breast cancer cells or tumor xenograft samples as described previously [19]. Briefly, total RNA was prepared using RNA STAT60 (Tel-Test). First strand complementary DNA was produced by reverse transcription (RT) of 1 ug total RNA using iScript cDNA synthesis kit (Bio-Rad). Q-PCR was carried out using the CFX96 real-time PCR detection system (Bio-Rad) in a final volume of 25 ul containing Taq polymerase, 1xTaq buffer, 125 uM dNTP, SYBR Green I (Molecular Probes) and genespecific primers. Amplification conditions were: 95uC (3 min), 40 cycles of 95uC (10 s) and 61uC (30 s). Q-PCR data were analyzed by CFX Manager 2.0 software (Bio-Rad), and normalized by beta-Actin (ACTB). The results were presented as ''Relative Expression'' as compared to that in the control shRNA cells, which was set as ''1''. Primers are listed in Table S2 in File S1.

Bisulfite treatment and methylation analysis
Bisulfite modification of DNA was performed using EZ Methylation-Gold kit (Zymo Research). Bisulfite genomic sequencing (BGS) was conducted using cells grown in DMEM plus 10% FBS as described [39][40][41][42]. Taq DNA polymerase (Zymo Research) was used for PCR amplification using specific primers (Table S2 in File S1). The PCR conditions were as follows: 1 cycle of 95uC for 10 min, then 40 cycles of 95uC for 45 s, 56uC for 1 min, and 72uC for 1 min; and 1 cycle of 72uC for 10 min. Amplified products were cloned into pCR4-Topo (Invitrogen), with 8 to 12 colonies randomly chosen and sequenced.

Supporting Information
File S1 Contains the following files: Figure S1, Table S1 and  Table S2. Figure S1. Validation of the polyclonal anti-PIAS1 antibody by immunofluorescence. MDA-MB231 cells containing control shRNA or PIAS1 shRNA2 were fixed by 3 different methods as indicated, followed by staining with polyclonal anti-PIAS1. FMA, formaldehyde. Table S1. Microarray analysis. Fold induction is defined as the ratio of the expression levels of a given gene in PIAS1 shRNA2 vs. control shRNA cells. Genes with greater than 10-fold induction under Stem Cell Media (SCM) condition are shown. Table S2. Primers used for Q-PCR, ChIP and methylation. (DOC)