The histone H3K9 demethylase KDM3A promotes anoikis by transcriptionally activating pro-apoptotic genes BNIP3 and BNIP3L

Epithelial cells that lose attachment to the extracellular matrix undergo a specialized form of apoptosis called anoikis. Here, using large-scale RNA interference (RNAi) screening, we find that KDM3A, a histone H3 lysine 9 (H3K9) mono- and di-demethylase, plays a pivotal role in anoikis induction. In attached breast epithelial cells, KDM3A expression is maintained at low levels by integrin signaling. Following detachment, integrin signaling is decreased resulting in increased KDM3A expression. RNAi-mediated knockdown of KDM3A substantially reduces apoptosis following detachment and, conversely, ectopic expression of KDM3A induces cell death in attached cells. We find that KDM3A promotes anoikis through transcriptional activation of BNIP3 and BNIP3L, which encode pro-apoptotic proteins. Using mouse models of breast cancer metastasis we show that knockdown of Kdm3a enhances metastatic potential. Finally, we find defective KDM3A expression in human breast cancer cell lines and tumors. Collectively, our results reveal a novel transcriptional regulatory program that mediates anoikis. DOI: http://dx.doi.org/10.7554/eLife.16844.001


Introduction
Epithelial cells that lose attachment to the extracellular matrix (ECM), or attach to an inappropriate ECM, undergo a specialized form of apoptosis called anoikis. Anoikis has an important role in preventing oncogenesis, particularly metastasis, by eliminating cells that lack proper ECM cues (Simpson et al., 2008;Zhu et al., 2001). Anoikis also functions to prevent the invasion of tumor cells into the luminal space, which is a hallmark of epithelial tumors (Debnath et al., 2002). In general, epithelial-derived cancers, such as breast cancer, develop resistance to anoikis (reviewed in Schwartz, 1997). Several signaling pathways have been shown to regulate anoikis (reviewed in Paoli et al., 2013). In particular, anoikis is suppressed by integrin signaling, which functions through focal adhesion kinase (FAK), an activator of the RAF/MEK/ERK pathway (King et al., 1997). FAK signaling is active in attached cells and is inactive following detachment (Frisch et al., 1996). Anoikis is also suppressed by integrin-mediated, ligand independent activation of the epidermal growth factor receptor (EGFR) signaling pathway (Moro et al., 1998), which, like FAK, also stimulates RAF/MEK/ ERK activity.
These cell signaling pathways have been found to regulate the levels of BIM (also called BCL2L11) and BMF, two pro-apoptotic members of the BCL2 family of apoptosis regulators previously shown to contribute to anoikis (Reginato et al., 2003;Schmelzle et al., 2007). However, depletion of BIM or BMF diminishes but does not completely prevent anoikis (Reginato et al., 2003;Schmelzle et al., 2007), suggesting the existence of other factors and regulatory pathways that can promote anoikis. Moreover, the basis of anoikis resistance remains to be determined and to date has not been linked to alterations in expression or activity of BIM or BMF.

Results and discussion
To investigate the possibility that there are additional factors and regulatory pathways that promote anoikis, we performed a large-scale RNA interference (RNAi) screen for genes whose loss of expression confer anoikis resistance. The screen was performed in MCF10A cells, an immortalized but nontransformed human breast epithelial cell line that has been frequently used to study anoikis (see, for example, Huang et al., 2010;Reginato et al., 2003;Schmelzle et al., 2007;Taube et al., 2006). A genome-wide human small hairpin RNA (shRNA) library comprising~62,400 shRNAs directed against 28,000 genes (Silva et al., 2003;Silva et al., 2005) was divided into 10 pools, which were packaged into retroviral particles and used to stably transduce MCF10A cells. Following selection, the cells were divided into two populations, one of which was plated on poly-2-hydroxyethylmethacrylate (HEMA)-coated plates for 10 days to inhibit cell attachment to matrix, and another that was cultured attached to matrix for 10 days as a control ( Figure 1A). Surviving cells were selected and shRNAs identified by deep sequencing. Bioinformatic analysis of the two populations identified 26 shRNAs whose abundance was significantly enriched >500-fold following detachment (Figure 1-eLife digest Epithelial cells line the inside of blood vessels, intestines and other organs throughout the body. Any epithelial cells that become detached from their natural surroundings die by a process called anoikis (a Greek word meaning "being without a home"). This process has an important role in preventing cancer from spreading around the body because it eliminates cells that are not in their proper environment. However, some cancers that start from epithelial cells, such as breast cancer, develop resistance to anoikis. Gaining a better understanding of the cellular factors that regulate anoikis, and how resistance develops, may reveal new drug targets for the treatment of breast cancer. Previous studies found proteins called BIM and BMF promote anoikis by inducing cell suicide. However, it is possible that other factors can also promote this process in different ways. Pedanou et al. performed a large-scale genetic screen in human breast epithelial cells and identified several new factors that promote anoikis.
Inside our cells, DNA is packaged around proteins called histones, which can influence whether a gene is switched on or off. One of the factors Pedanou et al. identified is a protein called KDM3A that can remove small chemical groups (known as methyl groups) from histones -a process that is known to switch on genes. Further experiments show that epithelial cells in their natural surroundings only produce low levels of KDM3A, but that the levels of this protein increase if these cells become detached. This promotes anoikis by activating two genes called BNIP3 and BNIP3L that induce cell suicide. However, KDM3A levels are low in human breast cancers, which suggests that these cancers become resistant to anoikis by preventing increases in KDM3A production.
Using a mouse model of breast cancer, Pedanou et al. found that switching off KDM3A in cancer cells increases their ability to move around the body. Collectively, these findings reveal a new mechanism that triggers anoikis in normal breast epithelial cells and is disabled during breast cancer development. Future challenges are to identify factors that directly regulate the production of KDM3A, and to understand how these factors are manipulated in breast cancer cells to cause anoikis resistance. Source data 1. List of 26 shRNAs, and the target genes, whose abundance was significantly enriched >500-fold following detachment of MCF10A cells. DOI: 10.7554/eLife.16844.004 Source data 2. Source data for Figure 1B. source data 1); such shRNAs presumably confer upon MCF10A cells a selective advantage by protecting them from undergoing anoikis.
To validate candidates isolated from the primary screen, we selected the top 20 most highly enriched shRNAs and analyzed them in an independent assay for their ability to confer resistance to anoikis. Briefly, MCF10A cells were transduced with a single shRNA, detached from matrix for 96 hr, and analysed for cell death by annexin V staining. As expected, knockdown of BIM, a positive control, decreased cell death following detachment compared to the control non-silencing (NS) shRNA ( Figure 1B and Figure 1-figure supplement 1). Of the 20 candidate shRNAs tested, five reduced the level of detachment-induced apoptosis compared to the NS shRNA, indicating they conferred anoikis resistance ( Figure 1B and Figure 1-figure supplement 1). Similar results were obtained using a second, unrelated shRNA directed against the same target gene (Figure 1-figure supplement 2). Quantitative RT-PCR (qRT-PCR) confirmed in all cases that expression of the target gene was decreased in the knockdown cell line (Figure 1-figure supplement 3).
One of the top scoring validated candidates was KDM3A (Figure 1-source data 1), a histone demethylase that specifically demethylates mono-methylated (me1) and di-methylated (me2) histone H3 lysine 9 (H3K9) (Yamane et al., 2006). H3K9 methylation is a transcriptional repressive mark, and the identification of KDM3A raised the intriguing possibility that induction of anoikis involves transcriptional activation of specific genes through H3K9me1/2 demethylation. Therefore, our subsequent experiments focused on investigating the role of KDM3A in anoikis.
We asked whether ectopic expression of KDM3A was sufficient to promote cell death in attached cells. MCF10A cells were transduced with a retrovirus expressing wild-type KDM3A, a catalytically inactive KDM3A mutant [KDM3A(H1120G/D1122N)] (Beyer et al., 2008) or, as a control, empty vector ( Figure 1-figure supplement 4), and then treated with puromycin for 10 days at which time viability was assessed by crystal violet staining. The results of Figure 1C show that ectopic expression of wild-type KDM3A but not KDM3A(H1120G/D1122N) greatly reduced MCF10A cell viability. Collectively, the results of Figure 1 demonstrate that KDM3A is necessary and sufficient for efficient induction of anoikis in breast epithelial cells.
We next examined the relationship between KDM3A expression and induction of anoikis. The immunoblot of Figure 2A shows that KDM3A protein levels were very low in attached MCF10A cells, but robustly increased in a time-dependent manner following detachment. The qRT-PCR analysis of Figure 2B shows that an increase in KDM3A expression following detachment was also detected at the mRNA level.
We next sought to understand the basis for the increase in KDM3A levels following detachment. As mentioned above, anoikis is suppressed by integrin signaling, which functions through FAK, a regulator of the RAF/MEK/ERK pathway (Frisch et al., 1996;King et al., 1997). Detachment causes a disruption in integrin-ECM contacts, resulting in a loss of FAK signaling in the detached cells (Frisch and Francis, 1994;Frisch et al., 1996), which we observed have elevated KDM3A levels (see Figures 2A and B). We therefore tested whether restoration of integrin signaling in detached cells would block the increase in KDM3A levels. The results of Figure 2C show that the addition of Matrigel basement membrane-like matrix, which restores integrin signaling, to detached cells markedly blocked the elevated levels of the BIM isoform BIM EL , as expected, and KDM3A. Treatment of MCF10A cells with a FAK inhibitor increased the levels of KDM3A protein ( Figure 2D) and mRNA ( Figure 2-figure supplement 1A). Thus, the increase in KDM3A levels upon detachment of MCF10A cells is due, at least in part, to the loss of integrin/FAK signaling.
We next analyzed the relationship between the EGFR signaling pathway and KDM3A levels. In the first set of experiments, we ectopically expressed either EGFR or a constitutively active MEK mutant, MEK2(S222D/S226D) (MEK2DD) (Voisin et al., 2008), both of which have been previously shown to block anoikis in detached cells (Reginato et al., 2003). Consistent with these previous results, Figure 2E shows that in detached MCF10A cells, expression of either EGFR or MEK2DD substantially decreased the level of BIM EL (Reginato et al., 2003). Expression of either EGFR or MEK2DD also decreased the levels of KDM3A in detached MCF10A cells. Conversely, KDM3A protein levels were increased in attached MCF10A cells treated with the EGFR inhibitor gefitinib (Barker et al., 2001;Ward et al., 1994) ( Figure 2F) or the MEK inhibitor U0126 (Favata et al., 1998) ( Figure 2G). Both gefitinib and U0126 treatment also resulted in increased KDM3A mRNA levels (Figure 2-figure supplement 1B,C).
The results described above suggest a model in which following detachment, the resulting increase in KDM3A demethylates H3K9me1/2 to stimulate expression of one or more pro-apoptotic genes. To test this model and identify pro-apoptotic KDM3A target genes, we took a candidatebased approach and analyzed expression of a panel of genes encoding pro-apoptotic BCL2 proteins (Boyd et al., 1994;Lomonosova and Chinnadurai, 2008;Matsushima et al., 1998) in attached MCF10A cells and detached cells expressing a NS or KDM3A shRNA. We sought to identify genes whose expression increased following detachment in control but not in KDM3A knockdown cells. We found that expression of the vast majority of genes encoding pro-apoptotic BCL2 proteins were   (Reginato et al., 2003;Schmelzle et al., 2007), expression of BIM and BMF were increased upon detachment. However, knockdown of KDM3A did not decrease expression of either BIM or BMF. By contrast, following detachment, expression of BNIP3 and BNIP3L increased, and were the only genes whose expression was diminished more than 2-fold by KDM3A knockdown ( Figure 3A and Figure 3-figure supplement 1). We therefore performed a series of experiments to determine whether BNIP3 and BNIP3L are critical KDM3A target genes that mediate anoikis.
In the first set of experiments we analyzed BNIP3 and BNIP3L protein levels during anoikis induction. The immunoblot of Figure 3B shows that BNIP3 and BNIP3L levels were very low in attached cells and substantially increased following detachment, with a time course similar to that of detachment-induced KDM3A expression (see Figure 2A). The chromatin immunoprecipitation (ChIP) experiment of Figure 3C shows that KDM3A was bound to the BNIP3 and BNIP3L promoters in detached but not attached cells. Moreover, the levels of H3K9me2 ( Figure 3D) and H3K9me1 (Figure 3-figure supplement 2) on the BNIP3 and BNIP3L promoters were greatly diminished following detachment, which was counteracted by knockdown of KDM3A. Conversely, overexpression of KDM3A but not KDM3A(H1120G/D1122N) in attached MCF10A cells resulted in decreased levels of H3K9me1 and H3K9me2 on the BNIP3 and BNIP3L promoters and increased expression of BNIP3 and BNIP3L (Figure 3-figure supplement 3). Finally, knockdown of BNIP3 or BNIP3L (Figure 3-figure supplement 4) resulted in decreased apoptosis following detachment ( Figure 3E and Figure 3-figure supplement 5). To further establish the pro-apoptotic role of BNIP3 and BNIP3L in MCF10A cells, we ectopically expressed BNIP3, BNIP3L or both in attached cells (Figure 3-figure supplement 6). Figure 3F shows that moderate cell death was observed upon ectopic expression of either BNIP3 or BNIP3L, but substantial cell death occurred in cells ectopically expressing both BNIP3 and BNIP3L. Collectively, these results establish BNIP3 and BNIP3L as critical KDM3A target genes that mediate anoikis ( Figure 3G).
We considered the possibility that decreased KDM3A expression may contribute to anoikis resistance in breast cancer cells and performed a series of experiments to test this idea. We first analyzed a panel of human breast cancer cell lines (BT549, MDA-MB-231, MCF7, SUM149 and T47D) comparing, as a control, anoikis-sensitive MCF10A cells. As expected, detachment-induced apoptosis was significantly diminished in breast cancer cell lines compared to MCF10A cells, indicative of anoikis   Figure 4B) and mRNA ( Figure 4C) levels was much lower than that observed in MCF10A cells. However, ectopic expression of KDM3A was sufficient to induce apoptosis in each of the five breast cancer cell lines ( Figure 4D). Collectively, these results indicate that anoikis-resistance of human breast cancer cells is due, at least in part, to inefficient induction of KDM3A following detachment.
We next analyzed KDM3A expression in human breast cancer patient samples. Interrogation of the Oncomine database (Rhodes et al., 2007) revealed decreased expression levels of KDM3A in several breast cancer datasets (Figure 4-figure supplement 2). To confirm these in silico results, we analyzed KDM3A expression by qRT-PCR in a series of human breast cancer patient samples. The results of Figure 4E show that compared to normal breast epithelium KDM3A expression was significantly decreased in a high percentage of breast cancers. Likewise, basal KDM3A expression levels were also diminished in most human breast cancer cell lines analyzed (Figure 4-figure supplement 3).
Finally, we performed a series of experiments to determine whether KDM3A affects metastatic potential. We first asked whether depletion of KDM3A would promote anoikis resistance in vivo using a mouse pulmonary survival assay. Briefly, immortalized but non-transformed mouse mammary epithelial CLS1 cells were stably transduced with an NS or Kdm3a shRNA (Figure 4-figure supplement 4) and injected into the tail vein of syngeneic mice. After 2 weeks, the lungs were harvested, dissociated into single cell suspensions, and plated in media containing puromycin to select for cells expressing the shRNA. The surviving colonies were visualized by crystal violet staining and quantified. The results of Figure 4F show that Kdm3a knockdown significantly increased the number of cells that survived in the mouse lung relative to the control NS shRNA.
In a second set of experiments, we used a well-characterized mouse breast cancer carcinoma progression series comprising isogenic cell lines with increasing metastatic potential: (1) non-invasive and non-metastatic 67NR cells, which form primary tumors, (2) invasive and non-metastatic 4T07 cells, which enter the circulation but fail to establish secondary tumors, and (3) highly metastatic 4T1 cells, which disseminate widely and colonize distant organ sites (Aslakson and Miller, 1992). qRT-PCR analysis revealed decreased Kdm3a expression in cell lines with greater metastatic potential (Figure 4-figure supplement 5). We expressed either a control NS shRNA or a Kdm3a shRNA in 67NR cells containing a luciferase reporter gene (Figure 4-figure supplement 6). Cells were injected into the tail veins of three syngeneic mice and pulmonary metastases were visualized by live animal imaging after 5 weeks. The results of Figure 4G show, as expected, that control 67NR cells failed to form pulmonary metastases in any of the three mice analyzed. By contrast, Kdm3a knockdown 67NR cells formed substantial pulmonary metastases in all three mice.
Finally, in a more stringent metastasis experiment, control and Kdm3a knockdown 4T07 cells (Figure 4-figure supplement 7), a non-metastatic mouse breast cancer cell line, were injected in the mammary fat pad of ten syngeneic mice. After 22 days the primary tumors were surgically removed and 8 weeks post-injection the animals were sacrificed and pulmonary tumors quantified. The growth of primary tumors formed by NS or Kdm3a knockdown cells was similar ( Figure 4H and Figure 4figure supplement 8). However, Kdm3a knockdown cells caused significantly increased metastatic burden in the lungs compared to control 4T07 cells ( Figure 4I and Figure 4-figure supplement 9). Consistent with our results, knockdown of Bnip3 has also been shown to cause increased metastasis in similar in vivo experiments (Manka et al., 2005). Collectively, these results show that KDM3A functions to prevent metastasis.
Based on the results presented above, we propose a model of anoikis induction that is illustrated in Figure 3G and discussed below. Following detachment of non-transformed cells, integrin signaling is decreased leading to transcriptional induction of KDM3A. The increased levels of KDM3A results in its recruitment to the pro-apoptotic genes BNIP3 and BNIP3L, where it promotes demethylation of inhibitory H3K9me1/2 marks and transcriptional activation of the two genes, resulting in anoikis induction. Consistent with this model, previous studies have shown that hypoxia results in transcriptional activation of KDM3A, BNIP3 and BNIP3L (Beyer et al., 2008;Sowter et al., 2001). We have found that in anoikis-resistant human breast cancer cell lines and tumors, KDM3A expression is defective, highlighting the importance of this pathway in promoting anoikis. Collectively, our results reveal a novel transcriptional regulatory program that mediates anoikis in non-transformed cells and is disabled during cancer development.
As described above, previous studies have shown that BIM and BMF are also effectors of anoikis (Reginato et al., 2003;Schmelzle et al., 2007). However, we have found that unlike BNIP3 and BNIP3L, BIM and BMF are not regulated by KDM3A. Thus, our results reveal that anoikis is promoted by multiple non-redundant pathways, which may help prevent the development of anoikis resistance.

RNA interference
The human shRNA mir pSM2 library (Open Biosystems/Thermo Scientific, Pittsburgh, PA) was obtained through the University of Massachusetts Medical School RNAi Core Facility (Worcester, MA). Retroviral pools were generated and used to transduce MCF10A cells as described previously (Gazin et al., 2007). Following puromycin selection, transduced cells were divided into two populations: one was plated on poly-HEMA-coated tissue culture plates (plates were coated with poly-HEMA (20 mg/ml) (Sigma-Aldrich), dried at room temperature overnight, and washed with phosphate buffered saline (PBS) before use) and grown for 10 days, and the other was grown for 10 days under normal tissue culture conditions. Cells that survived 10 days in suspension (a time point at which >95% of cells transduced with the control NS shRNA were killed) were seeded under normal tissue culture conditions to expand the population. shRNAs present in the surviving suspension population and the attached population were identified by deep sequencing at the University of Massachusetts Medical School Deep Sequencing Core Facility (Worcester, MA). The frequency of individual shRNAs in each sample was determined as described previously (Xie et al., 2012). The raw sequencing data have been uploaded to NCBI Gene Expression Omnibus and are accessible through GEO Series accession number GSE80144.
For stable shRNA knockdowns, 1 Â 10 5 cells were seeded in a six-well plate to 50% confluency and subsequently transduced with 200 ml lentiviral particles expressing shRNAs (obtained from Open Biosystems/Thermo Scientific through the UMMS RNAi Core Facility, listed in Supplementary file 1) in a total volume of 1 ml of appropriate media supplemented with 6 mg/ml polybrene (Sigma-Aldrich). Media was replaced after overnight incubation to remove the polybrene, and viral particles and cells were subjected to puromycin selection (2 mg/ml) for 3 days.

qRT-PCR
Total RNA was isolated and reverse transcription was performed as described (Gazin et al., 2007), followed by qRT-PCR using Power SYBR Green PCR Master Mix (Applied Biosystems, Grand Island, NY). RPL41 or GAPDH were used as internal reference genes for normalization. See Supplementary file 2 for primer sequences. Each sample was analyzed three independent times and the results from one representative experiment, with technical triplicates or quadruplicates, are shown.

Anoikis assays
Cells were placed in suspension in normal growth media in the presence of 0.5% methyl cellulose (Sigma Aldrich) (to avoid clumping of cells) on poly-HEMA-coated tissue culture plates. All anoikis assays were done at a cell density of 3 Â 10 5 cells/ml. Control cells were cultured under normal cell culture conditions. Cell death was measured by staining the cells with FITC-conjugated Annexin-V (ApoAlert, ClonTech) according to the manufacturer's instructions followed by analysis by flow cytometry (Flow Cytometry Core Facility, University of Massachusetts Medical School) at the indicated times. To restore integrin signaling in suspension, media was supplemented with 5% growthfactor-reduced Matrigel (BD Biosciences, San Diego, CA). Each sample was analyzed in biological triplicate.

Animal experiments
All animal protocols were approved by the Institution Animal Care and Use Committee (IACUC). Animal sample sizes were selected based on precedent established from previous publications.
In vivo anoikis assays CLS1 cells were stably transduced with either a NS or Kdm3a shRNA and selected with 2 mg/ml puromycin for 5 days. Stably transduced CLS1 cells (2 Â 10 5 ) were injected into the tail vein of 4-6 week old female BALB/c mice (Taconic Biosciences) (n = 4 mice per shRNA). Two weeks post injection the lungs were harvested, dissociated into single cell suspension, and plated onto tissue culture plates. Transduced CLS1 cells were selected for by treating the dissociated lung cells with 2 mg/ml puromycin. Surviving colonies were stained with crystal violet and quantified by counting. All experiments were performed in accordance with the Institutional Animal Care and Use Committee (IACUC) guidelines.
Pulmonary tumor assay 67NR cells were transduced with a NS or Kdm3a shRNA and selected with 2 mg/ml puromycin for 5 days. Stably transduced 67NR cells (2 Â 10 5 ) were injected into the tail vein of 6-8 week old female BALB/c mice (n = 3 mice per shRNA). Five weeks post injection, mice were given an intraperitoneal injection of D-Luciferin (100 mg/kg) (Gold Biotechnology, St. Louis, MO) and imaged on the Xenogen IVIS-100 (Caliper Life Sciences). Images were taken with Living Image software. All experiments were performed in accordance with the Institutional Animal Care and Use Committee (IACUC) guidelines.

Spontaneous metastasis assays
Female BALB/c mice (4-6 weeks) were purchased from Charles River Laboratories (Shrewsbury, MA). The mice were housed in facilities managed by the McGill University Animal Resources Centre (Montreal, Canada), and all animal experiments were conducted under a McGill University-approved Animal Use Protocol in accordance with guidelines established by the Canadian Council on Animal Care.
Spontaneous metastasis studies were carried out as previously described (Tabaries et al., 2011). Briefly, 4T07 cells expressing a NS or Kdm3a shRNA were first tested for mycoplasma contamination and found to be negative. Cells were then harvested from subconfluent plates, washed once with PBS, and resuspended (5 Â 10 3 cells) in 50 ml of a 50:50 solution of Matrigel (BD Biosciences) and PBS. This cell suspension was injected into the right abdominal mammary fat pad of BALB/c mice (n = 10 mice per shRNA) and measurements were taken beginning on day 7 post-injection. Animals that did not develop a primary tumor were excluded from the study. Tumor volumes were calculated using the following formula: pLW 2 /6, where L is the length and W is the width of the tumor. Tumors were surgically removed, using a cautery unit, once they reached a volume around 500 mm 3 , approximately 3 weeks post injection. Lungs were collected 8 weeks post-injection. Tumor burden in the lungs was quantified from four H&E stained step sections (200 mm/step). The number of lesions per section were counted using Imagescope software (Aperio, Vista, CA).

Statistics
All quantitative data were collected from experiments performed in at least triplicate, and expressed as mean ± standard deviation, with the exception of Figure 4H and I, which are expressed as mean ± SEM. Differences between groups were assayed using two-tailed Student's t test, except where noted above. Significant differences were considered when P<0.05.