HMG20A Is Required for SNAI1-Mediated Epithelial to Mesenchymal Transition

ORIGINAL ARTICLE HMG20A is required for SNAI1-mediated epithelial to mesenchymal transition

S Rivero1, M Ceballos-Chávez1, SS Bhattacharya2 and JC Reyes1

HMG20A is a high mobility group (HMG) domain containing protein homologous to HMG20B, a core subunit of the Lys-speciﬁc demethylase 1/REST co-repressor 1 (LSD1-CoREST) histone demethylase complex. Here, we show that HMG20A can replace HMG20B and, therefore, they are mutually exclusive subunits of the complex. Both proteins interact through a coiled-coil domain with BHC80, another subunit of the LSD1-CoREST complex. To investigate the functional differences between the two proteins, we performed transcriptomic analysis of HMG20A- and HMG20B-depleted cells. Analysis of the misregulated genes in HMG20A- knockdown cells evidenced a high proportion of genes related to the epithelial-to-mesenchymal transition (EMT) process. EMT occurs during embryonic development or during the course of malignant cancer progression and consists in the dynamic and reversible transitions between epithelial and mesenchymal phenotypes. We show that HMG20A together with LSD1 are required for SNAI1-dependent repression of epithelial genes and for (transforming growth factor β) TGF-β-triggered EMT. Importantly, HMG20A-depleted cells displayed reduced binding of LSD1 to epithelial gene promoters and increased methylation of lysine 4 of histone H3, suggesting a role of HMG20A in recruiting or in stabilizing the complex at the chromatin. SNAI1 and the TGF-β-related transcription factor SMAD4 were found to be associated with the LSD1-CoREST complex containing HMG20A. Furthermore, we show that HMG20A-depleted cells displayed reduced motility and invasion activity. Finally, we show that expression of HMG20A correlates positively with mesenchymal markers and negatively with epithelial markers in human tumor samples. Taken together, our data demonstrate that HMG20A is essential for the mesenchymal phenotype.

Oncogene (2015) 34, 5264–5276; doi:10.1038/onc.2014.446; published online 2 February 2015

INTRODUCTION gene expression and protein stability and by promoting its 1 Cellular plasticity is the ability of some differentiated cell types to migration to the nucleus. Both EMT and MET require extensive reorganization of the reversibly change their phenotype. An example of cellular 7,8 plasticity is the process by which epithelial cells can downregulate epigenetic information of the cells. In fact, master transcription epithelial characters and acquire mesenchymal characteristics. regulators of these processes recruit multiprotein complexes that fi This process, known as epithelial-to-mesenchymal transition introduce or remove modi cations of chromatin-associated (EMT), is a key process during embryonic development as well histones. For example, SNAI1 represses transcription of epithelial as in some pathological situations, including cancer and organ genes, such as CDH1 (encoding E-cadherin), by recruiting – chromatin-modifying machineries including the Polycomb repres- fibrosis.1 4 In human cancer, EMT is an important process sive complex 2 and the Lys-specific demethylase 1/REST co- conducive to tumor invasion and dissemination. The reverse – repressor 1 (LSD1-CoREST) complex.9 11 LSD1 is the first reported transformation, the mesenchymal-to-epithelial transition (MET), is 12 fi histone demethylase and interacts directly with the SNAG required for the formation of organs in the nal destinations of domain of SNAI1. LSD1 is involved in gene repression as a part of embryonic migratory cells and for establishment of overt 2,5 the LSD1-CoREST complex mediating the demethylation of metastases at distant sites in tumors. H3K4me1/213 and in gene activation associated with nuclear During EMT, cells perform an extensive reorganization of cell receptors through demethylation of H3K9me1/2.14 Consistent junction complexes, cytoskeletal architecture and interactions with its function as SNAI1 co-repressor, LSD1 is highly expressed in with the extracellular matrix. Cells increase their motility and estrogen receptor-negative tumors, which display mesenchymal invasion properties and become more resistant to drugs. These gene signatures.15 Furthermore, blocking SNAI1-LSD1 interaction transformations require large changes in gene expression suppresses the motility and invasiveness of cancer cells of controlled by master regulators, including SNAIL, TWIST and different origins.16 The core of the LSD1-CoREST complex 1 zinc-finger E-box-binding (ZEB) transcription factors. SNAI1 (also comprises CoREST, HDAC1-2, BHC80 (also called PHF21A) and called SNAIL1 or SNAIL) is one of the most important factors HMG20B (also called BRAF35).13,17–22 HMG20B contains a high involved in EMT.6 SNAI1 is a repressor of epithelial genes and an mobility group (HMG) domain in the amino terminal half of the activator of mesenchymal genes. Multiple signaling pathways protein and was first identified as an interactor of BRCA2.23 The including transforming growth factor β (TGF-β), WNT, Notch and role of HMG20B in the LSD1-CoREST complex is unclear.21 MAPKs cooperate in the initiation of EMT, by increasing SNAI1 HMG20A is a paralogous of HMG20B. Both proteins share the

1Molecular Biology Department, Centro Andaluz de Biología Molecular y Medicina Regenerativa (CABIMER), Consejo Superior de Investigaciones Científicas (CSIC), Seville, Spain and 2Cell Therapy and Regenerative Medicine Department, Centro Andaluz de Biología Molecular y Medicina Regenerativa (CABIMER), Seville, Spain. Correspondence: Dr JC Reyes, Molecular Biology Department, CABIMER-CSIC, Av. Americo Vespucio, Seville 41092, Spain. E-mail: [email protected] Received 30 July 2014; revised 28 November 2014; accepted 5 December 2014; published online 2 February 2015 HMG20A is required for EMT S Rivero et al 5265 HMG domain and a sequence with predicted coiled-coil structure targets HMG20B. If HMG20A binds the complex through HMG20B, in their carboxy terminus.24 It has previously been reported that then depletion of this protein would impair association of HMG20A HMG20A is a functional antagonist of HMG20B in the process of with the complex. However, knockdown of HMG20B did not affect neural differentiation.24,25 We have also shown that HMG20A the association of HMG20A with the complex (Figure 1d). These forms homodimers and heterodimers with HMG20B, but HMG20B data suggested that HMG20A and HMG20B bind independently to is not able to form homodimers. Furthermore, we showed that the complex. Then, we wondered whether the LSD1-CoREST HMG20A-HMG20B heterodimerization impairs sumoylation of complex can contain both factors at the same time or whether, HMG20B which is essential for its function.25 However, some alternatively, their interaction with the complex is mutually proteomic analyses have reported substoichiometric amounts of exclusive. To investigate these possibilities, the LSD1-CoREST HMG20A in the complex,17,18,20 while other works do not detect complex was immunoprecipitated using anti-LSD1 antibodies the protein as a subunit of LSD1-CoREST complex.13,19,21,22 To under normal conditions or upon overexpression of HA-HMG20A clarify this situation, we have investigated the presence of or Flag-HMG20B. Interestingly, overexpression of HMG20A dis- HMG20A in the LSD1-CoREST complex and its functional placed endogenous HMG20B from the LSD1-CoREST complex and consequences. Here, we show that HMG20A is able to substitute increased the level of HMG20A associated with the complex HMG20B in the LSD1-CoREST complex and therefore that HMG20A (Figure 1e). As a control, we verified that the mutant HMG20AΔcc and HMG20B are mutually exclusive subunits of the core complex. was not able to displace endogenous HMG20B from the complex Gene expression profiling of cells depleted of HMG20A or (Supplementary Figure S1). Conversely, overexpression of HMG20B HMG20B demonstrated that both proteins have common as well provoked the substitution of endogenous HMG20A by HMG20B in as specific targets. Interestingly, many differentially regulated the complex (Figure 1f). These data suggest that the interaction of genes in the absence of HMG20A were related to EMT and MET HMG20A or HMG20B with the complex is mutually exclusive. processes. Importantly, we show that HMG20A is required for Consistently, in vitro pull-down experiments showed that HMG20A SNAI1-dependent repression of epithelial genes and for TGF-β- interacts with BHC80 (Figure 1g), the same subunit of the LSD1- mediated EMT. The LSD1-CoREST complex containing HMG20A CoREST complex responsible for the interaction with HMG20B interacts with SNAI1 and with SMAD4. Consistent with these data, (Figure 1g and ref. 26). Furthermore, HMG20B was more efficiently HMG20A was required for mesenchymal phenotypes including retained by the GST-BHC80 coated beads than HMG20A (Figure 1g), cell motility and invasion activity. Taken together, these data which may account for the lower amount of HMG20A normally demonstrate that HMG20A is an important factor required for associated with the LSD1-CoREST complex. Therefore, our data promoting and maintaining mesenchymal characters. indicate that HMG20A can replace HMG20B in the LSD1-CoREST complex (Figure 1h). In addition, the fact that HMG20A can also 25 RESULTS form homodimers and heterodimers with HMG20B suggests the existence of a complex equilibrium to determine the proportion of HMG20A and HMG20B are alternative subunits of the LSD1- complexes containing HMG20A or HMG20B (Figure 1h). CoREST complex fi fi Several proteomic analysis identi ed HMG20A in the puri ed LSD1- HMG20A and HMG20B have common and independent target genes CoREST complex.17,18,20 However, HMG20A was not associated with To elucidate the functional differences between HMG20A and the complex in other studies13,19,21,22 and was found to form HMG20B, we performed RNA expression profiling using Affymetrix homodimers and heterodimers with HMG20B.25 To shed light on DNA microarrays of retinal pigment epithelial RPE1 cells transfected this controversy, the presence of endogenous HMG20A in the with control siRNA or with siRNAs that specifically target HMG20A or LSD1-CoREST complex was investigated by immunoprecipitation HMG20B (Figures 2a and b). Depletion of HMG20B provoked using anti-HMG20A antibodies in 293T cells. LSD1, CoREST, HDAC2 misregulation of 662 genes (P-valueo0.01 and lineal and HMG20B subunits of the core complex were co-precipitated change41.25-fold) among which 472 were upregulated and 190 together with HMG20A (Figure 1a). We have previously shown that were downregulated (Supplementary Table S1). Depletion of HMG20B interacts with the LSD1-CoREST complex through its o 25 HMG20A provoked misregulation of 416 genes (P-value 0.01 and coiled-coil domain. Therefore, we checked whether the homo- lineal change41.25-fold) among which 247 were upregulated and logous coiled-coil domain found in HMG20A is also involved in the 169 were downregulated (Supplementary Table S2). Misregulated interaction with the complex. To achieve this, we expressed a wild- expression of three genes affected by depletion of HMG20A or type HA-tagged version of HMG20A or a mutant version lacking 15 HMG20B was confirmed by RT-qPCR (Supplementary Figure S2). – amino acids (residues 252 266) of the coiled-coil region (HA- Interestingly, 138 genes were found to be commonly misexpressed Δ − HMG20A cc) and performed immunoprecipitation using anti-HA by depleting HMG20A or HMG20B (Figure 2c) (P-value=1.0×10 107, antibodies. LSD1, CoREST and HMG20B co-precipitated with wild- hypergeometric distribution). Most of them (110 genes) were Δ type HA-HMG20A but not with HA-HMG20A cc (Figure 1b), upregulated in both cases. These results suggest that HMG20A and indicating that HMG20A interacts with the LSD1-CoREST complex HMG20B have common but also specifictargets. through the coiled-coil domain. Next, we characterized in detail Gene ontology analysis of misregulated genes in HMG20A- how HMG20A and HMG20B bind the complex. HMG20B cannot depleted cells showed a marked enrichment in several functional 25 form homodimers; therefore, it should be able to bind to the categories related to cell adhesion, cell motion and extracellular complex as a monomer. To verify this, we overexpressed a Flag- matrix (Supplementary Figure S3a). However, HMG20B-dependent tagged version of HMG20B and performed immunoprecipitation genes only showed a moderated enrichment of one cell adhesion using anti-Flag antibodies. LSD1, CoREST and Flag-HMG20B were category (Supplementary Figure S3b). Furthermore, gene set co-precipitated, but not endogenous non-tagged HMG20B, indicat- enrichment analysis (GSEA)27 indicated that EMT-upregulated28 and ing the absence of complexes containing Flag-HMG20B-HMG20B TGF-β-upregulated gene signatures were enriched among the genes dimers (Figure 1c) and therefore, suggesting that HMG20B binds downregulated in HMG20A-depleted cells, strongly suggesting that the complex as a monomer. This is also consistent with proteomic HMG20A loss induces a reversion of the EMT signaling program – data.13,17 22 Given the fact that HMG20A and HMG20B can form (Figure 2d). Consistently, HMG20A depletion increased epithelial heterodimers through their coiled-coil domains,25 it would be markers such as CXADR (coxsackie virus and adenovirus receptor), possible that HMG20A binds the complex through heterodimeriza- CLDN12 (claudin 12) and several keratins (KRT8, KRT15 and KRT18) tion with HMG20B. To check this possibility, 293T cells were and decreased mesenchymal markers such as smooth muscle actin transfected either with control siRNA or with siRNA that specifically (ACTG2 and ACTA2), POSTN and PTGS2 (Supplementary Table S2).

© 2015 Macmillan Publishers Limited Oncogene (2015) 5264 – 5276 HMG20A is required for EMT S Rivero et al 5266 These data suggest that HMG20A is involved in the regulation of Depletion of HMG20A decreases SNAI1 gene expression the dynamic equilibrium between epithelial and mesenchymal SNAI1 is one of the master transcription regulators that control phenotypes. Therefore, we decided to focus on the role of HMG20A epithelial traits through direct repression of epithelial genes.1,6 in EMT. First, we investigated whether HMG20A depletion affected

Input IgG IP HA Input IgG HMG20A IP HA-HMG20A wt Δcc wt Δcc wt Δcc LSD1 LSD1 Flag-HMG20B Input IgG Flag LSD1 CoREST CoREST CoREST HDAC2 HMG20B HMG20B Flag-HMG20B HMG20B HA HMG20B

HMG20A

Input IgG IP LSD1

Input IgG IP LSD1 siRNA ControlHMG20B ControlHMG20BControlHMG20B HA-HMG20A -+-+-+ LSD1 LSD1

CoREST CoREST

HDAC2 HMG20B

HMG20B HA

HMG20A

Input IgG IP LSD1 Flag-HMG20B -+-+-+

LSD1 Input GST GST-BHC80

35S-HMG20B 14.1%+- 3.58 CoREST 35S-HMG20A 2.1%+- 0.83 HMG20A 35S-eIF3b 0.4%+- 0.06 HMG20B Flag-HMG20B HMG20B

LSD1-CoREST complex HMG20 HMG20A HMG20A HMG20B HMG20A

LSD1 LSD1 BHC80 BHC80 HDAC1/2 HDAC1/2 CoREST CoREST

HMG20B HMG20A HMG20B

siHMG20B (662) siHMG20A (416)

siRNA Control HMG20A siRNA Control HMG20B

HMG20A HMG20B 524 138 278

α-tubulin α-tubulin

p=1.0 x 10-107 1.2 n.s. siHMG20B (472) siHMG20A (247) siHMG20B (190) siHMG20A (169) 1.0

0.8 362 110 137 171 19 150 0.6

0.4

Relative Fold (mRNA) 0.2 up-regulated down-regulated 0.0 p=6.0 x 10-116 p=1.8 x 10-15

siControl siControl siHMG20A siHMG20B HMG20A HMG20B

Enrichment plot: Enrichment plot: Taubeup.grp Enrichment plot: TGF_UP.V1_UP TRANSFORMING_GROWTH_FACTOR_BETA_RECEPTOR_ SIGNALING_PATHWAY

NES= -1.4 NES= -1.27 NES= -1.64 p<0.0001 p<0.0001 p<0.0001

Figure 2. HMG20A is involved in regulation of EMT signaling program. (a, b) RPE1 cells were transfected with control siRNAs or siRNAs against HMG20A or HMG20B. After 72 h, levels of HMG20A and HMG20B proteins were determined by western blot (a) and levels of transcripts were determined by RT-qPCR (b). Values were relative to the level of HMG20A transcript. (c) Venn diagrams showing overlapping between genes misregulated in HMG20A or HMG20B knockdown cells. Upper panel: all misregulated genes. Lower left panel: only upregulated genes. Lower right panel: only downregulated genes. Probability of overlapping based on hypergeometric distribution is provided. (d) GSEA in control versus HMG20A knockdown cells showing negative enrichment of a published gene signature constituted by genes upregulated in EMT28 (left panel); and two TGF-β-related gene signatures (middle and right panels). NES, normalized enrichment score. Provided P-value estimates the statistical signiﬁcance of the enrichment score.

Figure 1. HMG20A interacts with the LSD1-CoREST complex. (a) HMG20A interacts with the LSD1-CoREST complex. Cell extracts from 293T cells were subjected to immunoprecipitation (IP) with anti-HMG20A or control (IgG) antibodies and analyzed by western blot for the presence of the indicated proteins. (b) HMG20AΔcc does not interact with the LSD1-CoREST complex. 293T cells were transfected with expression vectors encoding wild-type (wt) or mutant version (Δcc) of HA-HMG20A. Cell extracts were subjected to immunoprecipitation with anti-HA or control (IgG) antibodies and analyzed by western blot for the presence of the indicated proteins. (c) HMG20B binds the LSD1- CoREST complex as a monomer. 293T cells were transfected with an expression vector encoding Flag-HMG20B. Then, cell extracts were subjected to immunoprecipitation with anti-Flag or control (IgG) antibodies and analyzed by western blot for the presence of the indicated proteins. (d) HMG20B is not required for interaction of HMG20A with the LSD1-CoREST complex. 293T cells were transfected with control siRNAs or siRNAs against HMG20B. Then, cell extracts were subjected to immunoprecipitation with anti-LSD1 or control (IgG) antibodies and analyzed by western blot for the presence of the indicated proteins. (e, f) Composition of LSD1-CoREST complex upon overexpression of HMG20A (e) or HMG20B (f). 293T cells were transfected with expression vectors encoding HA (− ) or HA-HMG20A (+) (e) or Flag ( − ) or Flag- HMG20B (+) (f). Then, cell extracts were subjected to immunoprecipitation with anti-LSD1 or control (IgG) antibodies and analyzed by western blot for the presence of the indicated proteins. (g) BHC80 binds HMG20B and HMG20A in vitro.1μg of GST-BHC80 or GST proteins bound to glutathione-sepharose beads was incubated with in vitro translated 35S-labeled HMG20B, HMG20A or eIF3b as a control. Bound proteins and 20% of the input were subjected to SDS-PAGE. Gels were dried and autoradiographed. Numbers represent the percentage of bound protein with respect to 100% of 35S-labeled protein used in the assay. Numbers are the average of three independent experiments ± s.d. (h) Model of interchange of HMG20A and HMG20B subunits in the LSD1-CoREST complex.

© 2015 Macmillan Publishers Limited Oncogene (2015) 5264 – 5276 HMG20A is required for EMT S Rivero et al 5268 expression of the SNAI1 gene in RPE1 cells. Interestingly, RT-qPCR deficient cells. It has already been reported that expression of experiments demonstrated that depletion of HMG20A decreased E-cadherin in RPE cells is very low.30 However, the retinal cadherin SNAI1 expression to about 50% (Figure 3a), in agreement with the CDH4 (also called R-cadherin) was found to be upregulated. 1.48-fold reduction observed in the microarray experiment. Induction of all these genes was verified by RT-qPCR (Figure 4a). Western blotting and immunofluorescence experiments demon- Consistent with the fact that HMG20A forms part of the LSD1- strated that SNAI1 protein levels were also decreased upon CoREST complex, expression of these four genes was also depletion of HMG20A (Figures 3b and c). Similar results were increased upon depletion of LSD1 (Figure 4a). SNAI1 recruits the observed in ARPE19 cells, another retinal pigment epithelial cell LSD1-CoREST complex to several epithelial genes.9,10 Therefore, line (Supplementary Figure S4). Furthermore, depletion of LSD1 we investigated whether SNAI1 also controlled expression of the (Supplementary Figure S5a) also promoted a reduction in the level four HMG20A target genes analyzed. Figure 4a shows that of SNAI1 mRNA (Figure 3a), suggesting that both LSD1 and depletion of SNAI1 (Supplementary Figure S5b) increased the HMG20A cooperates to maintain high levels of SNAI1. Then, we amount of CXADR, CLDN12, KRT18 and CDH4 transcripts, suggest- investigated whether these factors regulate the activity of the ing that these four genes are targets of SNAI1. Therefore, our data SNAI1 promoter. To that end, we transfected a SNAI1 promoter 29 indicate that HMG20A represses epithelial gene targets of SNAI1. reporter construct in RPE1 cells. LSD1 expression increased Next, we analyzed whether the effect of HMG20A depletion on promoter activity by about twofold (Figure 3d). Expression of epithelial markers is exclusively due to the observed down- HMG20A alone did not cause induction. However, expression of regulation of SNAI1. To do this, RPE1 cells were transiently both factors together produced a synergistic activation. Then, we transfected at the same time with a plasmid that overexpresses performed ChIP experiments to investigate whether HMG20A and SNAI1 and with control siRNA or a siRNA that targets HMG20A.As LSD1 control SNAI1 expression directly or indirectly. We were expected, CXADR and CLDN12 genes were repressed to about 30% unable to detect HMG20A or LSD1 binding to the SNAI1 promoter of their original level upon SNAI1 expression. However, repression (data not shown), suggesting that the LSD1/CoREST complex was completely abolished by depletion of HMG20A, indicating affects indirectly the expression of SNAI1. that the repressive function of SNAI1 depends on HMG20A (Figure 4b). Therefore, our data suggest that HMG20A cooperates HMG20A represses epithelial markers with SNAI1 in epithelial genes repression. As denoted above, the transcriptomic analysis showed that typical Chromatin immunoprecipitation (ChIP) experiments demon- epithelial genes such as CXADR, CLDN12 and KRT18 were strated that HMG20A and LSD1 bind to the CXADR and CLDN12 upregulated in HMG20A-depleted cells. Expression of E-cadherin promoters (Figure 4c). As control of the ChIP experiment, we (CDH1), a typical epithelial marker, was not increased in HMG20A- verified that depletion of HMG20A reduced HMG20A occupancy.

siControl siHMG20A 1.2 SNAI1 1.0 0.8 0.6 SNAI1 0.4 0.2

Relative Fold (mRNA) 0.0 siControl + -- siHMG20A - + - siLSD1 - - + SNAI1 Luc

-869 +59 SNAI1 300 siRNA Control HMG20A 250 LSD1 200

SNAI1 150 HMG20A 100 50 α-tubulin

Relative luciferase activity 0 Flag + --- Flag-HMG20A - + - + Flag-LSD1 - - + + Figure 3. Depletion of HMG20A or LSD1 decreases SNAI1 gene expression. (a) SNAI1 mRNA expression was analyzed by RT-qPCR in RPE1 cells 72 h after transfection with control siRNAs (siControl) or siRNAs against HMG20A (siHMG20A). Data are the average of three independent experiments ± s.d. *Po0.05 versus control by ANOVA. (b) RPE1 cells were transfected with siControl or siHMG20A. Seventy-two hours after transfection a total lysate was prepared and analyzed by western blot for LSD1, SNAI1, HMG20A and α-tubulin. (c) Immunoﬂuorescence staining of SNAI1 in siControl or siHMG20A transfected cells. Bar, 10 μm. (d) Reporter assays showing HMG20A and LSD1 activation of SNAI1 promoter in RPE1 cells. RPE1 cells were transfected with empty vector (Flag) or expression vectors encoding Flag-LSD1 or Flag-HMG20A. Data are the average of two independent experiments ± s.d. *Po0.05, **Po0.01 versus control by ANOVA.

Oncogene (2015) 5264 – 5276 © 2015 Macmillan Publishers Limited HMG20A is required for EMT S Rivero et al 5269 SNAI1 CXADR CLDN12 KRT18 CDH4 1.2 6 7 4.5 4.0 1.0 5 6 4.0 3.5 3.5 3.0 0.8 4 5 3.0 2.5 4 2.5 0.6 3 2.0 3 2.0 0.4 2 1.5 1.5 2 1.0 1.0 0.2 1 1 0.5 0.5

Relative Fold (mRNA) 0.0 0 0 0.0 0.0 siControl + --- + --- + --- + --- + --- siHMG20A - + - - - + - - - + - - - + - - - + - - siLSD1 - - + - - - + - - - + - - - + - - - + - siSNAI1 - --+ - --+ ---+ - --+ - --+

SNAI1 CXADR CLDN12 1x104 6 3.5 8x103 5 3.0 2.5 4 4 2.0 3 3 1.5 2 2 1.0 1 1 0.5

Relative Fold (mRNA) 0 0 0.0 siControl + -+- + -+- + -+- siHMG20A - + - + - + - + - + - + Flag-SNAI1 - - + + - - + + - - + +

CXADR CLDN12 CXADR CLDN12 1.2 1.2 siControl 10 6 siHMG20A 9 siControl 1.0 1.0 8 5 siHMG20A 0.8 0.8 7 4 6 0.6 0.6 5 3 4 0.4 0.4 3 2 Relative Fold Relative Fold Relative Fold 0.2 Relative Fold 0.2 2 1 1 0.0 0.0 0 0 HMG20A LSD1 HMG20A LSD1 H3K4me2 H3K4me3 H3K4me2 H3K4me3 Figure 4. HMG20A, LSD1 and SNAI1 are repressors of epithelial genes. (a) SNAI1, CXADR, CLDN12, KRT18 and CDH4 mRNA expression was analyzed by RT-qPCR in RPE1 cells 72 h after transfection with the indicated siRNAs. Data are the average of three independent experiments ± s.d. *Po0.05, **Po0.01 versus control by ANOVA. (b) Twenty-four hours after transfection with the indicated siRNAs, RPE1 cells were transfected with an empty vector ( − ) or an expression vector encoding Flag-SNAI1 (+). SNAI1, CXADR and CLDN12 expression was analyzed by RT-qPCR. *Po0.05 at the indicated comparison by ANOVA. (c, d) Depletion of HMG20A is coupled with decreased occupancy of HMG20A and LSD1 (c) and elevated levels of H3K4me2 and H3K4me3 (d)atCXADR and CLDN12 promoters. RPE1 cells transfected with Control or HMG20A siRNAs were subjected to ChIP assays with the indicated antibodies. Relative fold values indicate occupancies relative to siControl set at 1. Values are average of multiple experiments (n ⩾2) ± s.d. *Po0.1, **Po0.05 compared with cells transfected with siControl, ANOVA.

Interestingly, HMG20A depletion decreased also LSD1 occupancy for 48 h and levels of SNAI1, CDH1 and CXADR mRNAs were at both promoters, suggesting that HMG20A is involved in determined by RT-qPCR. As expected, TGF-β-treated cells dis- recruitment of the LSD1-CoREST complex or its stabilization at played a strong induction of SNAI1 and downregulation of CDH1 the chromatin (Figure 4c). Next, we analyzed how HMG20A and CXADR expression. HMG20A depletion did not affect SNAI1 depletion affects the levels of methylation of H3K4. Consistently induction. In contrast, HMG20A depletion impaired downregula- with the reduced recruitment of the H3K4me2 demethylase LSD1, tion of CDH1 and CXADR mRNA levels (Figure 5b). Next, depletion of HMG20A strongly increased levels of H3K4me2 immunofluorescence of siControl- or siHMG20A-treated NMuMG (Figure 4d). Concomitantly, a strong increase in the level of cells using anti-CDH1 antibodies, in the absence or in the presence H3K4me3 was also observed. of TGF-β, was performed. TGF-β treatment strongly decreased CDH1 signal in siControl-treated cells. However, TGF-β-treated HMG20A is required for TGF-β-induced EMT cells depleted of HMG20A showed a significant level of CDH1 in Then, we investigated the role of HMG20A during the process of the cell-cell contacts. These results indicated that repression of EMT. For this, we used the well-established model of mouse epithelial markers that occurs during TGF-β-mediated EMT is mammary epithelial NMuMG cells treated with TGF-β.31 TGF-β is a impaired when the levels of HMG20A are reduced. We also pleiotropic signaling molecule that induces EMT in different cell investigated the effect of TGF-β in RPE1 cells. TGF-β treatment types.32–34 As a control, we verified that TGF-β treatment of during 48 h increased levels of SNAI1 and vimentin NMuMG cells for 48 h increased SNAI1 and decreased E-cadherin (Supplementary Figure S6a), indicating that TGF-β increased the (CDH1) protein expression. We also verified that levels of LSD1, mesenchymal phenotype of these cells. However, depletion of HMG20A and HMG20B did not change significantly during the HMG20A strongly increased expression of the epithelial markers process (Figure 5a). Next, NMuMG cells transfected with control CXADR and CLDN12 irrespectively of the presence of TGF-β siRNA or with a siRNA targeting HMG20A were treated with TGF-β (Supplementary Figure S6b and c). Taken together, our data

SNAI1 siControl

LSD1

HMG20B

HMG20A siHMG20A α-tubulin

CDH1/HMG20A/Dapi

SNAI1 CDH1 CXADR 35 1.2 1.2 30 1.0 1.0 25 0.8 0.8 20 0.6 0.6 15 0.4 0.4 10 5 0.2 0.2 Relative Fold (mRNA) Relative Fold (mRNA) 0 0.0 Relative Fold (mRNA) 0.0 TGF-β1 - + + TGF-β1 - + + TGF-β1 - + + siControl - + - siControl - + - siControl - + - siHMG20A - - + siHMG20A - - + siHMG20A - - + Figure 5. HMG20A is essential for repression of epithelial markers in TGF-β-induced EMT. (a) Western blot analysis of CDH1, SNAI1, LSD1, HMG20A, HMG20B and α-tubulin with proteins extracted from NMuMG cells treated with vehicle (− ) or 10 ng/ml TGF-β1 for 48 h (+). (b) Effect of siRNAs against HMG20A (siHMG20A) on SNAI1, CDH1 and CXADR mRNA levels during TGF-β-induced EMT. Data are the average of three independent experiments ± s.d. *Po0.05 versus control by ANOVA. (c) Immunoﬂuorescence staining of CDH1 and HMG20A in NMuMG cells transfected with siControl or siHMG20A treated with 10 ng/ml TGF-β1 or vehicle for 48 h. Nuclei were stained with DAPI. Bar, 10 μm.

demonstrate that HMG20A is essential for repression of epithelial co-precipitated with LSD1 and with HMG20A, both in RPE1 characters, both in NMuMG and in RPE1 cells. cells (data not shown) and in MDA-MB-231 cells (Figure 6b), indicating the existence of a LSD1-CoREST-HMG20A-SNAI1-SMAD4 SNAI1 and SMAD4 interact with LSD1-CoREST complexes complex. containing HMG20A SNAI1 recruits LSD1-CoREST complexes to the chromatin by a HMG20A is required for cell migration and invasiveness direct interaction between SNAI1 and LSD1.9,10 Therefore, we Our results suggest that HMG20A is important for repression decided to check whether HMG20A is present in endogenous of epithelial genes and consequently for the maintenance LSD1-CoREST complexes associated with SNAI1. To this end, we of the mesenchymal phenotype. High cell motility and invasive- performed immunoprecipitation of endogenous LSD1 or HMG20A ness are typical characteristics of mesenchymal cells.36,37 There- proteins from total extracts of RPE1 cells treated or non-treated fore, we studied whether HMG20A is also involved in cell with TGF-β and the presence of LSD1, CoREST, HMG20A and SNAI1 migration and invasiveness. Scratch assays were performed to in the precipitated complexes was determined. Figure 6a shows test whether loss of HMG20A affected migration of RPE1 cells that both HMG20A and SNAI1 were present in the LSD1 treated or not with TGF-β. RPE1 cells displayed high motility in the containing complexes. Conversely, LSD1 and SNAI1 were present absence of TGF-β, consistently with the relatively high basal level in the HMG20A containing complexes. Level of SNAI1 in the of SNAI1 in these cells (see Supplementary Figure S6a). Motility complexes was drastically increased in the presence of TGF-β. was signiﬁcantly enhanced in the presence of TGF-β (Figures 6c Complexes containing LSD1, CoREST, HMG20A and SNAI1 were and e). Importantly, depletion of HMG20A severely decreased cell also detected in MDA-MB-231 cells (Figure 6b). motility both in the absence and in the presence of TGF-β,in It has previously been shown that SNAI1 forms a complex with agreement with a role of HMG20A in promoting mesenchymal SMAD3 and SMAD4 transcription factors in breast epithelial cells characters (Figures 6c and e). We also investigated how cell and that this complex directly represses CXADR, CDH1, OCLN and migration is affected by HMG20A depletion in NMuMG cells. As CLDN3 genes during TGF-β-driven epithelial-to-mesenchymal shown in Figure 6d and quantiﬁed in Figure 6f, depletion of transition.35 Since we have shown that HMG20A interacts HMG20A also decreased TGF-β-induced cell motility in with SNAI1 and it is required for normal TGF-β-repression of NMuMG cells. CXADR and CDH1, we wondered whether SMAD4 protein was also To test cell invasiveness, we moved to the highly invasive and associated with LSD1-CoREST complexes that contain HMG20A. metastatic cell line MDA-MB-231. We monitored cell invasion by Immunoprecipitation experiments demonstrated SMAD4 using Matrigel-coated Boyden chambers. Depletion of HMG20A

Oncogene (2015) 5264 – 5276 © 2015 Macmillan Publishers Limited HMG20A is required for EMT S Rivero et al 5271 severely impaired invasiveness, to a similar extent as depletion of Expression of HMG20A in human cancer LSD1 and SNAI1, conﬁrming that HMG20A is essential for Analysis of publicly available gene expression sets using the mesenchymal characteristics (Figure 6g). ONCOMINE database38 revealed that HMG20A is signiﬁcantly

IP IP IP Input Control LSD1 HMG20A IP TGF-β1 -++ -+-+- LSD1 Input IgG LSD1 HMG20A LSD1 HMG20A HMG20A SNAI1 SNAI1 SMAD4 CoREST CoREST

siControl siHMG20A VehicleTGF-β1 Vehicle TGF-β1 0 h 24 h

siControl siHMG20A 0 h 24 h 0 h 24 h 1 β TGF-

200 500 120 180 450 160 400 100 140 350 80 120 300 100 250 60 80 n.s. 200 60 150 40 40 100 20 20 50

0 0 Invaded cells (% control) 0 Number of migrated cells Number of migrated cells TGF-β1 - +-+ TGF-β1 + + siControl + --- siControl + + - - siControl + - siHMG20A - + - - siHMG20A - - + + siHMG20A - + siLSD1 - - + - siSNAI1 - --+ Figure 6. SNAI1 and SMAD4 interact with LSD1-CoREST complexes containing HMG20A. (a) Whole-cell extracts from RPE1 cells treated with vehicle ( − ) or TGF-β1 (+) for 48 h were subjected to IP with anti-LSD1, anti-HMG20A or control (IgG) antibodies. Immunoprecipitated proteins and 3% of the input were analyzed by western blot with the indicated antibodies. In the case of HMG20A and SNAI1 two exposure times are shown. (b) SNAI1 and SMAD4 co-precipitate with the LSD1-CoREST complex in MDA-MB-231 cells. Whole-cell extracts were subjected to immunoprecipitation with anti-LSD1 antibody, anti-HMG20A or control (IgG) antibodies. Immunoprecipitated proteins and 3% of the input extracts were analyzed by western blot with the indicated antibodies. (c, d) HMG20A depletion inhibits cell migration in Scratch assays. RPE1 cells (c) or NMuMG cells (d) were transfected with siControl or siHMG20A. Twenty-four hours after transfection cells were plated onto culture inserts and treated with vehicle or with 10 ng/ml TGF-β1 as indicated. Culture inserts were displaced 24 h later. At this time point and 24 h later, phase-contrast pictures of the wounds at different locations were taken. (e, f) Quantification of representative scratch assays shown in (c) and (d), respectively. Cells in the scratch area were counted over 10 microscopic fields for each condition. The average numbers of migratory cells are plotted ± s.d. **Po0.01 versus control (except when indicated) by ANOVA. n.s., non significative. (g) HMG20A is required for MDA-MB-231 invasiveness. MDA-MB-231 cells were transfected with the indicated siRNAs and cell motility through Matrigel-coated filters was measured 18 h after plating. The migrating cells were stained, visualized by microscopy, and triplicate filters were counted in three individual experiments. Values are average (n ⩾3) ± s.d. **Po0.01 versus control by ANOVA.

© 2015 Macmillan Publishers Limited Oncogene (2015) 5264 – 5276 HMG20A is required for EMT S Rivero et al 5272 increased in three different sets of melanomas (from Po0.0001 to melanomas than in primary melanomas. Levels of HMG20A were P = 0.0008) (Figures 7a–c).39–41 Interestingly, levels of HMG20A also found to be elevated in aggressive mesothelioma (Po0.0003) were more signiﬁcantly increased in metastatic or malignant (Figure 7d),42 and in myxoid/round cell liposarcoma (Po0.0001)

Mesothelioma Melanoma Melanoma Melanoma (Riker et al., 2008) (Gordon et al., 2005) (Talantov et al., 2005) (Xu et al., 2008) P=0.0008 P=0.0003 P<0.0001 P=0.045 P=0.0005 P=0.0007 P=0.12 P<0.0001 Expression Expression Log2 expression Log2 expression

a g m kin ma nevi Lun leura astatic P Normal Primary Primary rmal s ma Malignant melanoMelanoma Basal cell melanoMet Malignant Benign No melanoma metastasis carcinoma melanoma Squamous cell helio carcinoma Pleural Mesot

Liposarcoma (Barretina et al., 2010) HMG20A-SNAI2 HMG20A-SMAD3 HMG20A-SMAD4 P<0.0001 Pearson:0.39 Pearson:0.68 Pearson:0.33 SNAI2 SMAD3 SMAD4 Log2 expression

HMG20A HMG20A HMG20A

Adipose TissueMyxoid/Round Cell Liposarcoma

HMG20A-ZEB1 HMG20A-ZEB2 HMG20A-FN1 HMG20A-CLDN3 Pearson:0.60 Pearson:0.64 Pearson:0.40 Pearson:-0.55 FN1 ZEB1 ZEB2 CLDN3

HMG20A HMG20A HMG20A HMG20A

HMG20A-KRT18 Pearson:-0.59 Gene name

Cancer type CLDN3 KRT18 KRT8 SMAD3 SMAD4 ZEB1 ZEB2 SNAI2 FN1 VIM Prostate Adenocarcinoma (45) -0.55 -0.59 -0.52 0.68 0.33 0.6 0.64 0.39 0.40 n.s. Prostate Adenocarcinoma (TCGA) -0.50 -0.55 -0.58 0.62 0.33 0.46 0.55 n.s. 0.55 0.41 KRT18 Uterine Carcinosarcoma (TCGA) n.s. -0.35 -0.30 n.s. 0.64 0.47 0.46 n.s. n.s. n.s. Kidney Renal Clear Cell Carcinoma (46 ) -0.33 -0.53 -0.50 n.s. 0.42 0.48 0.52 n.s. n.s. n.s.

HMG20A

Figure 7. Expression of HMG20A in human cancer. (a–e) Relative levels of HMG20A transcript in human samples from the indicated cancer types. Samples were grouped according to tissue origin or tumor grade. P-values were calculated using an unpaired t-test. (f–m) Relative mRNA levels of HMG20A and the corresponding levels of SNAI2 (f), SMAD3 (g), SMAD4 (h), ZEB1 (i), ZEB2 (j), FN1 (k), CLND3 (l) and KRT18 (m) transcripts in samples of prostate adenocarcinoma were plotted. Linear regression-fitted lines are shown. Pearson correlation coefficient is also shown. (n) Pearson coefficient of the correlations between expressions of HMG20A and the indicated genes in four different cancer data sets. Negative correlations are shown in red and positive correlations are shown in green. Pearson coefficient 4 − 0.3 and o0.3 was considered as non-significant (n.s.).

Oncogene (2015) 5264 – 5276 © 2015 Macmillan Publishers Limited HMG20A is required for EMT S Rivero et al 5273 (Figure 7e).43 We have also analyzed expression data from The sequence specificity (our unpublished results), it is possible that Cancer Genome Atlas (TCGA) and other studies through HMG20A contributes to stabilize the complex at the chromatin. cBioPortal.44 Interestingly, we observed that, in prostate 45 adenocarcinoma, levels of HMG20A mRNA positively correlated HMG20A is involved in EMT with those of the EMT-related transcription factors SNAI2, SMAD3, Transcriptomic analysis indicated that HMG20A and HMG20B have SMAD4, ZEB1 and ZEB2 and of the mesenchymal markers fi fi – both common and speci c targets. Many genes misregulated in bronectin (FN1) and vimentin (VIM) (Figures 7f k and n). HMG20A-depleted cells are related to the process of EMT. LSD1 Inversely, levels of HMG20A mRNA negatively correlated with – has previously been linked to EMT processes. In fact, LSD1 levels of KRT18, KRT8 and CLDN3 mRNAs (Figures 7l n). Similar interacts with the SNAG domain of SNAI1 and it has been shown positive and negative correlations, respectively, were also found in to be required for SNAI1-dependent repression of epithelial other prostate adenocarcinoma studies (TCGA), in uterine markers through its H3K4me2 demethylase activity.9,10 However, carcinosarcoma (TCGA) and in kidney renal clear cell carcinoma 47 46 other reports seem to directly contradict these data. Wang et al. (Figure 7n). Taken together, these data strongly support a role of found that LSD1 can inhibit the invasiveness of MDA-MB-231 HMG20A in the maintenance and/or the transition to the breast cancer cells and suppress their metastatic potential, and mesenchymal phenotype and suggest the involvement of this McDonald et al.17 reported that downregulation of LSD1 or factor in cancer metastasis. inhibition of its activity with pargyline increases expression of the mesenchymal marker vimentin in AML12 hepatocyte cells. We DISCUSSION found that HMG20A and LSD1 are essential factors controlling mesenchymal phenotype in RPE1 cells. Thus, knockdown of In this study, we show that HMG20A is a subunit of the LSD1- HMG20A or LSD1 upregulated epithelial markers such as CXADR, CoREST complex and that depletion of this protein impairs CLDN12, KRT18 and CDH4. Moreover, SNAI1 overexpression- recruitment of the complex to its targets. We find that HMG20A is mediated repression of CXADR and CLDN12 was impaired in required for SNAI1-dependent and TGF-β-dependent repression of HMG20A-depleted cells, indicating that SNAI1 requires HMG20A epithelial genes. Consistently SNAI1 and SMAD4 are found in to repress CXADR and CLDN12 genes. In agreement with these LSD1-CoREST complexes that contain HMG20A. In agreement with results, we found SNAI1 in HMG20A containing LSD1-CoREST these results, depletion of HMG20A strongly impairs cell migration complexes. Importantly, a similar level of downregulation of SNAI1 and invasiveness, two essential characteristics of mesenchymal and upregulation of epithelial genes was obtained by depletion of cells. Finally, we also analyzed cancer expression data sets and HMG20A or LSD1 (see Figures 3a and 4a), demonstrating that both found that expression of HMG20A correlates positively with factors are equally required for the function of the complex. We expression of mesenchymal markers and negatively with expres- also observed that levels of SNAI1 were downregulated in sion of epithelial markers. All together, our data demonstrate that HMG20A- or LSD1-depleted cells, suggesting a positive role of HMG20A is essential for the mesenchymal phenotype. the LSD1-CoREST complex in SNAI1 expression. However, we were unable to find HMG20A or LSD1 at the SNAI1 promoter, HMG20A is a subunit of the LSD1-CoREST complex suggesting an indirect effect. It is well known that levels of SNAI1 We have previously shown that HMG20A is able to form are affected by the epithelial-mesenchymal phenotype through homodimers and heterodimers with HMG20B through a coiled- several different signal transduction pathways. For example, coil region located in the carboxy terminal halves of both downregulation of CDH128,48 or expression of the metalloprotease proteins.25 In several proteomic analysis, HMG20A and HMG20B MMP349 is sufficient to trigger SNAI1 and EMT. Therefore, it is were found to copurify with LSD1,17,18,20 raising the question of possible that derepression of epithelial genes promoted by LSD1- how the interplay of these two proteins is within the LSD1-CoREST CoREST complex deficiency results in downregulation of SNAI1 complex. By knockdown and overexpression experiments fol- expression. lowed by immunoprecipitation of the complexes, we demonstrate TGF-β is an important player in the epithelial-mesenchymal that HMG20A and HMG20B are mutually exclusive subunits of the equilibrium. TGF-β treatment of different cell types increased complex. Both proteins use their coiled-coil regions to interact expression of SNAI1.1,32 In addition, TGF-β triggers phosphoryla- with the complex. Therefore, they use the same domain for tion of SMAD2/3 factors, which associate with SMAD4 and heterodimerization and for interaction with the complex, suggest- translocate to the nucleus where they control the expression of ing that heterodimerization inhibits interaction with the complex. target genes. It has been shown that SMAD3/SMAD4 form a In fact, the same mutations that inhibit dimerization also inhibit complex with SNAI1 and cooperate in the repression of CXADR, interaction with the complex (Figure 1).25 Both proteins directly CDH1 and OCLN epithelial genes.35 Interestingly, we were able to interact with the core subunit BHC80. Purification of the LSD1- immunoprecipitate SMAD4 by using HMG20A antibodies, sug- CoREST complex by using anti-LSD1 antibodies often brings down gesting the existence of a complex containing SNAI1, SMAD4, lower amounts of HMG20A than that of HMG20B,17,18,20 suggest- HMG20A and other components of the LSD1-CoREST complex. In ing that the proportion of complexes containing HMG20A is lower agreement with this data, a yeast two-hybrid high throughput than the complexes containing HMG20B, at least under the screening identified an interaction between HMG20A and reported conditions. Consistently, we observed that BHC80 seems SMAD4.50 Consistently with this interaction, HMG20A was required to display higher affinity for HMG20B than for HMG20A, in vitro. All for TGF-β-dependent repression of CXADR and CDH1 in the together, our experiments suggest that the amount of LSD1- epithelial cell line NMuMG. Furthermore, HMG20A was also CoREST complexes having HMG20A or HMG20B is controlled by a required for cell migration promoted by TGF-β and for invasive- delicate equilibrium between the formation of HMG20A-HMG20A ness of the metastatic cell line MDA-MB-231. homodimers, HMG20A-HMG20B heterodimers and the association EMT transdifferentiation have been related to cancer cell of each protein with the complex through interaction with BHC80 invasion and metastatic spread.2,5,8 The fact that HMG20A is (Figure 1h). What is the role of HMG20A in the complex? We show upregulated in three independent malignant melanoma studies here that depletion of HMG20A reduces LSD1 occupancy at target and the strong positive correlation between HMG20A and genes, suggesting that HMG20A is involved in recruiting the LSD1- mesenchymal markers expression levels, in several types of CoREST complex. Alternatively, since HMG20A contains an HMG tumors, opens the possibility to investigate the role of this protein domain that binds structured DNA without or with very low in cancer metastasis.

© 2015 Macmillan Publishers Limited Oncogene (2015) 5264 – 5276 HMG20A is required for EMT S Rivero et al 5274 Increasing evidences demonstrate that pure epithelial and and mouse monoclonal anti-CAR (Millipore). Secondary antibodies used mesenchymal phenotypes are the ends of a spectrum of transitory were Texas Red dye-conjugated goat anti-rabbit IgG (Jackson ImmunoR- cell states that can dynamically change during development or esearch, West Grove, PA, USA) and FITC-conjugated goat anti-mouse IgG cancer progression. At the chromatin level, these changes seem to (Molecular Probes, Life Technologies, Paisley, UK). be accompanied by a variety of progressively more stable epigenetic modifications driven by a multiplicity of histone- Protein expression and in vitro pull-down assays modifying enzymes.8 In agreement with this view, other GST and GST fusion constructs were transformed in Escherichia coli BL21 chromatin modifiers have been related to SNAI1-mediated (DE3). Production of the fusion proteins was induced for 3 h at 37 °C after repression, in addition to LSD1-CoREST, including Polycomb addition of 1 mM IPTG. Recombinant proteins were purified using proteins that introduce methylation of H3K27,11 SUV39H1 and Glutathione sepharose 4B beads (GE Healthcare) following the manufac- ’ G9a that mediate H3K9 methylation 51 and recently LOXL2, an turer s instructions. In vitro transcription/translation reactions were 52 performed with the TNT Quick Coupled Transcription/Translation System H3K4 deaminase. SNAI1 is also able to recruit DNA methyl- (Promega, Madison, WI, USA) in the presence of [35S]Methionine (Perkin- transferases to the E-cadherin promoter, to promote DNA μ 51 Elmer, Boston, MA, USA) using 1 g of template. methylation. Further investigation will be required to under- For pull-down experiments, GST or GST-BHC80 proteins loaded on stand how the choreography of these elements is executed. Glutathione sepharose 4B beads (GE Healthcare) were incubated with in vitro transcription/translation reaction mixtures containing [35S]-methionine-labeled HMG20A, HMG20B or eIF3b proteins, in 20 mM Tris-HCl, pH 7, MATERIALS AND METHODS 100 mM NaCl, 1 mM EDTA pH 8, 10% glycerol and 0.01% NP-40. After 12 h of Cells and treatments incubation at 4 °C, glutathione sepharose beads were collected by centrifugation and washed three times with 1 ml of the same buffer to Human retinal pigment epithelial cell line RPE1 (Clontech, Palo Alto, CA, remove unbound proteins. The beads were boiled in Laemmli sample USA) was grown in DMEM/F12 with 10% fetal bovine serum. Human breast buffer, and the samples were analyzed by SDS-PAGE. The gels were dried cancer cell line MDA-MB-231 and human embryonic kidney cell line 293T on Whatman 3 MM paper and subjected to autoradiography. were grown in DMEM supplemented with 10% fetal bovine serum. Normal murine mammary gland cells NMuMG were cultured in DMEM supplemented with 10% fetal bovine serum and 10 μg/ml insulin. Cells were Microarray expression analysis maintained at 37 °C in 5% CO2. Transient EMT was induced by the addition RPE1 cells were transfected with control, HMG20A or HMG20B siRNAs. of TGF-β1 (10 ng/ml; R&D Systems, Minneapolis, MN, USA) for 48 h. After 72 h, total RNA was isolated in triplicate from cells using the RNeasy Mini Kit (Qiagen, Hilden, Germany). Purity and quality of isolated RNA were Cloning and DNA constructs assessed by RNA 6000 Nano assay on a 2100 Bioanalyzer (Agilent Technologies, Santa 6 Clara, CA, USA). RNA (100 ng) was used for Mouse cDNAs of HMG20A and HMG20B were kindly provided by L production of end-labeled biotinylated ssDNA. Labeled ssDNA was Sumoy.53 HMG20A and HMG20B expression vectors were previously 25 hybridized to the GeneChip human Gene 1.0 ST array oligonucleotide described. GST-BHC80 construction was derived from pGEX-6P-3 (GE microarray (Affymetrix, Santa Clara, CA, USA) according to the manufac- Healthcare, Buckinghamshire, UK). Full-length human BHC80 cDNA was turer’s recommendations. The arrays were scanned using the GeneChip obtained from pCMV-BHC80-delta4 non-tag kindly provided by Dr Tadashi 26 Scanner 3000 7G (Affymetrix), and raw data were extracted from the Baba. Plasmids for in vitro transcription/translation reactions were scanned images and analyzed with the Affymetrix GeneChip Command constructed using pBlueScript SK(+). pGL3-hSNAI1 promoter was kindly 29 Console Software (Affymetrix). The raw array data were pre-processed and provided by A García de Herreros. Full-length human SNAI1 with a Flag 55 9 normalized using the Robust Multichip Average (RMA) method. Data epitope tag at its carboxyl end was kindly provided by J Lu. were further processed using oneChannelGUI.56 Genes differentially expressed in HMG20A- or HMG20B-depleted cells with respect to control Transfections and luciferase reporter assays cells more 1.25-fold (lineal change) and with P-valueo0.01 were selected for further analysis. Gene ontology functional categories were analyzed Transient transfections of cells with expression or reporter plasmids were using DAVID.57 GSEA was performed using the GSEA v2.0.14 software performed with Fugene (Roche, Hoffmann-La Roche Inc., Nutley, NJ, USA). (GSEA, Broad Institute, Cambridge, MA, USA) with 1000 phenotype For siRNA transfections, we used Oligofectamine (Invitrogen, Life permutations.27 Gene signature constitute by genes upregulated in EMT Technologies, Karlsrube, Germany) for RPE1 cells and RNAiMAX (Invitrogen) was obtained from Taube et al.28 The TGF-β related gene sets were for NMuMG cells. The following siRNA sequences were used: human and download from MSigDB. mouse HMG20A 5′-AGGCAAAUCUCAUAGGCAA-3′;humanHMG20B 5′-GAG CUUCGGCGCUUGCGGA-3′;humanLSD1 5′-AAGGAAAGCUAGAAGAAAA-3′; human SNAI1 5′-GGUGUGACUAACUAUGCAA-3′; siCt, 5′-CGUACGCGGAAU Scratch assays and invasion assays 54 ACUUCGA-3′. Luciferase assay was performed as previously described. Scratch assays were performed as previously described.58 Following transfection with either control or HMG20A siRNAs, an equal number of Antibodies, western blotting, immunoprecipitation and cells was plated into culture inserts (Ibidi, Martinsried, Germany) and immunohistochemistry incubated at 37 °C for 24 h. The next day, culture inserts were removed and the wound was imaged at 0 h, and again after 24 h. The Invasion assays Co-immunoprecipitations, immunohistochemistry and western blotting 59 μ 25 were performed as previously described by using 24-well Transwell 8 m were performed as previously described. The antibodies used for filters that were coated with 10 μg growth factor-reduced Matrigel immunoprecipitations and western blot were rat monoclonal anti-HA (Trevigen, Gaithersburg, MD, USA). Fibroblast conditioned medium, which (Roche), mouse monoclonal anti-Flag (M2, F1804, Sigma-Aldrich, St Louis, was obtained by a 24 h incubation of NIH3T3 cells with 50 μg ascorbic acid MO, USA), rabbit polyclonal anti-LSD1 (Abcam, Cambridge, UK), mouse in serum-free DME media, was used as a chemoattractant. Cell suspension monoclonal anti-LSD1 (AOF2, Sigma-Aldrich), rabbit polyclonal anti-CoREST was plated at 1 × 105 cells/0.5 ml DME, 0.1% BSA per well into the upper (Millipore, Billerica, MA, USA), mouse monoclonal anti-HDAC2 (3F3, compartment of the Boyden chamber for 18 h. Cells on top of the filters Millipore), mouse monoclonal anti-HMG20B (Abnova Corp., Taipai, Taiwan), were removed, and cells penetrating the filters were stained with 4´,6- rabbit polyclonal anti-HMG20A (Sigma-Aldrich) and rabbit polyclonal diamidino-2-phenylindole to stain nuclei. Filters were counted and invaded anti-SMAD4 (Millipore). As secondary antibodies, goat anti-mouse-HRP cells were expressed as the average number of migrated cells bound per and anti-rabbit-HRP (Sigma-Aldrich) were used. microscopic field over four fields per assay in triplicate experiments. Antibodies used in ChIP experiments were rabbit polyclonal anti-LSD1 (Abcam), rabbit polyclonal anti-HMG20A (Sigma-Aldrich), rabbit polyclonal to Histone H3 (tri methyl K4) (Abcam) and rabbit monoclonal to Histone RNA extraction and RT-PCR H3 (di methyl K4) (Abcam). Antibodies used in immunofluorescence Total RNA was prepared by using the RNAsy Kit (Qiagen), as described in were mouse monoclonal anti-SNAI1 (Cell Signaling Technology, Danvers, the manufacturer’s instructions, including Dnase I digestion to avoid MA, USA), rabbit polyclonal anti-HMG20A (Sigma-Aldrich), mouse anti-E- potential contaminations of DNA. The cDNA was generated from 100 ng of cadherin (BD Transduction Laboratories, BD Biosciences, San Jose, CA, USA) total RNA by using SuperScript First Strand Synthesis System (Invitrogen).

Oncogene (2015) 5264 – 5276 © 2015 Macmillan Publishers Limited HMG20A is required for EMT S Rivero et al 5275 In all, 2 μl of generated cDNA solution was used as a template for real-time 15 Lim S, Janzer A, Becker A, Zimmer A, Schule R, Buettner R et al. Lysine-specific PCR (qPCR). Quantification of gene products was performed by qPCR with demethylase 1 (LSD1) is highly expressed in ER-negative breast cancers and a the Applied Biosystems 7500 FAST Real-Time PCR System, using Applied biomarker predicting aggressive biology. Carcinogenesis 2010; 31:512–520. Biosystems Power SYBR Green Master Mix. Values were normalized to the 16 Ferrari-Amorotti G, Fragliasso V, Esteki R, Prudente Z, Soliera AR, Cattelani S et al. expression of GAPDH housekeeping gene. Each experiment was performed Inhibiting interactions of lysine demethylase LSD1 with snail/slug blocks cancer at least in triplicate. Sequences of all oligonucleotides used are listed in cell invasion. Cancer Res 2012; 73:235–245. Supplementary Table S3. 17 McDonald OG, Wu H, Timp W, Doi A, Feinberg AP. Genome-scale epigenetic reprogramming during epithelial-to-mesenchymal transition. Nat Struct Mol Biol 18 – ChIP assays 2011; :867 874. 18 Saleque S, Kim J, Rooke HM, Orkin SH. Epigenetic regulation of hematopoietic 54 fi ChIP assays were performed as previously described. Quanti cation of differentiation by Gfi-1 and Gfi-1b is mediated by the cofactors CoREST and LSD1. fi immunoprecipitated DNA was performed by qPCR. ChIP was quanti ed by Mol Cell 2007; 27:562–572. using three qPCR determinations. Provided data are the average of at least 19 Shi Y, Sawada J, Sui G, Affar el B, Whetstine JR, Lan F et al. Coordinated histone three independent experiments. Sequences of all oligonucleotides used modifications mediated by a CtBP co-repressor complex. Nature 2003; 422: are listed in Supplementary Table S4. 735–738. 20 Shi YJ, Matson C, Lan F, Iwase S, Baba T, Shi Y. Regulation of LSD1 histone Analysis of tumor data sets demethylase activity by its associated factors. Mol Cell 2005; 19:857–864. 21 Hakimi MA, Bochar DA, Chenoweth J, Lane WS, Mandel G, Shiekhattar R. Microarray gene expression data were obtained from ONCOMINE.38 A core-BRAF35 complex containing histone deacetylase mediates repression P-values were calculated using a two-tailed Student´s t-test by using of neuronal-specific genes. Proc Natl Acad Sci USA 2002; 99:7420–7425. GraphPad Prism version 5.0 (GraphPad Software, San Diego, CA, USA). 22 You A, Tong JK, Grozinger CM, Schreiber SL. CoREST is an integral component of RNA-seq expression data from TCGA (http://cancergenome.nih.gov/) were the CoREST- human histone deacetylase complex. Proc Natl Acad Sci USA 2001; obtained through cBioPortal.44 Regression plots and determination of 98: 1454–1458. Pearson coefficients were performed using GraphPad Prism version 5.0. 23 Marmorstein LY, Kinev AV, Chan GK, Bochar DA, Beniya H, Epstein JA et al. For all provided data, slops of linear regression equations were considered A human BRCA2 complex containing a structural DNA binding component significantly different from 0 with P valueo0.0001 using F-test. influences cell cycle progression. Cell 2001; 104: 247–257. 24 Wynder C, Hakimi MA, Epstein JA, Shilatifard A, Shiekhattar R. Recruitment of MLL ACKNOWLEDGEMENTS by HMG-domain protein iBRAF promotes neural differentiation. Nat Cell Biol 2005; 7: 1113–1117. We thank L Sumoy, T Baba, A García de Herreros and J Lu for reagents and plasmids. 25 Ceballos-Chavez M, Rivero S, Garcia-Gutierrez P, Rodriguez-Paredes M, Garcia- We thank Dr M Garcia-Dominguez and Dr JA Pintor-Toro for critical reading of the Dominguez M, Bhattacharya S et al. Control of neuronal differentiation by manuscript. We thank E Andújar and M Pérez from the CABIMER Genomic Unit for sumoylation of BRAF35, a subunit of the LSD1-CoREST histone demethylase microarray expression analysis. The results published here are in part based upon complex. Proc Natl Acad Sci USA 2012. data generated by the TCGA Research Network: http://cancergenome.nih.gov/. This 26 Iwase S, Januma A, Miyamoto K, Shono N, Honda A, Yanagisawa J. Character- work was funded by the Spanish Ministry of Economy and Competitiveness ization of BHC80 in BRAF-HDAC complex, involved in neuron-specific gene [BFU-2011-23442 to JCR] and a Juan de la Cierva grant to SR and by Junta de repression. Biochem Biophys Res Commun 2004; 322:601–608. Andalucía (P09-CTS-04967 to SSB). 27 Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci USA 2005; 102: REFERENCES 15545–15550. 1 Lamouille S, Xu J, Derynck R. Molecular mechanisms of epithelial-mesenchymal 28 Taube JH, Herschkowitz JI, Komurov K, Zhou AY, Gupta S, Yang J et al. Core transition. Nat Rev Mol Cell Biol 2014; 15: 178–196. epithelial-to-mesenchymal transition interactome gene-expression signature is 2 Nieto MA. Epithelial plasticity: a common theme in embryonic and cancer cells. associated with claudin-low and metaplastic breast cancer subtypes. Proc Natl Science 2013; 342: 1234850. Acad Sci USA 2010; 107: 15449–15454. 3 Nieto MA. The ins and outs of the epithelial to mesenchymal transition in health 29 Barbera MJ, Puig I, Dominguez D, Julien-Grille S, Guaita-Esteruelas S, Peiro S et al. and disease. Annu Rev Cell Dev Biol 2011; 27:347–376. Regulation of Snail transcription during epithelial to mesenchymal transition of 4 Kalluri R, Weinberg RA. The basics of epithelial-mesenchymal transition. J Clin tumor cells. Oncogene 2004; 23: 7345–7354. Invest 2009; 119: 1420–1428. 30 McKay BS, Irving PE, Skumatz CM, Burke JM. Cell-cell adhesion molecules and the 5 Tsai JH, Yang J. Epithelial-mesenchymal plasticity in carcinoma metastasis. Genes development of an epithelial phenotype in cultured human retinal pigment Dev 2013; 27: 2192–2206. epithelial cells. Exp Eye Res 1997; 65:661–671. 6 Barrallo-Gimeno A, Nieto MA. The Snail genes as inducers of cell movement and 31 Miettinen PJ, Ebner R, Lopez AR, Derynck R. TGF-beta induced transdifferentiation survival: implications in development and cancer. Development 2005; 132: of mammary epithelial cells to mesenchymal cells: involvement of type I recep- 3151–3161. tors. J Cell Biol 1994; 127: 2021–2036. 7 Wu CY, Tsai YP, Wu MZ, Teng SC, Wu KJ. Epigenetic reprogramming and 32 Massague J. TGFbeta signalling in context. Nat Rev Mol Cell Biol 2012; 13:616–630. post-transcriptional regulation during the epithelial-mesenchymal transition. 33 Li H, Wang H, Wang F, Gu Q, Xu X. Snail involves in the transforming growth factor Trends Genet 2012; 28:454–463. beta1-mediated epithelial-mesenchymal transition of retinal pigment 8 Tam WL, Weinberg RA. The epigenetics of epithelial-mesenchymal plasticity epithelial cells. PLoS One 2011; 6: e23322. in cancer. Nat Med 2013; 19: 1438–1449. 34 Xu J, Lamouille S, Derynck R. TGF-beta-induced epithelial to mesenchymal tran- 9 Lin T, Ponn A, Hu X, Law BK, Lu J. Requirement of the histone demethylase LSD1 sition. Cell Res 2009; 19:156–172. in Snai1-mediated transcriptional repression during epithelial-mesenchymal 35 Vincent T, Neve EP, Johnson JR, Kukalev A, Rojo F, Albanell J et al. A SNAIL1- transition. Oncogene 2010; 29: 4896–4904. SMAD3/4 transcriptional repressor complex promotes TGF-beta mediated 10 Lin Y, Wu Y, Li J, Dong C, Ye X, Chi YI, Evers BM, Zhou BP. The SNAG domain of epithelial-mesenchymal transition. Nat Cell Biol 2009; 11: 943–950. Snail1 functions as a molecular hook for recruiting lysine-specific demethylase 1. 36 Yilmaz M, Christofori G. EMT the cytoskeleton, and cancer cell invasion. Cancer EMBO J 2010; 29: 1803–1816. Metastasis Rev 2009; 28:15–33. 11 Herranz N, Pasini D, Diaz VM, Franci C, Gutierrez A, Dave N et al. Polycomb 37 Yilmaz M, Christofori G. Mechanisms of motility in metastasizing cells. Mol Cancer complex 2 is required for E-cadherin repression by the Snail1 transcription factor. Res 2010; 8:629–642. Mol Cell Biol 2008; 28: 4772–4781. 38 Rhodes DR, Yu J, Shanker K, Deshpande N, Varambally R, Ghosh D. ONCOMINE: 12 Shi Y, Lan F, Matson C, Mulligan P, Whetstine JR, Cole PA et al. Histone deme- a cancer microarray database and integrated data-mining platform. Neoplasia thylation mediated by the nuclear amine oxidase homolog LSD1. Cell 2004; 119: 2004; 6:1–6. 941–953. 39 Xu L, Shen SS, Hoshida Y, Subramanian A, Ross K, Brunet JP et al. Gene expression 13 Lee MG, Wynder C, Cooch N, Shiekhattar R. An essential role for CoREST in changes in an animal melanoma model correlate with aggressiveness of human nucleosomal histone 3 lysine 4 demethylation. Nature 2005; 437: 432–435. melanoma metastases. Mol Cancer Res 2008; 6:760–769. 14 Metzger E, Wissmann M, Yin N, Muller JM, Schneider R, Peters AH et al. LSD1 40 Talantov D, Mazumder A, Yu JX, Briggs T, Jiang Y, Backus J et al. Novel genes demethylates repressive histone marks to promote androgen-receptor- associated with malignant melanoma but not benign melanocytic lesions. Clin dependent transcription. Nature 2005; 437:436–439. Cancer Res 2005; 11: 7234–7242.

© 2015 Macmillan Publishers Limited Oncogene (2015) 5264 – 5276 HMG20A is required for EMT S Rivero et al 5276 41 Riker AI, Enkemann SA, Fodstad O, Liu S, Ren S, Morris C et al. The gene expression 50 Colland F, Jacq X, Trouplin V, Mougin C, Groizeleau C, Hamburger A et al. Functional profiles of primary and metastatic melanoma yields a transition point of tumor proteomics mapping of a human signaling pathway. Genome Res 2004; 14: 1324–1332. progression and metastasis. BMC Med Genomics 2008; 1: 13. 51 Dong C, Wu Y, Yao J, Wang Y, Yu Y, Rychahou PG et al. G9a interacts with Snail 42 Gordon GJ, Rockwell GN, Jensen RV, Rheinwald JG, Glickman JN, Aronson JP et al. and is critical for Snail-mediated E-cadherin repression in human breast cancer. Identification of novel candidate oncogenes and tumor suppressors in malignant J Clin Invest 2012; 122:1469–1486. pleural mesothelioma using large-scale transcriptional profiling. Am J Pathol 2005; 52 Millanes-Romero A, Herranz N, Perrera V, Iturbide A, Loubat-Casanovas J, 166: 1827–1840. Gil J et al. Regulation of heterochromatin transcription by Snail1/LOXL2 during 43 Barretina J, Taylor BS, Banerji S, Ramos AH, Lagos-Quintana M, Decarolis PL et al. epithelial-to-mesenchymal transition. Mol Cell 2013; 52:746–757. Subtype-specific genomic alterations define new targets for soft-tissue sarcoma 53 Sumoy L, Carim L, Escarceller M, Nadal M, Gratacos M, Pujana MA et al. HMG20A therapy. Nat Genet 2010; 42: 715–721. and HMG20B map to human chromosomes 15q24 and 19p13.3 and constitute a 44 Gao J, Aksoy BA, Dogrusoz U, Dresdner G, Gross B, Sumer SO et al. Integrative distinct class of HMG-box genes with ubiquitous expression. Cytogenet Cell Genet analysis of complex cancer genomics and clinical profiles using the cBioPortal. 2000; 88:62–67. Sci Signal 2013; 6: pl1. 54 Subtil-Rodriguez A, Reyes JC. BRG1 helps RNA polymerase II to overcome 45 Barbieri CE, Baca SC, Lawrence MS, Demichelis F, Blattner M, Theurillat JP et al. a nucleosomal barrier during elongation, in vivo. EMBO Rep 2010; 11: 751–757. Exome sequencing identifies recurrent SPOP, FOXA1 and MED12 mutations in 55 Irizarry RA, Bolstad BM, Collin F, Cope LM, Hobbs B, Speed TP. Summaries of prostate cancer. Nat Genet 2012; 44:685–689. Affymetrix GeneChip probe level data. Nucleic Acids Res 2003; 31: e15. 46 Cancer Genome Atlas Research Network Comprehensive molecular character- 56 Sanges R, Cordero F, Calogero RA. oneChannelGUI: a graphical interface to Bio- ization of clear cell renal cell carcinoma. Nature 2013; 499:43–49. conductor tools, designed for life scientists who are not familiar with R language. 47 Wang Y, Zhang H, Chen Y, Sun Y, Yang F, Yu W et al. LSD1 is a subunit of the Bioinformatics 2007; 23: 3406–3408. NuRD complex and targets the metastasis programs in breast cancer. Cell 2009; 138: 57 Huang da W, Sherman BT, Lempicki RA. Systematic and integrative analysis 660–672. of large gene lists using DAVID bioinformatics resources. Nat Protoc 2009; 4: 48 Onder TT, Gupta PB, Mani SA, Yang J, Lander ES, Weinberg RA. Loss of E-cadherin 44–57. promotes metastasis via multiple downstream transcriptional pathways. Cancer 58 Liang CC, Park AY, Guan JL. In vitro scratch assay: a convenient and inexpensive Res 2008; 68: 3645–3654. method for analysis of cell migration in vitro. Nat Protoc 2007; 2:329–333. 49 Radisky DC, Levy DD, Littlepage LE, Liu H, Nelson CM, Fata JE et al. Rac1b and 59 Hazan RB, Phillips GR, Qiao RF, Norton L, Aaronson SA. Exogenous expression of reactive oxygen species mediate MMP-3-induced EMT and genomic instability. N-cadherin in breast cancer cells induces cell migration, invasion, and metastasis. Nature 2005; 436:123–127. J Cell Biol 2000; 148: 779–790.

Supplementary Information accompanies this paper on the Oncogene website (http://www.nature.com/onc)