Epithelial-Mesenchymal Transition Cross-Journal Focus EMT Epithelial-Mesenchymal Transition

EMBO Molecular Medicine

cross-journal focus EMT Epithelial-Mesenchymal Transition cross-journal focus EMT Epithelial-Mesenchymal Transition

EDITORS David del Alamo Editor [email protected] | T +49 6221 8891 309

David received his PhD. from the Madrid’s Autónoma University where he studied proximal-distal patterning in Drosophila with Fernando Díaz-Benjumea. As a postdoc, he continued working with Drosophila, first in Marek Mlodzik’s lab (Mount Sinai School of Medicine, New York) on the mechanisms of epithelial planar cell polarity generation, and then with François Schweisguth (Institut Pasteur, Paris) where he focused on the modulation of Notch signalling in lateral inhibition. David joined The EMBO Journal in 2011.

Nonia Pariente Senior Editor [email protected] | T +49 6221 8891 305

Nonia joined EMBO Reports in August 2007. She studied biochemistry and molecular biology in Madrid’s Autónoma University, where she also gained her PhD on the generation of new antiviral strategies against RNA viruses. She did a four-year post-doc at UCLA focusing on the development of new strategies for gene therapy.

Roberto Buccione Editor [email protected] | T +49 6221 8891 310

Roberto Buccione completed his PhD at the University of l’Aquila, Italy studying the process of oogenesis in mammals. After continuing these studies as a post-doctoral researcher at the EMBO Jackson Laboratory, Bar Harbor ME, USA, he joined the Mario Negri Sud research institute in S. Molecular Maria Imbaro, Italy, where he lead a research group focused on the cell biology of tumour cell Medicine invasion. He joined EMBO Molecular Medicine as a Scientific Editor in October 2012.

Maria Polychronidou Editor [email protected] | T +49 6221 8891 410

Maria received her PhD from the University of Heidelberg, where she studied the role of nuclear membrane proteins in development and aging. During her post-doctoral work, she focused on the analysis of tissue-specific regulatory functions of Hox transcription factors using a combination of computational and genome-wide methods.

emboj.embopress.org | embor.embopress.org | embomolmed.embopress.org | msb.embopress.org full articles

The EMBO Journal Genome-wide identification of miR-200 targets reveals a regulatory network controlling cell invasion. Bracken CP, Li X, Wright JA, Lawrence DM, Pillman KA, Salmanidis M, Anderson MA, Dredge BK, Gregory PA, Tsykin A, Neilsen C, Thomson DW, Bert AG, Leerberg JM, Yap AS, Jensen KB, Khew-Goodall Y, Goodall GJ. DOI: 10.15252/embj.201488641 | Published 28.07.2014 MiR-133 promotes cardiac reprogramming by directly repressing Snai1 and silencing fibroblast signatures. Muraoka N, Yamakawa H, Miyamoto K, Sadahiro T, Umei T, Isomi M, Nakashima H, Akiyama M, Wada R, Inagawa K, Nishiyama T, Kaneda R, Fukuda T, Takeda S, Tohyama S, Hashimoto H, Kawamura Y, Goshima N, Aeba R, Yamagishi H, Fukuda K, Ieda M. DOI: 10.15252/embj.201387605 | Published 11.06.2014

EMBO Reports Review Dedifferentiation and reprogramming: origins of cancer stem cells. Friedmann-Morvinski D, Verma IM. DOI: 10.1002/embr.201338254 | Published 14.02.2014 Article E2F1 induces miR-224/452 expression to drive EMT through TXNIP downregulation. Knoll S, Fürst K, Kowtharapu B, Schmitz U, Marquardt S, Wolkenhauer O, Martin H, Pützer BM. DOI: 10.15252/embr.201439392 | Published 23.10.2014

EMBO Molecular Medicine Epithelial-mesenchymal transition spectrum quantification and its efficacy in deciphering survival and drug responses of cancer patients. Tan TZ, Miow QH, Miki Y, Noda T, Mori S, Huang RY, Thiery JP. DOI: 10.15252/emmm.201404208 | Published 11.09.2014

Molecular Systems Biology A computational study of the Warburg effect identifies metabolic targets inhibiting cancer migration. Yizhak K, Le Dévédec SE, Rogkoti VM, Baenke F, de Boer VC, Frezza C, Schulze A, van de Water B, Ruppin E. DOI: 10.15252/msb.20134993 | Published 01.08.2014

For further reading please see inside back cover

Article

Genome-wide identification of miR-200 targets reveals a regulatory network controlling cell invasion

Cameron P Bracken1,2, Xiaochun Li1, Josephine A Wright1, David M Lawrence1, Katherine A Pillman1, Marika Salmanidis1, Matthew A Anderson1, B Kate Dredge3, Philip A Gregory1,2, Anna Tsykin1, Corine Neilsen1, Daniel W Thomson1, Andrew G Bert1, Joanne M Leerberg4, Alpha S Yap4, Kirk B Jensen3, Yeesim Khew-Goodall1,3,*,† & Gregory J Goodall1,2,3,**,†

Abstract Introduction

The microRNAs of the miR-200 family maintain the central charac- MicroRNAs and their targets are important components in the regu- teristics of epithelia and inhibit tumor cell motility and invasive- latory networks that maintain cell phenotype and control cell differ- ness. Using the Ago-HITS-CLIP technology for transcriptome-wide entiation. Although microRNAs typically act as mild modulators of identification of direct microRNA targets in living cells, along gene expression, exerting only a modest inhibitory effect on individ- with extensive validation to verify the reliability of the approach, ual targets, conceivably they can broadly refine gene expression we have identified hundreds of miR-200a and miR-200b targets, patterns because each microRNA may target several hundred different providing insights into general features of miRNA target site mRNAs. Thus, one microRNA can potentially influence a biological selection. Gene ontology analysis revealed a predominant effect process by having a coordinated effect on multiple components of miR-200 targets in widespread coordinate control of actin of a network or pathway. However, due to the uncertainties in cytoskeleton dynamics. Functional characterization of the miR-200 predicting or experimentally identifying the spectrum of targets of targets indicates that they constitute subnetworks that underlie individual miRNAs, there are few confirmed examples of broad the ability of cancer cells to migrate and invade, including coor- network regulation by a miRNA. dinate effects on Rho-ROCK signaling, invadopodia formation, The miR-200 family of microRNAs acts as enforcers of the epithe- MMP activity, and focal adhesions. Thus, the miR-200 family lial phenotype. They are expressed in most, if not all, epithelial cells maintains the central characteristics of the epithelial phenotype and their expression must be turned off for epithelial to mesenchy- by acting on numerous targets at multiple levels, encompassing mal transition (EMT) to occur (Burk et al, 2008; Gregory et al, both cytoskeletal effectors that control actin filament organiza- 2008a; Korpal et al, 2008; Park et al, 2008). EMT involves a tion and dynamics, and upstream signals that locally regulate morphological change whereby immobile epithelial cells acquire the cytoskeleton to maintain cell morphology and prevent cell pro-invasive mesenchymal characteristics that are key to various migration. developmental processes and are drivers of metastatic progression in cancer (Nieto, 2013). Two major targets of miR-200 in controlling Keywords cytoskeleton; HITS-CLIP; invadopodia; microRNA; miR-200 EMT are the transcription repressors, ZEB1 and ZEB2 (formerly Subject Categories Cancer; Cell Adhesion, Polarity & Cytoskeleton; RNA known as SIP1) (Gregory et al, 2008a; Park et al, 2008). Further- Biology more, the ZEB proteins are strong repressors of transcription of the DOI 10.15252/embj.201488641 | Received 2 April 2014 | Revised 6 June 2014 | miR-200 genes, producing a double negative feedback loop that is Accepted 12 June 2014 | Published online 28 July 2014 central to the control of EMT in in vitro models (Bracken et al, The EMBO Journal (2014) 33: 2040–2056 2008; Burk et al, 2008; Parker et al, 2009; Gregory et al, 2011) and has been implicated in contributing to cancer progression through promotion of EMT in various epithelial-derived cancers. However, like all microRNAs, the miR-200 family members are predicted to

1 Centre for Cancer Biology, SA Pathology and University of South Australia, Adelaide, SA, Australia 2 Discipline of Medicine, University of Adelaide, Adelaide, SA, Australia 3 School of Molecular and Biomedical Science, University of Adelaide, Adelaide, SA, Australia 4 Division of Molecular Cell Biology, Institute for Molecular Bioscience, University of Queensland, St Lucia, Brisbane, Qld, Australia *Corresponding author. Tel: +61 8 8222 3410; Fax: +61 8 8232 4092; Email: [email protected] **Corresponding author. Tel: +61 8 8222 3430; Fax +61 8 8232 4092; Email: [email protected] †These authors contributed equally to the work

2040 The EMBO Journal Vol 33 | No 18 | 2014 ª 2014 The Authors Cameron P Bracken et al miR-200 targets a cytoskeletal regulatory network The EMBO Journal The EMBO Journal miR-200 targets a cytoskeletal regulatory network Cameron P Bracken et al

A have many other targets. Some of these may also play central roles cancer cell invasion and metastasis. In particular, we show that the Figure 1. Genome-wide identification of miR-200 targets by Ago-HITS- in mediating the epithelial response to miR-200, while others may formation of invadopodia, which relies on rearrangement of the CLIP. provide a more subtle role in the sculpting of phenotype by actin cytoskeleton and provides a site for localized secretion of A Percentages of total reads and read peaks derived from Ago-HITS-CLIP and mapped relative to their genomic locations. miR-200. enzymes to degrade the extracellular matrix, is regulated by miR-200 B, C Histograms displaying read peaks across the CFL2 and MPRIP 30UTRs. The The miR-200 family consists of 5 miRNAs that are closely related at multiple points in the pathway. y-axis shows the number of overlapping unique sequencing reads in sequence, but are predicted to comprise two distinct classes in comprising the peak, and the x-axis indicates the position of the peak terms of their targets, with miR-200a and miR-141 sharing identical within the 30UTR. The locations of potential seed sites (black arrows, seed regions and miR-200b, miR-200c, and miR-429 constituting a Results 8-mers; purple arrows, 7-mers; yellow arrows, 6-mers; asterisk, central- paired) are indicated. Sequence alignments of miR-200b to the target separate targeting class. These 5 microRNAs arise from two precursor sites identified by Ago-HITS-CLIP are shown below. genes encoded at separate genomic loci. The expression of the 5 Global identification of transcripts targeted by miR-200 family members is highly correlated, being coexpressed in essentially all epithelial cells and absent in mesenchymal cells. While the The 5 members of the miR-200 family are coregulated and have a all interaction sites within 30UTRs. Examples of various non- targeting of ZEB1 and ZEB2 by miR-200 has been extensively inves- controlling influence on the epithelial versus mesenchymal cell canonical miR-200 interactions are shown in Supplementary Fig S2. B tigated, and several additional targets have been identified, a state. They are closely related to each other in sequence, but fall Because the miR-200a and miR-200b seed regions differ at one comprehensive unbiased investigation of direct target genes, and into 2 classes with respect to their seed regions, indicating they are position, they are each equivalent to a seed mismatch of the other. We the consequences for cell phenotype of the targeting of these genes, likely to have two distinct sets of mRNA targets. We investigated found that 15% of seed-based interaction sites for each miR was also a has not been reported. miR-200a as representative of the miR-200a/miR-141 seed class and site of interaction for the other family member, demonstrating there is MicroRNAs function as the specificity component of the protein– miR-200b as representative of the miR-200b/miR-200c/miR-429 a high degree of specificity in seed region interactions, but confirming RNA complex known as RISC (RNA-induced silencing complex). seed class. We performed Ago-HITS-CLIP on control and transfected that mismatch interactions can occur at a proportion of sites. The microRNA provides sequence-specific binding of RISC to MDA-MB-231 breast cancer cells, which have very low endogenous Targeting by miRNAs typically reduces the level of the target specific mRNA targets, resulting in decreased efficiency of trans- levels of the miR-200 family (Gregory et al, 2008a). We confirmed mRNA (Grimson et al, 2007; Hafner et al, 2010). By microarray lation and an increased rate of mRNA degradation (Carthew & the successful precipitation of Ago by immunoblot analysis (Supple- analysis, we found a strong bias toward downregulation by miR-200 Sontheimer, 2009). While miRNAs are now relatively easy to mentary Fig S1A) and subjected the barcoded cDNAs (Supplemen- of mRNAs with miR-200 interaction sites identified by Ago-HITS- discover and measure, the key to understanding their functions tary Fig S1B) to Illumina sequencing. From the mapped small RNA CLIP, with the greatest bias being for transcripts containing perfect remains the identification of their gene targets, which until recently reads, 223 unique miRNAs were identified with miR-27a being 8-mer seed sites within 30UTRs (Supplementary Table S1), consis- has largely been achieved via computational prediction followed by the most abundant. Upon transfection, Ago-bound miR-200a and tent with previous reports that longer seed matches correlate with individual experimental verification, or miRNA manipulation in miR-200b were each increased to levels similar to miR-27a, confirming stronger target repression (Grimson et al, 2007). These biases were conjunction with approaches such as microarray or proteomic the loading of Ago with the transfected miRNAs at physiological also present, but progressively less strong, for 7-mer and 6-mer profiling. In silico target prediction is limited by our incomplete levels (Supplementary Fig S1C). sites. Transcripts targeted through seed sites in the coding region understanding of targeting ‘rules’ due largely to an inability to From the pool of immunoprecipitated mRNAs, reads were were less strongly biased toward downregulation, in agreement with reliably model the influences of RNA secondary structure and mapped across 12,814 genes. Sites bound to Ago were identified by studies of let-7, miR-1, and miR-124 (Lim et al, 2005; Forman & RNA-binding proteins that interfere with potential target sites. clustering overlapping reads into peaks, setting a threshold of at Coller, 2010). Pronounced RNA destabilization via seed-mismatch Approaches based on mRNA expression analysis can only identify least 3 non-identical overlapping reads for designation as a peak. sites was rare, indicating that although seed-mismatch sites are rela- targets that are destabilized at the RNA level, cannot identify the Although 43% of the individual reads mapped to introns and inter- C tively frequent sites of miRNA association, they rarely have a precise site of targeting, and are unable to differentiate direct from genic regions, this drops to 14% after the clustering of individual pronounced effect on mRNA abundance. This observation, coupled indirect targets, while proteomic approaches are limited in their reads into overlapping peaks. Conversely, reads mapping to 30UTRs with the rarity of other non-canonical miRNA interaction sites, led sensitivity and also do not differentiate direct from indirect targets. and coding regions increase from 55% of individual reads to 83% us to focus on the effects of miR-200 on genes targeted through seed

A considerable methodological improvement has been the devel- of peaks, suggesting the clustering of reads into peaks reduces matches in 30UTRs. opment of the Ago-HITS-CLIP (Argonaute High Throughput background sequencing noise (Fig 1A) and identifies prominent The HITS-CLIP procedure indicates sites of interaction between Sequencing after Cross-Linked Immunoprecipitation) procedure, in interaction hotspots. To identify transcripts targeted by miR-200, we Ago and mRNA, but additional data are needed to determine the which RNA–protein complexes are stabilized by UV cross-linking in located read peaks from cells transfected with miR-200a or strength of repression that this interaction confers on the mRNA. To live cells, followed by direct immunoprecipitation and purification miR-200b that were absent in samples from untransfected cells and better assess the functional effects of the miRNA-target interactions of miRNA-loaded RISC, enabling the identification of directly associ- from cells transfected with a control miRNA mimic. Because the identified by the Ago-HITS-CLIP, we selected 56 genes that represent ated target transcripts on a global scale by massively parallel majority of interactions occur through base pairing to the miRNA a range of perfect seed matches (6-mer, 7-mer, and 8-mer) and seed- sequencing (Chi et al, 2009). By capturing the RISC complex in the ‘seed region’, we initially identified such interaction sites by cross- mismatch sites for further verification. Given the reported predomi- act of binding to mRNA in living cells, and incorporating several referencing peaks that were dependent on miR-200 transfection with nance of 30UTR targeting and our own data showing the expression purification steps, Ago-HITS-CLIP has the advantages of being all the potential cognate 6-, 7-, or 8-mer seed sites. This identified of genes possessing 30UTR sites are more likely to be downregulated highly stringent, requires no a priori predictions of the identity or 917 and 1,194 transcripts that directly bound miR-200a and by miRNAs (Supplementary Table S1), we restricted our selection of locations of binding sites, and avoids non-specific Ago–RNA interac- miR-200b, respectively (Supplementary Table S1). Of these, ~65% genes to those with interaction sites within their 30UTRs and tested tions that may otherwise occur in vitro (Riley et al, 2012). We have map to 30UTRs, 33% to coding regions, and 2% to 50UTRs, with 9% the capacity of miR-200 to regulate these sites by the cloning of full- applied this procedure to identify many transcripts that interact with representing 8-mers, 27% 7-mers and the remaining 64% 6-mers. length 30UTRs into luciferase reporter constructs. Full-length 30UTRs, miR-200, including a number of non-canonical targets such as Many potential seed matches were not sites of interaction, highlight- sometimes interact through seed sites that contain mismatches or rather than isolated minimal target sites, were used in order to more central-paired and seed-mismatch interactions, and also find novel ing the utility of Ago-HITS-CLIP in distinguishing functional from bulges (Chi et al, 2012; Loeb et al, 2012) or via ‘central-pairing’ fully replicate the endogenous site of targeting. The luciferase target sites in previously misannotated transcripts. We find that predicted sites. For example, only a small subset of the potential which involves the central region of the miRNA but not the seed reporters confirmed that the majority of targets identified by regulators of actin cytoskeleton dynamics are strongly enriched seed sites in the CFL2 and MPRIP mRNAs are actually engaged by region (Shin et al, 2010). We found examples of seed-bulge and Ago-HITS-CLIP are indeed repressed by miR-200 (Fig 2A) and, as among the targets of both miRNAs, indicating that the miR-200 miR-200 at a discernable level (Fig 1B and C). central-paired interactions with miR-200, but these were rare, cumu- indicated by our microarray data (Supplementary Table S1), demon- family imposes coordinated control of functional networks that are In addition to canonical target interactions whereby the miRNA latively accounting for only ~3% of target sites. In contrast, peaks strated that repression is particularly strong when mediated through central to cell shape and motility, and of crucial importance in seed region perfectly base pairs with a target transcript, miRNAs attributable to seed mismatches were abundant, constituting ~35% of perfect 8-mer seed sites (Fig 2C). For example, we found miR-200b

ª 2014 The Authors The EMBO Journal Vol 33 | No 18 | 2014 2041 2042 The EMBO Journal Vol 33 | No 18 | 2014 ª 2014 The Authors Cameron P Bracken et al miR-200 targets a cytoskeletal regulatory network The EMBO Journal The EMBO Journal miR-200 targets a cytoskeletal regulatory network Cameron P Bracken et al

confers statistically significant repression of 84% of the 8-mer and with its targeting of the Rho activators ARHGEF3 and NET1. In 7-mer sites selected, with most of these targets (81%) being strongly addition, the concomitant direct downregulation of ROCK2 levels A repressed to a level < 75% of the control. Of the 6-mer sites tested, by miR-200 (Fig 2) might amplify its effect on RhoA-ROCK signaling. 48% were significantly repressed, although only 13% of these were In contrast, we did not observe any significant effect of miR-200 on suppressed by more than 75%. The same hierarchical effect (8-mer the activation of cdc42 or Rac1 (data not shown). > 7-mer > 6-mer seed matches) on endogenous mRNA levels was also observed by qPCR measurement of the effect of miR-200 on the MiR-200 and its targets regulate cell morphology, migration, endogenous mRNAs (Fig 2B and D). The targeting of a subset of and invasion genes possessing 8-mer miR-200 seed sites, selected on the basis of their important roles in cytoskeletal remodeling, was further As indicated by gene ontology analysis, the pathway with the largest confirmed by Western blot analysis (Fig 2E). Although individual cohort of miR-200 targets is that associated with cytoskeletal remod- targets may be repressed through seed-mismatch sites, in neither the eling which is consistent with previous work implicating miR-200 in luciferase reporters, qPCR or microarray data were seed mismatches controlling cell motility and cell shape changes (Burk et al, 2008; found to generally mediate functionally significant repression. Gregory et al, 2008a). We also verified that, in breast cancer cells, The unbiased nature of Ago-HITS-CLIP allows the identification miR-200 promoted the rearrangement of the actin cytoskeleton from of interaction sites that are missed by prediction tools due to misan- stress fibers to cortical actin, (Supplementary Fig S4A) and inhibited notation of transcripts. For example, we found that a prominent cell migration and invasion (Supplementary Fig S4B–D). We then miR-200b target site resides downstream of the annotated end of the used a real-time invasion assay to compare the effect of knockdown EGFR mRNA (Fig 3A). Inspection of ESTs aligned to this region of several individual miR-200 targets with the effect of miR-200 itself. supports the existence of a long UTR form. We designed qPCR Knockdown of MPRIP, ABL2, and WIPF1 each inhibited invasion assays to quantitate expression of the short and long forms, and (Fig 5A and B), consistent with the suppressive effects of miR-200 B found that the long form is predominant in MDA-MB-231 cells and and with the previously ascribed roles of these genes as cytoskeletal is significantly reduced in level in response to miR-200b (Fig 3B). regulators. Examination of the cells at the invading front of the 3D We cloned the long and short UTRs into luciferase reporters and invasion assay revealed that miR-200 inhibited the formation of verified that miR-200b significantly repressed activity of the reporter membrane protrusions (Fig 5C), which are characteristic of migratory with the long 30UTR, but not the short form (Fig 3C). This suggests cells and are sites of invadopodia formation (Yu & Machesky, 2012). that EGFR expression, which can have important consequences in tumors, can be subjected to regulation in response to miR-200, but MiR-200 and its targets regulate invadopodia formation can made independent of miR-200 by alternative polyadenylation. Cell invasion is the collective outcome of a number of processes HITS-CLIP reveals a network of cytoskeletal regulators, including dependent on extensive cytoskeletal rearrangements and assembly components of the Rho signaling pathway, that are miR-200 of actin bundles, including the assembly of invadopodia. These are target genes actin-rich membrane protrusions specialized for the delivery of proteases that degrade the extracellular matrix at specific sites To assess whether any functional systems might be coordinately (Sibony-Benyamini & Gil-Henn, 2012) to facilitate movement regulated by miR-200, we conducted gene ontology analysis of the through the matrix. Given that miR-200 suppresses cell invasion miR-200 HITS-CLIP targets. This revealed a striking over-representa- (Fig 5B), we examined the potential for miR-200 to inhibit the tion of genes associated with cytoskeletal remodeling (Fig 4A). formation of invadopodia. Invadopodia can be identified by the Importantly, cytoskeletal remodeling was the pathway most highly colocalization in dense puncta of actin with cortactin and Tks5, targeted by both miR-200a and miR-200b, and this was even more two proteins essential for initiation of invadopodia formation, pronounced when restricted to genes with binding sites within their (Courtneidge et al, 2005; Bowden et al, 2006). Sites of invadopodial CD E 30UTRs (Supplementary Fig S3), the most functionally optimal site activity can also be identified by monitoring the local pericellular of targeting. Most of the other significantly enriched ontologies are degradation of a fluorescently tagged gelatin matrix onto which the the functional outcomes of cytoskeletal remodeling, such as cell cells have been plated (Grass et al, 2012). Utilizing each of these adhesion, and processes with which miR-200 has been previously assays, we found that miR-200 significantly reduces invadopodia implicated including EMT, TGF-b, and Wnt signaling. formation (Fig 5D–G). We also investigated a number of miR-200 A number of the miR-200 targets identified by Ago-HITS-CLIP are targets for their role in invadopodia formation, including WIPF1, upstream regulators or downstream effectors of the Rho-ROCK MPRIP, and Abl2 which significantly inhibited invasion when signaling pathway, which is implicated in tumor progression and knocked down (Fig 5B), as well as cofilin2 (CFL2), moesin (MSN), invasion (Sahai & Marshall, 2003; Croft et al, 2004; Fig 4B). RhoA and LGR4 which have all been previously implicated in promoting itself was not a strong direct target, but the RhoA activating guanine invasion or metastasis (Ono et al, 2008; Estecha et al, 2009; He et al, exchange factors (GEFs), ARHGEF3 and NET1 (ARHGEF8), and the 2010; Haynes et al, 2011; Garcia et al, 2012; Gil-Henn et al, 2012; inactivating GTPase-activating protein (GAP), ARHGAP29, were Erkutlu et al, 2013). We found that knockdown of MPRIP and WIPF1 identified as targets (Fig 2). In light of their opposing roles on Rho phenocopied miR-200 in inhibiting invadopodia formation, while activity, we determined the net effect of miR-200 by measuring the knocking down ABL2 reduced invadopodia numbers by about 33% levels of both active and total Rho following transfection with (Fig 5H). CFL2 knockdown also phenocopied miR-200 (Fig 5H), miR-200a, miR-200b, or both. MiR-200b decreased the levels of consistent with a recent study reporting the role of ADF/cofilin in active Rho without affecting total Rho levels (Fig 4C), consistent regulating membrane recycling essential for invadopodia formation

ª 2014 The Authors The EMBO Journal Vol 33 | No 18 | 2014 2043 2044 The EMBO Journal Vol 33 | No 18 | 2014 ª 2014 The Authors Cameron P Bracken et al miR-200 targets a cytoskeletal regulatory network The EMBO Journal The EMBO Journal miR-200 targets a cytoskeletal regulatory network Cameron P Bracken et al

A Figure 2. Validation of miR-200 targets by luciferase reporter assay and qPCR. ◀ A, B Relative activities of full-length 30UTR-luciferase reporters (A) and relative mRNA levels as measured by qPCR (B) of selected miR-200 targets in MDA-MB-231 cells transfected with control or miRNA mimics. Colored boxes below the histograms indicate the nature of the target site identified by Ago-HITS-CLIP (MM represents mismatch sites). Errors bars show s.e.m. n ≥ 3;*P < 0.05; one-tailed Student’s t-test. C, D Pooled data from the luciferase reporter assays (C) and qPCRs (D) grouped according to the nature of the target site and represented as box plots. Significance (*P < 0.05) was calculated by two-tailed t-test of mean = 1. E Immunoblot analysis of selected targets. Relative quantitation is shown in right panel. Source data are available online for this figure.

(Hagedorn et al, 2014). However, MSN1 which we had shown to Tks5, indicating that the effects of miR-200 on invadopodia mimic miR-200 in inhibiting invasion when knocked down, and to formation were not due to effects on the initiation steps (Supple- rescue the inhibition of metastasis by miR-200 when reintroduced mentary Fig S5). Maturation of invadopodia is accompanied by local (Li et al, 2013), had no detectable effect on invadopodia formation. release of MMPs, principally MMP9 and MMP14 (also known as Invadopodia formation was also unaffected by knockdown of LGR4 MT1-MMP) (Poincloux et al, 2009). We assessed MMP secretion B (Fig 5H). using an in-gel zymase assay, which indicated that MMP9 is strongly downregulated by miR-200, whereas MMP2, which is not MiR-200 targets regulate individual steps in associated with invadopodia function, was not detectably affected invadopodia formation by miR-200 (Fig 6A). Furthermore, MMP9 was not indicated by Ago-HITS-CLIP to be a miR-200 target and this was verified using a Invadopodia formation proceeds in several steps, including their luciferase reporter assay (Fig 6B), suggesting that the decrease in initiation by Src-mediated phosphorylation of cortactin and Tks5, secreted MMP9 could be due to the loss of invadopodia formation. disassembly of adjacent focal adhesions, assembly of fresh actin fila- However, we found that miR-200 strongly repressed the level of ments at the site of invadopodial protrusion, and delivery of matrix MMP9 mRNA (Fig 6C), suggesting that miR-200 indirectly repressed metalloproteases (MMPs) to the invadopodia for membrane inser- MMP9 synthesis independently of its association with invadopodia. tion or localized secretion (reviewed in Murphy & Courtneidge, Similarly, we found that although MMP14 was not a direct target, 2011). We assessed the effect of miR-200 on each of these steps. miR-200 also repressed both MMP14 protein and mRNA levels Phosphoprotein immunoblot analysis showed that miR-200 did not (Fig 6D and E). The direct miR-200 targets responsible for decreasing affect Src activation, or the phosphorylation state of cortactin or synthesis of MMP9 and MMP14 are yet to be identified.

A B

Figure 3. Ago-HITS-CLIP reveals miR-200b targeting of EGFR via an extended 30UTR.

indicate EGFR 30UTRs annotated in Refseq and ESTs from GenBank. B Measurement by qPCR of total and long UTR forms of EGFR mRNA in MDA-MB-231 cells transfected with control or miRNA mimics. n = 3;*P < 0.05; one-tailed Student’s t-test.

C Relative activity of luciferase reporters with long and short forms of the EGFR 30UTR in MDA-MB-231 cells transfected with control or miRNA mimics. n = 3;*P < 0.05; one-tailed Student’s t-test.

ª 2014 The Authors The EMBO Journal Vol 33 | No 18 | 2014 2045 2046 The EMBO Journal Vol 33 | No 18 | 2014 ª 2014 The Authors Cameron P Bracken et al miR-200 targets a cytoskeletal regulatory network The EMBO Journal The EMBO Journal miR-200 targets a cytoskeletal regulatory network Cameron P Bracken et al

A B

Figure 3. Ago-HITS-CLIP reveals miR-200b targeting of EGFR via an extended 30UTR.

A Histogram displaying read peaks across the EGFR 30UTR. The y-axis shows the number of overlapping unique sequencing reads comprising the peak, and the x-axis indicates the position of the peak within the 30UTR. The locations of potential seed sites (black arrows, 8-mers; purple arrows, 7-mers; yellow arrows, 6-mers; asterisk, central-paired) are indicated. Sequence alignment of miR-200b to the target site identified by Ago-HITS-CLIP is shown below. Black bars below the histograms indicate EGFR 30UTRs annotated in Refseq and ESTs from GenBank. B Measurement by qPCR of total and long UTR forms of EGFR mRNA in MDA-MB-231 cells transfected with control or miRNA mimics. n = 3;*P < 0.05; one-tailed Student’s t-test.

C Relative activity of luciferase reporters with long and short forms of the EGFR 30UTR in MDA-MB-231 cells transfected with control or miRNA mimics. n = 3;*P < 0.05; one-tailed Student’s t-test.

Figure 4. MiR-200 targets multiple genes within cytoskeletal regulatory networks that regulate actin nucleation, branching, and actomyosin contractility. A BC ◀ A Gene GO pathway analysis of miR-200 target genes identified by Ago-HITS-CLIP. B Relationships between direct miR-200 targets identified by Ago-HITS-CLIP and validated by qPCR and luciferase reporter activity are indicated. Blue circles denote direct miR-200 targets. Gray circles denote network components not identified as miR-200 targets. (P) indicates phosphorylation. Numbers denote references: 1) (Thiesen et al, 2000); 2) (Alberts & Treisman, 1998); 3) (Saras et al, 1997); 4) (Zhang et al, 2005); 5) (Cuenda & Dorow, 1998); 6) (Derijard et al, 1995); 7) (Wang et al, 1997); 8) (Fukata et al, 1998); 9) (Yang et al, 1998); 10) (Arber et al, 1998); 11) (Frenette et al, 2012); 12) (Piekny & Glotzer, 2008); 13) (Totsukawa et al, 2000); 14) (Hu et al, 2005); 15) (Miller et al, 2010); 16) (Boyle & Koleske, 2007); 17) (Sasahara et al, 2002); 18) (Noy et al, 2012); 19) (Krzewski et al, 2006); 20) (Kaneko et al, 2000); 21) (Ho et al, 2004); 22) (Singh et al, 2011); 23) (Kawano et al, 1999); 24) (Surks et al, 2005); 25) (Ono et al, 2008); 26) (Zhang et al, 2007); 27) (Mu et al, 2012); 28) (Turtoi et al, 2013); 29) (Kovacs et al, 2006); 30) (Kovacs et al, 2011). C Effect of miR-200 on Rho activity, measured using the Rho Pull-Down assay. A representative experiment is shown on the left and relative Rho activity from 3 independent experiments is quantitated on the right. *P < 0.05; Student’s t-test.

assessed the effect of miR-200 and its targets on focal adhesions. We strongest for targets that contain 8-mer sites in the 30UTR and is found that the expression of miR-200 increased the number and progressively less strong for 7-mer and 6-mer sites in the 30UTR and width of focal adhesions (Fig 7) and decreased their dynamic for sites in the coding region (CDS) (Supplementary Fig S6A and B). D rearrangements (Supplementary Movies S1A and S1B). Whereas Such bias was not present when the miR-200 targets were correlated knockdown of WIPF1, CFL2, and MPRIP did not affect focal against a control miRNA (Supplementary Fig S7). This hierarchy of adhesions, knockdown of ABL2 phenocopied miR-200, increasing sites is in good agreement with our assessment of the functional focal adhesion number and size (Fig 7). Together, our data indicate effects of miR-200 on a panel of Ago-HITS-CLIP identified targets that miR-200 acts through overlapping subsets of targets to regulate (Fig 2C and D). Since microRNAs can repress translation in addition cell invasion at multiple independent levels, including decreasing to their effect on mRNA abundance (Carthew & Sontheimer, 2009), numbers of invadopodia, decreasing the synthesis of invadopodia- the inverse correlations we observe in these analyses may be even associated MMPs, and stabilizing cell-matrix adhesion. stronger at the protein level. Together, these data underscore the capacity of Ago-HITS-CLIP to identify functionally significant targets MiR-200 targets identified by HITS-CLIP reflect functionally and are consistent with the notion that miRNAs regulate networks significant targets in primary breast tumors through the combination of both major effects on key genes and cumulative subtle effects on a wide range of targets. Expression of the miR-200 family can become dysregulated in cancer, driving inappropriate EMT/MET processes that underlie cancer invasion and metastatic recolonization (Gregory et al, 2008b; Discussion Korpal et al, 2011; Creighton et al, 2013). To assess the potential for the involvement of the miR-200 target genes that we identified by We have found hundreds of interaction sites for miR-200a and

invadopodia formation and invasion (Fig 5), we first compared the matches (8-mer and 7-mers) within 30UTRs responsible for stronger levels of each of these to miR-200 in a panel of human breast cancer effects of miRNAs on specific targets, while a plethora of shorter cell lines. This indicated a pronounced reciprocal relationship of seed-matched and non-canonical sites (Chi et al, 2012; Loeb et al, each of these genes to miR-200 (Fig 8A–C), with the more epithelial 2012) provide more subtle regulatory effects on a broader range of cell lines expressing high levels of miR-200 and low levels of the 3 genes. In view of the uncertainty regarding the efficacy of miRNA- targets, whereas the more mesenchymal cell lines showed the binding sites in coding regions (Gu et al, 2009; Forman & Coller, H reciprocal pattern of expression. 2010), it is interesting that the effect of miR-200 on CDS sites (at To more broadly assess the responses of miR-200 targets, we least for 8-mer seed matches) is sufficient to produce a bias toward calculated Spearman correlation coefficients for the relationship negative correlations across the multiple cell lines within the NCI-60 between miR-200 and all of the targets identified by HITS-CLIP across panel, supporting the functionality of CDS sites. 934 breast cancers from the TCGA dataset and 59 cell lines from the Our identification of functional sites revealed a model of hierar- NCI-60 panel (Fig 8D and E). Because miR-200c possesses an identi- chical network regulation that is the most extensive validation to cal seed region to miR-200b, but is expressed from a separate gene date of the concept that individual miRNAs can broadly target whole locus, we also included miR-200c in the analysis. We observed a networks. It was noteworthy, however, that although miR-200 distinct bias toward inverse correlation, supporting the notion that affects both invadopodia and focal adhesions, this is through the cytoskeletal miR-200 targets we identify here reflect functionally distinct subnetworks. Despite their impact on invadopodia, WIPF1

ª 2014 The Authors The EMBO Journal Vol 33 | No 18 | 2014 2047 2048 The EMBO Journal Vol 33 | No 18 | 2014 ª 2014 The Authors Cameron P Bracken et al miR-200 targets a cytoskeletal regulatory network The EMBO Journal The EMBO Journal miR-200 targets a cytoskeletal regulatory network Cameron P Bracken et al

Focal adhesions are specialized structures in which integrins relevant targets in vivo. The bias is slightly stronger in the NCI-60 provide an adhesive link between the ECM and the actin cytoskeleton dataset than in the TCGA dataset, probably because the mixture of (Wehrle-Haller, 2012). As the dynamic assembly and disassembly of cell types in tumor samples introduces some unrelated variation, focal adhesions are central to cell motility and disassembly of focal which the cell lines of the NCI-60 panel are not affected by. In both adhesions is an obligatory step in invadopodia formation, we also datasets, it is evident that the tendency toward inverse correlation is assessed the effect of miR-200 and its targets on focal adhesions. We strongest for targets that contain 8-mer sites in the 30UTR and is found that the expression of miR-200 increased the number and progressively less strong for 7-mer and 6-mer sites in the 30UTR and width of focal adhesions (Fig 7) and decreased their dynamic for sites in the coding region (CDS) (Supplementary Fig S6A and B). D rearrangements (Supplementary Movies S1A and S1B). Whereas Such bias was not present when the miR-200 targets were correlated knockdown of WIPF1, CFL2, and MPRIP did not affect focal against a control miRNA (Supplementary Fig S7). This hierarchy of adhesions, knockdown of ABL2 phenocopied miR-200, increasing sites is in good agreement with our assessment of the functional focal adhesion number and size (Fig 7). Together, our data indicate effects of miR-200 on a panel of Ago-HITS-CLIP identified targets that miR-200 acts through overlapping subsets of targets to regulate (Fig 2C and D). Since microRNAs can repress translation in addition cell invasion at multiple independent levels, including decreasing to their effect on mRNA abundance (Carthew & Sontheimer, 2009), numbers of invadopodia, decreasing the synthesis of invadopodia- the inverse correlations we observe in these analyses may be even associated MMPs, and stabilizing cell-matrix adhesion. stronger at the protein level. Together, these data underscore the capacity of Ago-HITS-CLIP to identify functionally significant targets MiR-200 targets identified by HITS-CLIP reflect functionally and are consistent with the notion that miRNAs regulate networks significant targets in primary breast tumors through the combination of both major effects on key genes and cumulative subtle effects on a wide range of targets. Expression of the miR-200 family can become dysregulated in cancer, driving inappropriate EMT/MET processes that underlie cancer invasion and metastatic recolonization (Gregory et al, 2008b; Discussion Korpal et al, 2011; Creighton et al, 2013). To assess the potential for the involvement of the miR-200 target genes that we identified by We have found hundreds of interaction sites for miR-200a and

Ago-HITS-CLIP, we assessed whether variations in miR-200 levels miR-200b, of which the majority were in 30UTRs of mRNAs. Expres- between cell lines and between breast cancer patients are reflected sion profile analysis, luciferase reporter assays, and correlation of E FG by the reciprocal variation of the network of cytoskeletal regulators expression between miR-200 and its targets in NCI-60 cell lines and that we had identified as targets of miR-200. Because CFL2, MPRIP, the TCGA breast cancer dataset validate the capacity of Ago-HITS- and WIPF1 were especially strong responders to miR-200 in lucifer- CLIP to find functionally significant targets. We further confirm the ase reporter assays (Fig 2) and were found to promote both hierarchical nature of target site effectiveness, with longer seed invadopodia formation and invasion (Fig 5), we first compared the matches (8-mer and 7-mers) within 30UTRs responsible for stronger levels of each of these to miR-200 in a panel of human breast cancer effects of miRNAs on specific targets, while a plethora of shorter cell lines. This indicated a pronounced reciprocal relationship of seed-matched and non-canonical sites (Chi et al, 2012; Loeb et al, each of these genes to miR-200 (Fig 8A–C), with the more epithelial 2012) provide more subtle regulatory effects on a broader range of cell lines expressing high levels of miR-200 and low levels of the 3 genes. In view of the uncertainty regarding the efficacy of miRNA- targets, whereas the more mesenchymal cell lines showed the binding sites in coding regions (Gu et al, 2009; Forman & Coller, H reciprocal pattern of expression. 2010), it is interesting that the effect of miR-200 on CDS sites (at To more broadly assess the responses of miR-200 targets, we least for 8-mer seed matches) is sufficient to produce a bias toward calculated Spearman correlation coefficients for the relationship negative correlations across the multiple cell lines within the NCI-60 between miR-200 and all of the targets identified by HITS-CLIP across panel, supporting the functionality of CDS sites. 934 breast cancers from the TCGA dataset and 59 cell lines from the Our identification of functional sites revealed a model of hierar- NCI-60 panel (Fig 8D and E). Because miR-200c possesses an identi- chical network regulation that is the most extensive validation to cal seed region to miR-200b, but is expressed from a separate gene date of the concept that individual miRNAs can broadly target whole locus, we also included miR-200c in the analysis. We observed a networks. It was noteworthy, however, that although miR-200 distinct bias toward inverse correlation, supporting the notion that affects both invadopodia and focal adhesions, this is through the cytoskeletal miR-200 targets we identify here reflect functionally distinct subnetworks. Despite their impact on invadopodia, WIPF1

A B Figure 5. MiR-200 inhibits invasion and invadopodia formation. ◀ MDA-MB-231 cells were transfected with control or miR-200a and -200b mimics. A Representative real-time images of a 3D scratch wound cell invasion assay. B Quantification of 3D cell invasion assays of cells transfected with control or miR-200 mimics (top panels) and control or siRNAs to miR-200 targets (bottom panels). Slopes of the time-course plots are shown at right. C High magnification images showing the morphology of cells at the invasive front from the 3D invasion assay shown in (A). D Invadopodia formation in MDA-MB-231 cells transfected with miR-200 mimics as monitored by the colocalization of cortactin and F-actin and degradation of a (green) FITC-conjugated gelatin matrix upon which control or miR-200 transfected cells are plated. Representative images are shown with arrows indicating sites of colocalization. Scale bars represent 0.01 mm. E-G Quantitation of cells possessing invadopodia as defined by the colocalization of (E) actin and cortactin (150 cells counted/variable/experiment), (F) actin and Tks5 (150 cells counted/variable/experiment) and (G) fluorescent gelatin degradation (250 cells counted/variable/experiment). n = 3;*P < 0.05, **P < 0.01, ***P < 0.001, Student’s t-test. H Effect of miR-200 and siRNA-mediated knockdown of selected miR-200 targets on invadopodia formation as measured by the colocalization of F-actin with cortactin.

and CLF2 depletion did not promote focal adhesions. Instead, focal Rho signaling is an important upstream regulator of the cytoskel- adhesions were influenced by another miR-200 target, ABL2, which eton and is thought to be dysregulated in cancer through the is consistent with recent evidence that Abl supports focal adhesions network of GEFs and GAPs that control the Rho GTPase cycle, at the cell periphery (Peacock et al, 2010). Increased focal adhesion rather than by mutation of Rho itself (Bos et al, 2007). We found size and number, such as was promoted by miR-200, may inhibit that miR-200 inhibits the Rho signaling pathway through both the cell migration by preventing adhesive release and cell translocation. targeting of the Rho activators, ARHGEF3 and NET1, and the targeting Thus, one mechanism for miR-200 to inhibit cell migration may be of multiple downstream elements in the Rho signaling pathway: the through stabilization of focal adhesions. This yields a model where Rho effector ROCK2; a number of direct ROCK substrates such as miR-200 enforces the epithelial phenotype, and prevents invasion, the non-muscle myosins MYH9 and MYH10, which are necessary through distinct subnetworks that coordinately inhibit invadopodia for contractility; ANLN, which mediates the interaction between formation and stabilize focal adhesions. myosin and RhoA; and the myosin regulators MPRIP, MYPT1, and Figure 7. Effects of miR-200 and its target genes on focal adhesions. A Representative images showing the effect of miR-200 and miR-200 target gene knockdown on focal adhesions as indicated by vinculin staining. Scale bars represent 0.01 mm. A B B The average number of focal adhesions per cell, average focal adhesion length (lm), and % of focal adhesions with a length > 0.05 lm was quantitated in control, miR-200 transfected, and target gene knockdown cells. 100 (for miR transfections) or 50 (for siRNA transfections) cells pooled from 3-4 separate experiments were counted.

MYLK, which control actomyosin contractility through regulating miR-200 controls invasion and further demonstrating how miRNAs myosin light chain phosphorylation. ROCK also phosphorylates function as master regulators, affecting multiple levels of a regula- other miR-200-targeted genes including the ERM family members tory hierarchy. moesin and radixin, which tether actin filaments to the cell This work provides the first demonstration of a global regulatory membrane; CFL2 (indirectly via LIMK), which promotes reorganiza- network directly regulated by miR-200 which strongly involves, but tion of actin filaments; and several members of the mitogen- is not exclusively limited to, a network of signals and effectors that activated protein kinase (MAPK) cascade, MAP3K7, MEKK1, and mediate the impact of Rho in tumor cells. This ultimately influences MAP2K4, which mediate JNK and p38 MAPK activation in response the epithelial–mesenchymal plasticity of cells and their ability to to growth factor signaling. Many of these aforementioned miR-200 invade and metastasize. C DE targets have been reported as positive regulators of cell migration, invasion, and metastasis (Wang et al, 2004; Ono et al, 2008; Safina et al, 2008; Estecha et al, 2009; Su et al, 2009; Wong et al, 2009; He Materials and Methods et al, 2010; Haynes et al, 2011; Garcia et al, 2012; Gil-Henn et al, 2012; Erkutlu et al, 2013). Cell lines and cell culture In addition to directly targeting multiple pathway components, miRNAs can further coordinate networks by targeting transcription All cell lines were cultured in Dulbecco’s Modified Eagles Medium factors. We identified a number of transcription factors, some of (DMEM; Invitrogen) supplemented with 10% fetal bovine serum which have been previously identified either as miR-200 targets or (FBS) with the exception of MDA-MB-435 and BT-549. MDA-MB- have been implicated in EMT, invasion, and cancer progression, 435 were maintained in Alpha Modified Eagles Medium (aMEM; including ZEB1, SUZ12, STAT5B, E2F3, TCF12, CTNNB1, and Invitrogen) supplemented with 5% FBS. BT-549 were maintained in several SMADs (Gregory et al, 2008a; Iliopoulos et al, 2010; RPMI 1640 (Invitrogen) supplemented with 10% FBS. Figure 6. MiR-200 represses matrix metalloprotease expression and function. Williams et al, 2012; Xia et al, 2012; Peng et al, 2012; Chen et al, A Gelatin zymography of culture supernatants from cells transfected with control or miR-200 mimics. Migration of MMP9 and MMP2 is indicated on the gel and enzyme activity quantitated on the right. 2013; Gal et al, 2008). Targeting such transcription factors extends Isolation of RNA and Real-Time PCR

B Relative activity of luciferase-MMP930UTR reporter in MDA-MB-231 cells transfected with control or miRNA mimics. Error bars represent SEM. the network influenced by miR-200, providing the capacity to C Relative MMP9 mRNA levels in MDA-MB-231 cells transfected with control or miR-200 mimics. control additional processes indirectly. Interestingly, though we Total RNA was extracted using TRIzol (Invitrogen) according to the D Relative MMP14 levels in MDA-MB-231 cells transfected with control or miRNA mimics. n = 3;*P < 0.05, Student’s t-test. Error bars represent SD. have not defined the precise mechanism, MMPs respond strongly manufacturer’s instructions, and real-time PCR performed using E Immunoblot analysis of MMP14 in cells transfected with control and miR-200 mimics. but indirectly to miR-200, providing another avenue through which primers as listed (Supplementary Table S2). MicroRNA PCRs were

ª 2014 The Authors The EMBO Journal Vol 33 | No 18 | 2014 2049 2050 The EMBO Journal Vol 33 | No 18 | 2014 ª 2014 The Authors Cameron P Bracken et al miR-200 targets a cytoskeletal regulatory network The EMBO Journal The EMBO Journal miR-200 targets a cytoskeletal regulatory network Cameron P Bracken et al

A D sodium pyrophosphate, 5 mM EDTA, 50 mM sodium fluoride, 1% ice-cold PBS by scraping and subjected to UV cross-linking at 254 nm Triton X-100 with protease inhibitor cocktail) and 50 lg total protein (Stratalinker). Cell pellets were lysed (0.1% SDS, 0.5% deoxycho- fractionated on 7.5%–15% SDS–polyacrylamide gels (dependent late, 0.5% NP-40 with protease inhibitors, Roche) for 10 mins on ice upon protein size). Proteins were transferred onto nitrocellulose followed by RQ1 DNAse (Promega) at 37°C for 15 min with shak- membranes and probed with the following primary antibodies: ing. RNAse A/T1 (Ambion) was then added for further 8 min, prior anti-CFL2 (Abnova; 110-06848), MPRIP (Sigma; HPA022901), MSN to the addition of EDTA (30 mM). Pellets were then spun (92,000 g) B (Cell Signaling; Q480), WIPF1 (Santa Cruz; sc25533), ROCK2 and the lysate subjected to immunoprecipitation for 2 h with a pan- (Research Diagnostics; AF4790), CRKL (Cell Signaling; 32H4), ABL2 anti-Ago antibody (2A8, kind gift of Zissimos Mourelatos) conju- (Sigma; WH0000027M9), MT1-MMP (Millipore; MAB3314), a-Tubu- gated to protein-A dynabeads (Invitrogen) using bridging rabbit lin (Abcam; ab7291). Blots were then probed with secondary anti- anti-mouse IgG (Jackson Immunolabs). Pellets were then succes- bodies (goat anti-mouse Alexa594 and goat anti-rabbit Alexa488, sively washed (0.1% SDS, 0.5% deoxycholate, 0.5% NP-40 in Li-Cor) and visualized using the Li-Cor Odyssey. 1× PBS; 0.1% SDS, 0.5% deoxycholate, 0.5% NP-40 in 5× PBS; 50 mM Tris pH 7.5, 10 mM MgCl , 0.5% NP-40), and on-bead phos- C 2 Immunofluorescence phatase treatment performed for 30 min with antarctic phosphatase (New England Biolabs) in the presence of superasin RNAse inhibitor

Coverslips were prepared using 0.2% porcine gelatine, as for (Ambion). The 30 RNA linker (CAGACGACGAGCGGG) was labeled E fluorescent matrix degradation assay. Cells were fixed at 37°C in with P32 using T4-PNK (NEB) and ligated on-bead for 1 h at 16°C 4% paraformaldehyde (PFA) in cytoskeleton stabilization buffer with T4 RNA ligase (Fermentas). Beads were then washed as previ-

(10 mM PIPES pH 6.8, 100 mM KCl, 300 mM sucrose, 2 mM ous and treated with PNK to ligate the 50 RNA linker

EGTA, 2 mM MgCl2) for 15 min, permeabilized with 0.1% Triton (AGGGAGGACGAUGCGGxxxG, with ‘X’ representing different X-100/PBS for 10 min, and blocked with 5% FBS/PBS (Cortactin nucleotides for barcoding). Beads were resuspended in 4× LDS ab) or 3% BSA/PBS (Tks5 ab) for 20 min. Samples were incu- Novex loading buffer with 4% 2-mercaptoethanol, incubated at 70°C bated with 1:1,000 Cortactin primary antibody (Upstate 05-180 for 10 min and the supernatant loaded on Novex NuPAGE 4–12% mouse monoclonal) in 0.1% Triton X-100/PBS, 1:500 TKS5 Bis-Tris acrylamide gels (Bio-Rad). After running, the Ago–RNA primary antibody (Millipore cat# 09-268) or vinculin primary anti- complexes were then transferred to nitrocellulose and exposed to body (Chemicon, cat#MAB3574) in 3% BSA/PBS primary antibod- film at 80°C for 3 days. Complexes running at ~110 kDa (AGO + � ies overnight at 4°C and with secondary antibodies and phalloidin miRNA) and ~130 kDa (AGO + mRNA fragments) were then excised for 2 h. After washing, coverslips were mounted with DAKO fluo- with a scalpel and resuspended (100 mM Tris pH 7.5, 50 mM NaCl, rescent mounting medium. Each experiment was repeated at least 10 mM EDTA, 4 mg/ml proteinase K) for 20 min at 37°C. The three times. sample was incubated for an additional 20 min in the presence of 3.5 M urea and RNA isolated by a phenol–chloroform extraction. Immunoprecipitation Sample was then run on a 10% denaturing (1:19) polyacrylamide gel and exposed to film with an intensifying screen at 80°C for 5 days. � Figure 8. Expression levels of miR-200 target genes correlate negatively with miR-200 across cell lines and breast cancers. Seventy-two hours after transfection, cells were washed with PBS Thin bands corresponding to the Ago–miRNA (~110 kD) and AGO A-C Expression of miR-200a and miR-200b (A), Western blot of the miR-200 target genes CFL2, WIPF1, and MPRIP (B), and quantified data of Western blot shown in (B) containing 100 lM Na3VO4 and lysed in lysis buffer (50 mM mRNA fragments (~130 kDa) were excised, crushed, and eluted at across a panel of breast cancer cell lines (C). D, E Correlations between miR-200 family members and their respective targets in (D) 59 cell lines from the NCI-60 cancer cell line panel and (E) 934 breast cancers Tris–HCl pH 7.5, 150 mM NaCl, 1% Triton X-100, 10 mM NaF, 37°C for 1 h (1 M NaOAc, pH 5.2, 1 mM EDTA). RNA was then from the Cancer Genome Atlas (TCGA) dataset. For each target identified by HITS-CLIP, the Spearman correlation coefficient and its q value (equivalent to the 100 lM Na3VO4, 2 mM DTT, protease inhibitors). Protein extracts precipitated overnight with ethanol, centrifuged, and dried. RNA P-value but with correction for multiple testing) was calculated and these were plotted as log q (on y-axis) versus correlation coefficient (on x-axis). Separate were centrifuged at 16,000 g for 15 min at 4°C, and protein was was then resuspended in 8 lHO, primer added (TCCCGCTCGTCG � l 2 plots are shown for all identified miR-targets (all) and for 30UTR 8-mer targets. miR-200a targets are shown correlated against miR-200a levels, and miR-200b quantitated with Bradford assay. 500 lg protein lysate was incu- TCTG) and reverse transcription performed using SuperScriptIII targets are shown correlated against the levels of miR-200b and miR-200c, which share the same seed sequence. To indicate the bias towards negative bated mixing at 4°C for 2 h with 2 g Cortactin (Upstate) or (Invitrogen). PCR was then performed with the above primer and an correlations in each plot, the median correlation coefficient is indicated by a vertical red line and the 25th and 75th percentiles are coloured, with negative l correlations in red and positive correlations in blue. The dotted vertical line shows the zero correlation point. 4G10 phosphotyrosine (Cell Signalling) antibodies. Primary anti- additional primer (ACGGAGGACGATGCGG) for 25 cycles. PCR prod- Source data are available online for this figure. bodies were precipitated by incubation with 50 ll Protein G uct was then run on a 10% native (1:29) polyacrylamide gel, stained Dynal beads (Invitrogen) for 2 h at 4°C. Immunoprecipitates were with Sybr Gold (Qiagen) and bands excised over a UV light box. The electrophoresed on 10% SDS–PAGE gels and immunoblotted for DNA was then precipitated using isopropanol and a final 10 cycle performed using TaqMan microRNA assays (Applied Biosystems). passive lysis buffer (Promega) and luciferase activity measured with phosphotyrosine (4G10, Cell Signalling), Cortactin (Upstate) or PCR performed with the following primers: AATGATACGGCGAC- Real-time PCR data for mRNA are expressed relative to the averaged the Dual-Luciferase Reporter Assay System (Promega) using the Tks5 (Millipore). CACCGACTATGGATACTTAGTCAGGGAGG ACGATGCGG, CAAG- values for GAPDH, RPL32, RPL39, and HPRT. PCR data for miRNA TD-20/20 Luminometer (Turner Designs). All reporter assays are CAGAAGACGGCATACGATCCCGCTCGTCGTC TG. Reactions were are normalized to U6. shown as relative luciferase activities (averaged ratios of Renilla/ Rho activation then run on 2% metaphor agarose/TBE gels and bands (~115 and Firefly luciferase SEM from at least three separate experiments). In 135 bp) excised corresponding to the linker sequence + RNA CLIP � Transfection and reporter assays Fig 3 where plasmids were not cotransfected, RNAiMAX (Invitrogen) Levels of active and total Rho were determined using the Active Rho tag. Samples were finally purified using quick-spin columns was used to transfect miRNA mimics. For siRNA experiments, cells Pull-Down and Detection kit as per manufacturer’s instruction (Qiagen) and subjected to Illumina GAII 35-bp read-length deep For reporter assays, cells were plated in 24-well plates and cotrans- were transfected with 20 nM Smart Pool siRNAs (Dharmacon) using (Thermo Scientific, cat#16116). sequencing (GeneWorks). fected using Lipofectamine 2000 (Invitrogen) with 200 ng firefly Lipofectamine RNAiMAX and analyzed after 72 h. luciferase reporter plasmid and 5 ng pCI-neo-hRL Renilla plasmid Argonaute:miRNA immunoprecipitation Alignment of sequencing data

(Pillai et al, 2005) into which 30UTRs were cloned (see Supplemen- Western Blotting tary Table S3 for cloning primers). In Fig 2A, cells were cotrans- MDA-MB-231 cells were grown in 20 × 10 cm dishes and transfected In-house scripts used for analysis may be accessed at https:// fected with 20 nM scrambled, miR-200a, or miR-200b microRNA Whole-cell extracts were prepared from transfected cells by Triton with 60 nM miRNA mimic (Ambion or GenePharma) using HiPerfect bitbucket.org/sacgf/bracken_hits-clip_2013, and were written in perl mimic (GenePharma). After 48 h of incubation, cells were lysed in X-100 lysis (50 mM Hepes, pH 7.5, 150 mM sodium chloride, 10 mM transfection reagent (Qiagen). After 24 h, cells were suspended in and Python using Biopython (Cock et al, 2009) and HTSeq

ª 2014 The Authors The EMBO Journal Vol 33 | No 18 | 2014 2051 2052 The EMBO Journal Vol 33 | No 18 | 2014 ª 2014 The Authors Cameron P Bracken et al miR-200 targets a cytoskeletal regulatory network The EMBO Journal The EMBO Journal miR-200 targets a cytoskeletal regulatory network Cameron P Bracken et al

(10 mM PIPES pH 6.8, 100 mM KCl, 300 mM sucrose, 2 mM ous and treated with PNK to ligate the 50 RNA linker

(http://www-huber.embl.de/users/anders/HTSeq). Prior to align- Invasion and migration assays washed in Developing Buffer (50 mM Tris–HCl pH 7.4; 5 mM Chen WS, Chen CC, Chen LL, Lee CC, Huang TS (2013) Secreted heat shock ment, perl scripts were used to filter reads for average quality and CaCl2) for 30 min, and then incubated in Developing Buffer protein 90a (HSP90a) induces nuclear factor-jB-mediated TCF12 protein homopolymeric tracts longer than 12 nucleotides, trim linker For Boyden chamber invasion assays, 72 h after transfection, shaking 60 rpm, 37°C, 24 h. Gels were stained in 0.1% Brilliant expression to down-regulate E-cadherin and to enhance colorectal cancer sequences, and separate sequences by barcode. The resulting 17–30 2 × 105 cells were plated into each BD Biocoat Matrigel Invasion Blue/40% methanol/10% acetic acid overnight and destained in cell migration and invasion. J Biol Chem 288: 9001 – 9010 nucleotide sequences were aligned against the human genome Chamber (8.0 lM PET) for 20 h. Cells were fixed with 10% buffered 25% methanol/7% acetic acid. Each experiment was repeated at Chi SW, Hannon GJ, Darnell RB (2012) An alternative mode of microRNA (NCBI36/hg18 build, downloaded from UCSC: http://genome.ucs- formalin, washed extensively with PBS, permeabilized with 0.1% least three times. target recognition. Nat Struct Mol Biol 19: 321 – 327 c.edu) using Bowtie (Langmead et al, 2009). The alignment parame- Triton X-100, stained with DAPI, washed extensively with PBS, and Chi SW, Zang JB, Mele A, Darnell RB (2009) Argonaute HITS-CLIP decodes ters allowed a maximum of 1 mismatch and reported a maximum of mounted with DAKO mounting media. Images were taken of each Supplementary information for this article is available online: microRNA-mRNA interaction maps. Nature 460: 479 – 486 15 alignments per read. filter, and all cells were counted. Identical protocols were used for http://emboj.embopress.org Cock PJ, Antao T, Chang JT, Chapman BA, Cox CJ, Dalke A, Friedberg I, migration assays using Transwells (Costar; 8.0 lM PET) without Hamelryck T, Kauff F, Wilczynski B, de Hoon MJ (2009) Biopython: freely Analysis of sequencing data Matrigel. For 3D scratch wound invasion assays, 24 h after transfec- Acknowledgements available Python tools for computational molecular biology and tion, 2 × 105 cells were resuspended in 4.5 mg/ml Matrigel Base- We thank Zissimos Mourelatos for the kind gift of the Argonaute antibody, bioinformatics. Bioinformatics 25: 1422 – 1423 Reference gene annotation in gtf format was downloaded from ment Membrane Matrix (BD) in serum-free DMEM and plated into Narelle Mancini for her technical assistance, and all our laboratory collea- Courtneidge SA, Azucena EF, Pass I, Seals DF, Tesfay L (2005) The SRC UCSC (http://genome.ucsc.edu/cgi-bin/hgTables; Group: ‘Genes each well. Matrigel was set for 1 h at 37°C then covered in 500 ll gues for their support and advice. This work was supported by Fellowships substrate Tks5, podosomes (invadopodia), and cancer cell invasion. Cold and Genes Prediction Tracks’; Track: ‘Refseq Genes’). To generate complete media. When cells reached confluence, wells were from the National Health and Medical Research Council of Australia to AY Spring Harb Symp Quant Biol 70: 167 – 171 the reference control, reads were combined from separate HITS- scratched. The scratch was filled with 4.5 mg/ml Matrigel, set for (631383) and GJG (1026191), from the National Breast Cancer Foundation Creighton CJ, Gibbons DL, Kurie JM (2013) The role of epithelial-mesenchymal CLIP experiments from untransfected and scrambled pre-miR trans- 1 h at 37°C, and then covered in 500 ll complete media. The rate of and Florey Foundation to CPB, from the Cancer Council to PAG and by transition programming in invasion and metastasis: a clinical perspective. fected cells. Given the size of fragments isolated from the HITS- wound closure was monitored by live-cell imaging using an Incu- grants GTN1008327 and GTN1034633 from the National Health and Medical Cancer Manag Res 5: 187 – 195 CLIP immunoprecipitation is ~40–60 nt, mapped reads were Cyte (Essen Biosciences) for up to 3–5 days. All assays were Research Council of Australia, the National Breast Cancer Foundation Croft DR, Sahai E, Mavria G, Li S, Tsai J, Lee WM, Marshall CJ, Olson MF extended 40 nt 30 from the 50 end of the aligned read. Duplicate performed in triplicates. Australia, the Association for International Cancer Research and the Kids’ (2004) Conditional ROCK activation in vivo induces tumor cell alignments were removed to produce a unique set of alignments, Cancer Project. dissemination and angiogenesis. Cancer Res 64: 8994 – 9001 leaving 483,237 alignments for miR-200a, ~644,442 alignments for Invadopodia assays Cuenda A, Dorow DS (1998) Differential activation of stress-activated protein miR-200b, and ~560,900 alignments for the control samples. For Author contributions kinase kinases SKK4/MKK7 and SKK1/MKK4 by the mixed-lineage kinase-2 analysis of the read alignment locations, regions of the genome Invadopodia were identified using 3 criteria; cortactin-actin and CPB, KBJ, ASY, YK-G, and GJG conceived the project; CBP, JAW, XL, BKD, MAA, and mitogen-activated protein kinase kinase (MKK) kinase-1. Biochem J 333 were defined using the reference annotation and analysis was Tks5-actin colocalizations and degradation of fluorescent gelatin MS, CN, and DWT conducted experiments; JML and ASY contributed reagents; (Pt 1): 11 – 15 performed using Python scripts. matrix. AGB assisted with manuscript preparation; CPB, JAW, XL, DL, PAG, AT, KAP, and Derijard B, Raingeaud J, Barrett T, Wu IH, Han J, Ulevitch RJ, Davis RJ (1995) Peaks were defined using scripts as follows: read depth per base Cortactin-actin and Tks5-actin colocalizations were performed by YK-G analyzed the data; and CPB, ASY, YK-G, and GJG wrote the manuscript. Independent human MAP-kinase signal transduction pathways defined by position per strand was calculated for each sample. The stranded immunofluorescent staining with cortactin or Tks5 antibodies and MEK and MKK isoforms. Science 267: 682 – 685 read depth of each transfection sample was compared to the control phalloidin (Alexa fluor 633-conjugated) as described in Immuno- Conflict of interest Erkutlu I, Cigiloglu A, Kalender ME, Alptekin M, Demiryurek AT, Suner A, sample at the same interval. Peaks were called where depth in the fluorescence, after which approximately 300 cells per variable were The authors declare that they have no conflict of interest. Ozkaya E, Ulasli M, Camci C (2013) Correlation between Rho-kinase transfection sample was at least three reads, and the depth in the scored for the presence or absence of cortactin-actin (or Tks5-actin) pathway gene expressions and development and progression of control sample was is zero or where depth was at least threefold colocalization. Three independent experiments were performed. glioblastoma multiforme. Tumour Biol 34: 1139 – 1144 greater in treatment than control. Annotation of peaks was References Estecha A, Sanchez-Martin L, Puig-Kroger A, Bartolome RA, Teixido J, performed by comparing the genomic location with annotated genes Fluorescent matrix degradation assay Samaniego R, Sanchez-Mateos P (2009) Moesin orchestrates cortical to assign each peak a gene name and genetic region. For each peak, Alberts AS, Treisman R (1998) Activation of RhoA and SAPK/JNK signalling polarity of melanoma tumour cells to initiate 3D invasion. J Cell Sci 122: the most likely microRNA target site was identified by searching the Acid-washed (20% nitric acid, 30 min) coverslips were coated with pathways by the RhoA-specific exchange factor mNET1. EMBO J 17: 3492 – 3501 reference sequence for seed matches according to various targeting 0.01% poly-L-lysine (Sigma) for 15 min and fixed in 0.5% glutaral- 4075 – 4085 Forman JJ, Coller HA (2010) The code within the code: microRNAs target rules. The miR-200a-3p and miR-200b-3p seed sequences were dehyde/PBS for 10 min. Coverslips were inverted onto 80 ll drop- Arber S, Barbayannis FA, Hanser H, Schneider C, Stanyon CA, Bernard O, coding regions. Cell Cycle 9: 1533 – 1541 obtained from miRBase (miR-200a-3p: CAGTGTTA, miR-200b-3p: lets of warmed 1:8 0.1% Oregon Green 488 conjugate-gelatin Caroni P (1998) Regulation of actin dynamics through phosphorylation of Frenette P, Haines E, Loloyan M, Kinal M, Pakarian P, Piekny A (2012) An CAGTATTA). Sequence matching was performed using Python regu- (Invitrogen): 0.2% porcine gelatin for 10 min. Coverslips were cofilin by LIM-kinase. Nature 393: 805 – 809 anillin-Ect2 complex stabilizes central spindle microtubules at the cortex lar expressions. For each annotated transcript which overlaps a washed in PBS then incubated with shaking for 3 min in 5 mg/ml Bos JL, Rehmann H, Wittinghofer A (2007) GEFs and GAPs: critical elements in during cytokinesis. PLoS ONE 7:e34888 peak, the mRNA sequence was analyzed computationally to identify NaBH4 in PBS. After rinsing in PBS, coverslips were incubated at the control of small G proteins. Cell 129: 865 – 877 Fukata Y, Kimura K, Oshiro N, Saya H, Matsuura Y, Kaibuchi K (1998) potential microRNA seed sites. Seed sites were classified from best- 37°C in complete medium for 2 h. 1 × 105 cells were seeded on each Bowden ET, Onikoyi E, Slack R, Myoui A, Yoneda T, Yamada KM, Mueller SC Association of the myosin-binding subunit of myosin phosphatase and to worst-match in the following order: 8-mer perfect match, 7-mer coverslip in duplicate, incubated for 16 h, and processed for immu- (2006) Co-localization of cortactin and phosphotyrosine identifies moesin: dual regulation of moesin phosphorylation by Rho-associated perfect match, 6-mer perfect match, central-paired (miRNA nt 4–15) nofluorescence. Images were taken for at least 200 cells per sample. active invadopodia in human breast cancer cells. Exp Cell Res 312: kinase and myosin phosphatase. J Cell Biol 141: 409 – 418 with 2 mismatches, 8-mer with 1 mismatch, 7-mer with 1 mismatch, The percentage of cells with invadopodia was calculated as the 1240 – 1253 Gal A, Sjöblom T, Fedorova L, Imreh S, Beug H, Moustakas A (2008) Sustained 7-mer with single nucleotide insertion or deletion (seed-bulge) inter- number of cells above dark holes in fluorescent matrix normalized Boyle SN, Koleske AJ (2007) Use of a chemical genetic technique to identify TGF beta exposure suppresses Smad and non-Smad signalling in action. The best seed site which occurred within 150 bp was to total cell number (counted by DAPI staining for nuclei). Each myosin IIb as a substrate of the Abl-related gene (Arg) tyrosine kinase. mammary epithelial cells, leading to EMT and inhibition of growth arrest assigned to the peak. experiment was repeated at least three times. Biochemistry 46: 11614 – 11620 and apoptosis. Oncogene 27: 1218 – 1230 For analysis of reads by location in the gene, distances from the Bracken CP, Gregory PA, Kolesnikoff N, Bert AG, Wang J, Shannon MF, Goodall Garcia E, Jones GE, Machesky LM, Anton IM (2012) WIP: WASP- midpoint of each read to the nearest annotated stop codon and tran- Gelatin zymography GJ (2008) A double-negative feedback loop between ZEB1-SIP1 and the interacting proteins at invadopodia and podosomes. Eur J Cell Biol 91: 869 – 877 script end were calculated and grouped according to the length of microRNA-200 family regulates epithelial-mesenchymal transition. Cancer Gil-Henn H, Patsialou A, Wang Y, Warren MS, Condeelis JS, Koleske AJ (2012) the feature (eg mRNA or 30UTR). Cumulative counts were used to Conditioned media were collected from cells at 80–90% confluency Res 68: 7846 – 7854 Arg/Abl2 promotes invasion and attenuates proliferation of breast cancer produce a stacked bar chart. For gene graphs, arrows indicating (96 h after transfection). 2× Laemmli buffer containing no reducing Burk U, Schubert J, Wellner U, Schmalhofer O, Vincan E, Spaderna S, Brabletz T in vivo. Oncogene 32: 2622 – 2630 potential miRNA targets were drawn using the seed matching agent was added to samples of conditioned media and incubated (2008) A reciprocal repression between ZEB1 and members of the miR-200 Grass GD, Bratoeva M, Toole BP (2012) Regulation of invadopodia formation method described above. Only peaks (and all unique reads at sites at room temperature for 15 min. Samples were analyzed using family promotes EMT and invasion in cancer cells. EMBO Rep 9: 582 – 589 and activity by CD147. J Cell Sci 125: 777 – 788 corresponding to locations with a peak in the control, miR-200a or 10% gelatin PAGE gels. Gels were first incubated in Renaturing Carthew RW, Sontheimer EJ (2009) Origins and mechanisms of miRNAs and Gregory PA, Bert AG, Paterson EL, Barry SC, Tsykin A, Farshid G, Vadas MA, miR-200b transfections) are shown. Buffer (2.5% Triton X-100 in Milli Q water) for 30 min twice, siRNAs. Cell 136: 642 – 655 Khew-Goodall Y, Goodall GJ (2008a) The miR-200 family and miR-205

ª 2014 The Authors The EMBO Journal Vol 33 | No 18 | 2014 2053 2054 The EMBO Journal Vol 33 | No 18 | 2014 ª 2014 The Authors Cameron P Bracken et al miR-200 targets a cytoskeletal regulatory network The EMBO Journal The EMBO Journal miR-200 targets a cytoskeletal regulatory network Cameron P Bracken et al

regulate epithelial to mesenchymal transition by targeting ZEB1 and SIP1. Kovacs EM, Verma S, Thomas SG, Yap AS (2011) Tuba and N-WASP function Poincloux R, Lizarraga F, Chavrier P (2009) Matrix invasion by tumour of MLC phosphorylation for assembly of stress fibers and focal adhesions Nat Cell Biol 10: 593 – 601 cooperatively to position the central lumen during epithelial cyst cells: a focus on MT1-MMP trafficking to invadopodia. J Cell Sci 122: in 3T3 fibroblasts. J Cell Biol 150: 797 – 806 Gregory PA, Bracken CP, Bert AG, Goodall GJ (2008b) MicroRNAs as regulators morphogenesis. Cell Adh Migr 5: 344 – 350 3015 – 3024 Turtoi A, Blomme A, Bellahcene A, Gilles C, Hennequiere V, Peixoto P, Bianchi of epithelial-mesenchymal transition. Cell Cycle 7: 3112 – 3118 Krzewski K, Chen X, Orange JS, Strominger JL (2006) Formation of a WIP-, Riley KJ, Yario TA, Steitz JA (2012) Association of Argonaute proteins and E, Noel A, De Pauw E, Lifrange E, Delvenne P, Castronovo V (2013) Gregory PA, Bracken CP, Smith E, Bert AG, Wright JA, Roslan S, Morris M, WASp-, actin-, and myosin IIA-containing multiprotein complex in microRNAs can occur after cell lysis. RNA 18: 1581 – 1585 Myoferlin is a key regulator of EGFR activity in breast cancer. Cancer Res Wyatt L, Farshid G, Lim YY, Lindeman GJ, Shannon MF, Drew PA, activated NK cells and its alteration by KIR inhibitory signaling. J Cell Biol Safina A, Ren MQ, Vandette E, Bakin AV (2008) TAK1 is required for TGF-beta 73: 5438 – 5448 Khew-Goodall Y, Goodall GJ (2011) An autocrine TGF-{beta}/ZEB/miR-200 173: 121 – 132 1-mediated regulation of matrix metalloproteinase-9 and metastasis. Wang L, Pan Y, Dai JL (2004) Evidence of MKK4 pro-oncogenic activity in signaling network regulates establishment and maintenance of Langmead B, Trapnell C, Pop M, Salzberg SL (2009) Ultrafast and Oncogene 27: 1198 – 1207 breast and pancreatic tumors. Oncogene 23: 5978 – 5985 epithelial-mesenchymal transition. Mol Biol Cell 22: 1686 – 1698 memory-efficient alignment of short DNA sequences to the human Sahai E, Marshall CJ (2003) Differing modes of tumour cell invasion have Wang W, Zhou G, Hu MC, Yao Z, Tan TH (1997) Activation of the Grimson A, Farh KK, Johnston WK, Garrett-Engele P, Lim LP, Bartel DP (2007) genome. Genome Biol 10:R25 distinct requirements for Rho/ROCK signalling and extracellular hematopoietic progenitor kinase-1 (HPK1)-dependent, stress-activated MicroRNA targeting specificity in mammals: determinants beyond seed Li X, Roslan S, Johnstone CN, Wright JA, Bracken CP, Anderson M, Bert AG, proteolysis. Nat Cell Biol 5: 711 – 719 c-Jun N-terminal kinase (JNK) pathway by transforming growth factor pairing. Mol Cell 27: 91 – 105 Selth LA, Anderson RL, Goodall GJ, Gregory PA, Khew-Goodall Y Saras J, Franzen P, Aspenstrom P, Hellman U, Gonez LJ, Heldin CH (1997) beta (TGF-beta)-activated kinase (TAK1), a kinase mediator of TGF beta Gu S, Jin L, Zhang F, Sarnow P, Kay MA (2009) Biological basis for restriction (2013) MiR-200 can repress breast cancer metastasis through A novel GTPase-activating protein for Rho interacts with a PDZ signal transduction. J Biol Chem 272: 22771 – 22775 of microRNA targets to the 3’ untranslated region in mammalian mRNAs. ZEB1-independent but moesin-dependent pathways. Oncogene 33: domain of the protein-tyrosine phosphatase PTPL1. J Biol Chem 272: Wehrle-Haller B (2012) Assembly and disassembly of cell matrix adhesions. Nat Struct Mol Biol 16: 144 – 150 4077 – 4088 24333 – 24338 Curr Opin Cell Biol 24: 569 – 581 Hafner M, Landthaler M, Burger L, Khorshid M, Hausser J, Berninger P, LimLP,LauNC,Garrett-EngeleP,GrimsonA,SchelterJM,CastleJ,BartelDP,Linsley Sasahara Y, Rachid R, Byrne MJ, de la Fuente MA, Abraham RT, Ramesh N, Williams KC, Renthal NE, Condon JC, Gerard RD, Mendelson CR (2012) Rothballer A, Ascano M Jr, Jungkamp AC, Munschauer M, Ulrich A, Wardle PS,JohnsonJM(2005)MicroarrayanalysisshowsthatsomemicroRNAs Geha RS (2002) Mechanism of recruitment of WASP to the immunological MicroRNA-200a serves a key role in the decline of progesterone receptor GS, Dewell S, Zavolan M, Tuschl T (2010) Transcriptome-wide downregulatelargenumbersoftargetmRNAs.Nature433:769 – 773 synapse and of its activation following TCR ligation. Mol Cell 10: function leading to term and preterm labor. Proc Natl Acad Sci USA 109: identification of RNA-binding protein and microRNA target sites by Loeb GB, Khan AA, Canner D, Hiatt JB, Shendure J, Darnell RB, Leslie CS, 1269 – 1281 7529 – 7534 PAR-CLIP. Cell 141: 129 – 141 Rudensky AY (2012) Transcriptome-wide miR-155 Binding Map Reveals Shin C, Nam JW, Farh KK, Chiang HR, Shkumatava A, Bartel DP (2010) Wong CC, Wong CM, Tung EK, Man K, Ng IO (2009) Rho-kinase 2 is Hagedorn EJ, Kelley LC, Naegeli KM, Wang Z, Chi Q, Sherwood DR (2014) Widespread Noncanonical MicroRNA Targeting. Mol Cell 48: 760 – 770 Expanding the microRNA targeting code: functional sites with centered frequently overexpressed in hepatocellular carcinoma and involved in ADF/cofilin promotes invadopodial membrane recycling during cell Miller MM, Lapetina S, MacGrath SM, Sfakianos MK, Pollard TD, Koleske AJ pairing. Mol Cell 38: 789 – 802 tumor invasion. Hepatology 49: 1583 – 1594 invasion in vivo. J Cell Biol 204: 1209 – 1218 (2010) Regulation of actin polymerization and adhesion-dependent cell Sibony-Benyamini H, Gil-Henn H (2012) Invadopodia: the leading force. Eur J Xia H, Ng SS, Jiang S, Cheung WK, Sze J, Bian XW, Kung HF, Lin MC (2012) Haynes J, Srivastava J, Madson N, Wittmann T, Barber DL (2011) Dynamic edge protrusion by the Abl-related gene (Arg) tyrosine kinase and Cell Biol 91: 896 – 901 miR-200a-mediated downregulation of ZEB2 and CTNNB1 differentially actin remodeling during epithelial-mesenchymal transition depends on N-WASp. Biochemistry 49: 2227 – 2234 Singh DK, Sarkar J, Raghavan A, Reddy SP, Raj JU (2011) Hypoxia modulates inhibits nasopharyngeal carcinoma cell growth, migration and invasion. increased moesin expression. Mol Biol Cell 22: 4750 – 4764 Mu Y, Gudey SK, Landstrom M (2012) Non-Smad signaling pathways. Cell the expression of leucine zipper-positive MYPT1 and its interaction with Biochem Biophys Res Commun 391: 535 – 541 He M, Cheng Y, Li W, Liu Q, Liu J, Huang J, Fu X (2010) Vascular endothelial Tissue Res 347: 11 – 20 protein kinase G and Rho kinases in pulmonary arterial smooth muscle Yang N, Higuchi O, Ohashi K, Nagata K, Wada A, Kangawa K, Nishida E, growth factor C promotes cervical cancer metastasis via up-regulation Murphy DA, Courtneidge SA (2011) The ‘ins’ and ‘outs’ of podosomes and cells. Pulm Circ 1: 487 – 498 Mizuno K (1998) Cofilin phosphorylation by LIM-kinase 1 and its role in and activation of RhoA/ROCK-2/moesin cascade. BMC Cancer 10: 170 invadopodia: characteristics, formation and function. Nat Rev Mol Cell Biol Su F, Li H, Yan C, Jia B, Zhang Y, Chen X (2009) Depleting MEKK1 expression Rac-mediated actin reorganization. Nature 393: 809 – 812 Ho HY, Rohatgi R, Lebensohn AM, Le M, Li J, Gygi SP, Kirschner MW (2004) 12: 413 – 426 inhibits the ability of invasion and migration of human pancreatic cancer Yu X, Machesky LM (2012) Cells assemble invadopodia-like structures and Toca-1 mediates Cdc42-dependent actin nucleation by activating the Nieto MA (2013) Epithelial plasticity: a common theme in embryonic and cells. J Cancer Res Clin Oncol 135: 1655 – 1663 invade into matrigel in a matrix metalloprotease dependent manner in N-WASP-WIP complex. Cell 118: 203 – 216 cancer cells. Science 342: 1234850 Surks HK, Riddick N, Ohtani K (2005) M-RIP targets myosin phosphatase to the circular invasion assay. PLoS ONE 7:e30605 Hu H, Bliss JM, Wang Y, Colicelli J (2005) RIN1 is an ABL tyrosine kinase Noy E, Fried S, Matalon O, Barda-Saad M (2012) WIP Remodeling Actin stress fibers to regulate myosin light chain phosphorylation in vascular Zhang L, Deng M, Parthasarathy R, Wang L, Mongan M, Molkentin JD, activator and a regulator of epithelial-cell adhesion and migration. Curr behind the Scenes: how WIP Reshapes Immune and Other Functions. Int J smooth muscle cells. J Biol Chem 280: 42543 – 42551 Zheng Y, Xia Y (2005) MEKK1 transduces activin signals in keratinocytes Biol 15: 815 – 823 Mol Sci 13: 7629 – 7647 Thiesen S, Kubart S, Ropers HH, Nothwang HG (2000) Isolation of two novel to induce actin stress fiber formation and migration. Mol Cell Biol 25: Iliopoulos D, Lindahl-Allen M, Polytarchou C, Hirsch HA, Tsichlis PN, Struhl K Ono R, Matsuoka J, Yamatsuji T, Naomoto Y, Tanaka N, Matsui H, Matsushita human RhoGEFs, ARHGEF3 and ARHGEF4, in 3p13-21 and 2q22. Biochem 60 – 65 (2010) Loss of miR-200 inhibition of Suz12 leads to polycomb-mediated M(2008) M-RIP, a novel target of JNK signaling and a requirement for Biophys Res Commun 273: 364 – 369 Zhang X, Pickin KA, Bose R, Jura N, Cole PA, Kuriyan J (2007) Inhibition of the repression required for the formation and maintenance of cancer stem human cancer cell invasion. Int J Mol Med 22: 199 – 203 Totsukawa G, Yamakita Y, Yamashiro S, Hartshorne DJ, Sasaki Y, Matsumura F EGF receptor by binding of MIG6 to an activating kinase domain interface. cells. Mol Cell 39: 761 – 772 Park SM, Gaur AB, Lengyel E, Peter ME (2008) The miR-200 family determines (2000) Distinct roles of ROCK (Rho-kinase) and MLCK in spatial regulation Nature 450: 741 – 744 Kaneko T, Amano M, Maeda A, Goto H, Takahashi K, Ito M, Kaibuchi K (2000) the epithelial phenotype of cancer cells by targeting the E-cadherin Identification of calponin as a novel substrate of Rho-kinase. Biochem repressors ZEB1 and ZEB2. Genes Dev 22: 894 – 907 Biophys Res Commun 273: 110 – 116 Parker LP, Taylor DD, Kesterson J, Metzinger DS, Gercel-Taylor C (2009) Kawano Y, Fukata Y, Oshiro N, Amano M, Nakamura T, Ito M, Matsumura F, Modulation of microRNA associated with ovarian cancer cells by genistein. Inagaki M, Kaibuchi K (1999) Phosphorylation of myosin-binding subunit Eur J Gynaecol Oncol 30: 616 – 621 (MBS) of myosin phosphatase by Rho-kinase in vivo. J Cell Biol 147: Peacock JG, Couch BA, Koleske AJ (2010) The Abl and Arg non-receptor 1023 – 1038 tyrosine kinases regulate different zones of stress fiber, focal adhesion, Korpal M, Ell BJ, Buffa FM, Ibrahim T, Blanco MA, Celia-Terrassa T, Mercatali and contractile network localization in spreading fibroblasts. Cytoskeleton L, Khan Z, Goodarzi H, Hua Y, Wei Y, Hu G, Garcia BA, Ragoussis J, (Hoboken) 67: 666 – 675 Amadori D, Harris AL, Kang Y (2011) Direct targeting of Sec23a by Peng C, Li N, Ng Y-K, Zhang J, Meier F, Theis FJ, Merkenschlager M, Chen W, miR-200s influences cancer cell secretome and promotes metastatic Wurst W, Prakash N (2012) A unilateral negative feedback loop between colonization. Nat Med 17: 1101 – 1108 miR-200 microRNAs and Sox2/E2F3 controls neural progenitor cell-cycle Korpal M, Lee ES, Hu G, Kang Y (2008) The miR-200 family inhibits exit and differentiation. J Neurosci 32: 13292 – 13308 epithelial-mesenchymal transition and cancer cell migration by direct Piekny AJ, Glotzer M (2008) Anillin is a scaffold protein that links RhoA, actin, targeting of E-cadherin transcriptional repressors ZEB1 and ZEB2. J Biol and myosin during cytokinesis. Curr Biol 18: 30 – 36 Chem 283: 14910 – 14914 Pillai RS, Bhattacharyya SN, Artus CG, Zoller T, Cougot N, Basyuk E, Bertrand Kovacs EM, Makar RS, Gertler FB (2006) Tuba stimulates intracellular E, Filipowicz W (2005) Inhibition of translational initiation by Let-7 N-WASP-dependent actin assembly. J Cell Sci 119: 2715 – 2726 MicroRNA in human cells. Science 309: 1573 – 1576

ª 2014 The Authors The EMBO Journal Vol 33 | No 18 | 2014 2055 2056 The EMBO Journal Vol 33 | No 18 | 2014 ª 2014 The Authors Cameron P Bracken et al miR-200 targets a cytoskeletal regulatory network The EMBO Journal The EMBO Journal miR-200 targets a cytoskeletal regulatory network Cameron P Bracken et al

ª 2014 The Authors The EMBO Journal Vol 33 | No 18 | 2014 2055 2056 The EMBO Journal Vol 33 | No 18 | 2014 ª 2014 The Authors The EMBO Journal MiR-133 promotes reprogramming by Snai1 repression Naoto Muraoka et al Article

Song et al, 2012). The in vivo iCMs were more fully reprogrammed cTnT expression in MEFs, similar to that induced in CFs (Fig 1A than their cultured counterparts, suggesting the presence of unde- and B, Supplementary Fig S1H; Ieda et al, 2010). fined factors that enhance reprogramming. Identification of such Next, we introduced these miRNAs along with GMT into MEFs to MiR-133 promotes cardiac reprogramming by potent reprogramming factors could provide new insights into the investigate whether miRNAs promote cardiac reprogramming. We mechanisms of cardiac reprogramming. found that the number of aMHC-GFP+ cells activating a cardiac MicroRNAs (miRNAs) can suppress the expression of hundreds reporter was increased by approximately twofold and that the directly repressing Snai1 and silencing + of genes, primarily through binding to the 30-untranslated region number of cTnT cells expressing the endogenous cardiac-specific (UTR) of target mRNAs, and thus play important roles in cell fate gene was increased by approximately sixfold by the addition of fibroblast signatures decisions. Embryonic stem cell-specific miRNAs enhanced the repro- miR-1, 133, or four miRs to the GMT transduction (Fig 1C and D). gramming efficiency of fibroblasts into induced pluripotent stem In contrast, addition of miR-208 or miR-499 had no substantial Naoto Muraoka1,2, Hiroyuki Yamakawa1,2, Kazutaka Miyamoto1,2, Taketaro Sadahiro1,2, Tomohiko Umei1, cells (iPSCs; Judson et al, 2009; Subramanyam et al, 2011), and effects, suggesting that the miRNA effects were specific. Among more recently, Jayawardena et al (2012) reported that a combina- them, miR-133 mimics showed the greatest effects, and thereby, we 1 1 1 1 1,2 1,2 Mari Isomi , Hanae Nakashima , Mizuha Akiyama , Rie Wada , Kohei Inagawa , Takahiko Nishiyama , tion of muscle-specific miRNAs (miR-1, 133, 208, 499) alone repro- used miR-133 in subsequent studies. We determined the dose Ruri Kaneda1,2, Toru Fukuda3, Shu Takeda3, Shugo Tohyama2, Hisayuki Hashimoto2, Yoshifumi grammed neonatal mouse CFs into cardiomyocyte-like cells dependency of miR-133-mediated cardiac induction and found that (Jayawardena et al, 2012). However, it remains unclear whether 15 nM of miR-133 was sufficient (Fig 1E and F). Addition of JAK 4 5 6 7 2 1,2,8,* Kawamura , Naoki Goshima , Ryo Aeba , Hiroyuki Yamagishi , Keiichi Fukuda & Masaki Ieda other types of fibroblasts could also be converted into iCMs inhibitor I, which reported to increase cardiac induction, to GMT/ by miRNAs. Moreover, the global transcriptional changes and miR-133 did not augment the reprogramming efficiency (Supple- mechanistic basis of cardiac reprogramming by miRNAs remain mentary Fig S1I and J; Jayawardena et al, 2012). FACS analyses unknown. demonstrated that expression of another cardiac marker, sarcomeric Abstract Introduction Here, we show that miR-133a (miR-133) promoted cardiac repro- a-actinin (a-actinin), was also increased by addition of miR-133 to gramming in mouse embryonic fibroblasts (MEFs), adult mouse GMT (Fig 1G). Immunostaining for cardiac markers, including Fibroblasts can be directly reprogrammed into cardiomyocyte-like Direct reprogramming of mature cells from one lineage to another cardiac fibroblasts, and human cardiac fibroblasts (HCFs) in combi- a-actinin, cTnT, and atrial natriuretic peptide (ANP), demonstrated cells (iCMs) by overexpression of cardiac transcription factors or without passing through a stem cell state has emerged as a new nation with GMT or GMTMM transduction. We found that miR-133 that GMT/miR-133 strongly enhanced cardiac protein expression, microRNAs. However, induction of functional cardiomyocytes is strategy for generating cell types of interest and may hold great suppressed the fibroblast programs by directly repressing Snai1, a and the iCMs had well-defined sarcomeric structures, similar to inefficient, and molecular mechanisms of direct reprogramming potential for regenerative medicine. Thus far, neurons, cardiomyo- master regulator of epithelial-to-mesenchymal transition (EMT), and neonatal cardiomyocytes (Fig 1H and I, Supplementary Fig S1B). remain undefined. Here, we demonstrate that addition of miR-133a cytes, hepatocytes, blood precursor cells, and neural progenitors thereby promoted cardiac reprogramming. Thus, miR-133 improved cardiac induction from MEFs in combina- (miR-133) to Gata4, Mef2c, and Tbx5 (GMT) or GMT plus Mesp1 and were successfully induced from fibroblasts by overexpression of tion with GMT transduction. Myocd improved cardiac reprogramming from mouse or human lineage-specific transcription factor (Ieda et al, 2010; Szabo et al, fibroblasts by directly repressing Snai1, a master regulator of 2010; Vierbuchen et al, 2010; Sekiya & Suzuki, 2011; Han et al, Results MiR-133 rapidly and efficiently induces functional epithelial-to-mesenchymal transition. MiR-133 overexpression 2012; Wada et al, 2013). Suppression of the starting-cell signature is cardiomyocyte-like cells from MEFs in combination with GMT generated sevenfold more beating iCMs from mouse a recognized characteristic of cell fate conversion, although the MiR-133 promotes cardiac induction in MEFs in combination with Gata4/Mef2c/Tbx5 embryonic fibroblasts and shortened the duration to induce beat- molecular mechanisms underlying this process and its importance with Gata4/Mef2c/Tbx5 ing cells from 30 to 10 days, compared to GMT alone. Snai1 knock- during direct reprogramming remain poorly understood (Marro To investigate the effects of miR-133 on cardiac reprogramming in down suppressed fibroblast genes, upregulated cardiac gene et al, 2011; Muraoka & Ieda, 2014). We first investigated whether miR-1, 133, 208, and 499 alone, more detail, we next compared the time courses of reprogramming expression, and induced more contracting iCMs with GMT trans- It was reported that induced cardiomyocyte-like cells (iCMs) can shown previously to induce cardiac reprogramming in neonatal between GMT and GMT/miR-133 induction. FACS analyses revealed duction, recapitulating the effects of miR-133 overexpression. In be directly generated from mouse fibroblasts by the combination of mouse CFs, could also generate iCMs from MEFs, which have a that GMT/miR-133 induced significantly more a-MHC-GFP and contrast, overexpression of Snai1 in GMT/miR-133-transduced cells transcription factors, Gata4, Mef2c, and Tbx5 (GMT), GMT plus distinct embryonic origin compared to CFs. We used MEFs from cTnT expression in the MEFs by as early as day 3 than GMT alone, maintained fibroblast signatures and inhibited generation of beat- Hand2 (GHMT), or Mef2c, Myocd, and Tbx5 in vitro (Ieda et al, aMHC promoter-driven EGFP transgenic mice (aMHC-GFP), in with the numbers peaking at day 7, and remaining higher even at ing iCMs. MiR-133-mediated Snai1 repression was also critical for 2010; Protze et al, 2012; Song et al, 2012). Recently, we and others which no cardiomyocytes or cardiac progenitor cells (CPCs) were 4 weeks after transduction (Fig 2A, Supplementary Fig S2A). The cardiac reprogramming in adult mouse and human cardiac fibro- reported that iCMs can be directly generated from human fibroblasts detected by immunofluorescence, fluorescence-activated cell sorting iCMs were less proliferative than non-converted fibroblasts and blasts. Thus, silencing fibroblast signatures, mediated by miR-133/ by overexpression of GMT plus Mesp1 and Myocd (GMTMM) or (FACS), and quantitative RT-PCR (qRT-PCR) analyses (Supplemen- decreased in percentage relative to the total number of cells over Snai1, is a key molecular roadblock during cardiac reprogramming. other combinations of reprogramming factors (Fu et al, 2013; Nam tary Fig S1A–E; Ieda et al, 2010). The transfection efficiency by time in culture. qRT-PCR demonstrated that the expression of et al, 2013; Wada et al, 2013). However, induction of functional miRNA mimics was 97%, but none of the miRNA mimics induced cardiac genes, Actn2 (sarcomeric a-actinin), Myh6 (a-myosin heavy Keywords cardiomyocyte; microRNA; reprogramming; Snai1; transcription cardiomyocytes in vitro was inefficient and slow, possibly hindering aMHC-GFP or cardiac troponin T (cTnT) expression in MEFs when chain), Ryr2 (ryanodine receptor 2), and Tnni3 (cardiac troponin I), factor our investigations of the molecular events during cardiac repro- used individually or as a pool (four miRs) after 1 week of transfec- was upregulated, while the expression of fibroblast genes, Subject Categories Development & Differentiation; Stem Cells gramming (Chen et al, 2012; Srivastava & Ieda, 2012; Addis & tion (Fig 1A and B, Supplementary Fig S1F and G; Jayawardena Col1a1 (collagen 1a1) and Fn1 (fibronectin 1), was significantly DOI 10.15252/embj.201387605 | Received 16 December 2013 | Revised 18 April Epstein, 2013). We and others also showed that endogenous mouse et al, 2012). In contrast, GMT transduction induced aMHC-GFP and downregulated from day 3 in the FACS-sorted a-MHC-GFP+ cells 2014 | Accepted 5 May 2014 | Published online 11 June 2014 cardiac fibroblasts (CFs) can be converted into iCMs in vivo by gene The EMBO Journal (2014) 33: 1565–1581 transfer of GMT or GHMT (Inagawa et al, 2012; Qian et al, 2012;

Figure 1. MiR-133 promotes Gata4/Mef2c/Tbx5-induced cardiac reprogramming. A, B FACS analyses for aMHC-GFP+ and cTnT+ cells 1 week after GMT transduction or miRNA transfection. Quantitative data are shown in (B) (n = 3). 1 Department of Clinical and Molecular Cardiovascular Research, Keio University School of Medicine, Shinjuku-ku, Tokyo, Japan + + ▸ C, D FACS analyses for aMHC-GFP and cTnT cells 1 week after GMT and miRNA transduction. Quantitative data are shown in (D) (n = 3). 2 Department of Cardiology, Keio University School of Medicine, Shinjuku-ku, Tokyo, Japan E, F Dose dependency of miR-133-mediated cardiac induction with GMT. Quantitative data are shown in (F) (n = 3). 3 Department of Physiology and Cell Biology, Tokyo Medical and Dental University, Bunkyo-ku, Tokyo, Japan + G FACS analyses for a-actinin cells 1 week after transduction. 4 Japan Biological Informatics Consortium (JBiC), Koto-ku, Tokyo, Japan H Immunocytochemisty for aMHC-GFP, a-actinin, and DAPI. GMT/miR-133 induced abundant aMHC-GFP and a-actinin expression 2 weeks after transduction. High- 5 Molecular Profiling Research Center for Drug Discovery, National Institute of Advanced Industrial Science and Technology, Koto-ku, Tokyo, Japan 6 Division of Cardiovascular Surgery, Keio University School of Medicine, Shinjuku-ku, Tokyo, Japan magnification views in insets show sarcomeric organization. GMT/miR-133 induced cTnT and ANP expression 2 weeks after transduction. Insets are high- 7 Department of Pediatrics, Keio University School of Medicine, Shinjuku-ku, Tokyo, Japan magnification views. 8 JST, CREST, Shinjuku-ku, Tokyo, Japan Data information: All data are presented as means SEM. *P < 0.05, **P < 0.01 versus relevant control. Scale bars represent 100 lm. *Corresponding author. Tel: +81 3 5843 6702; Fax: +81 3 5363 3875; E-mail: [email protected] �

ª 2014 The Authors The EMBO Journal Vol 33 | No 14 | 2014 1565 1566 The EMBO Journal Vol 33 | No 14 | 2014 ª 2014 The Authors The EMBO Journal MiR-133 promotes reprogramming by Snai1 repression Naoto Muraoka et al Article

ª 2014 The Authors The EMBO Journal Vol 33 | No 14 | 2014 1565 1566 The EMBO Journal Vol 33 | No 14 | 2014 ª 2014 The Authors Naoto Muraoka et al MiR-133 promotes reprogramming by Snai1 repression The EMBO Journal The EMBO Journal MiR-133 promotes reprogramming by Snai1 repression Naoto Muraoka et al

induced with GMT/miR-133 (Fig 2B). Non-sorted samples also structures after 7 days of infection, which normally takes AB revealed comparable results (Supplementary Fig S2B). Immunocyto- 2 weeks with GMT alone (Fig 2C). qRT-PCR and immunostaining chemistry demonstrated that the GMT/miR-133-iCMs showed sarcomeric for the genes specific to nodal, atrial, and ventricular myocytes

A B

C D

D E H

E F G

H IJ

revealed that most iCMs were atrial-type myocytes with either trans- sixfold more cells exhibited Ca2+ flux with GMT/miR-133 induction duction (Supplementary Fig S2C and D). Functionally, a subset of than with GMT alone (Fig 2D and E, Supplementary Movie S1). MEF-derived iCMs showed spontaneous Ca2+ oscillations, and Notably, cell contraction started from 10 days after GMT/miR-133

ª 2014 The Authors The EMBO Journal Vol 33 | No 14 | 2014 1567 1568 The EMBO Journal Vol 33 | No 14 | 2014 ª 2014 The Authors Naoto Muraoka et al MiR-133 promotes reprogramming by Snai1 repression The EMBO Journal The EMBO Journal MiR-133 promotes reprogramming by Snai1 repression Naoto Muraoka et al

A B

C D

D E H

E F G

H IJ

transduction, which generally took 4 weeks with GMT. The number were significantly enriched for GO terms associated with cardiac of beating cells increased over time, with sevenfold more contractile function and development, while the downregulated genes were cells achieved compared to using GMT alone (Fig 2F and G, Supple- significantly enriched for the GO terms associated with fibroblast mentary Fig S2E, and Supplementary Movies S2 and S3). We did not signatures, such as cell adhesion, cell proliferation, and collagen observe any beating cells in untreated MEF cultures, excluding the fibril organization (Fig 3C and D). Scatter plot analyses at day 3 of possibility of cardiomyocyte contamination. EdU incorporation transduction demonstrated that the upregulated genes induced by assays revealed that the increase in beating iCMs with GMT/miR-133 GMT/miR-133 were significantly enriched in heart compared to CD transduction was not due to cell proliferation (Supplementary Fig MEFs (P = 7.7E-21), while the downregulated genes in GMT/miR- S2F and G). These results suggest that miR-133 promoted the speed 133-iCMs were highly enriched in MEFs (P = 1.2E-34), indicating and efficiency of cardiac reprogramming in combination with GMT. that miR-133 induced cardiac gene programs and extinguished fibro- Mesp1-GFP mice, in which the progeny of multipotent CPCs can blast signatures from the early stages of reprogramming (Fig 3E). be traced by fluorescence, were used to determine the route of Microarray and qRT-PCR analyses revealed that cardiac genes cardiac reprogramming (Saga et al, 1999; Kawamoto et al, 2000). related to different functions, such as sarcomere structures (Actn2, We isolated Mesp1-GFP–/Thy1+ MEFs by FACS and transduced the Myh6, Tnni3), mitochondrial metabolism (Ppargc1a), and ion chan- cells with GMT/miR-133 (Fig 2H). The resulting cTnT+ cells did not nels (Ryr2, Slc8a1, Kcnd2, and Scn5a), were upregulated, while express GFP, suggesting that the iCMs were generated from MEFs fibroblast genes, Snai1, Col1a1, Col1a2, Fn1, and Postn, were without passing through a mitotic Mesp1+-CPC state (Fig 2I). More- downregulated in the GMT/miR-133-iCMs compared to the over, a later CPC marker, isl1, was not induced by GMT/miR-133 GMT-iCMs at day 7 (Fig 3F and G). Epithelial genes, such as cdh1 during reprogramming (Fig 2J). These results indicated that the (E-cadherin), Dsp, Pkp1, Ctnnb1, F2r, and Ocln, were not upregulated, iCMs were directly generated from fibroblasts by GMT/miR-133. suggesting miR-133 did not induce mesenchymal-to-epithelial transition (MET) process. Thus, miR-133 silenced fibroblast signatures in E F MiR-133 suppresses fibroblast signatures in concert with cardiac parallel with cardiac gene activation from the early stages of repro- gene activation gramming.

state toward a cardiac-like phenotype by GMT or GMT/miR-133 binding sites within the 30UTR (Fig 4B). Snai1 is a basic helix-loop- transduction at all stages (Fig 3A). Next, to identify the genes that helix transcription factor, known as a master regulator of EMT, and were regulated by miR-133, we focused on the genes that were induces mesenchymal programs and fibrogenesis during develop- differentially expressed between GMT and GMT/miR-133 transduc- ment and disease (Rowe et al, 2009; Li et al, 2010). In luciferase

tion at all stages. Among 23,474 probes, 46 genes were upregulated, reporter assays with a construct containing a full-length Snai1 30UTR and 129 genes were downregulated by at least 1.5-fold by GMT/ sequence, miR-133 transfection strongly repressed the luciferase miR-133 induction, consistent with the function of miRNAs, typi- activity by 60%. Mutations of either predicted miR-133-binding site miR-133 expression was induced by GMT transduction in MEFs and addition of miR-133 to GMT, consistent with the array data (Figs 4E

cally diminishing the expression of their mRNA targets (Fig 3B, in the Snai1 30UTR reduced the responsiveness to miR-133, which further upregulated by miR-133 overexpression (Fig 4D). Inversely, and 3F). Western blot analyses also demonstrated that Snai1 protein Supplementary Table S1). Gene ontology (GO) analyses was almost absent with mutations of both sites, suggesting direct the mRNA expression of Snai1 was high in MEFs, significantly expression was strongly downregulated in MEFs by transduction demonstrated that upregulated genes in the GMT/miR-133-iCMs binding of miR-133 to both sites (Fig 4C). qRT-PCR confirmed that downregulated by GMT and further reduced by 60% with the with miR-133 alone or GMT/miR-133 (Fig 4F). These results

ª 2014 The Authors The EMBO Journal Vol 33 | No 14 | 2014 1569 1570 The EMBO Journal Vol 33 | No 14 | 2014 ª 2014 The Authors Naoto Muraoka et al MiR-133 promotes reprogramming by Snai1 repression The EMBO Journal The EMBO Journal MiR-133 promotes reprogramming by Snai1 repression Naoto Muraoka et al

AB Figure 2. MiR-133 enhances generation of functional iCMs. A Time course of aMHC-GFP and cTnT expression by GMT or GMT/miR-133 transduction in MEFs. See also Supplementary Fig S2A. ◂ + B qRT-PCR for cardiac and fibroblast gene expression in aMHC-GFP cells by GMT or GMT/miR-133 transduction (n = 4). Data were normalized against the values of GMT-iCMs at day 7 (Actn2, Myh6, Ryr2, Tnni3) or MEFs at day 0 (Col1a1, Fn1). See also Supplementary Fig S2B. C GMT/miR-133 induced expression of a-actinin with sarcomeric organization 1 week after transduction. D, E Spontaneous Ca2+ oscillations observed in MEF-derived iCMs (arrows) after 4 weeks of induction, corresponding to Supplementary Movie S1. Rhod-3 images at Ca2+ max and min are shown in the upper panels and Rhod-3 intensity trace is shown in the lower panel (D). Total number of Ca2+ oscillation+ cells in 10 randomly selected fields per well is shown in (E) (n = 3). F Spontaneously beating GMT/miR-133 iCMs 4 weeks after transduction (arrows), corresponding to Supplementary Movie S2. See also Supplementary Fig S2E and Movie S3. G Number of spontaneously beating cells in each well after transduction of mock, GMT, or GMT/miR-133 at the indicated time. + + H, I Mesp1-GFP�/Thy1 MEFs were sorted and transduced with GMT or GMT/miR-133 (H). All cTnT cells were negative for Mesp1-GFP (I). J qRT-PCR for isl1 expression in the cells transduced with GMT or GMT/miR-133 (n = 3). Data information: All data are presented as means SEM. *P < 0.05, **P < 0.01 versus relevant control. Scale bars: 100 lm (C, F); 5 s (D). � transduction, which generally took 4 weeks with GMT. The number were significantly enriched for GO terms associated with cardiac of beating cells increased over time, with sevenfold more contractile function and development, while the downregulated genes were cells achieved compared to using GMT alone (Fig 2F and G, Supple- significantly enriched for the GO terms associated with fibroblast mentary Fig S2E, and Supplementary Movies S2 and S3). We did not signatures, such as cell adhesion, cell proliferation, and collagen observe any beating cells in untreated MEF cultures, excluding the fibril organization (Fig 3C and D). Scatter plot analyses at day 3 of possibility of cardiomyocyte contamination. EdU incorporation transduction demonstrated that the upregulated genes induced by assays revealed that the increase in beating iCMs with GMT/miR-133 GMT/miR-133 were significantly enriched in heart compared to CD transduction was not due to cell proliferation (Supplementary Fig MEFs (P = 7.7E-21), while the downregulated genes in GMT/miR- S2F and G). These results suggest that miR-133 promoted the speed 133-iCMs were highly enriched in MEFs (P = 1.2E-34), indicating and efficiency of cardiac reprogramming in combination with GMT. that miR-133 induced cardiac gene programs and extinguished fibro- Mesp1-GFP mice, in which the progeny of multipotent CPCs can blast signatures from the early stages of reprogramming (Fig 3E). be traced by fluorescence, were used to determine the route of Microarray and qRT-PCR analyses revealed that cardiac genes cardiac reprogramming (Saga et al, 1999; Kawamoto et al, 2000). related to different functions, such as sarcomere structures (Actn2, We isolated Mesp1-GFP–/Thy1+ MEFs by FACS and transduced the Myh6, Tnni3), mitochondrial metabolism (Ppargc1a), and ion chan- cells with GMT/miR-133 (Fig 2H). The resulting cTnT+ cells did not nels (Ryr2, Slc8a1, Kcnd2, and Scn5a), were upregulated, while express GFP, suggesting that the iCMs were generated from MEFs fibroblast genes, Snai1, Col1a1, Col1a2, Fn1, and Postn, were without passing through a mitotic Mesp1+-CPC state (Fig 2I). More- downregulated in the GMT/miR-133-iCMs compared to the over, a later CPC marker, isl1, was not induced by GMT/miR-133 GMT-iCMs at day 7 (Fig 3F and G). Epithelial genes, such as cdh1 during reprogramming (Fig 2J). These results indicated that the (E-cadherin), Dsp, Pkp1, Ctnnb1, F2r, and Ocln, were not upregulated, iCMs were directly generated from fibroblasts by GMT/miR-133. suggesting miR-133 did not induce mesenchymal-to-epithelial transition (MET) process. Thus, miR-133 silenced fibroblast signatures in E F MiR-133 suppresses fibroblast signatures in concert with cardiac parallel with cardiac gene activation from the early stages of repro- gene activation gramming.

Next, to investigate the mechanistic basis of miR-133-mediated MiR-133 directly represses Snai1 expression during cardiac reprogramming, we analyzed the global gene expression cardiac reprogramming profiles of iCMs induced with GMT or GMT/miR-133 by microarray. We FACS-sorted a-MHC-GFP+ cells at 3, 7, and 18 days after trans- We then searched for potential direct mRNA targets of miR-133 duction, before and after the GMT/miR-133-transduced cells started during cardiac reprogramming. Expression of Ccnd2, Cdc42, Hand2, contractions, and compared the differential gene expressions RhoA, and Srf, shown previously as the direct targets of miR-133, between MEFs and hearts. Although the vast majority of a-MHC- was not significantly altered in GMT-miR-133-iCMs compared to GFP+ cells were only partially reprogrammed iCMs, with few capa- GMT-iCMs, as shown by microarray (Fig 4A; Liu & Olson, 2010). G ble of beating, the heatmap image of microarray data revealed a Using the miRNA target prediction program, we identified Snai1 as shift in the global gene expression patterns of the iCMs from a MEF a putative direct target of miR-133 with two conserved miR-133- state toward a cardiac-like phenotype by GMT or GMT/miR-133 binding sites within the 30UTR (Fig 4B). Snai1 is a basic helix-loop- transduction at all stages (Fig 3A). Next, to identify the genes that helix transcription factor, known as a master regulator of EMT, and were regulated by miR-133, we focused on the genes that were induces mesenchymal programs and fibrogenesis during develop- differentially expressed between GMT and GMT/miR-133 transduc- ment and disease (Rowe et al, 2009; Li et al, 2010). In luciferase tion at all stages. Among 23,474 probes, 46 genes were upregulated, reporter assays with a construct containing a full-length Snai1 30UTR and 129 genes were downregulated by at least 1.5-fold by GMT/ sequence, miR-133 transfection strongly repressed the luciferase miR-133 induction, consistent with the function of miRNAs, typi- activity by 60%. Mutations of either predicted miR-133-binding site miR-133 expression was induced by GMT transduction in MEFs and addition of miR-133 to GMT, consistent with the array data (Figs 4E cally diminishing the expression of their mRNA targets (Fig 3B, in the Snai1 30UTR reduced the responsiveness to miR-133, which further upregulated by miR-133 overexpression (Fig 4D). Inversely, and 3F). Western blot analyses also demonstrated that Snai1 protein Supplementary Table S1). Gene ontology (GO) analyses was almost absent with mutations of both sites, suggesting direct the mRNA expression of Snai1 was high in MEFs, significantly expression was strongly downregulated in MEFs by transduction demonstrated that upregulated genes in the GMT/miR-133-iCMs binding of miR-133 to both sites (Fig 4C). qRT-PCR confirmed that downregulated by GMT and further reduced by 60% with the with miR-133 alone or GMT/miR-133 (Fig 4F). These results

Figure 3. MiR-133 silences fibroblast signatures and activates cardiac programs. A B C ◂ A Heat-map image of microarray data illustrating the global gene expression pattern of MEFs, iCMs, and hearts. The iCMs were sorted as aMHC-GFP+ cells after 3 (D3), 7 (D7), and 18 (D18) days of GMT or GMT/miR-133 transduction. Differentially expressed genes between MEFs and hearts are shown (n = 1). B Differentially expressed genes between GMT-iCMs and GMT/miR-iCMs are shown (n = 1). See also Supplementaray Table S1. C, D GO analyses of the upregulated (C) and downregulated (D) genes in GMT/miR-iCMs at all stages. Top 10 GO categories are shown. Cardiac (C) and fibroblast- related (D) GO terms are shown in red. E The upregulated and downregulated genes in GMT/miR-iCMs at day 3 were analyzed by scatter plots. F The relative mRNA expression of cardiomyocyte, fibroblast, and epithelial cell genes in D7 GMT/miR-iCMs compared to D7 GMT-iCMs by microarray. G The relative mRNA expression of D7 GMT/miR-iCMs compared to D7 GMT-iCMs was determined by qRT-PCR (n = 3). Data information: Data were normalized by the values of GMT-iCMs. All data are presented as means SEM (G). *P < 0.05, **P < 0.01 versus relevant control. �

D EFG suggested that miR-133 directly targets Snai1, resulting in reduced and F). These results suggest that Snai1 repression is critical for expression of this protein during reprogramming. silencing fibroblast programs and activating the cardiac phenotype Next, to investigate the possible contribution of Snai1 during in MEFs induced with GMT/miR-133. Snai1 overexpression also cardiac reprogramming, we suppressed Snai1 expression with inhibited both the induction of a-MHC-GFP and cTnT expression in siRNA in GMT-transduced MEFs (Fig 4G). qRT-PCR at day 7 of GMT/miR-133-transduced fibroblasts, as shown by FACS analyses transduction demonstrated that knockdown of Snai1 in the presence (Fig 5D and E), and the expression of endogenous cardiac proteins, of GMT strongly downregulated expression of multiple fibroblast a-actinin, cTnT, and ANP, in GMT/miR-133-transduced cells at genes, including Fn1, Postn, and Snai2, to levels comparable with 4 weeks, as revealed by immunocytochemistry (Fig 5F–H, Supple- those affected by GMT/miR-133 (Fig 4H). Intriguingly, inhibition of mentary Fig S3G). Consistent with this, constitutive expression of H Snai1 concomitantly upregulated a panel of cardiac genes related to Snai1 strongly suppressed spontaneous Ca2+ oscillations in GMT/ different functions, such as sarcomere structures (Actn2, Ttn), gap miR-133-transduced cells and inhibited generation of beating iCMs junctions (Gja1), hormones (Nppa), and ion channels (Ryr2 and at 4 weeks, overriding the positive effects of miR-133 (Fig 5I). These Kcnd2), in GMT-transduced cells (Fig 4H, Supplementary Fig S3A). results suggested that downregulation of Snai1 is critical for FACS analyses demonstrated that knockdown of Snai1 significantly suppressing fibroblast profiles and cardiac reprogramming in MEFs increased the induction of a-MHC-GFP+ cells and cTnT+ cells from induced with GMT/miR-133. MEFs in combination with GMT, but not with GMT/miR-133 (Fig 4I I J and J, Supplementary Fig S3B). Immunostaining for cardiac mark- MiR-133-induced Snai1 suppression is critical for cardiac ers, including a-actinin, cTnT, and ANP, demonstrated that Snai1 reprogramming in adult mouse cardiac fibroblasts suppression increased cardiac protein expression in combination with GMT after 4 weeks (Fig 4K–M, Supplementary Fig S3C). Snai1 We next investigated whether the miR-133-mediated suppression of suppression significantly increased spontaneous Ca2+ oscillations Snai1 also plays critical roles in cardiac reprogramming in adult and cell contractions in GMT-transduced cells, although not to the mouse CFs. Similar to the MEF cultures, we did not detect contami- levels seen with miR-133 overexpression (Fig 4N, Supplementary nation of cardiomyocytes in adult CF cultures derived from a-MHC- Fig S3D, Supplementary Movie S4). These results suggested that GFP mice. Transfection of miR-133 alone did not induce cardiac suppression of Snai1 reduced fibroblast profiles and concomitantly reprogramming in adult CFs either with or without the JAK inhibitor K L promoted cardiac induction in GMT-transduced fibroblasts, which (Supplementary Fig S3H). We then introduced miR-133 with GMT recapitulated the effects of miR-133 overexpression. and found significantly increased induction of a-MHC-GFP+ and cTnT+ cells from adult CFs by FACS at 1 week (Fig 6A and B). qRT- Overexpression of Snai1 maintains fibroblast signatures and PCR demonstrated that miR-133 upregulated multiple cardiac genes disrupts cardiac reprogramming in and concomitantly downregulated fibroblast gene expression in GMT/miR-133-transduced MEFs adult CFs in combination with GMT (Fig 6C). Immunocytochemistry and Ca2+ imaging also showed that addition of miR-133 to GMT We next asked whether suppression of Snai1 is a consequence of or promoted cardiac induction from adult CFs (Fig 6D–G, Supplemen- is required for cardiac reprogramming induced with GMT/miR-133. tary Movie S5). These results suggested that miR-133 promotes To address this, we restored Snai1 expression in GMT/miR-133- cardiac reprogramming in adult CFs, but with lower reprogramming transduced MEFs by overexpression of Snai1 cDNA without the efficiency compared to that in MEFs. MN 30UTR and investigated whether Snai1 restoration could counteract Next, we analyzed the link between miR-133 and Snai1 during the effects of miR-133-mediated cardiac reprogramming (Fig 5A). reprogramming in adult CFs. The mRNA expression of Snai1 was Microarray analyses revealed that among 46 genes upregulated by high in adult CFs, downregulated by GMT, and further reduced by GMT/miR-133 (Fig 3B), 39 were suppressed by Snai1 overexpres- 50% with the addition of miR-133 to GMT, which was inversely sion. In contrast, 105 out of 129 genes downregulated by miR-133 correlated with the expression of miR-133 (Fig 6H). qRT-PCR addition (Fig 3B) were upregulated by Snai1 restoration, suggesting demonstrated that knockdown of Snai1 upregulated a panel of most portions of the transcriptional changes effected by miR-133 cardiac genes related to multiple functions and repressed fibroblast were mediated via Snai1 suppression (Fig 5B). The genes upregulated genes in GMT-transduced adult CFs (Fig 6I). Immunostaining and downregulated by Snai1 overexpression were identified as revealed that inhibition of Snai1 increased the induction of a-MHC- fibroblast- and cardiac-related genes, respectively, and qRT-PCR GFP+ cells and a-actinin+ cells from adult CFs in combination with analyses confirmed the array data (Fig 5C, Supplementary Fig S3E GMT (Fig 6J and K). These results suggested that suppressing Snai1

ª 2014 The Authors The EMBO Journal Vol 33 | No 14 | 2014 1571 1572 The EMBO Journal Vol 33 | No 14 | 2014 ª 2014 The Authors Naoto Muraoka et al MiR-133 promotes reprogramming by Snai1 repression The EMBO Journal The EMBO Journal MiR-133 promotes reprogramming by Snai1 repression Naoto Muraoka et al

could in turn suppress fibroblast programs and promote cardiac qRT-PCR, FACS analyses, and immunocytochemistry revealed that induction in adult CFs. We then also overexpressed the Snai1 cDNA Snai1 overexpression inhibited GMTMM/miR-133-mediated cardiac DE without 30UTR in GMT/miR-133-transduced adult CFs and found induction (Fig 7A, B, H, I and J, Supplementary Fig S4D and E), reduced a-MHC-GFP+ and cTnT+ cells, as measured by FACS whereas Snail knockdown increased the induction of cardiac genes (Fig 6A and B). qRT-PCR revealed downregulated expression of in combination with GMTMM as measured by qRT-PCR (Supple- cardiac genes and upregulated expression of multiple fibroblast mentary Fig S4F). FACS analyses and immunostaining also demon- genes by Snai1 overexpression, suggesting that Snai1 repression is strated that knockdown of Snai1 significantly increased expression critical for silencing fibroblast signatures and inducing cardiac of a-actinin+ cells in HCFs in combination with GMTMM, suggest- reprogramming in adult CFs (Fig 6C). Immunocytochemistry and ing that suppressing Snai1 expression promotes cardiac induction functional studies also revealed that Snai1 overexpression strongly (Supplementary Fig S4G–I). Taken together, these results suggest inhibited cardiac reprogramming in GMT/miR-133-transduced adult that miR-133-mediated Snai1 suppression is also crucial for human CFs (Fig 6D–F). cardiac reprogramming.

FG MiR-133-mediated Snai1 repression is also critical in human cardiac reprogramming Discussion

We next tested whether miR-133-mediated Snai1 suppression also Direct reprogramming is characterized as a process that progres- plays important roles in human cardiac reprogramming using HCFs. sively activates the target cell program and concomitantly Transfection of miR-133 alone or 4miRs with or without JAKI-1 did suppresses the starting-cell profile without passing through a stem not induce cardiac reprogramming (Supplementary Fig S4A). In cell state by overexpression of lineage-specific transcription factors contrast, using new lentiviral vectors to transduce the genes effi- or miRNAs. In this study, we demonstrated that overexpression of ciently into HCFs (Supplementary Fig S4B), we demonstrated that miR-133 in the presence of these transcription factors promoted transduction of lentiviral GMTMM induced cTnT+ and a-actinin+ cardiac reprogramming in mouse and human, partly by suppressing cells in 2–8% of HCFs (Fig 7A and B, Supplementary Fig S4C). The Snai1, a master regulator of EMT, and silencing the fibroblast induction rate increased to 23–27% of HCFs with addition of program. miR-133 to GMTMM (Fig 7A and B, Supplementary Fig S4C). Micro- The miR-133 family comprises three miRNAs: miR-133a-1, miR- array analyses further revealed that GMTMM or GMTMM/miR-133 133a-2, and miR-133b. MiR-133a-1 and miR-133a-2 have identical transduction upregulated 1,270 cardiac-enriched genes and down- sequences and are expressed in cardiac and skeletal muscles. regulated 1,111 fibroblast-related genes in HCFs, suggesting global Indeed, mice lacking both miR-133a-1 and miR-133a-2 undergo H I transcriptional changes toward cardiac fate induced by the repro- embryonic or neonatal death due to heart defects, suggesting critical gramming factors (Fig 7C). Differential gene expression analyses roles of miR-133a in cardiogenesis (Liu et al, 2008). Consistent with between GMTMM-HCFs and GMTMM/miR-133-HCFs showed that this, addition of miR-133a to cardiac reprogramming factors addition of miR-133 upregulated 399 genes and downregulated 264 increased cardiac reporter and gene expression, shifted the global genes (Fig 7D). GO term analyses associated the upregulated genes gene expression profile of the iCMs toward a cardiac fate, and gener- with cardiac function and the downregulated genes with fibroblast ated more functional iCMs in direct reprogramming. signatures, suggesting that miR-133 promotes cardiac reprogram- Our findings support that combining lineage-specific transcrip- ming in HCFs (Fig 7E). Snai1 mRNA expression was suppressed by tion factors and miRNAs provides a powerful approach to improve GMTMM and further reduced by 40% with GMTMM/miR-133 in direct conversion of fibroblasts into another lineage. Yoo et al human cardiac reprogramming (Fig 7F). Microarray analyses (2011) reported that a combination of neural-specific miRNAs, miR- demonstrated that the transcriptional changes effected by miR-133 9/9*, and miR-124, and neurogenic transcription factors directly cardiac reprogramming in human fibroblasts (Nam et al, 2013). mechanisms and importance of suppressing fibroblast signatures were largely counteracted by Snai1 overexpression, suggesting the reprogrammed human fibroblasts into functional neuronal cells However, the mechanistic basis underlying direct reprogramming during direct reprogramming remained largely unknown. miR-133-induced cardiac reprogramming was mainly mediated (Yoo et al, 2011). Recently, Nam et al (2013) reported that addition by the transcription factor/miRNA combinations was not estab- In the present study, we found that miR-133 suppressed a large through Snai1 suppression (Fig 7D and G). Consistent with this, of miR-1 and miR-133 to Gata4, Hand2, Myocd, and Tbx5 promoted lished in these previous studies. Moreover, the molecular set of fibroblast genes and concomitantly activated cardiac gene

ª 2014 The Authors The EMBO Journal Vol 33 | No 14 | 2014 1573 1574 The EMBO Journal Vol 33 | No 14 | 2014 ª 2014 The Authors Naoto Muraoka et al MiR-133 promotes reprogramming by Snai1 repression The EMBO Journal The EMBO Journal MiR-133 promotes reprogramming by Snai1 repression Naoto Muraoka et al

Figure 4. Repression of Snai1 silences fibroblast profile and promotes cardiac reprogramming. A B A The relative mRNA expression of potential direct targets of miR-133 in D7 GMT/miR-iCMs compared to D7 GMT-iCMs by microarray. ◂ B Snai130UTR contains two predicted miR-133a binding sites. Both are conserved among species, shown in red. C MiR-133a directly repressed WT Snai130UTR in luciferase assay, and the repression was abolished when both of binding sites were mutated (n = 3). D Relative miR-133a expression in MEFs, GMT-iCMs, GMT/miR133-iCMs, and hearts (n = 3). E Relative mRNA expression of Snai1 in MEFs, GMT-iCMs, GMT/miR133-iCMs, and hearts (n = 3). F Western blot analyses for Snai1 expression in MEFs, MEFs transfected with miR-133 alone, and MEFs transduced with GMT and GMT/miR-133. G Relative mRNA expression of Snai1 in MEFs and MEFs transfected with siRNA against Snai1 (5, 15, 100 nM) (n = 3). H Relative mRNA expression of cardiac (Actn2, Ttn, Gja1, Nppa) and fibroblast genes (Fn1, Postn, Snai2) in MEFs transduced with GMT, GMT/si-Snai1, or GMT/miR-133 (n = 3). See also Supplementary Fig S3A. I, J FACS analyses for aMHC-GFP+ and cTnT+ cells 1 week after GMT transduction with si-Snai1 or miR-133 transfection. Quantitative data are shown in (J) (n = 3). K–M Immunocytochemisty for aMHC-GFP, a-actinin, cTnT, and DAPI. Snai1 suppression increased cardiac protein expression in GMT-transduced cells (M, n = 5). See also Supplementary Fig S3C. C N Total number of Ca2+ oscillation+ cells in 10 randomly selected fields per well is shown (the left panel, n = 8). Spontaneously beating cells were counted in each well after 4 weeks of infection (the right panel, n = 3). See also Supplementary Fig S3D and Movie S4. Data information: All data are presented as means SEM. *P < 0.05, **P < 0.01 versus relevant control. Scale bars, 100 lm. � could in turn suppress fibroblast programs and promote cardiac qRT-PCR, FACS analyses, and immunocytochemistry revealed that induction in adult CFs. We then also overexpressed the Snai1 cDNA Snai1 overexpression inhibited GMTMM/miR-133-mediated cardiac DE without 30UTR in GMT/miR-133-transduced adult CFs and found induction (Fig 7A, B, H, I and J, Supplementary Fig S4D and E), reduced a-MHC-GFP+ and cTnT+ cells, as measured by FACS whereas Snail knockdown increased the induction of cardiac genes (Fig 6A and B). qRT-PCR revealed downregulated expression of in combination with GMTMM as measured by qRT-PCR (Supple- cardiac genes and upregulated expression of multiple fibroblast mentary Fig S4F). FACS analyses and immunostaining also demon- genes by Snai1 overexpression, suggesting that Snai1 repression is strated that knockdown of Snai1 significantly increased expression critical for silencing fibroblast signatures and inducing cardiac of a-actinin+ cells in HCFs in combination with GMTMM, suggest- reprogramming in adult CFs (Fig 6C). Immunocytochemistry and ing that suppressing Snai1 expression promotes cardiac induction functional studies also revealed that Snai1 overexpression strongly (Supplementary Fig S4G–I). Taken together, these results suggest inhibited cardiac reprogramming in GMT/miR-133-transduced adult that miR-133-mediated Snai1 suppression is also crucial for human CFs (Fig 6D–F). cardiac reprogramming.

FG MiR-133-mediated Snai1 repression is also critical in human cardiac reprogramming Discussion

programs during cardiac reprogramming when used in combination damaged hearts (Qian et al, 2012; Song et al, 2012). Further in vitro with GMT or GMTMM. Among many predicted targets of miR-133, and in vivo studies are thus needed to progress our understanding we identified Snai1 as a novel direct target of the miRNA and of molecular mechanisms underlying cardiac reprogramming and demonstrated that Snai1 repression silences fibroblast programs and apply this new technology to future regenerative therapies. promotes cardiac reprogramming, recapitulating the effects of D EF miR-133 overexpression. In contrast, overexpression of Snai1 inhibited suppression of fibroblast genes and activation of cardiac Materials and Methods programming induced with GMT/miR-133. Thus, silencing fibroblast signatures, mediated by miR-133/Snai1, could be a key Generation of aMHC-GFP and Mesp1-GFP mice molecular roadblock during direct cardiac reprogramming. Intrigu- ingly, this process is similar to the MET, a critical step during the Transgenic mice overexpressing GFP under the control of an a-MHC reprogramming of fibroblasts into iPSCs by the Yamanaka factors promoter were generated as described previously (Ieda et al, 2010). (Li et al, 2010; Samavarchi-Tehrani et al, 2010). During iPSC gener- Mesp1-GFP mice were obtained by crossing Mesp1-Cre mice and ation, Oct4 and Sox2 suppress Snai1, while Klf4 induces epithelial CAG-CAT-EGFP reporter mice (Saga et al, 1999; Kawamoto et al, genes including E-cadherin (Li et al, 2010). Snai1 downregulation 2000). The Keio University Ethics Committee for Animal Experi- and E-cadherin upregulation cooperatively suppress EMT signals ments approved all experiments in this study. and activate an epithelial program, leading to MET and iPSC generation. Similar to our findings, blocking MET by overexpression of Quantitative RT–PCR Snai1 or addition of a Snai1 activator (transforming growth factor b) impaired iPSC generation, whereas inducing MET by addition of Total RNA was isolated from cells, and qRT-PCR was performed on a G H transforming growth factor b inhibitors promoted iPSC induction (Li StepOnePlusTM (Applied Biosystems) with TaqMan probes (Applied et al, 2010; Samavarchi-Tehrani et al, 2010). Thus, it is conceivable Biosystems). TaqMan probes: Actc1 (Mm01333821_m1, Hs00606316_ that silencing fibroblast signatures by repressing Snai1 might be a m1), Myh6 (Mm00440354_m1, Hs00411908_m1), Ryr2 (Mm00465877_ common pathway in the early phase of reprogramming from fibro- m1, Hs00892842_m1), Gja1 (Mm00439105_m1), Nppa (Mm01255747_g1, blasts to another lineage and that manipulation of this pathway Hs00383230_g1), Actn2 (Mm00473657_m1, Hs00153809_m1), Kcnd2 could be a new target to enhance direct reprogramming in general. (Mm0116732_m1), Slc8a1 (Mm01232254_m1, Hs01062258_m1), Scn5a To our knowledge, this is the first study demonstrating a molecular (Mm00451971_m1), Tnni3 (Mm00437164_m1), Tnnt2 (Hs00165960_m1), I mechanism of direct cardiac reprogramming. Ttn (Mm00621005_m1, Hs00399225_m1), Myl2 (Mm00440384_m1, While we found that the miR-133/Snai1 pathway is critical for Hs00166405_m1), Myl7 (Mm00491655_m1), Hcn4 (Mm01176086_m1), cardiac reprogramming, the iCM population was heterogeneous and isl1 (Mm00517585_m1), Col1a1 (Mm00801666_g1, Hs00164004_m1), most of the cells remained as partially reprogrammed cardiac cells Col1a2 (Mm00483888_m1), Col3a1 (Mm01254476_m1), Col5a2 in culture (Supplementary Table S2). Moreover, the reprogramming (Mm00483675_m1), Fn1 (Mm01256744_m1, Hs00365052_m1), Snai1 efficiency of adult CFs was low compared with MEFs in this study (Mm00441533_g1, Hs00195591_m1), Snai2 (Mm00441531_m1), Ddr2 and our previous data using neonatal CFs (Ieda et al, 2010). Differ- (Mm0000445615_m1), Postn (Mm00450111_m1, Hs00170815_m1), ences between mouse lines used and transcriptional and epigenetic Hsa-miR-133a (1102119-Q). mRNA levels were normalized by JK differences between fibroblasts might have contributed to the lower comparison to Gapdh (Mm99999915_g1, Hs02758991_g1). miRNA reprogramming efficiency in adult CFs. Because miR-133 has numer- levels were normalized by comparison to Rnu6b (0811824-H) ous predicted targets, and knockdown of Snai1 increased cardiac snRNA. induction, but not to the level observed with miR-133 overexpression, some additional pathways might be involved in miR-133-medi- Molecular cloning, retroviral infection, lentiviral infection, and ated cardiac reprogramming (Liu & Olson, 2010). Nevertheless, miRNA mimics transfection given that the in vivo environment might be more permissive than culture dishes for reprogramming, GMT/miR-133 transduction or To construct the pMXs retroviral vectors, we amplified the coding Snai1 knockdown with GMT in vivo might be sufficient to repair regions of GFP, Gata4, Mef2c, Tbx5, and Snai1 by PCR and

ª 2014 The Authors The EMBO Journal Vol 33 | No 14 | 2014 1575 1576 The EMBO Journal Vol 33 | No 14 | 2014 ª 2014 The Authors Naoto Muraoka et al MiR-133 promotes reprogramming by Snai1 repression The EMBO Journal The EMBO Journal MiR-133 promotes reprogramming by Snai1 repression Naoto Muraoka et al

Figure 5. Overexpression of Snai1 inhibits cardiac reprogramming. A B A Western blot analyses for Snai1 expression in MEFs and MEFs transduced with GMT, GMT/miR-133, and GMT/miR-133 with Snai1 overexpression. ◂ + B Heat-map image of microarray data for GMT-, GMT/miR-, and GMT/miR/Snai1-iCMs sorted as aMHC-GFP cells after 1 week of transduction (left panel, n = 1). Differentially expressed genes between GMT-iCMs and GMT/miR-iCMs are shown (see also Fig 3B). Thirty-nine genes out of 46 upregulated genes were suppressed by Snai1 overexpression, while 105 genes out of 129 downregulated genes were increased with Snai1 transduction (right panel). C The relative mRNA expression of D7 GMT/miR-iCMs compared to D7 GMT-iCMs was determined by qRT-PCR (n = 3). Relative mRNA expression of cardiac (Actn2, Myh6, Ryr2) and fibroblast genes (Col1a1, Fn1, Postn) in MEFs transduced with GMT and GMT/miR-133 with or without Snai1 overexpression (n = 3). D, E FACS analyses for aMHC-GFP + and cTnT+ cells 1 week after GMT and GMT/miR-133 transduction with or without Snai1 overexpression. Quantitative data are shown in (E) (n = 3). F–H Immunocytochemisty for aMHC-GFP, a-actinin, cTnT, and DAPI. Snai1 overexpression suppressed cardiac protein expression in GMT/miR-133-transduced cells (H, n = 5). I Numbers of Ca2+ oscillation+ cells in 10 randomly selected fields per well are shown (left panel, n = 8). Number of spontaneously beating cells in each well after C 4 weeks of infection (right panel, n = 3). Data information: All data are presented as means SEM. *P < 0.05, **P < 0.01 versus relevant control. Scale bars, 100 lm. Æ programs during cardiac reprogramming when used in combination damaged hearts (Qian et al, 2012; Song et al, 2012). Further in vitro with GMT or GMTMM. Among many predicted targets of miR-133, and in vivo studies are thus needed to progress our understanding we identified Snai1 as a novel direct target of the miRNA and of molecular mechanisms underlying cardiac reprogramming and demonstrated that Snai1 repression silences fibroblast programs and apply this new technology to future regenerative therapies. promotes cardiac reprogramming, recapitulating the effects of D EF miR-133 overexpression. In contrast, overexpression of Snai1 inhibited suppression of fibroblast genes and activation of cardiac Materials and Methods programming induced with GMT/miR-133. Thus, silencing fibroblast signatures, mediated by miR-133/Snai1, could be a key Generation of aMHC-GFP and Mesp1-GFP mice molecular roadblock during direct cardiac reprogramming. Intrigu- ingly, this process is similar to the MET, a critical step during the Transgenic mice overexpressing GFP under the control of an a-MHC reprogramming of fibroblasts into iPSCs by the Yamanaka factors promoter were generated as described previously (Ieda et al, 2010). (Li et al, 2010; Samavarchi-Tehrani et al, 2010). During iPSC gener- Mesp1-GFP mice were obtained by crossing Mesp1-Cre mice and ation, Oct4 and Sox2 suppress Snai1, while Klf4 induces epithelial CAG-CAT-EGFP reporter mice (Saga et al, 1999; Kawamoto et al, genes including E-cadherin (Li et al, 2010). Snai1 downregulation 2000). The Keio University Ethics Committee for Animal Experi- and E-cadherin upregulation cooperatively suppress EMT signals ments approved all experiments in this study. and activate an epithelial program, leading to MET and iPSC generation. Similar to our findings, blocking MET by overexpression of Quantitative RT–PCR Snai1 or addition of a Snai1 activator (transforming growth factor b) impaired iPSC generation, whereas inducing MET by addition of Total RNA was isolated from cells, and qRT-PCR was performed on a G H transforming growth factor b inhibitors promoted iPSC induction (Li StepOnePlusTM (Applied Biosystems) with TaqMan probes (Applied et al, 2010; Samavarchi-Tehrani et al, 2010). Thus, it is conceivable Biosystems). TaqMan probes: Actc1 (Mm01333821_m1, Hs00606316_ that silencing fibroblast signatures by repressing Snai1 might be a m1), Myh6 (Mm00440354_m1, Hs00411908_m1), Ryr2 (Mm00465877_ common pathway in the early phase of reprogramming from fibro- m1, Hs00892842_m1), Gja1 (Mm00439105_m1), Nppa (Mm01255747_g1, blasts to another lineage and that manipulation of this pathway Hs00383230_g1), Actn2 (Mm00473657_m1, Hs00153809_m1), Kcnd2 could be a new target to enhance direct reprogramming in general. (Mm0116732_m1), Slc8a1 (Mm01232254_m1, Hs01062258_m1), Scn5a To our knowledge, this is the first study demonstrating a molecular (Mm00451971_m1), Tnni3 (Mm00437164_m1), Tnnt2 (Hs00165960_m1), I mechanism of direct cardiac reprogramming. Ttn (Mm00621005_m1, Hs00399225_m1), Myl2 (Mm00440384_m1, While we found that the miR-133/Snai1 pathway is critical for Hs00166405_m1), Myl7 (Mm00491655_m1), Hcn4 (Mm01176086_m1), cardiac reprogramming, the iCM population was heterogeneous and isl1 (Mm00517585_m1), Col1a1 (Mm00801666_g1, Hs00164004_m1), most of the cells remained as partially reprogrammed cardiac cells Col1a2 (Mm00483888_m1), Col3a1 (Mm01254476_m1), Col5a2 in culture (Supplementary Table S2). Moreover, the reprogramming (Mm00483675_m1), Fn1 (Mm01256744_m1, Hs00365052_m1), Snai1 efficiency of adult CFs was low compared with MEFs in this study (Mm00441533_g1, Hs00195591_m1), Snai2 (Mm00441531_m1), Ddr2 and our previous data using neonatal CFs (Ieda et al, 2010). Differ- (Mm0000445615_m1), Postn (Mm00450111_m1, Hs00170815_m1), ences between mouse lines used and transcriptional and epigenetic Hsa-miR-133a (1102119-Q). mRNA levels were normalized by JK differences between fibroblasts might have contributed to the lower comparison to Gapdh (Mm99999915_g1, Hs02758991_g1). miRNA reprogramming efficiency in adult CFs. Because miR-133 has numer- levels were normalized by comparison to Rnu6b (0811824-H) ous predicted targets, and knockdown of Snai1 increased cardiac snRNA. induction, but not to the level observed with miR-133 overexpression, some additional pathways might be involved in miR-133-medi- Molecular cloning, retroviral infection, lentiviral infection, and ated cardiac reprogramming (Liu & Olson, 2010). Nevertheless, miRNA mimics transfection given that the in vivo environment might be more permissive than culture dishes for reprogramming, GMT/miR-133 transduction or To construct the pMXs retroviral vectors, we amplified the coding Snai1 knockdown with GMT in vivo might be sufficient to repair regions of GFP, Gata4, Mef2c, Tbx5, and Snai1 by PCR and

subcloned them into respective pMXs vectors for transfection into were collected by centrifugation and resuspended in DMEM/10% Plat-E cells using Fugene 6 (Roche) to generate retroviruses (Ieda FBS (Thermo Scientific, SV30014.03) for culturing at 37°C in 5%

et al, 2010). We generated lentiviral vectors by subcloning human CO2. For isolation of mouse adult cardiac fibroblasts, aMHC-GFP TG Gata4, Mef2c, Tbx5, Mesp1, Myocd, and Snai1 (HuPEX, AIST) into adult mouse hearts were minced into small pieces < 1 mm3 in size. the CSII-CMV-RfA plasmid (RIKEN BRC) using the Gateway system The explants were plated on gelatin-coated dishes and cultured for (Invitrogen). To generate the lentiviruses, we transfected the vectors 10–14 days in explant medium (IMDM with L-Glutamate and into HEK293 cells with pCAG-HIVgp and pCMV-VSV-G-RSV-Rev 25 mM HEPES (Gibco, 12440-053)/20% FBS). Migrated fibroblasts plasmids (RIKEN BRC) using Lipofectamine 2000 (Invitrogen). The were harvested and filtered with 40-lm cell strainers (BD) to avoid CSII-CMV-Venus vector was used to determine transduction effi- contamination with tissue fragments. The aMHC-GFP–/Thy1+ CFs ciency. Virus-containing supernatants were collected after 48 h and were FACS sorted and plated at a density of 104/cm2 for the retro- used for transduction. Synthetic mimics of mature miRNAs (Thermo virus transduction. E FG Scientific) and the siRNA pool (Thermo Scientific) were transfected Human atrial tissues were obtained from patients undergoing simultaneously into cells with Lipofectamine 2000 (Invitrogen). cardiac surgery (age 1 month to 80 year; average age, 35 year) with Synthetic mimics of mature miRNAs and siRNA pool: miRIDIAN informed consent in conformation with the guidelines of the Keio microRNA Mouse mmu-miR-1-Mimic (C-310376-07-0020), miRIDI- University Ethics Committee. For isolation of human cardiac fibro- AN microRNA Mouse mmu-miR-133a-Mimic (C-310407-07-0020), blasts, human hearts were minced into small pieces < 1 mm3 in miRIDIAN microRNA Mouse mmu-miR-208a-3p-Mimic (C-310501- size. The explants were plated on gelatin-coated dishes and cultured 05-0020), miRIDIAN microRNA Mouse mmu-miR-499-Mimic (C- for 14 days in the explant medium. Migrated fibroblasts were 310727-01-0020), miRIDIAN microRNA Mimic Negative Control #2 harvested, filtered with 40-lm cell strainers (BD), and plated at a (CN-002000-01-05), siGENOME Mouse Snai1 siRNA - SMARTpool density of 5 × 103/cm2 for the virus transduction. For all experi- (M-062765-00-0020), siGENOME Human Snai1 siRNA - SMARTpool ments, we used fibroblasts of early passage number (P1-3). The (M-010847-00-0020). After 24 h, the medium was replaced with Keio Center for Clinical Research approved all of the human experi- D-MEM (high glucose) with L-Glutamate and Phenol Red (Wako, ments in this study (20100131). 044-29765)/Medium199 with Earle’s Salts, L-Glutamate and 22 g/l Sodium Bicarbonate (Gibco, 11150-059)/10% Hyclone Characterized FACS analyses and sorting FBS (Thermo Scientific, SV30014.03) medium and changed every – 2 3 days. JAK inhibitor I (1 nM, EMD Biosciences) treatment was For GFP expression analyses, cells were harvested from culture dishes H initiated 2 days after transfection and continued daily for 7 days. and analyzed on a FACSCalibur (BD Biosciences) with FlowJo software. For aMHC-GFP/cTnT expression, cells were fixed with 4% PFA Cell culture for 15 min, permealized with saponin, and stained with anti-cTnT and anti-GFP antibodies, followed by secondary antibodies conjugated with For MEF isolation, embryos isolated from 12.5-day pregnant mice Alexa 488 and 647, respectively. For a-actinin or cTnT expression, cells were washed with PBS, and the head and visceral tissues were care- were stained with anti-a-actinin or cTnT antibody, followed by second- fully removed. The remaining parts of the embryos were washed in ary antibody conjugated with Alexa 647. For iCM sorting, cells were fresh PBS, minced using a pair of scissors, transferred into a sorted as aMHC-GFP + cells, and for Mesp1-GFP–/Thy1+ cell sorting, 0.25 mM trypsin/1 mM EDTA solution (3 ml per embryo), and cells were incubated with APC-conjugated anti-Thy1 antibody I J incubated at 37°C for 20 min. An additional 3 ml of trypsin/EDTA (eBioscience) and sorted by FACS Aria. solution was then added, and the mixture was further incubated at 37°C for 20 min. After trypsinization, an equal amount of medium Immunocytochemistry (6 ml of DMEM containing 10% FBS per embryo) was added and pipetted up and down a few times to help tissue dissociation. After Cells were fixed in 4% paraformaldehyde for 15 min at room incubation of the tissue/medium mixture for 5 min at room temper- temperature, blocked by 5% serum, and incubated with primary ature, the supernatant was transferred into a new tube and cells antibodies against sarcomeric a-actinin (Sigma Aldrich), vimentin

ª 2014 The Authors The EMBO Journal Vol 33 | No 14 | 2014 1577 1578 The EMBO Journal Vol 33 | No 14 | 2014 ª 2014 The Authors Naoto Muraoka et al MiR-133 promotes reprogramming by Snai1 repression The EMBO Journal The EMBO Journal MiR-133 promotes reprogramming by Snai1 repression Naoto Muraoka et al

Figure 6. MiR-133-mediated Snai1 repression is critical for cardiac reprogramming in adult cardiac fibroblasts. A B ◂ A, B FACS analyses for aMHC-GFP + and cTnT+ cells in adult CFs 1 week after GMT and GMT/miR-133 transduction with or without Snai1 overexpression. Quantitative data are shown in (B) (n = 3). C Relative mRNA expression of cardiac (Actn2, Myh6, Gja1, Scn5a) and fibroblast genes (Col1a1, Col3a1, Fn1) in adult CFs transduced with GMT and GMT/miR-133 with or without Snai1 overexpression (n = 3). D, E Immunocytochemisty for aMHC-GFP and a-actinin with DAPI staining in GMT, GMT/miR-133, or GMT/miR-133/Snai1-transduced adult CFs 4 weeks after transduction. High-magnification views in insets show sarcomeric organization. Quantitative data are shown in (E) (n = 5). F, G Total number of Ca2+ oscillation+ cells in 10 randomly selected fields per well is shown in (F) (n = 3). Spontaneous Ca2+ oscillations observed in adult CF-derived GMT/miR-133-iCMs (arrows in G), corresponding to Supplementary Movie S5. The Rhod-3 images and intensity trace are shown in (G). H Relative Snai1 mRNA and miR-133a expression in adult CFs, GMT-iCMs, and GMT/miR133-iCMs (n = 3). I Relative mRNA expression of cardiac (Actn2, Actc1, Slc8a1, Nppa) and fibroblast genes (Col1a2, Col5a2, Snai2) in adult CFs transduced with GMT, GMT/si-Snai1, or C D GMT/miR-133 (n = 3). J, K Immunocytochemisty for aMHC-GFP and a-actinin in GMT or GMT/si-Snai1 transduced adult CFs 4 weeks after transduction. High-magnification views in insets show sarcomeric organization. Quantitative data are shown in (K) (n = 5). Data information: All data are presented as means SEM. *P < 0.05, **P < 0.01 versus relevant control. Scale bars, 100 lm. Æ subcloned them into respective pMXs vectors for transfection into were collected by centrifugation and resuspended in DMEM/10% Plat-E cells using Fugene 6 (Roche) to generate retroviruses (Ieda FBS (Thermo Scientific, SV30014.03) for culturing at 37°C in 5% et al, 2010). We generated lentiviral vectors by subcloning human CO2. For isolation of mouse adult cardiac fibroblasts, aMHC-GFP TG Gata4, Mef2c, Tbx5, Mesp1, Myocd, and Snai1 (HuPEX, AIST) into adult mouse hearts were minced into small pieces < 1 mm3 in size. the CSII-CMV-RfA plasmid (RIKEN BRC) using the Gateway system The explants were plated on gelatin-coated dishes and cultured for (Invitrogen). To generate the lentiviruses, we transfected the vectors 10–14 days in explant medium (IMDM with L-Glutamate and into HEK293 cells with pCAG-HIVgp and pCMV-VSV-G-RSV-Rev 25 mM HEPES (Gibco, 12440-053)/20% FBS). Migrated fibroblasts plasmids (RIKEN BRC) using Lipofectamine 2000 (Invitrogen). The were harvested and filtered with 40-lm cell strainers (BD) to avoid CSII-CMV-Venus vector was used to determine transduction effi- contamination with tissue fragments. The aMHC-GFP–/Thy1+ CFs ciency. Virus-containing supernatants were collected after 48 h and were FACS sorted and plated at a density of 104/cm2 for the retro- used for transduction. Synthetic mimics of mature miRNAs (Thermo virus transduction. E FG Scientific) and the siRNA pool (Thermo Scientific) were transfected Human atrial tissues were obtained from patients undergoing simultaneously into cells with Lipofectamine 2000 (Invitrogen). cardiac surgery (age 1 month to 80 year; average age, 35 year) with Synthetic mimics of mature miRNAs and siRNA pool: miRIDIAN informed consent in conformation with the guidelines of the Keio microRNA Mouse mmu-miR-1-Mimic (C-310376-07-0020), miRIDI- University Ethics Committee. For isolation of human cardiac fibro- AN microRNA Mouse mmu-miR-133a-Mimic (C-310407-07-0020), blasts, human hearts were minced into small pieces < 1 mm3 in miRIDIAN microRNA Mouse mmu-miR-208a-3p-Mimic (C-310501- size. The explants were plated on gelatin-coated dishes and cultured 05-0020), miRIDIAN microRNA Mouse mmu-miR-499-Mimic (C- for 14 days in the explant medium. Migrated fibroblasts were 310727-01-0020), miRIDIAN microRNA Mimic Negative Control #2 harvested, filtered with 40-lm cell strainers (BD), and plated at a (CN-002000-01-05), siGENOME Mouse Snai1 siRNA - SMARTpool density of 5 × 103/cm2 for the virus transduction. For all experi- (M-062765-00-0020), siGENOME Human Snai1 siRNA - SMARTpool ments, we used fibroblasts of early passage number (P1-3). The (M-010847-00-0020). After 24 h, the medium was replaced with Keio Center for Clinical Research approved all of the human experi- D-MEM (high glucose) with L-Glutamate and Phenol Red (Wako, ments in this study (20100131). 044-29765)/Medium199 with Earle’s Salts, L-Glutamate and 22 g/l Sodium Bicarbonate (Gibco, 11150-059)/10% Hyclone Characterized FACS analyses and sorting FBS (Thermo Scientific, SV30014.03) medium and changed every – 2 3 days. JAK inhibitor I (1 nM, EMD Biosciences) treatment was For GFP expression analyses, cells were harvested from culture dishes H initiated 2 days after transfection and continued daily for 7 days. and analyzed on a FACSCalibur (BD Biosciences) with FlowJo software. For aMHC-GFP/cTnT expression, cells were fixed with 4% PFA Cell culture for 15 min, permealized with saponin, and stained with anti-cTnT and anti-GFP antibodies, followed by secondary antibodies conjugated with For MEF isolation, embryos isolated from 12.5-day pregnant mice Alexa 488 and 647, respectively. For a-actinin or cTnT expression, cells were washed with PBS, and the head and visceral tissues were care- were stained with anti-a-actinin or cTnT antibody, followed by second- fully removed. The remaining parts of the embryos were washed in ary antibody conjugated with Alexa 647. For iCM sorting, cells were fresh PBS, minced using a pair of scissors, transferred into a sorted as aMHC-GFP + cells, and for Mesp1-GFP–/Thy1+ cell sorting, 0.25 mM trypsin/1 mM EDTA solution (3 ml per embryo), and cells were incubated with APC-conjugated anti-Thy1 antibody I J incubated at 37°C for 20 min. An additional 3 ml of trypsin/EDTA (eBioscience) and sorted by FACS Aria. solution was then added, and the mixture was further incubated at 37°C for 20 min. After trypsinization, an equal amount of medium Immunocytochemistry (6 ml of DMEM containing 10% FBS per embryo) was added and pipetted up and down a few times to help tissue dissociation. After Cells were fixed in 4% paraformaldehyde for 15 min at room incubation of the tissue/medium mixture for 5 min at room temper- temperature, blocked by 5% serum, and incubated with primary ature, the supernatant was transferred into a new tube and cells antibodies against sarcomeric a-actinin (Sigma Aldrich), vimentin

fixed with 4% paraformaldehyde for 15 min, permeabilized, incu- upstream of the luciferase gene. A 755-bp Snai1 30UTR fragment Zhang B et al (2010) A mesenchymal-to-epithelial transition initiates and bated with anti-cTnT antibody followed by secondary antibody containing miR-133a-binding sites was amplified by PCR and subcl- Acknowledgements is required for the nuclear reprogramming of mouse fibroblasts. Cell Stem conjugated with Alexa 546 (for immunocytochemistry) or 647 (for oned into the modified pGL3-Basic vector. The activities of firefly We are grateful to members of the Fukuda lab, S. Mikami, H. Mochizuki, and Cell 7: 51 – 63 FACS), and then incubated with the EdU reaction cocktail following luciferase and renilla luciferase in the control vector were deter- H. Miyoshi (RIKEN BRC), for discussion and reagents. M. I. was supported by Liu N, Bezprozvannaya S, Williams AH, Qi X, Richardson JA, Bassel-Duby R, the manufacturer’s instructions (Invitrogen). mined by the dual-luciferase reporter assay (Promega). Mutations of research grants from JST CREST, JSPS, the Mitsubishi Foundation, Banyu Life Olson EN (2008) microRNA-133a regulates cardiomyocyte proliferation and the AGGGGACCA miR-133a seed binding sequence were Science, the Uehara Memorial Foundation, Senshin Medical Research suppresses smooth muscle gene expression in the heart. Genes Dev 22: Ca2+ imaging and counting beating cells constructed through PCR-based mutagenesis (Stratagene). The Foundation, AstraZeneca, and Takeda Science Foundation, and N. M. was 3242 – 3254 miRNA target prediction program (http://cbio.mskcc.org/cgi-bin/ supported by research grants from Japan Heart Foundation Research Grant, Liu N, Olson EN (2010) MicroRNA regulatory networks in cardiovascular Ca2+ imaging was performed according to the standard protocol. mirnaviewer/mirnaviewer.pl) was used to identify putative targets. Keio University Medical Science Fund, and Keio University Grant-in-Aid for development. Dev Cell 18: 510 – 525 Briefly, cells were labeled with Rhod-3 (Invitrogen) for 1 h at room Encouragement of Young Scientists. Marro S, Pang ZP, Yang N, Tsai MC, Qu K, Chang HY, Sudhof TC, Wernig M temperature, washed, and incubated for an additional 1 h to allow Gene microarray analyses (2011) Direct lineage conversion of terminally differentiated hepatocytes de-esterification of the dye. Rhod-3 labeled cells were analyzed at Author contributions to functional neurons. Cell Stem Cell 9: 374 – 382 37°C by LSM 510 META confocal microscopy (Carl Zeiss). Imaging Mouse genome-wide gene expression analyses were performed NM and MI designed the experiments. NM, HYamak, KM, TS, TM, MI, HN, MA, Muraoka N, Ieda M (2014) Direct reprogramming of fibroblasts into myocytes of the Ca2+ oscillations was possible for only a short time due to using 3D-Gene Mouse Oligo chip 25k (Toray Industries Inc.). For RW, KI, YK, RA, HYamag, and NG carried out the experiments. NM, TN, RK, TF, to reverse fibrosis. Annu Rev Physiol 76: 21 – 37 the increasing background fluorescence from the medium, and thus, efficient hybridization, this microarray is three-dimensional and is STa, STo, HH, and KF analyzed the data. NM and MI wrote the paper. Nam YJ, Song K, Luo X, Daniel E, Lambeth K, West K, Hill JA, DiMaio JM, Baker the measurements were taken within 30 min after changing to the constructed with a well as the space between the probes and cylin- LA, Bassel-Duby R, Olson EN (2013) Reprogramming of human fibroblasts Tyrode’s buffer. The Ca2+ oscillation+ cells were counted in 10 der stems with 70-mer oligonucleotide probes on the top. RNA was Conflict of interest toward a cardiac fate. Proc Natl Acad Sci USA 110: 5588 – 5593 randomly selected fields per well in at least three independent extracted from MEFs, GMT-, GMT/miR-133-, or GMT/miR-133/ The authors declare that they have no conflict of interest. Protze S, Khattak S, Poulet C, Lindemann D, Tanaka EM, Ravens U (2012)A experiments, and a minimum of 1,000 cells were counted in total. Snai1-induced aMHC-GFP+ cells, neonatal mouse heart tissues, new approach to transcription factor screening for reprogramming of The total number of Ca2+ oscillation+ cells in 10 randomly selected HCFs, GMTMM-, GMTMM/miR-133-, GMTMM/miR-133/Snai1- fibroblasts to cardiomyocyte-like cells. J Mol Cell Cardiol 53: 323 – 332 fields per well is shown. transduced HCFs using ReliaPrepTM RNA Cell Miniprep System References Qian L, Huang Y, Spencer CI, Foley A, Vedantham V, Liu L, Conway SJ, Fu JD, For counting beating cells, we seeded 50,000 fibroblasts per well (Promega). Total RNA was labeled with Cy5 using the Amino Allyl Srivastava D (2012) In vivo reprogramming of murine cardiac fibroblasts on 12-well plates, performed cell transductions, and then monitored MessageAMP II aRNA Amplification Kit (Applied Biosystems), and Addis RC, Epstein JA (2013) Induced regeneration–the progress and into induced cardiomyocytes. Nature 485: 593 – 598 cell contraction. The number of spontaneously contracting cells was the hybridization was performed using the supplier’s protocols promise of direct reprogramming for heart repair. Nat Med 19: Rowe RG, Li XY, Hu Y, Saunders TL, Virtanen I, Garcia de Herreros A, Becker manually counted in each well in at least three independent experi- (www.3d-gene.com). Hybridization signals were scanned using 3D- 829 – 836 KF, Ingvarsen S, Engelholm LH, Bommer GT, Fearon ER, Weiss SJ (2009) ments. The number of beating cells per well is shown. The measure- Gene Scanner (Toray Industries Inc.) and processed by extraction Chen JX, Krane M, Deutsch MA, Wang L, Rav-Acha M, Gregoire S, Engels MC, Mesenchymal cells reactivate Snail1 expression to drive three-dimensional ments and calculations were made in a blinded manner. (Toray Industries Inc.). The raw data of each spot were normalized Rajarajan K, Karra R, Abel ED, Wu JC, Milan D, Wu SM (2012) Inefficient invasion programs. J Cell Biol 184: 399 – 408

ª 2014 The Authors The EMBO Journal Vol 33 | No 14 | 2014 1579 1580 The EMBO Journal Vol 33 | No 14 | 2014 ª 2014 The Authors Naoto Muraoka et al MiR-133 promotes reprogramming by Snai1 repression The EMBO Journal The EMBO Journal MiR-133 promotes reprogramming by Snai1 repression Naoto Muraoka et al

Figure 7. MiR-133/Snai1 pathway is critical for human cardiac reprogramming. by substitution with a mean intensity of the background signal reprogramming of fibroblasts into cardiomyocytes using gata4, mef2c, and ◂ A, B FACS analyses for cTnT+ cells in HCFs 1 week after GMTMM, GMTMM/miR-133, and GMTMM/miR-133/Snai1 transduction. Quantitative data are shown in (B) determined by the combined signal intensities of all blank spots at tbx5. Circ Res 111: 50 – 55 (n = 4). 95% confidence intervals. Raw data intensities > 2 standard devia- Fu JD, Stone NR, Liu L, Spencer CI, Qian L, Hayashi Y, Delgado-Olguin P, C Heat-map image of microarray data illustrating the global gene expression pattern of HCFs, GMTMM-HCFs, and GMTMM/miR-133-HCFs after 1 week of tions (SD) of the background signal intensity were considered to Ding S, Bruneau BG, Srivastava D (2013) Direct reprogramming of transduction (n = 1, left panel). The differentially expressed genes between GMTMM- or GMTMM/miR-133-HCFs and HCFs are shown. Cardiac genes were be valid. Detected signals for each gene were normalized by the human fibroblasts toward a cardiomyocyte-like state. Stem Cell Rep 1: upregulated and fibroblast genes were downregulated by transduction of the reprogramming factors, as shown in the scatter plot analyses (right panel). D 399 genes were upregulated and 264 genes were downregulated by the addition of miR-133 to GMTMM. global normalization method. Heatmap images for differentially 235 – 247 E GO term analyses of the upregulated and downregulated genes, shown in (D). Cardiac- and fibroblast-related GO terms are shown. expressed genes (more than 2-fold or 1.5-fold difference) were Han DW, Tapia N, Hermann A, Hemmer K, Hoing S, Arauzo-Bravo MJ, Zaehres F The mRNA expression of Snai1 in HCFs, GMTMM-HCFs, and GMTMM/miR-133-HCFs (n = 3). processed using the Cluster 2.0 software, and the results were H, Wu G, Frank S, Moritz S, Greber B, Yang JH, Lee HT, Schwamborn JC, G Snai1 restoration counteracted the effects of miR-133. 309 out of 399 upregulated genes were suppressed by Snai1 overexpression, while 214 out of 264 displayed with the TreeView program (http://rana.lbl.gov/eisen/). Storch A, Scholer HR (2012) Direct reprogramming of fibroblasts into downregulated genes were increased with Snai1. See also Fig 7D. Scatter plot analyses were processed using Microsoft Excel. Gene neural stem cells by defined factors. Cell Stem Cell 10: 465 – 472 H Relative mRNA expression of cardiac (Myh6, Actn2, Ttn, Nppa) and fibroblast genes (Col1a1, Fn1, Postn) in GMTMM-, GMTMM/miR-133-, and GMTMM/miR-133/ Snai1-HCFs after 1 week of transduction (n = 3). ontology (GO) analysis was performed using GeneCodis. This Ieda M, Fu JD, Delgado-Olguin P, Vedantham V, Hayashi Y, Bruneau BG, I, J Immunocytochemisty for a-actinin and DAPI. Snai1 overexpression suppressed cardiac protein expression in GMTMM/miR-133-transduced HCFs (J, n = 10). High- method computes hypergeometric P-values for over- or under- Srivastava D (2010) Direct reprogramming of fibroblasts into functional magnification view in inset shows sarcomeric organization. representation of each GO term in the specified ontology for the cardiomyocytes by defined factors. Cell 142: 375 – 386 Data information: All data are presented as means SEM. *P < 0.05, **P < 0.01 versus relevant control. Scale bars, 100 lm. gene set of interest. Moderated t-statistics and the associated Inagawa K, Miyamoto K, Yamakawa H, Muraoka N, Sadahiro T, Umei T, Wada � P-values were calculated by the Welch t-test using Microsoft Excel. R, Katsumata Y, Kaneda R, Nakade K, Kurihara C, Obata Y, Miyake K, Differential gene expression was defined using the statistics/thresh- Fukuda K, Ieda M (2012) Induction of cardiomyocyte-like cells in infarct (Progen), Collagen 1 (Millipore), GFP (Invitrogen), cTnT (Thermo Western blotting old combination. hearts by gene transfer of Gata4, Mef2c, and Tbx5. Circ Res 111: Scientific), ANP (Millipore), Myl7 (Synaptic Systems), or Nkx2.5 1147 – 1156 (Santa Cruz), and subsequently with secondary antibodies conju- Lysates were prepared by homogenization of cells in RIPA buffer Statistical analyses Jayawardena TM, Egemnazarov B, Finch EA, Zhang L, Payne JA, Pandya K, gated with Alexa 488 or 546 (Molecular Probes), followed by DAPI and run on SDS–PAGE to separate proteins prior to the immunoblot Zhang Z, Rosenberg P, Mirotsou M, Dzau VJ (2012) MicroRNA-mediated in counterstaining (Invitrogen). The percentage of cells immunoposi- analyses. After transfer to nitrocellulose membranes, immunodetec- Differences between groups were examined for statistical signifi- vitro and in vivo direct reprogramming of cardiac fibroblasts to tive for GFP, a-actinin, cTnT, and ANP were counted in six tion was performed with antibodies to Snai1 (Abcam) and Gapdh cance using Student’s t-test or ANOVA. P-values of < 0.05 were cardiomyocytes. Circ Res 110: 1465 – 1473 randomly selected fields per well in three independent experiments, (Cell Signaling Technology), followed by the appropriate HRP- regarded as significant. Judson RL, Babiarz JE, Venere M, Blelloch R (2009) Embryonic stem and 500–1,000 cells were counted in total. The measurements and conjugated secondary antibodies (Cell Signaling Technology). The cell-specific microRNAs promote induced pluripotency. Nat Biotechnol 27: calculations were made in a blinded manner. antibody-bound proteins were visualized by chemiluminescence Data deposition 459 – 461 detection (ECL, Amersham). Kawamoto S, Niwa H, Tashiro F, Sano S, Kondoh G, Takeda J, Tabayashi K, EdU labeling assay Microarray data are deposited in GEO with accession number Miyazaki J (2000) A novel reporter mouse strain that expresses enhanced Dual-luciferase reporter assay GSE56913. green fluorescent protein upon Cre-mediated recombination. FEBS Lett For the experiments assessing cell proliferation, 10 lM EdU was 470: 263 – 268 added to the culture medium after 2 weeks of transduction and For construction of the Snai1 30UTR reporter, the CMV promoter Supplementary information for this article is available online: Li R, Liang J, Ni S, Zhou T, Qing X, Li H, He W, Chen J, Li F, Zhuang Q, maintained throughout the culture for a further 2 weeks. Cells were was subcloned into the promoterless pGL3-Basic vector (Promega) http://emboj.embopress.org Qin B, Xu J, Li W, Yang J, Gan Y, Qin D, Feng S, Song H, Yang D, fixed with 4% paraformaldehyde for 15 min, permeabilized, incu- upstream of the luciferase gene. A 755-bp Snai1 30UTR fragment Zhang B et al (2010) A mesenchymal-to-epithelial transition initiates and bated with anti-cTnT antibody followed by secondary antibody containing miR-133a-binding sites was amplified by PCR and subcl- Acknowledgements is required for the nuclear reprogramming of mouse fibroblasts. Cell Stem conjugated with Alexa 546 (for immunocytochemistry) or 647 (for oned into the modified pGL3-Basic vector. The activities of firefly We are grateful to members of the Fukuda lab, S. Mikami, H. Mochizuki, and Cell 7: 51 – 63 FACS), and then incubated with the EdU reaction cocktail following luciferase and renilla luciferase in the control vector were deter- H. Miyoshi (RIKEN BRC), for discussion and reagents. M. I. was supported by Liu N, Bezprozvannaya S, Williams AH, Qi X, Richardson JA, Bassel-Duby R, the manufacturer’s instructions (Invitrogen). mined by the dual-luciferase reporter assay (Promega). Mutations of research grants from JST CREST, JSPS, the Mitsubishi Foundation, Banyu Life Olson EN (2008) microRNA-133a regulates cardiomyocyte proliferation and the AGGGGACCA miR-133a seed binding sequence were Science, the Uehara Memorial Foundation, Senshin Medical Research suppresses smooth muscle gene expression in the heart. Genes Dev 22: Ca2+ imaging and counting beating cells constructed through PCR-based mutagenesis (Stratagene). The Foundation, AstraZeneca, and Takeda Science Foundation, and N. M. was 3242 – 3254 miRNA target prediction program (http://cbio.mskcc.org/cgi-bin/ supported by research grants from Japan Heart Foundation Research Grant, Liu N, Olson EN (2010) MicroRNA regulatory networks in cardiovascular Ca2+ imaging was performed according to the standard protocol. mirnaviewer/mirnaviewer.pl) was used to identify putative targets. Keio University Medical Science Fund, and Keio University Grant-in-Aid for development. Dev Cell 18: 510 – 525 Briefly, cells were labeled with Rhod-3 (Invitrogen) for 1 h at room Encouragement of Young Scientists. Marro S, Pang ZP, Yang N, Tsai MC, Qu K, Chang HY, Sudhof TC, Wernig M temperature, washed, and incubated for an additional 1 h to allow Gene microarray analyses (2011) Direct lineage conversion of terminally differentiated hepatocytes de-esterification of the dye. Rhod-3 labeled cells were analyzed at Author contributions to functional neurons. Cell Stem Cell 9: 374 – 382 37°C by LSM 510 META confocal microscopy (Carl Zeiss). Imaging Mouse genome-wide gene expression analyses were performed NM and MI designed the experiments. NM, HYamak, KM, TS, TM, MI, HN, MA, Muraoka N, Ieda M (2014) Direct reprogramming of fibroblasts into myocytes of the Ca2+ oscillations was possible for only a short time due to using 3D-Gene Mouse Oligo chip 25k (Toray Industries Inc.). For RW, KI, YK, RA, HYamag, and NG carried out the experiments. NM, TN, RK, TF, to reverse fibrosis. Annu Rev Physiol 76: 21 – 37 the increasing background fluorescence from the medium, and thus, efficient hybridization, this microarray is three-dimensional and is STa, STo, HH, and KF analyzed the data. NM and MI wrote the paper. Nam YJ, Song K, Luo X, Daniel E, Lambeth K, West K, Hill JA, DiMaio JM, Baker the measurements were taken within 30 min after changing to the constructed with a well as the space between the probes and cylin- LA, Bassel-Duby R, Olson EN (2013) Reprogramming of human fibroblasts Tyrode’s buffer. The Ca2+ oscillation+ cells were counted in 10 der stems with 70-mer oligonucleotide probes on the top. RNA was Conflict of interest toward a cardiac fate. Proc Natl Acad Sci USA 110: 5588 – 5593 randomly selected fields per well in at least three independent extracted from MEFs, GMT-, GMT/miR-133-, or GMT/miR-133/ The authors declare that they have no conflict of interest. Protze S, Khattak S, Poulet C, Lindemann D, Tanaka EM, Ravens U (2012)A experiments, and a minimum of 1,000 cells were counted in total. Snai1-induced aMHC-GFP+ cells, neonatal mouse heart tissues, new approach to transcription factor screening for reprogramming of The total number of Ca2+ oscillation+ cells in 10 randomly selected HCFs, GMTMM-, GMTMM/miR-133-, GMTMM/miR-133/Snai1- fibroblasts to cardiomyocyte-like cells. J Mol Cell Cardiol 53: 323 – 332 fields per well is shown. transduced HCFs using ReliaPrepTM RNA Cell Miniprep System References Qian L, Huang Y, Spencer CI, Foley A, Vedantham V, Liu L, Conway SJ, Fu JD, For counting beating cells, we seeded 50,000 fibroblasts per well (Promega). Total RNA was labeled with Cy5 using the Amino Allyl Srivastava D (2012) In vivo reprogramming of murine cardiac fibroblasts on 12-well plates, performed cell transductions, and then monitored MessageAMP II aRNA Amplification Kit (Applied Biosystems), and Addis RC, Epstein JA (2013) Induced regeneration–the progress and into induced cardiomyocytes. Nature 485: 593 – 598 cell contraction. The number of spontaneously contracting cells was the hybridization was performed using the supplier’s protocols promise of direct reprogramming for heart repair. Nat Med 19: Rowe RG, Li XY, Hu Y, Saunders TL, Virtanen I, Garcia de Herreros A, Becker manually counted in each well in at least three independent experi- (www.3d-gene.com). Hybridization signals were scanned using 3D- 829 – 836 KF, Ingvarsen S, Engelholm LH, Bommer GT, Fearon ER, Weiss SJ (2009) ments. The number of beating cells per well is shown. The measure- Gene Scanner (Toray Industries Inc.) and processed by extraction Chen JX, Krane M, Deutsch MA, Wang L, Rav-Acha M, Gregoire S, Engels MC, Mesenchymal cells reactivate Snail1 expression to drive three-dimensional ments and calculations were made in a blinded manner. (Toray Industries Inc.). The raw data of each spot were normalized Rajarajan K, Karra R, Abel ED, Wu JC, Milan D, Wu SM (2012) Inefficient invasion programs. J Cell Biol 184: 399 – 408

Saga Y, Miyagawa-Tomita S, Takagi A, Kitajima S, Miyazaki J, Inoue T (1999) reprogramming of human fibroblasts to induced pluripotent stem cells. MesP1 is expressed in the heart precursor cells and required for the Nat Biotechnol 29: 443 – 448 formation of a single heart tube. Development 126: 3437 – 3447 Szabo E, Rampalli S, Risueno RM, Schnerch A, Mitchell R, Fiebig-Comyn Samavarchi-Tehrani P, Golipour A, David L, Sung HK, Beyer TA, Datti A, A, Levadoux-Martin M, Bhatia M (2010) Direct conversion of Woltjen K, Nagy A, Wrana JL (2010) Functional genomics reveals a human fibroblasts to multilineage blood progenitors. Nature 468: BMP-driven mesenchymal-to-epithelial transition in the initiation of 521 – 526 somatic cell reprogramming. Cell Stem Cell 7: 64 – 77 Vierbuchen T, Ostermeier A, Pang ZP, Kokubu Y, Sudhof TC, Wernig M (2010) Sekiya S, Suzuki A (2011) Direct conversion of mouse fibroblasts to Direct conversion of fibroblasts to functional neurons by defined factors. hepatocyte-like cells by defined factors. Nature 475: 390 – 393 Nature 463: 1035 – 1041 Song K, Nam YJ, Luo X, Qi X, Tan W, Huang GN, Acharya A, Smith CL, Wada R, Muraoka N, Inagawa K, Yamakawa H, Miyamoto K, Sadahiro T, Tallquist MD, Neilson EG, Hill JA, Bassel-Duby R, Olson EN (2012) Heart Umei T, Kaneda R, Suzuki T, Kamiya K, Tohyama S, Yuasa S, Kokaji K, Aeba repair by reprogramming non-myocytes with cardiac transcription factors. R, Yozu R, Yamagishi H, Kitamura T, Fukuda K, Ieda M (2013) Induction of Nature 485: 599 – 604 human cardiomyocyte-like cells from fibroblasts by defined factors. Proc Srivastava D, Ieda M (2012) Critical factors for cardiac reprogramming. Circ Natl Acad Sci USA 110: 12667 – 12672 Res 111: 5 – 8 Yoo AS, Sun AX, Li L, Shcheglovitov A, Portmann T, Li Y, Lee-Messer C, Subramanyam D, Lamouille S, Judson RL, Liu JY, Bucay N, Derynck R, Blelloch Dolmetsch RE, Tsien RW, Crabtree GR (2011) MicroRNA-mediated R(2011) Multiple targets of miR-302 and miR-372 promote conversion of human fibroblasts to neurons. Nature 476: 228 – 231

ª 2014 The Authors The EMBO Journal Vol 33 | No 14 | 2014 1581 Review

Dedifferentiation and reprogramming: origins of cancer stem cells Dinorah Friedmann-Morvinski & Inder M Verma*

Abstract nes, offers a new insight into cancer stem cells (CSCs). They may be the product of dedifferentiation of somatic cells following oncogenic Regenerative medicine aims to replace the lost or damaged cells in insult. Cancer cells are the ultimate survivors and will exploit and the human body through a new source of healthy transplanted subvert the cellular machinery to achieve that goal, by proliferation, cells or by endogenous repair. Although human embryonic stem dedifferentiation, and even transdifferentiation. Here, we will high- cells were first thought to be the ideal source for cell therapy and light some human cancers that may be the product of somatic cell tissue repair in humans, the discovery by Yamanaka and colleagues reprogramming and as such may even pose some risk to the applica- revolutionized the field. Almost any differentiated cell can be sent tion of iPSCs in regenerative medicine. back in time to a pluripotency state by expressing the appropriate transcription factors. The process of somatic reprogramming using Yamanaka factors, many of which are oncogenes, offers a glimpse From ESC to iPSC into how cancer stem cells may originate. In this review we discuss the similarities between tumor dedifferentiation and somatic cell Embryonic stem cells have attracted special attention by virtue of reprogramming and how this may pose a risk to the application of their pluripotent nature, that is, the ability to self-renew indefinitely this new technology in regenerative medicine. in culture while retaining the capacity to differentiate into nearly all cell types in the body. Since their isolation, ESCs have been regarded Keywords cancer stem cells; dedifferentiation; somatic reprogramming; as gold standard for their unique properties and extraordinary poten- tumor plasticity tial in regenerative medicine. Despite these important properties and DOI 10.1002/embr.201338254 | Received 21 November 2013 | Revised 14 remarkable qualities, ESC-based therapy has many limitations in January 2014 | Accepted 21 January 2014 | Published online 14 February 2014 treating human diseases. ESCs are derived from the inner cell mass EMBO Reports (2014) 15, 244–253 of pre-implantation embryos, and hence patient- specific, and cannot be used as a general cell source for transplantation to other patients in See the Glossary for abbreviations used in this article. need due to the risk of immune rejection. Historically, the solution to overcome this obstacle comes from seminal frog studies when Briggs and King [5] demonstrated the reversal of cell differentiation Introduction by transplantation of a viable cell nucleus into an enucleated frog egg. They succeeded in producing normal swimming tadpoles of A new branch of medicine will develop that attempts to change the Rana pipiens by transplanting the nuclei of embryo (blastula) cells. course of chronic disease and in many instances will regenerate Later, similar experiments were carried out with eggs of the South tired and failing organ systems [1,2]. African frog Xenopus laevis using nuclei from fully differentiated Many human diseases are caused by deficits in the quantity or cells [6]. Collectively, these results challenged the unidirectional functionality of particular cells. These diseases include neurodegen- developmental model: Cells now can go back in time, dedifferentiate erative disorders, certain types of blindness and deafness, diabetes, by changes in nuclear gene expression while maintaining their gen- and some types of liver and heart disease. The idea that degenera- ome intact. Somatic cell nuclear transfer (SCNT), or nuclear repro- tive as well as genetic diseases can be treated by regenerating the gramming, is the technology by which the nucleus of the donor diseased organ or replacing the “damaged” cells with “healthy” new somatic cell is removed and transferred into an enucleated oocyte, cells is fast becoming a reality since the first isolation of human where undefined factors in the cytoplasm of this oocyte are able to embryonic stem cells (ESCs) and generation of induced pluripotent reprogram the somatic donor nucleus to a pluripotent state. stem cells (iPSCs) from terminally differentiated somatic cells [3,4]. In a remarkable experiment, Takahashi and Yamanaka [3] Remarkably, the process of dedifferentiation or reprogramming of demonstrated that introduction of mere four genes (Oct-3/4, Sox2, the somatic cells by Yamanaka factors, many of which are oncoge- c-Myc, and KLF4) into an adult mouse fibroblast population can

Laboratory of Genetics, The Salk Institute for Biological Studies, La Jolla, CA, USA *Corresponding author. Tel: +1 858 453 4100 x1462; Fax: +1 858 558 7454; E-mail: [email protected]

244 EMBO reports Vol 15 | No 3 | 2014 ª 2014 The Authors Dinorah Friedmann-Morvinski & Inder M Verma Dedifferentiation and reprogramming: origins of cancer stem cells EMBO reports EMBO reports Dedifferentiation and reprogramming: origins of cancer stem cells Dinorah Friedmann-Morvinski & Inder M Verma

reprogramming of somatic cells to iPSC and how these modifi- Glossary cations determine the cell fate. Bmi1 B-lymphoma Mo-MLV insertion region 1 homolog Interestingly, the acquisition of stem cell properties has also been CNS central nervous system reported in the case of normal differentiated cells in certain organs. Cre Cre recombinase The non-stem cell compartment seems to be the source of a new CSC cancer stem cell pool of cells with stem-like characteristics very similar to the endog- Dll1 Delta-like gene 1 enous stem cell counterparts in the organ. Dot1l DOT1-like, histone H3 methyltransferase Tumor progression EGFR epidermal growth factor receptor Examples of this interconversion between stem and non-stem EMT epithelial–mesenchymal transition cell compartments have been shown in a subpopulation of basal-like ESC embryonic stem cell human mammary epithelial cells [13], a rare subpopulation of Differentiation markers GBM glioblastoma somatic cells from breast tissue [14]. Differentiated airway epithelial H3K27me3 histone H3 trimethyl Lys27 cells have also been reported to dedifferentiate to a stem-like state H3K9me2 histone H3 dimethyl Lys9 Progenitor/stem cell markers H3K9me3 histone H3 trimethyl Lys9 [15], and upon tissue damage, Dll1+ intestinal secretory progenitor iPSC induced pluripotent stem cell cells can also acquire cell plasticity and regain stemness [16]. If Klf4 Kruppel-like factor 4 terminally differentiated cells can regain stem cell traits to maintain Fig 1. Kinetic expression of differentiation markers along tumor progression. Kras Ki-ras2 Kirsten rat sarcoma viral oncogene a balanced equilibrium between non-stem and stem cell compart- Glioblastoma tumors induced by oncogenic lentivirus either in neurons or in glia in the cortex initially express differentiation markers (e.g., Tuj1 and GFAP, respectively), but as tumor progresses, these markers decrease and stem/progenitor markers become predominantly expressed (like nestin and Sox2) [20]. LTR long terminal repeat ments or to be able to regenerate damage tissue, it is fair to assume MET mesenchymal–epithelial transition Nanog Nanog homeobox that this process can be adopted in disease states like cancer. NF-jB nuclear factor kappa-light-chain-enhancer of activated Glioblastoma (GBM), the most common and aggressive subtype transcriptions factors [3], then, following the same line of reasoning B cells of the malignant gliomas, is characterized by intense proliferation, and in theory, intestinal epithelial cells (IEC) can dedifferentiate to a NSC neural stem cell Dedifferentiation in cancer cells invasion, and intratumor heterogeneity. A decade ago, Ronald progenitor/stem cell state given an appropriate transcription factor, Oct-3/4 octamer-binding transcription factor 3/4 DePinho’s group demonstrated that the combined loss of p16INK4a for example Wnt signaling, is strongly activated. In the study of Ras rat sarcoma ARF Setdb1 Set domain, bifurcated 1 methyltransferase There are multiple levels of heterogeneity associated with cancer, and p19 enables mature astrocyte dedifferentiation in response Florian Greten’s group, the combination of an oncogenic hit like Kras and this heterogeneity is one of the hallmarks of cancers arising in to EGFR activation [19]. Moreover, transduction of Ink4a/Arf( / ) and the activation of NF-jB induces the stabilization of b-catenin Sox2 sex-determining region Y-related box2 À À Suv39h1 suppressor of variegation 3–9 homolog 1 several organs. Genetic heterogeneity in the majority of the cancers neural stem cells (NSCs) or astrocytes with constitutively active and thereof the activation of the b-catenin/tcf transcription complex, TDEC tumor-derived endothelial cell is reflected by genome instability, and in addition to these genetic EGFR induces a common high-grade glioma phenotype. These find- leading to the conversion of non-stem IEC into IEC with stem cell Tet1 Ten-eleven translocation methylcytosine dioxygenase 1 alterations, the state of the cell may be changed epigenetically. ings identify neural stem cells and astrocytes as equally permissive properties [22]. The implication of NF-jB activation in the context TGF-b transforming growth factor beta Phenotypic heterogeneity refers to the diverse functional properties compartments for gliomagenesis. The identification of TUJ1-positive of inflammation and cancer has already been shown previously TNF-a tumor necrosis factor alpha TUJ1 neuron-specific class III beta-tubulin and expression of different lineage markers that tumor cells can neurons in the tumors originating from the transformed astrocytes [23]. But in this study, both Kras and TNF-a-dependent NF-jB acti- Wnt Wingless integrase-1 adopt along cancer progression. Based on cell surface markers, we suggested that dedifferentiation may be so complete as to generate a vation enhances b-catenin/Tcf-mediated transcriptional activity and ZEB1 zinc finger E-box-binding homeobox 1 can identify distinct subpopulations of neoplastic cells within the pluripotent cell with the potential to make neurons as well as glia. induces dedifferentiation of non-stem IEC into tumor-initiating cells, same tumors, suggesting that irrespective of their genetic altera- More recently, our group showed that GBM can originate from a supporting a model of bidirectional interconvertibility rather than tions, cancer cells may exist in different states of differentiation variety of cells in the brain, including terminally differentiated the strict unidirectional model of the stem differentiation hierarchy generate colonies with the characteristics of ESCs. These colonies [17]. The latest has been defined as intratumor heterogeneity, to cortical astrocytes and neurons [20]. Transduction by oncogenic (reviewed by [24]). It remains unclear whether this plasticity of were capable of differentiation to endodermal, ectodermal, and explain the variations found within individual tumors, and intertu- Cre-inducible lentiviruses in the cortex of synapsinI-Cre or GFAP- tumor cells is specific to certain types of cancer, how frequently this mesodermal lineages upon transplantation in immunodeficient moral heterogeneity refers to the molecular differences that occur Cre transgenic mice, which drive the expression of Cre specifically process of interconversion occurs in vivo, and what is the mecha- mice. The authors termed them as induced pluripotent cells, or between tumors initiated in the same organ, which allows the in neurons and glial cells, respectively, induced the formation of nism that regulates the dynamic equilibrium that exists between iPSCs. In the subsequent years, a variety of approaches and a classification of these tumors in different subtypes and may even gliomas. Interestingly, these tumors mostly expressed markers of non-CSC and CSC. Recently, Robert Weinberg’s group addressed very wide variety of differentiated cells have now been used to represent biologically distinct disease entities [18]. progenitor/neural stem cells, nestin and Sox2. In a study aimed to some of these questions and found that the switch between non- generate iPSCs [7–9]. Cancer stem cell can be defined as the cells within a tumor that follow the kinetic expression of some of these markers during tumor CSC to CSC state is frequently common in certain types of breast The accumulated knowledge in the mechanisms and pathways possess the capacity to self-renew and to cause the heterogeneous development, we observed that at early stages, differentiation mark- cancer, and proposed a mechanism responsible for this transition. involved in cellular reprogramming and induced pluripotency lineages of cancer cells that comprise the tumor. CSCs are thus a ers are progressively diminished, while nestin, a marker of NSC, Based on the notion that the epithelial–mesenchymal transition clearly shows the important connections between protein and tran- biologically unique subpopulation of cells that can perpetuate indef- undetectable a few days after transduction, increased significantly (EMT) generates cells with stem cell properties [25], they found scriptional networks and how these factors affect the chromatin initely as oppose to the bulk of cells that reside in the tumor, and with tumor progression (Fig 1). We proposed that oncogenic- ZEB1, a transcription factor known to be involved in the EMT pro- landscape. Changes in chromatin structure clearly affect global gene are mostly insensitive to currently used cancer therapies. The CSC induced dedifferentiation of mature cells in the brain to a stem-/ gram, to be required for the conversion from non-CSC to CSC and expression directly contributing to cell fate transitions. The key plu- model assumes that this unique subpopulation of cells sustain progenitor-like state leads to heterogeneous glioma tumors (Fig 2). also for the maintenance of the CSC-like activity [26]. More specifi- ripotency transcription factors described before are intertwined with malignant growth by means of their ability to self-renew and the The genetically acquired plasticity of these cells allows progression cally, the chromatin configuration in a bivalent/poised state associ- chromatin-associated factors to form sophisticated networks with possibility to give rise to progeny with self-limited proliferative and maintenance of this aggressive tumor and even formation of its ated with the ZEB1 promoter enables de novo generation of CSCs complex regulatory interactions responsible for the maintenance of capacity. This suggests a hierarchical organization where CSCs are own blood vessels by transdifferentiation [21]. These data also from non-CSC populations. Not surprisingly, the tumor microenvi- stemness and differentiation states. Chromatin marks linked to gene responsible for the generation of the heterogeneity found within supported the view originally proposed by Ronald DePinho and his ronment provides the stimuli, in this case by the secretion of TGF-b activation, particularly HK4me3, are highly enriched in genes tumors. Although CSCs exhibit the stem cell properties of self- group [19] that dysregulation of specific genetic pathways, rather that enhances the rate of transition from non-CSC to CSC by induc- expressed in ESCs. Conversely, upon differentiation, cells acquire renewal and differentiation, they do not necessarily originate from than cell of origin, dictates the emergence and phenotype of high- ing ZEB1 expression. The implications of tumor plasticity can chromatin marks associated with transcriptional repression such as the transformation of normal tissue stem cells [18]. grade gliomas. go beyond the primary tumor and play an important role in tumor H3K27me3 [10]. Not surprisingly, chromatin markers of iPSCs Several recent studies now suggest that not all cancers strictly The bidirectional conversion between CSCs and non-CSCs was metastasis. It has been suggested that CSCs are endowed with resemble those found in ESCs [11]. Chromatin changes in both conform to the unidirectional hierarchical CSC model, and enter- also found in intestinal tumors, where an inflammatory stroma, multiple traits that are required for most of the steps related to the promoters and enhancers are early events in response to the core tain the possibility of “tumor cell plasticity”, where non-CSC can through NF-jB activation, enhances Wnt signaling and leads to the invasion–metastasis cascade [27] and therefore might be key players transcription factors during the reprogramming process [12]. We dedifferentiate and acquire CSC-like properties under certain reprogramming process. If any differentiated cell can be repro- in metastatic tumors. Although it still needs to be proven, non-CSC are still learning how the epigenetic landscape is reset during conditions as demonstrated by examples below: grammed to a pluripotent state through the right combination of too may leave a primary tumor and, upon arrival to the secondary

ª 2014 The Authors EMBO reports Vol 15 | No 3 | 2014 245 246 EMBO reports Vol 15 | No 3 | 2014 ª 2014 The Authors Dinorah Friedmann-Morvinski & Inder M Verma Dedifferentiation and reprogramming: origins of cancer stem cells EMBO reports EMBO reports Dedifferentiation and reprogramming: origins of cancer stem cells Dinorah Friedmann-Morvinski & Inder M Verma

Lenti-OKSM Teratomas features in more differentiated tumor cells [33]. In this scenario, Self- myofibroblasts create a niche that support dedifferentiation of colon renewing cancer cells to a CSC phenotype, and this dynamic conversion Normal Neural stem cell iPS colonies toward stemness is regulated by the intensity of Wnt activity. As we discussed previously, Wnt activity was an important cell-intrinsic feature responsible for the reprogramming of IEC into CSC. But the Progenitor cell results of Medema’s group suggest that stroma cells can not only sustain high Wnt activity in the surrounding CSC but can also Gliomas induce reprogramming of differentiated tumor cells into CSC by secreting extrinsic factors and activating their Wnt signaling path- Astrocytes way. An inflammatory tumor microenvironment can also shape the Tumorspheres response of the tumor cells to adoptive cell transfer therapy (ACT). TNF-a secreted by macrophages and T cells in the tumor microenvironment can induce an interconversion between differentiated and Lenti-Hras-shp53 dedifferentiated melanoma tumor cells [34]. This induced inflammation plasticity allows the tumor cells to escape immune surveillance Fig 2. Parallel between glioma cancer stem cells and induced and in the context of ACT explains relapse of the tumors after the pluripotent stem cell. initial remission. Changes in the differentiation state of the tumor An astrocyte transduced with LV-Hras-shp53 dedifferentiates/reprograms to a cells also change the lineage markers expressed on the surface of Oligodendrocyte Endothelial cell progenitor/stem cell state, leading to tumorsphere formation. These Neuron tumorspheres when transplanted orthotopically in the brain of mice lead to these tumor cells, generating a mechanism of resistance and escape Astrocyte glioma tumors. The same astrocytes transduced with a lentivector carrying the from cytotoxic T-cell recognition. four transcription factors (Oct-4, Klf-4, Sox2, and myc) reprogram and form iPS colonies than when transplanted s.c. into mice develop teratomas. Self- renewing Versatility of cancer stem cells: transdifferentiation GBM Cancer stem cell site, dedifferentiate and create a new pool of CSCs, hence becoming Stem cells have acquired the plasticity of not only self-sustenance founders of new metastatic colonies. and differentiation into expected lineages, but can also transdiffer- The study from Oleksi Petrenko’s laboratory used a model of entiate into other lineages and cell types. Despite the fact that GBMs conditional expression of oncogenic KrasG12D in mice to show the are highly vascularized, inhibitors of angiogenesis, like avastin, are phenotypic changes at the molecular and cellular level required in not very effective [35,36]. Several laboratories have discovered that the process of transformation. Their results support the concept of in GBMs, the cancer cells can transdifferentiate to endothelial cells ? tumor plasticity and the existence of a controlled balance between (TDECs), leading to the formation of blood vessels, which are differentiated tumor cells and tumor cells in a stem-like state [28]. functional [21,37,38]. In some human GBMs, over 70% of the The acquisition of CSC characteristics induced by oncogenic Ras endothelial cells forming the blood vessels were derived from cancer and one of its downstream targets, Myc, is essential to initiate the cells [37]. More recently, it has been shown that in human GBMs, malignant transformation. According to their results, genome repro- the cancer cells can transdifferentiate to pericytes [39]. In some gramming and dedifferentiation are important early steps in pancre- ways, it is not surprising, because in mice, NSCs have been shown atic tumor initiation, progression, and even predispose early-stage to differentiate into endothelial cells which then presumably provide pancreatic tumor cells to metastatic spread. Activating Kras muta- the niche for NSCs to differentiate to various cells of CNS lineage tion occurs at a frequency of 90% in pancreatic cancer and is the [40]. In the context of glioma tumors, the transdifferentiation pro- TDEC most frequent mutation among all cancers [29–32]. Kras signals via cess seems to be reversible, with tumor endothelial cells being able downstream effector pathways, and in their study, the transcription to reverse to a CSC state (Yasushi Soda and Amy Rommel, personal Transformed Transformed neurons factor Myc is implicated in the regulation of self-renewal and devel- communication; Fig 3). Clearly, the emphasis will now be to under- oligodendrocytes Transformed opment of metastatic pancreatic cancer cells. Collectively, their data stand the molecular mechanisms of transdifferentiation. Although astrocytes support the studies described before in this section, wherein any cell there is not as yet evidence of transdifferentiation of CSCs to other in the tumor, regardless of its differentiation state, has the potential cell types, it may be prudent to further analyze other cell types, to become a CSC given that an appropriate oncogenic insult induces especially the stromal cells surrounding the tumors, some of which Fig 3. A model for the generation of malignant gliomas. this plasticity. The big challenge in the field now will be to elucidate may actually be of tumor origin. Normal mechanism of neuronal differentiation: Neural stem cell can self-renew, go through an intermediate progenitor cell, and differentiate into oligodendrocytes, the mechanism responsible for this process. astrocytes, neurons, and endothelial cells. In the formation of glioblastoma, the transformed neurons, astrocytes, and possibly oligodendrocytes can dedifferentiate/ Although we have focused so far on tumor reprogramming as a reprogram to become cancer stem cells (CSCs), which can then continue to self-proliferate and differentiate to more transformed neurons and astrocytes. The transformed consequence mostly of cell-intrinsic changes, the tumor microenvi- Parallels between reprogramming of somatic cells and neurons and astrocytes can also transdifferentiate into endothelial cells (TDECs), which can again dedifferentiate to CSCs. ronment plays a very important role in this process. To this extent, dedifferentiation in cancer two independent studies showed that cancer stemness can be regulated by extrinsic factors generated in the tumor microenvironment. The concept that differentiated cells become “dedifferentiated” in where a more specialized cell type looses expression of lineage- generation also implies a reversion from a differentiated state to a The study of Jean Paul Medema’s group suggested that colorectal cancer has been controversial, but the latest studies described above specific genes of specialized tissue function, in favor of expression pluripotent state, and following this reprogramming, the differenti- stroma cells surrounding the tumor, specifically myofibroblasts, by have provided sufficient data supporting the existence of this pro- of the more primitive signature of the related tissue development. ated somatic cells acquire unlimited proliferating properties and secreting hepatocyte growth factor can induce nuclear b-catenin cess and even the requirement of this transition for tumorigenesis. Indeed, in some types of cancer, the alterations in differentiated self-renewal capacity. As discussed below, this is only one of the localization, Wnt signaling activity, and the generation of stem cell In this context, dedifferentiation should apply only to a situation cells result in reversion to a stem cell phenotype. The process of iPS many characteristics that iPSCs share with cancer development.

ª 2014 The Authors EMBO reports Vol 15 | No 3 | 2014 247 248 EMBO reports Vol 15 | No 3 | 2014 ª 2014 The Authors Dinorah Friedmann-Morvinski & Inder M Verma Dedifferentiation and reprogramming: origins of cancer stem cells EMBO reports EMBO reports Dedifferentiation and reprogramming: origins of cancer stem cells Dinorah Friedmann-Morvinski & Inder M Verma

Oncogenic transformation frequently involves de novo acquisi- required to achieve nuclear reprogramming, evidenced by the loss eliminated. Because stem cells have unique expression of genes tion of developmental programs, analogous to cellular reprogram- of promoter methylation in key pluripotency genes during iPSC required for self-perpetuation, like chromatin remodelers, perhaps Sidebar A. In need of answers ming, and yields cells with unlimited self-renewal potential, a generation [3]. Several studies have reported de novo DNA methyla- they could be the targets of drug therapy. In preliminary experi- (i) Do all cancers have CSCs? feature shared with iPSCs. This implies that similar pathways can tion during reprogramming of differentiated cells to iPSC. DNA ments, we have shown that if GBM cells are transduced with an (ii) Do all CSCs originate by dedifferentiation or be associated with both the induction of pluripotency and oncogene- methyltransferases (DNMTs) are involved in the establishment and shRNA targeting Bmi1, an essential gene for self-replication, upon reprogramming? sis. The appropriation of specific ESC-associated regulators and gene maintenance of DNA methylation, and high expression of DNMTs transplantation, these cells are unable to form tumors. In GBMs, (iii) Does the microenvironment influence the maintenance or frequency expression pathways by poorly differentiated solid tumors has been has been reported during the induction of reprogramming as well additional molecules that prevent differentiation to various CNS of CSCs? described [30]. Indeed, molecular analysis of gene sets associated as in ESCs [53]. Tet1 proteins facilitate the hydroxylation of 5-meth- lineages can also be targets of therapy, though blocking normal (iv) What is the mechanism of transdifferentiation? with ESC identity in various human tumor types highlights the fact ylcytosine (5mC) to 5-hydroxymethylcytosine (5hmC), and this differentiation process in the brain may also have deleterious side (v) Do CSCs contribute toward tumor invasiveness? (vi) What is the role of epigenetic changes in tumor plasticity? that tumorigenesis can hijack embryonic pathways of tissue devel- modification in DNA methylation plays an important role during effects. In case of GBMs, approaches to block transdifferentiation to opment. Another program that is being replayed in the evolution of the reprogramming process. The enrichment of 5hmC in the Oct-4 endothelial cells may also be another avenue to explore as a thera- primary tumors toward metastatic phenotypes and, as previously loci facilitates DNA demethylation and the transcriptional reactiva- peutic agent. In other CSCs, it will be very beneficial to find specific discussed, shared common transcription players in tumor dediffer- tion required for the induction of reprogramming by the core tran- cell surface proteins/receptors which can serve as therapeutic whether ESCs and iPSCs are distinct cell types. Some groups argued entiation is the EMT. The EMT and the reverse process, termed the scription factors (OKSM) [54]. Abnormal patterns of genomic targets. Identification of unique drugable targets (like kinases, tran- that these two populations are undistinguishable [64–66]; others mesenchymal–epithelial transition (MET), play central roles in methylation in cancer are characterized by global losses of genomic scription factors) in CSC will be very helpful in eliminating them have reported that they differ in their molecular signature [67–70], embryogenesis [25]. Some of the transcription factors orchestrating methylation and hypermethylation, predominantly in CpG islands, from the tumors. In addition to interfering with the proliferation of DNA methylation [51,70–72], and their potential for differentiation EMTs have been found to confer malignant traits. Furthermore, it a well-recognized epigenetic event in cancer [52]. Both inactivation CSCs, alteration of microenvironment should also be considered to [73]. Yamanaka’s group recently reported a subset of iPSC lines that has been shown previously that EMT can reprogram differentiated and higher expression of DNMTs have been reported in cancer, and prevent both formation and proliferation of CSC. Because many of have aberrant gene expression and defective potential in neural dif- mammary epithelial cells into a less differentiated epithelial stem previous studies have suggested that such altered expressions of these interfering strategies will also have an impact on normal stem ferentiation [74]. They performed a large-scale analysis of human cell with mesenchymal traits, establishing a link between EMT and DNMTs could partly explain the abnormal methylation patterns cells, it will be important to ensure relative safety of the treated iPSCs and ESCs and found that although they have overlapping vari- the acquisition of stem cell properties [25]. observed in cancer cells [55]. Chromatin regulators (CRs) have also patients. ations in gene expression and DNA methylation, some iPSC clones When focusing on dedifferentiation processes and comparing been involved both in cellular reprogramming and in oncogenesis. retained a significant number of undifferentiated cells, even after those with dedifferentiation leading to tumor cells, and the relevant Like the transcription factors described before, CRs have also been neural differentiation, and formed teratomas when transplanted in role that CSCs play in tumor malignancy and growth, it is inescap- implicated in tumorigenesis, either acting as oncogenes or as tumor The risks and limitations of iPS-based cell therapy vivo. These differentiation-defective iPSC clones express high levels able to appreciate the similarities between somatic cell reprogram- suppressor genes. CRs are associated with both repressive and of LTRs of endogenous retroviruses and retain a substantial number ming and tumorigenesis. Each of the iPSC reprogramming factors active chromatin states. Epigenetic silencing is associated with the The first therapeutic success using iPSCs was reported for the mouse of undifferentiated cells after in vitro directed neural differentiation. has established roles in oncogenesis. Oct-4 plays a driving role in following histone modifications: H3K27me3, H3K9me2, and model of sickle cell anemia, a blood disorder disease [62]. The Clearly, prior to applications in regenerative medicine, these defec- initiating germ cell tumors and has been proposed to be a useful H3K9me3. As an example, inhibition of CRs that catalyze H3K9 defective b-globin gene was corrected by homologous recombina- tive iPSC clones need to be identified and eliminated. marker for germ cell tumors such as seminomas and embryonal methylation, including Suv39h1, Setdb1, and G9a, leads to a higher tion in a mutant iPSC line and the transplantation of these cells into carcinomas [41]. Although Oct-4 is highly expressed in seminomas, reprogramming efficiency [56,57] and all three have established the mutant mice cured the disease. This is a very good example and other non-germ-cell-originated tumors show detectable levels com- roles in different malignancies [58,59]. There are also several exam- model of iPSC-mediated regenerative medicine: a genetic disorder Concluding remarks pared to their normal cell counterparts, like breast carcinomas and ples of CRs involved in active chromatin states that play important disease caused by a single defective gene that can be corrected by papillary carcinomas of the thyroid [42], as well as esophageal cell roles both in reprogramming and in cancer (e.g., MLL and Dot1l replacement in autologous cells. The first limitation that comes to It has long been known that many cancer cells show markers and carcinoma [43] and prostate cancer [44]. The notion that Oct-4 [60]). In our GBM model system, another common feature between mind when thinking of autologous iPSC for individualized medicine properties of ESCs, and some of these have often been targets of induction affects epigenetic regulations and contributes to the main- our dedifferentiated tumorspheres and iPSCs is their chromatin is the associated high medical costs, the lack of large-scale culture therapy. The discovery of CSCs further points to this notion. In tenance of undifferentiated proliferating cells [45] may provide a state. It is well accepted that ESCs as well as iPSC have an “open” technologies, and the timeframe needed to prepare the cells for organismal development, events are deterministic and move for- possible link between transcription factor-mediated reprogramming chromatin, while differentiated cells have a “close” chromatin [10]. transplantation (crucial, for example, for spinal cord injuries). ward in one direction, without ability to reverse the process. The and oncogenesis. Sox2 is amplified in lung and esophagus cancer Using a qRT-PCR designed in our laboratory to detect highly repeti- Another important aspect when considering using iPSC in the clinics famous Waddington‘s landscape for development visualizes the and is an essential driver of CSCs subpopulations in GBM, breast tive DNA elements in the murine pericentric heterochromatin (e.g., is the quality of these cells, mostly derived from somatic cells of developmental history of a cell in an embryo, “by a ball rolling cancer, and Ewing sarcoma [46,47]. A large variety of human malig- minor and major satellites [61]), we showed that both dedifferenti- aged individuals. The risk that comes with this source of cells is the down the ‘landscape’ making several ‘choices’ as to which way to nancies express high levels of MYC. Its expression may explain the ated tumorspheres and NSCs have a relaxed chromatin that resulted incidence of spontaneously occurring tumors, which commonly go—just as the developing embryo is influenced down certain observation that most of the mice generated with iPSC clones spon- in derepression of normally silenced genes in the heterochromatin increases exponentially with aging. Although it has been reported ‘paths’ by various genetic and environmental factors—and by the taneously developed tumors [48]. Myc is an important transcrip- regions [20]. that epigenetic changes and telomerase activity in cells of aging indi- time it reaches the bottom of the landscape, it will have made sev- tional regulator in ESC, and it significantly promotes the process of There are many parallels between reprogramming and cancer. viduals can be reversed during the reprogramming process [63], eral such choices”. The ball eventually lands at the bottom, signal- iPSC derivation. Its role as a global amplifier of gene expression not The similarities between the process of reprogramming cells to iPCS somatic mutations and chromosomal aberrations acquired by these ing that being pushed upward will be difficult, thereby hinting that surprisingly also drives a wide range of malignant programs [49]. and differentiated tumor cells to CSCs suggest that some of these cells are not corrected in the reprogramming process. These abnor- the process is essentially one directional. The discovery of Yama- The list can go on including KLF4, Nanog, Lin28, and other pluripo- mechanisms, like epigenetic resetting, can render cells in a suscepti- malities may lead to iPSCs with reduced functionality and higher naka and colleagues, however, shows that terminally differentiated tency factors and transcription factors that mediate direct lineage ble state where genetic alterations are only the next step toward risk of developing cancer. cells can be pushed upwards, going back to the original pluripo- conversion, emphasizing the link between reprogramming and transformation and tumor progression. Understanding the mecha- Other problems, mostly associated with the first generation of tent cell, a situation created by oncogenic insult to terminally dif- oncogenesis [50,51]. nisms governing cellular reprogramming and induced pluripotency human iPSCs, were the integration site of retroviral vectors, the risk ferentiated cells like glia or neuron in the case of GBM (Fig 3). Changes in the epigenetic landscape have also been implicated may shed light into deciphering the processes involved in of insertional mutagenesis and hence the risk of tumorigenicity, and The convergence and commonality of CSCs and iPSCs opens a in both reprogramming and oncogenic transformation. Epigenetics tumorigenesis. the use of undefined serum-containing media to support iPSC gener- new avenue to develop therapeutic approaches to combat recur- can be defined as the external modifications to DNA that regulate ation. In addition, the use of oncogenic transgenes, such as MYC, ring cancers. gene expression without changes in the underlying DNA sequence. can also increase the risk of tumor development. As mentioned Two major epigenetic regulations are DNA methylation and histone Prospects of eliminating cancer stem cells above, new and safer technologies for the generation of iPSCs have Acknowledgments modifications. DNA methylation is a relatively stable epigenetic emerged in the past few years that diminished these risks. We thank Jamie Simon for the graphic design. IMV is an American Cancer Soci- modification that mediates silencing of repetitive elements and Treatments of tumors, which have resident stem cells, will not be Although it is well accepted today that iPSCs are pluripotent, the ety Professor of Molecular Biology and holds the Irwin and Joan Jacobs Chair certain gene promoters [52]. Changes in DNA methylation are very effective, leading to recurrence, unless the stem cells are also findings in the past few years have been controversial in regard to in Exemplary Life Science. This work was supported in part by grants from the

ª 2014 The Authors EMBO reports Vol 15 | No 3 | 2014 249 250 EMBO reports Vol 15 | No 3 | 2014 ª 2014 The Authors Dinorah Friedmann-Morvinski & Inder M Verma Dedifferentiation and reprogramming: origins of cancer stem cells EMBO reports EMBO reports Dedifferentiation and reprogramming: origins of cancer stem cells Dinorah Friedmann-Morvinski & Inder M Verma

NIH (HL053670), Cancer Center Core Grant (P30 CA014195-38), Ipsen, the 18. Visvader JE (2011) Cells of origin in cancer. Nature 469: 314 – 322 35. Jain RK, di Tomaso E, Duda DG, Loeffler JS, Sorensen AG, Batchelor TT reveals an important role of DNA methylation and hydroxymethylation Leona M. and Harry B. Helmsley Charitable Trust, and the H.N. and Frances C. 19. Bachoo RM, Maher EA, Ligon KL, Sharpless NE, Chan SS, You MJ, Tang (2007) Angiogenesis in brain tumours. Nat Rev Neurosci 8: 610 – 622 in reprogramming. Cell Stem Cell 12: 453 – 469 Berger Foundation. The content of this report is solely the responsibility of the Y, DeFrances J, Stover E, Weissleder R et al (2002) Epidermal growth 36. Vredenburgh JJ, Desjardins A, Herndon JE 2nd, Dowell JM, Reardon DA, 55. Linhart HG, Lin H, Yamada Y, Moran E, Steine EJ, Gokhale S, Lo G, Cantu authors and does not necessarily represent the official views of the National factor receptor and Ink4a/Arf: convergent mechanisms governing termi- Quinn JA, Rich JN, Sathornsumetee S, Gururangan S, Wagner M et al E, Ehrich M, He T et al (2007) Dnmt3b promotes tumorigenesis in vivo Institutes of Health. nal differentiation and transformation along the neural stem cell to (2007) Phase II trial of bevacizumab and irinotecan in recurrent malig- by gene-specific de novo methylation and transcriptional silencing. astrocyte axis. Cancer Cell 1: 269 – 277 nant glioma. Clin Cancer Res 13: 1253 – 1259 Genes Dev 21: 3110 – 3122 Conflict of interest 20. Friedmann-Morvinski D, Bushong EA, Ke E, Soda Y, Marumoto T, Singer 37. Ricci-Vitiani L, Pallini R, Biffoni M, Todaro M, Invernici G, Cenci T, Maira 56. Onder TT, Kara N, Cherry A, Sinha AU, Zhu N, Bernt KM, Cahan P, Mar- The authors declare that they have no conflict of interest. O, Ellisman MH, Verma IM (2012) Dedifferentiation of neurons and as- G, Parati EA, Stassi G, Larocca LM et al (2010) Tumour vascularization carci BO, Unternaehrer J, Gupta PB et al (2012) Chromatin-modifying trocytes by oncogenes can induce gliomas in mice. Science 338: via endothelial differentiation of glioblastoma stem-like cells. Nature enzymes as modulators of reprogramming. Nature 483: 598 – 602 1080 – 1084 468: 824 – 828 57. Chen J, Liu H, Liu J, Qi J, Wei B, Yang J, Liang H, Chen Y, Wu Y, Guo L References 21. Soda Y, Marumoto T, Friedmann-Morvinski D, Soda M, Liu F, Michiue H, 38. Wang R, Chadalavada K, Wilshire J, Kowalik U, Hovinga KE, Geber A, et al (2013) H3K9 methylation is a barrier during somatic cell repro- Pastorino S, Yang M, Hoffman RM, Kesari S et al (2011) Transdifferenti- Fligelman B, Leversha M, Brennan C, Tabar V (2010) Glioblastoma gramming into iPSCs. Nat Genet 45: 34 – 42 1. Kaiser LR (1992) The future of multihospital systems. Top Health Care ation of glioblastoma cells into vascular endothelial cells. Proc Natl Acad stem-like cells give rise to tumour endothelium. Nature 468: 829 – 833 58. Shinkai Y, Tachibana M (2011) H3K9 methyltransferase G9a and the Financ 18: 32 – 45 Sci USA 108: 4274 – 4280 39. Cheng L, Huang Z, Zhou W, Wu Q, Donnola S, Liu JK, Fang X, Sloan AE, Mao related molecule GLP. Genes Dev 25: 781 – 788 2. Lysaght MJ, Crager J (2009) Origins. Tissue Eng Part A 15: 1449 – 1450 22. Schwitalla S, Fingerle AA, Cammareri P, Nebelsiek T, Goktuna SI, Ziegler Y, Lathia JD et al (2013) Glioblastoma stem cells generate vascular peri- 59. Lakshmikuttyamma A, Scott SA, DeCoteau JF, Geyer CR (2010) Reexpres- 3. Takahashi K, Yamanaka S (2006) Induction of pluripotent stem cells PK, Canli O, Heijmans J, Huels DJ, Moreaux G et al (2013) Intestinal cytes to support vessel function and tumor growth. Cell 153: 139 – 152 sion of epigenetically silenced AML tumor suppressor genes by SUV39H1 from mouse embryonic and adult fibroblast cultures by defined factors. tumorigenesis initiated by dedifferentiation and acquisition of stem-- 40. Wurmser AE, Nakashima K, Summers RG, Toni N, DAmour KA, Lie DC, inhibition. Oncogene 29: 576 – 588 Cell 126: 663 – 676 cell-like properties. Cell 152: 25 – 38 Gage FH (2004) Cell fusion-independent differentiation of neural stem 60. Suva ML, Riggi N, Bernstein BE (2013) Epigenetic reprogramming in 4. Yu J, Vodyanik MA, Smuga-Otto K, Antosiewicz-Bourget J, Frane JL, Tian 23. Karin M, Greten FR (2005) NF-kappaB: linking inflammation and immu- cells to the endothelial lineage. Nature 430: 350 – 356 cancer. Science 339: 1567 – 1570 S, Nie J, Jonsdottir GA, Ruotti V, Stewart R et al (2007) Induced pluripo- nity to cancer development and progression. Nat Rev Immunol 5: 41. Gidekel S, Pizov G, Bergman Y, Pikarsky E (2003) Oct-3/4 is a 61. Zhu Q, Pao GM, Huynh AM, Suh H, Tonnu N, Nederlof PM, Gage FH, tent stem cell lines derived from human somatic cells. Science 318: 749 – 759 dose-dependent oncogenic fate determinant. Cancer Cell 4: 361 – 370 Verma IM (2011) BRCA1 tumour suppression occurs via heterochroma- 1917 – 1920 24. Gupta PB, Chaffer CL, Weinberg RA (2009) Cancer stem cells: mirage or 42. Madjd Z, Hashemi F, Shayanfar N, Farahani E, Zarnani A, Shrifi A, Ak- tin-mediated silencing. Nature 477: 179 – 184 5. Briggs R, King TJ (1952) Transplantation of living nuclei from blastula reality? Nat Med 15: 1010 – 1012 bari M (2009) OCT-4, embryonic Stem cell marker expressed in breast, 62. Hanna J, Wernig M, Markoulaki S, Sun CW, Meissner A, Cassady JP, cells into enucleated frogs eggs. Proc Natl Acad Sci USA 38: 455 – 463 25. Mani SA, Guo W, Liao MJ, Eaton EN, Ayyanan A, Zhou AY, Brooks M, brain and thyroid carcinomas compared to testicular carcinoma. Iran J Beard C, Brambrink T, Wu LC, Townes TM et al (2007) Treatment of 6. Gurdon JB (1962) The developmental capacity of nuclei taken from Reinhard F, Zhang CC, Shipitsin M et al (2008) The epithelial-mesenchy- Cancer Prev 2: 167 – 173 sickle cell anemia mouse model with iPS cells generated from autolo- intestinal epithelium cells of feeding tadpoles. J Embryol Exp Morphol mal transition generates cells with properties of stem cells. Cell 133: 43. Zhou X, Huang GR, Hu P (2011) Over-expression of Oct4 in human gous skin. Science 318: 1920 – 1923 10: 622 – 640 704 – 715 esophageal squamous cell carcinoma. Mol Cells 32: 39 – 45 63. Marion RM, Strati K, Li H, Tejera A, Schoeftner S, Ortega S, Serrano M, 7. Sommer CA, Mostoslavsky G (2013) The evolving field of induced pluripo- 26. Chaffer CL, Marjanovic ND, Lee T, Bell G, Kleer CG, Reinhardt F, DAlessio 44. de Resende MF, Chinen LT, Vieira S, Jampietro J, da Fonseca FP, Vassal- Blasco MA (2009) Telomeres acquire embryonic stem cell characteristics tency: recent progress and future challenges. J Cell Physiol 228: 267 – 275 AC, Young RA, Weinberg RA (2013) Poised chromatin at the ZEB1 pro- lo J, Campos LC, Guimaraes GC, Soares FA, Rocha RM (2013) Prognosti- in induced pluripotent stem cells. Cell Stem Cell 4: 141 – 154 8. Malik N, Rao MS (2013) A review of the methods for human iPSC deri- moter enables breast cancer cell plasticity and enhances tumorigenic- cation of OCT4 isoform expression in prostate cancer. Tumour Biol 34: 64. Bock C, Kiskinis E, Verstappen G, Gu H, Boulting G, Smith ZD, Ziller M, vation. Methods Mol Biol 997: 23 – 33 ity. Cell 154: 61 – 74 2665 – 2673 Croft GF, Amoroso MW, Oakley DH et al (2011) Reference Maps of 9. Gonzalez F, Boue S, Izpisua Belmonte JC (2011) Methods for making 27. Thiery JP, Acloque H, Huang RY, Nieto MA (2009) Epithelial-mesenchy- 45. Hochedlinger K, Yamada Y, Beard C, Jaenisch R (2005) Ectopic expres- human ES and iPS cell variation enable high-throughput characteriza- induced pluripotent stem cells: reprogramming a la carte. Nat Rev mal transitions in development and disease. Cell 139: 871 – 890 sion of Oct-4 blocks progenitor-cell differentiation and causes dysplasia tion of pluripotent cell lines. Cell 144: 439 – 452 Genet 12: 231 – 242 28. Ischenko I, Zhi J, Moll UM, Nemajerova A, Petrenko O (2013) Direct in epithelial tissues. Cell 121: 465 – 477 65. Guenther MG, Frampton GM, Soldner F, Hockemeyer D, Mitalipova M, 10. Orkin SH, Hochedlinger K (2011) Chromatin connections to pluripotency reprogramming by oncogenic Ras and Myc. Proc Natl Acad Sci USA 110: 46. Riggi N, Suva ML, De Vito C, Provero P, Stehle JC, Baumer K, Cironi L, Jaenisch R, Young RA (2010) Chromatin structure and gene expression and cellular reprogramming. Cell 145: 835 – 850 3937 – 3942 Janiszewska M, Petricevic T, Suva D et al (2010) EWS-FLI-1 modulates programs of human embryonic and induced pluripotent stem cells. Cell 11. Mikkelsen TS, Hanna J, Zhang X, Ku M, Wernig M, Schorderet P, Bernstein 29. Guerra C, Schuhmacher AJ, Canamero M, Grippo PJ, Verdaguer L, miRNA145 and SOX2 expression to initiate mesenchymal stem cell Stem Cell 7: 249 – 257 BE, Jaenisch R, Lander ES, Meissner A (2008) Dissecting direct reprogram- Perez-Gallego L, Dubus P, Sandgren EP, Barbacid M (2007) Chronic pan- reprogramming toward Ewing sarcoma cancer stem cells. Genes Dev 24: 66. Newman AM, Cooper JB (2010) Lab-specific gene expression signatures ming through integrative genomic analysis. Nature 454: 49 – 55 creatitis is essential for induction of pancreatic ductal adenocarcinoma 916 – 932 in pluripotent stem cells. Cell Stem Cell 7: 258 – 262 12. Koche RP, Smith ZD, Adli M, Gu H, Ku M, Gnirke A, Bernstein BE, Meiss- by K-Ras oncogenes in adult mice. Cancer Cell 11: 291 – 302 47. Sarkar A, Hochedlinger K (2013) The sox family of transcription factors: 67. Chin MH, Mason MJ, Xie W, Volinia S, Singer M, Peterson C, Ambart- ner A (2011) Reprogramming factor expression initiates widespread tar- 30. Ben-Porath I, Thomson MW, Carey VJ, Ge R, Bell GW, Regev A, Weinberg versatile regulators of stem and progenitor cell fate. Cell Stem Cell 12: sumyan G, Aimiuwu O, Richter L, Zhang J et al (2009) Induced pluripo- geted chromatin remodeling. Cell Stem Cell 8: 96 – 105 RA (2008) An embryonic stem cell-like gene expression signature in 15 – 30 tent stem cells and embryonic stem cells are distinguished by gene 13. Chaffer CL, Brueckmann I, Scheel C, Kaestli AJ, Wiggins PA, Rodrigues poorly differentiated aggressive human tumors. Nat Genet 40: 499 – 507 48. Yamanaka S (2012) Induced pluripotent stem cells: past, present, and expression signatures. Cell Stem Cell 5: 111 – 123 LO, Brooks M, Reinhardt F, Su Y, Polyak K et al (2011) Normal and neo- 31. Gidekel Friedlander SY, Chu GC, Snyder EL, Girnius N, Dibelius G, Crowley future. Cell Stem Cell 10: 678 – 684 68. Marchetto MC, Yeo GW, Kainohana O, Marsala M, Gage FH, Muotri AR plastic nonstem cells can spontaneously convert to a stem-like state. D, Vasile E, DePinho RA, Jacks T (2009) Context-dependent transformation 49. Dang CV (2012) MYC on the path to cancer. Cell 149: 22 – 35 (2009) Transcriptional signature and memory retention of human-in- Proc Natl Acad Sci USA 108: 7950 – 7955 of adult pancreatic cells by oncogenic K-Ras. Cancer Cell 16: 379 – 389 50. Morris SA, Daley GQ (2013) A blueprint for engineering cell fate: current duced pluripotent stem cells. PLoS ONE 4: e7076 14. Roy S, Gascard P, Dumont N, Zhao J, Pan D, Petrie S, Margeta M, Tlsty 32. Anokye-Danso F, Trivedi CM, Juhr D, Gupta M, Cui Z, Tian Y, Zhang Y, technologies to reprogram cell identity. Cell Res 23: 33 – 48 69. Ghosh Z, Wilson KD, Wu Y, Hu S, Quertermous T, Wu JC (2010) Persis- TD (2013) Rare somatic cells from human breast tissue exhibit exten- Yang W, Gruber PJ, Epstein JA et al (2011) Highly efficient miRNA-medi- 51. Daley GQ (2008) Common themes of dedifferentiation in somatic cell tent donor cell gene expression among human induced pluripotent sive lineage plasticity. Proc Natl Acad Sci USA 110: 4598 – 4603 ated reprogramming of mouse and human somatic cells to pluripoten- reprogramming and cancer. Cold Spring Harb Symp Quant Biol 73: stem cells contributes to differences with human embryonic stem cells. 15. Tata PR, Mou H, Pardo-Saganta A, Zhao R, Prabhu M, Law BM, Vinarsky cy. Cell Stem Cell 8: 376 – 388 171 – 174 PLoS ONE 5: e8975 V, Cho JL, Breton S, Sahay A et al (2013) Dedifferentiation of committed 33. Vermeulen L, De Sousa EMF, van der Heijden M, Cameron K, de Jong JH, 52. Baylin SB, Jones PA (2011) A decade of exploring the cancer epige- 70. Ohi Y, Qin H, Hong C, Blouin L, Polo JM, Guo T, Qi Z, Downey SL, Manos epithelial cells into stem cells in vivo. Nature 503: 218 – 223 Borovski T, Tuynman JB, Todaro M, Merz C, Rodermond H et al (2010) nome—biological and translational implications. Nat Rev Cancer 11: PD, Rossi DJ et al (2011) Incomplete DNA methylation underlies a tran- 16. van Es JH, Sato T, van de Wetering M, Lyubimova A, Nee AN, Gregorieff Wnt activity defines colon cancer stem cells and is regulated by the 726 – 734 scriptional memory of somatic cells in human iPS cells. Nat Cell Biol 13: A, Sasaki N, Zeinstra L, van den Born M, Korving J et al (2012) Dll1 + microenvironment. Nat Cell Biol 12: 468 – 476 53. Stadtfeld M, Nagaya M, Utikal J, Weir G, Hochedlinger K (2008) Induced 541 – 549 secretory progenitor cells revert to stem cells upon crypt damage. Nat 34. Landsberg J, Kohlmeyer J, Renn M, Bald T, Rogava M, Cron M, Fatho M, pluripotent stem cells generated without viral integration. Science 322: 71. Deng J, Shoemaker R, Xie B, Gore A, LeProust EM, Antosiewicz-Bourget J, Cell Biol 14: 1099 – 1104 Lennerz V, Wolfel T, Holzel M et al (2012) Melanomas resist T-cell ther- 945 – 949 Egli D, Maherali N, Park IH, Yu J et al (2009) Targeted bisulfite sequenc- 17. Visvader JE, Lindeman GJ (2012) Cancer stem cells: current status and apy through inflammation-induced reversible dedifferentiation. Nature 54. Gao Y, Chen J, Li K, Wu T, Huang B, Liu W, Kou X, Zhang Y, Huang H, ing reveals changes in DNA methylation associated with nuclear repro- evolving complexities. Cell Stem Cell 10: 717 – 728 490: 412 – 416 Jiang Y et al (2013) Replacement of Oct4 by Tet1 during iPSC induction gramming. Nat Biotechnol 27: 353 – 360

ª 2014 The Authors EMBO reports Vol 15 | No 3 | 2014 251 252 EMBO reports Vol 15 | No 3 | 2014 ª 2014 The Authors Dinorah Friedmann-Morvinski & Inder M Verma Dedifferentiation and reprogramming: origins of cancer stem cells EMBO reports EMBO reports Dedifferentiation and reprogramming: origins of cancer stem cells Dinorah Friedmann-Morvinski & Inder M Verma

72. Lister R, Pelizzola M, Kida YS, Hawkins RD, Nery JR, Hon G, Anto- inhibition on proliferation of human gastric carcinoma cells. Oncogene siewicz-Bourget J, OMalley R, Castanon R, Klugman S et al (2011) Hot- 30: 956 – 966 spots of aberrant epigenomic reprogramming in human induced 74. Koyanagi-Aoi M, Ohnuki M, Takahashi K, Okita K, Noma H, Sawamura pluripotent stem cells. Nature 471: 68 – 73 Y, Teramoto I, Narita M, Sato Y, Ichisaka T et al (2013) Differentia- 73. Asciutti S, Akiri G, Grumolato L, Vijayakumar S, Aaronson SA (2011) tion-defective phenotypes revealed by large-scale analyses of human Diverse mechanisms of Wnt activation and effects of pathway pluripotent stem cells. Proc Natl Acad Sci USA 110: 20569 – 20574

ª 2014 The Authors EMBO reports Vol 15 | No 3 | 2014 253 Scientific Report

E2F1 induces miR-224/452 expression to drive EMT through TXNIP downregulation

Susanne Knoll1, Katharina Fürst1, Bhavani Kowtharapu1, Ulf Schmitz2, Stephan Marquardt1, Olaf Wolkenhauer2, Hubert Martin3 & Brigitte M Pützer1,*

Abstract uncontrolled cell proliferation and DNA damage, initiate activation of E2F1 as a mediator of apoptosis. However, E2F1 loses its tumor Malignant melanoma is highly lethal due to its aggressive invasive suppressor function in highly aggressive, apoptosis-resistant tumor properties and metastatic dissemination. The transcription factor types like malignant melanoma and contributes to tumor progres- E2F1 is crucial for melanoma progression through poorly under- sion [3,6,7]. stood mechanisms. Here, we show that the miR-224/miR-452 clus- The exact mechanism of E2F1-induced metastasis is not comple- ter is significantly increased in advanced melanoma and invasive/ tely understood. A promising approach to elucidate the molecular metastatic cell lines that express high levels of E2F1. miR-224/miR- issues involved in the acquisition of an enhanced aggressive pheno- 452 expression is directly activated by E2F1 through transactiva- type essential for E2F1-related tumor progression and metastasis of tion of the GABRE gene. Ectopic expression of miR-224/miR-452 in melanoma is the investigation of dysregulated microRNAs (miRs). less aggressive cells induces EMT and cytoskeletal rearrangements MiRs represent a large class of non-protein-coding RNAs, which act and enhances migration/invasion. Conversely, miR-224/miR-452 as negative gene regulators [8]. They bind their target mRNA depletion in metastatic cells induces the reversal of EMT, inhibition sequence either with perfect or imperfect complementary, resulting of motility, loss of the invasive phenotype and an absence of lung in RNA degradation or inhibition of translation [8]. Detailed descrip- metastases in mice. We identify the metastasis suppressor TXNIP tions about miRNA biogenesis and mode of action are given in as new target of miR-224/miR-452 that induces feedback inhibition numerous reviews. According to bioinformatic analysis, each of E2F1 and show that miR-224/452-mediated downregulation of miRNA is able to control hundreds of target mRNAs, allowing this TXNIP is essential for E2F1-induced EMT and invasion. The E2F1- class to manipulate almost every cellular scenario [9]. MiRs are key miR-224/452-TXNIP axis constitutes a molecular signature that players in carcinogenesis, as they participate in events that are predicts patient survival and may help to set novel therapies. impaired in cancer, for example proliferation and apoptosis [10]. Since different reports hint toward an interaction of oncogenic Keywords E2F1 transcription factor; epithelial-mesenchymal transition; miRNAs and E2F1 [11], an involvement of miRNAs in E2F1-induced melanoma metastasis; miRNA cluster; thioredoxin-interacting protein cancer metastasis is conceivable. To date, numerous miRs with ther- Subject Categories Cancer; Cell Adhesion, Polarity & Cytoskeleton; apeutic potential have been identified in a variety of tumor types RNA Biology [12]. An interesting example is miR-224, which is well known for its DOI 10.15252/embr.201439392 | Received 1 September 2014 | Revised 18 tumorigenic function in hepatocellular carcinoma (HCC) [13]. This September 2014 | Accepted 19 September 2014 miRNA is significantly upregulated during HCC development [14,15] and contributes to tumor-relevant processes like migration, invasion, proliferation and apoptosis inhibition [13,16]. In line with these reports, miR-224 is also increased in other tumor types, like Introduction clear cell renal cell carcinoma (ccRCC) [17,18], colorectal cancer (CRC) [19–21] and glioma [22], having an impact on tumor progres- Malignant melanoma as a highly aggressive tumor is characterized sion. Nevertheless, miR-224 overexpression is also described as an by strong metastasis and a pronounced chemoresistance [1,2]. indication for a more favorable prognosis in medullary thyroid carci- Reasons for the poor therapeutic suggestibility in advanced stages of noma (MTC) patients, since tumors with high content of miR-224 this cancer type are defects in apoptotic signaling pathways [2]. Our did not show any node metastases but biochemical healing in post- group has shown first that the E2F1 transcription factor (TF) is key treatment [23]. Similar findings apply to medulloblastoma, particu- to driving metastasis of melanoma cells [3]. E2F1 acts as tumor larly as enhanced miR-224 expression improves responsiveness to suppressor or oncogene [4,5]. Cellular stress signals, such as radiation and decreases proliferation as well as the transforming

1 Institute of Experimental Gene Therapy and Cancer Research, Rostock University Medical Center, Rostock, Germany 2 Systems Biology and Bioinformatics, University of Rostock, Rostock, Germany 3 Department of Neuropathology, University Hospital Charité, Berlin, Germany *Corresponding author. Tel: +49 381 494 5066/68; Fax: +49 381 494 5062; E-mail: [email protected]

ª 2014 The Authors EMBO reports 1 EMBO reports E2F1-miR-224/452-TXNIP axis promotes EMT Susanne Knoll et al Susanne Knoll et al E2F1-miR-224/452-TXNIP axis promotes EMT EMBO reports

carcinoma (OSCC) [25]. However, apart from the last-mentioned (ER) were treated with tamoxifen (4-OHT). In parallel, we examined 20 40 40 publications, there are considerably more, suggesting a tumorigenic miR-224/452 expression in invasive/metastatic SK-Mel-147 cells 1 15 30 30 function of miR-224. Interestingly, miR-224 is part of a miRNA clus- after E2F1 knockdown using lentiviral pLKO.1-shE2F1 that ensures miR-224 20 10 20 0,5

ter [24]. Together with miR-452, it is located intronic in the GABRE efficient depletion of E2F1 in melanoma cells. E2F1 activation and normalized

for bladder cancer determination [33]. Moreover, this miRNA is Mel-147 cells (Fig 1C, top panel) showing elevated miR-224/452 -20 highly expressed in esophageal cancer tissue [34], in prostate cancer levels (Fig 1A). Furthermore, decreased expression of the host gene stem/progenitor cells [35], as well as in neural crest cells [36]. In the after E2F1 ablation in invasive/metastatic SK-Mel-147 and upregula- latter, miR-452 was shown to have an influence on an epithelial- tion of GABRE upon E2F1 activation in non-metastatic SK-Mel- C D GABRE promoter E mesenchymal signaling pathway in the first pharyngeal arch [36]. 29.ER-E2F1 indicates its transcriptional co-regulation with miR-224/ * Regarding putative oncogenic activities of miR-224 in various 452 by E2F1 (Fig 1C, lower panel). In silico analysis of putative tran- +1 9,00E+06 malignancies and first hints about overexpression of miR-452 in scription factor binding sites in the GABRE promoter revealed three 8,00E+06

some kinds of human cancer, we investigated the role of the whole E2F elements in the region 812 bp upstream of the transcriptional GABRE~miR-224/452 cluster in melanoma progression. Here, we show that expression of start site (GABRE 1-759-770; GABRE 2-358-391 containing 2 E2F-motifs), 7,00E+06 E2F1 GABRE-1 GABRE-2 miR-224/452 in invasive/metastatic melanoma is controlled by which were confirmed using chromatin immunoprecipitation (ChIP) 6,00E+06 E2F1 causing a decrease of the metastasis suppressor TXNIP that (Fig 1D). To verify transcriptional activation by E2F1, the GABRE 5,00E+06 blocks E2F1 in a regulatory loop. Our results demonstrate a novel GABRE promoter region was cloned into the pGL3-basic reporter E2F1 binding site Actin transcription factor (E2F1)-miRNA axis that is activated during plasmid. Luciferase assay revealed a clear induction under E2F1 4,00E+06 melanoma progression and promotes reversible phenotypic changes co-transfection in a concentration-dependent manner, whereas RLU/mg protein 3,00E+06 toward epithelial-mesenchymal transition (EMT) and invasion. E2F1-mutants E(-TA) and E123 did not stimulate the promoter 2,00E+06 (Fig 1E). Thus, in advanced tumors with high levels of E2F1, the IP: IgG E2F1 SK-Mel-147 SK-Mel-29. transcription factor leads to the induction of miR-224/452 by trans- ER-E2F1 4OHT - + - + 1,00E+06 Results activating their host gene GABRE. GABRE-2 0,00E+00 E2F1 E2F1 induces miR-224/452 expression during MiR-224/452 cluster is an essential mediator of EMT and GABRE-1 Actin melanoma development melanoma invasion E2F1 Apaf-1 GABRE First, we analyzed the expression levels of potentially oncogenic According to the cellular context, miR-224 and miR-452 show onco- Input GAPDH E2F1 miRs in primary and metastatic patient samples and found a high genic as well as tumor suppressive properties. Therefore, functional content of miR-224 and miR-452 in the latter (Fig 1A, upper panel). analysis of the entire cluster is required to clearly assign potential These results were confirmed in established clinically relevant mela- cancer-promoting effects to one specific miR or the cooperation of Figure 1.E2F1 regulates miR-224 and miR-452 by transactivating their host gene GABRE. noma cell systems including SK-Mel-28, SK-Mel-29, SK-Mel-103 and both. To this end, we constructed lentiviral vectors for efficient and A Quantification of E2F1, miR-224 and miR-452 expression determined by real-time PCR and TaqMan® MicroRNA single assays in primary tumor samples and non- SK-Mel-147 (Fig 1A, lower panel). As described by Alla et al [3], the stable expression of individual candidates or the entire cluster in invasive melanoma cell lines (SK-Mel-28, SK-Mel-29) versus metastases and highly metastatic SK-Mel-103 and SK-Mel-147 cells. behavior of non-invasive SK-Mel-28/SK-Mel-29 and invasive/meta- SK-Mel-29 (Fig 2) and SK-Mel-28 cells (Supplementary Fig S2). BE2F1 activates miR-224/452. Normalized miRNA expression is shown in stable SK-Mel-29.ER-E2F1 cells treated with 4OHT or solvent (control; middle panel) static SK-Mel-103/SK-Mel-147 cells that correspond to primary Functional assays were performed in comparison with parental compared to parental SK-Mel-29, and in SK-Mel-147 cells expressing control or E2F1-specific shRNA. E2F1, ER-E2F1 and survivin expression was verified by Western blot. Actin served as a loading control. tumors and metastases with low or high E2F1 expression, respec- control cell lines expressing a randomly generated scrambled C MiR-224/452 cluster is co-expressed with its host gene GABRE. GABRE and E2F1 expression in metastatic (SK-Mel-103 and -147) versus non-metastatic (SK-Mel-28 tively, is clearly E2F1 dependent and thus represents an optimal sequence (Scr). MiR expression levels were validated by and -29) cells was verified by Western blot. GABRE transcript levels were analyzed after E2F1 knockdown in SK-Mel-147 and activation in SK-Mel-29.ER-E2F1. model system to study the interaction between the transcription TaqManMicroRNA single assays (Fig 2A). Stable expression of miR- D Scheme of the GABRE promoter showing relevant E2F-binding sites (GABRE-1, GABRE-2) upstream of the transcriptional start site (+1). Binding of E2F1 was verified by ChIP in SK-Mel-29.ER-E2F1 cells (treated with 4OHT (+), or solvent ( )) using E2F1 or IgG antibody. APAF-1 promoter served as a positive control. factor and the identified miRs. 224/452 in non-invasive SK-Mel-29 resulted in increased migration À EE2F1 directly activates GABRE. Relative luciferase activities (RLU) measured 36 h after co-transfection of GABRE promoter construct and increasing amounts (0.5, 1, Since miR-224/452 expression correlates with E2F1 levels in and invasion (Fig 2B and C; Supplementary Fig S1A and B). The 2 lg) of E2F1 expression plasmid or E2F mutants E(-TA) and E123 (2 lg). Protein expression was verified by Western blot. Actin served as a loading control. primary versus metastatic patient samples as well as in invasive same effect was observed in miR-expressing SK-Mel-28 cells Data information: Bar graphs are represented as means SD, n = 3,*P ≤ 0.05, two-sided Student’s t-test. Box-whisker plots: Boxes indicate the 25 and 75% quartile Æ versus non-invasive melanoma cells, regulation of miR-224/452 by (Supplementary Fig S2). In addition, we analyzed the impact of surrounding the median (n = 4), *P ≤ 0.05, **P ≤ 0.01, two-sided Student’s t-test. The lines represent the minimum and maximum transcript levels. E2F1 is conceivable (Fig 1A). In order to verify the effect of E2F1 on miR-224/452 on epithelial-mesenchymal transition, a highly

2 EMBO reports ª 2014 The Authors ª 2014 The Authors EMBO reports 3 EMBO reports E2F1-miR-224/452-TXNIP axis promotes EMT Susanne Knoll et al Susanne Knoll et al E2F1-miR-224/452-TXNIP axis promotes EMT EMBO reports

capability of medulloblastoma cells [24]. Furthermore, underexpres- miR expression, SK-Mel-29.ER-E2F1 cells stably expressing E2F1 A B SK-Mel-29 SK-Mel-29.ER-E2F1 SK-Mel-147 n.s. * * sion of miR-224 was identified as a biomarker of oral squamous cell fused to the ligand-binding domain of the murine estrogen receptor 25 50 50 1,5 carcinoma (OSCC) [25]. However, apart from the last-mentioned (ER) were treated with tamoxifen (4-OHT). In parallel, we examined 20 40 40 publications, there are considerably more, suggesting a tumorigenic miR-224/452 expression in invasive/metastatic SK-Mel-147 cells 1 15 30 30 function of miR-224. Interestingly, miR-224 is part of a miRNA clus- after E2F1 knockdown using lentiviral pLKO.1-shE2F1 that ensures miR-224 20 10 20 0,5 ter [24]. Together with miR-452, it is located intronic in the GABRE efficient depletion of E2F1 in melanoma cells. E2F1 activation and normalized

expression level 5 10 10 gene, which encodes the epsilon subunit of the gamma-aminobutyric knockdown was confirmed by detection of the E2F1 target survivin. expression level relative normalized 0 0 0 acid (GABA) A receptor [24]. In most cases, closely arranged As shown in Fig 1B, the miR-224/452 cluster is strongly upregulated 0 miRs are under the control of common regulatory elements and upon E2F1 activation in SK-Mel-29.ER-E2F1 cells compared to -10 primary primary primary n.s. metastatic metastatic metastatic * * therefore possess similar biological functions [26]. MiR-452 is parental SK-Mel-29, whereas knockdown of the transcription factor 25 25 1,2 dysregulated during diabetic wound healing [27] and decreased in SK-Mel-147 cells led to decreased expression of both miRs. 20 20 1 E2F1 miR-452 miR-224 0,8 upon cigarette smoking in alveolar macrophages [28]. In the course MiRNAs are located inside or outside of genes. Intergenic miRs 15 15 * * 0,6 miR-452 of myoblast differentiation, a reduction of miR-452 was recorded are produced from their own transcriptional units. In contrast, 14 100 10 10 0,4 [29]. Furthermore, miR-452 is inhibited by the tumorigenic SOX2 miRNAs in introns of protein-coding genes are usually regulated by 12 expression level 5 5 80 relative normalized 0,2 gene in glioblastoma cells [30,31] and assumed to have a tumor- the gene promoter and processed with the precursor mRNA. This 10 0 0 0 suppressive potential in medulloblastomas [24]. Intriguingly, miR- often results in co-expression of miR and host gene [37]. Since 8 60 E2F1 ER-E2F1 E2F1 452 and miR-452* are described as prognostic markers in urothelial miR-224 and miR-452 are located in the gamma-aminobutyric acid 6 40 carcinoma, since strong expression of these miRs correlates with the (GABA) A receptor epsilon gene (GABRE), we examined whether 4 Survivin Survivin Survivin 20 incidence of lymph node metastases and an unfavorable prognosis this gene is regulated through E2F1 (Fig 1C). GABRE protein expres- 2 Actin Actin Actin for patients [32]. MiR-452 was further validated as a urinary marker sion clearly correlates with the level of E2F1 in SK-Mel-103 and SK- normalized expression level 0 0 for bladder cancer determination [33]. Moreover, this miRNA is Mel-147 cells (Fig 1C, top panel) showing elevated miR-224/452 -20 highly expressed in esophageal cancer tissue [34], in prostate cancer levels (Fig 1A). Furthermore, decreased expression of the host gene stem/progenitor cells [35], as well as in neural crest cells [36]. In the after E2F1 ablation in invasive/metastatic SK-Mel-147 and upregula- latter, miR-452 was shown to have an influence on an epithelial- tion of GABRE upon E2F1 activation in non-metastatic SK-Mel- C D GABRE promoter E mesenchymal signaling pathway in the first pharyngeal arch [36]. 29.ER-E2F1 indicates its transcriptional co-regulation with miR-224/ * Regarding putative oncogenic activities of miR-224 in various 452 by E2F1 (Fig 1C, lower panel). In silico analysis of putative tran- +1 9,00E+06 malignancies and first hints about overexpression of miR-452 in scription factor binding sites in the GABRE promoter revealed three 8,00E+06 some kinds of human cancer, we investigated the role of the whole E2F elements in the region 812 bp upstream of the transcriptional GABRE~miR-224/452 cluster in melanoma progression. Here, we show that expression of start site (GABRE 1-759-770; GABRE 2-358-391 containing 2 E2F-motifs), 7,00E+06 E2F1 GABRE-1 GABRE-2 miR-224/452 in invasive/metastatic melanoma is controlled by which were confirmed using chromatin immunoprecipitation (ChIP) 6,00E+06 E2F1 causing a decrease of the metastasis suppressor TXNIP that (Fig 1D). To verify transcriptional activation by E2F1, the GABRE 5,00E+06 blocks E2F1 in a regulatory loop. Our results demonstrate a novel GABRE promoter region was cloned into the pGL3-basic reporter E2F1 binding site Actin transcription factor (E2F1)-miRNA axis that is activated during plasmid. Luciferase assay revealed a clear induction under E2F1 4,00E+06 melanoma progression and promotes reversible phenotypic changes co-transfection in a concentration-dependent manner, whereas RLU/mg protein 3,00E+06 toward epithelial-mesenchymal transition (EMT) and invasion. E2F1-mutants E(-TA) and E123 did not stimulate the promoter 2,00E+06 (Fig 1E). Thus, in advanced tumors with high levels of E2F1, the IP: IgG E2F1 SK-Mel-147 SK-Mel-29. transcription factor leads to the induction of miR-224/452 by trans- ER-E2F1 4OHT - + - + 1,00E+06 Results activating their host gene GABRE. GABRE-2 0,00E+00 E2F1 E2F1 induces miR-224/452 expression during MiR-224/452 cluster is an essential mediator of EMT and GABRE-1 Actin melanoma development melanoma invasion E2F1 Apaf-1 GABRE First, we analyzed the expression levels of potentially oncogenic According to the cellular context, miR-224 and miR-452 show onco- Input GAPDH E2F1 miRs in primary and metastatic patient samples and found a high genic as well as tumor suppressive properties. Therefore, functional content of miR-224 and miR-452 in the latter (Fig 1A, upper panel). analysis of the entire cluster is required to clearly assign potential These results were confirmed in established clinically relevant mela- cancer-promoting effects to one specific miR or the cooperation of Figure 1.E2F1 regulates miR-224 and miR-452 by transactivating their host gene GABRE. noma cell systems including SK-Mel-28, SK-Mel-29, SK-Mel-103 and both. To this end, we constructed lentiviral vectors for efficient and A Quantification of E2F1, miR-224 and miR-452 expression determined by real-time PCR and TaqMan® MicroRNA single assays in primary tumor samples and non- SK-Mel-147 (Fig 1A, lower panel). As described by Alla et al [3], the stable expression of individual candidates or the entire cluster in invasive melanoma cell lines (SK-Mel-28, SK-Mel-29) versus metastases and highly metastatic SK-Mel-103 and SK-Mel-147 cells. behavior of non-invasive SK-Mel-28/SK-Mel-29 and invasive/meta- SK-Mel-29 (Fig 2) and SK-Mel-28 cells (Supplementary Fig S2). BE2F1 activates miR-224/452. Normalized miRNA expression is shown in stable SK-Mel-29.ER-E2F1 cells treated with 4OHT or solvent (control; middle panel) static SK-Mel-103/SK-Mel-147 cells that correspond to primary Functional assays were performed in comparison with parental compared to parental SK-Mel-29, and in SK-Mel-147 cells expressing control or E2F1-specific shRNA. E2F1, ER-E2F1 and survivin expression was verified by Western blot. Actin served as a loading control. tumors and metastases with low or high E2F1 expression, respec- control cell lines expressing a randomly generated scrambled C MiR-224/452 cluster is co-expressed with its host gene GABRE. GABRE and E2F1 expression in metastatic (SK-Mel-103 and -147) versus non-metastatic (SK-Mel-28 tively, is clearly E2F1 dependent and thus represents an optimal sequence (Scr). MiR expression levels were validated by and -29) cells was verified by Western blot. GABRE transcript levels were analyzed after E2F1 knockdown in SK-Mel-147 and activation in SK-Mel-29.ER-E2F1. model system to study the interaction between the transcription TaqManMicroRNA single assays (Fig 2A). Stable expression of miR- D Scheme of the GABRE promoter showing relevant E2F-binding sites (GABRE-1, GABRE-2) upstream of the transcriptional start site (+1). Binding of E2F1 was verified by ChIP in SK-Mel-29.ER-E2F1 cells (treated with 4OHT (+), or solvent ( )) using E2F1 or IgG antibody. APAF-1 promoter served as a positive control. factor and the identified miRs. 224/452 in non-invasive SK-Mel-29 resulted in increased migration À EE2F1 directly activates GABRE. Relative luciferase activities (RLU) measured 36 h after co-transfection of GABRE promoter construct and increasing amounts (0.5, 1, Since miR-224/452 expression correlates with E2F1 levels in and invasion (Fig 2B and C; Supplementary Fig S1A and B). The 2 lg) of E2F1 expression plasmid or E2F mutants E(-TA) and E123 (2 lg). Protein expression was verified by Western blot. Actin served as a loading control. primary versus metastatic patient samples as well as in invasive same effect was observed in miR-expressing SK-Mel-28 cells Data information: Bar graphs are represented as means SD, n = 3,*P ≤ 0.05, two-sided Student’s t-test. Box-whisker plots: Boxes indicate the 25 and 75% quartile Æ versus non-invasive melanoma cells, regulation of miR-224/452 by (Supplementary Fig S2). In addition, we analyzed the impact of surrounding the median (n = 4), *P ≤ 0.05, **P ≤ 0.01, two-sided Student’s t-test. The lines represent the minimum and maximum transcript levels. E2F1 is conceivable (Fig 1A). In order to verify the effect of E2F1 on miR-224/452 on epithelial-mesenchymal transition, a highly

conserved genetic program that enables epithelial tumor cells to in melanoma [41], is downregulated (Fig 2D). Furthermore, the miR-452-depleted SK-Mel-147 cells to invade Matrigel (Fig 3B; and its upregulation under knockdown of these miRs in metastatic migrate from the existing cell layer into surrounding tissues. This miR-224/452 cluster induces changes in the actin cytoskeleton Supplementary Fig S3A) and migrate across the scratched area SK-Mel-147 both on RNA (Fig 4B, upper panel) and protein level process includes loss of intercellular tight junctions and cell polarity, toward a more aggressive phenotype as evident from phalloidin/ (Fig 3C) was markedly reduced. A significantly impaired motility (Fig 4B, lower panel). Consistent with our data demonstrating that and the acquisition of mesenchymal properties [38–40]. As reported TRITC staining (Fig 2E). under miRZIP-224/452 expression was also observed in aggressive miR-224 and miR-452 are regulated by E2F1, knockdown of the previously, knockdown of endogenous E2F1 in metastatic melanoma To directly assess the activity of endogenous miR-224/452 on the SK-Mel-103 cells (Supplementary Fig S4). Similar to miR-224/452 transcription factor resulted in an increase of TXNIP, whereas target cells restored expression of the epithelial marker E-cadherin [3]. In migratory/invasive capacity in the presence of high E2F1 levels, overexpression studies in SK-Mel-29, the strongest effect was level declined in response to E2F1 activation through 4OHT addition accordance, non-invasive SK-Mel-29 cells that stably express miR- SK-Mel-147 cells were stably transduced with lentiviral vectors achieved by the ablation of both miRs. In addition, knockdown of (Fig 4C). Furthermore, luciferase activity of the pMiR-Report

overexpression, whereas ZEB2, shown to be a differentiation factor cells expressing control miRZIP-Scr, the ability of miR-224- and E-cadherin and downregulation of vimentin as key epithelial and 30 UTR together with miRZIP plasmids against miR-224/452 in SK- mesenchymal markers was also confirmed in miR-224/452-depleted Mel-147 cells and saw a comparably strong, in this case stimulating

SK-Mel-147 cells by immunofluorescence staining (Fig 3E). These effect on TXNIP 30 UTR reporter activity after miR ablation (Fig 4D, ACB * miR-224/452 knockdown cells showed clear changes in the actin right panel). MiR-224/452 expression was confirmed by TaqMan miR-452 miR-224 cytoskeleton toward a less aggressive phenotype (Fig 3E, lower Single Assays shown in Supplementary Fig S5. To further demon- 2 * 14 * panel). Whereas parental control cells exhibit actin-bearing strate a direct interaction of miR-224/452 with the TXNIP 30 UTR, 1,8 6000 12 membrane ruffles and thin F-actin filaments oriented parallel to the site-directed mutagenesis of the miR-224/452 binding sites was 1,6 n.s. major cell axis as a typical feature of migrating cells, such character- performed. These mutations completely abolished the repressive 5000 1,4 10 istic structures are not visible after depletion of miR-224/452. More- effects of miR-224 and miR-452 on TXNIP 30 UTR (Fig 4E) compared over, miR-depleted cells displayed actin staining predominantly at to the positive controls with intact miR-452 (left) and miR-224 4000 1,2 8 cell–cell contacts suggesting stronger cell–cell adhesions and (right) binding sites. In sum, these results indicate that TXNIP is a 1 3000 impaired motility [42]. Finally, in compliance with these morpho- major target of the E2F1-miR-224/452 module. 0,8 6 logical changes and the impaired migration/invasion, aggressive n-fold invasion 2000 0,6 4 melanoma cells completely failed to form metastases in vivo when Loss of TXNIP promotes EMT and invasion by the E2F1-miR-224/ 0,4 miR-224/452 expression is blocked (Fig 3F). Notably, while inva- 452 axis 1000 2 0,2 sive growth is largely abrogated after E2F1 depletion in SK-Mel- relative quantity of wound healing

normalized expression level 0 0 0 147.miR-Scr cells, those cells in which miR-224, miR-452 or both Based on the results that TXNIP is a target of miR-224/452 and their are re-expressed retain their aggressive behavior also in the absence function in melanoma progression, we speculated whether this -1000 of E2F1 (Fig 3G; Supplementary Fig S3B), pointing out their autono- oncogenic behavior is dependent on TXNIP inhibition. For this

mous oncogenic capabilities. The expression levels of miR-224/452 purpose, we expressed TXNIP (without 30 UTR) in metastatic mela- are shown in Supplementary Fig S3C. noma cells with low endogenous levels of this protein and analyzed changes in invasion and migration. Indeed, similar to the ablation of D E MiR-224/452 regulates metastasis suppressor TXNIP miR-224/452, overexpression of TXNIP in SK-Mel-147 led to an effi- E-cadherin Vimentin F-actin F-actin cient reduction of migrating/invading cells (Fig 5A and B; Supple- To further investigate the mechanism by which miR-224/452 mentary Fig S6A and B). Furthermore, Western blot analysis I promotes tumor progression, we analyzed putative targets of this indicated a gain of E-cadherin expression and loss of mesenchymal cluster (Fig 4A). In order to identify shared target genes of miR-224 markers at levels comparable to miR-224/452-depleted cells (Fig 5C). ZEB1 I and miR-452, we extracted from the starBase database (v1.0) [43] The effect of TXNIP on melanoma cell invasion became even Argonaute-target interaction sites that match computationally more apparent in miR-224/miR-452-positive SK-Mel-29 cells, which E-cadherin predicted target sites of one of the two miRNAs. The data in star- showed enhanced migratory properties (Fig 2). Ectopic overexpres- Vimentin Base are derived from high-throughput CLIP-Seq experiments. sion of TXNIP in these cells suppressed miR-224/452-mediated II II Those targets in which binding sites for both miRNAs exist were invasiveness to the level detected for SK-Mel-29.miR-Scr control Slug considered for further analysis. Furthermore, we extracted predicted cells (Fig 5D; Supplementary Fig S6C). mutual targets of both miRNAs from the miRror Suite [44], a Web Importantly, increased invasion of less aggressive melanoma Actin service that integrates predictions from twelve complementary algo- cells by overexpression of E2F1 was significantly reduced upon ZEB2 miR-224/452 miR-Scr rithms and computes targets that are regulated by a set of miRNAs knockdown of miR-224 and miR-452. As demonstrated in Fig 6A in a coordinated fashion. From these targets, we considered only and Supplementary Fig S7A, this is due to the recovery of TXNIP Actin those with approved AGO binding sites based on starBase. In total, expression upon miR-224/452 inhibition. In support of the rele- we received a set of 20 target genes (Supplementary Table S1) most vance of TXNIP downregulation during E2F1-miR-224/452-induced

Figure 2. MiR-224/452 cluster induces EMT, migration and invasion of less aggressive melanoma cells. likely being targeted by both miR-224 and miR-452. From this set, malignant progression, E2F1-induced invasion of SK-Mel-29 cells A Stable expression of miR-224 and miR-452 in SK-Mel-29 cells was verified by TaqMan® MicroRNA single assays. we selected those genes with relevance in cancer based on their is also markedly reduced when TXNIP (without 30 UTR) is B, C MiR-224 and miR-452 induce migration and invasion. Stable miR-expressing SK-Mel-29 cells were subjected to scratch (B) and Boyden chamber assay (C). associated Gene Ontology terms [45]. After GO filtering, a set of re-expressed (Fig 6B, left; Supplementary Fig S7B). Moreover, D MiR-224/452 induces EMT. Expression levels of epithelial (E-cadherin) and mesenchymal markers (vimentin, ZEB1 and Slug) as well as ZEB2 in stable targets were subjected for experimental validation (Fig 4A). The depletion of endogenously high E2F1 in SK-Mel-147 cells which is SK-Mel-29.miR-224, SK-Mel-29.miR-452 and SK-Mel-29.miR-224/452 were verified by Western blot compared to miR-Scr cells. Actin was used for equal loading. most notable one was thioredoxin-interacting protein TXNIP (also associated with the upregulation of TXNIP protein expression E MiR-224/452 expression leads to cytoskeleton rearrangements toward an aggressive phenotype. Stable miR-224/452-expressing SK-Mel-29 were stained for named as VDUP-1 or TBP-2), which is often downregulated in (Fig 6B, right) severely reduced their invasive capacity, whereas E-cadherin, vimentin (Cy3) and F-actin (Phalloidin/TRITC). Nuclei were counter-stained with DAPI, and fluorescence was visualized by confocal laser scanning microscopy (CLSM). Scale bar, 20 lm. cancer and known as a proapoptotic factor and metastasis suppres- knockdown of TXNIP in these cells is needed to increase invasivity Data information: Bar graphs are represented as means SD, n = 3 (A, B) or n = 5 (C), *P ≤ 0.05, two-sided Student’s t-test. sor [46,47]. Validation experiments revealed a clear inhibition of again (Fig 6B, right; Supplementary Fig S7B). Similar effects Æ TXNIP upon miR-224 and miR-452 overexpression in SK-Mel-29, were observed for the E2F1-mediated EMT phenotype where the

4 EMBO reports ª 2014 The Authors ª 2014 The Authors EMBO reports 5 EMBO reports E2F1-miR-224/452-TXNIP axis promotes EMT Susanne Knoll et al Susanne Knoll et al E2F1-miR-224/452-TXNIP axis promotes EMT EMBO reports

224/452 show loss of E-cadherin. These cells attain a mesenchymal encoding shRNAs against miR-224 (miRZIP-224) and miR-452 (miR- miR-224/452 in SK-Mel-147 caused an increase of epithelial E-cadherin construct containing the 30 UTR of TXNIP was vigorously reduced state associated with increased levels of Slug, ZEB1 and vimentin ZIP-452) to specifically ablate miRNA expression. Efficient knock- and ZEB2, whereas mesenchymal vimentin, ZEB1 and Slug decreased after miR-224 and/or miR-452 co-transfection in SK-Mel-29 cells that are substantially upregulated through miR-224 and/or miR-452 down of miR-224/452 expression is shown in Fig 3A. Compared to compared to miRZIP-Scr control cells (Fig 3D). Upregulation of (Fig 4D, left panel). In addition, we transfected pMiR-Report-TXNIP- overexpression, whereas ZEB2, shown to be a differentiation factor cells expressing control miRZIP-Scr, the ability of miR-224- and E-cadherin and downregulation of vimentin as key epithelial and 30 UTR together with miRZIP plasmids against miR-224/452 in SK- mesenchymal markers was also confirmed in miR-224/452-depleted Mel-147 cells and saw a comparably strong, in this case stimulating

C D A Figure 3. Depletion of miR-224/452 abolishes migration/invasion and lung metastasis. A Stable knockdown of miR-224/452 by specific miRZip anti-microRNA constructs was verified by TaqMan® MicroRNA single assays. miR-452 miR-224 ZIP-Scr ZIP-224 ◀ B, C Detection of the invasive and migratory potential of miR-224/452-depleted SK-Mel-147 cells by (B) Boyden chamber and (C) wound closure assay 24 h after seeding/scratching. Fold changes were calculated relative to SK-Mel-147.ZIP-Scr (set as 1). Representative images of three independent scratch assays for each ZIP- 1,2 ZEB1 miR are shown.

1 0h Vimentin D, E Ablation of miR-224/452 in SK-Mel-147 reverses EMT. Expression of the indicated proteins was measured by Western blot (D). Changes in E-cadherin (Cy3) and vimentin (Cy3) expression as well as cytoskeleton rearrangement (Phalloidin/TRITC) as visualized by confocal laser scanning microscopy (CLSM) (E). 0,8 E-cadherin F Knockdown of miR-224/452 leads to complete lack of pulmonary metastases. Representative hematoxylin and eosin sections of lungs (a–d). Visible metastatic tumor infiltrates with atypic large cell shape und nuclei with high mitotic activity in the miRZIP-Scr group (a, b). Normal lung structure without tumor infiltrates 0,6 Slug in mice (n = 5 per group) injected with miR-224/452 ablated SK-Mel-147 (c, d). Magnification (×20) of the marked area is shown (b, d). G MiR-224/452 expression is sufficient to induce tumor invasion. The effect of miR-224/452 overexpression in shE2F1-treated SK-Mel-147 was investigated by Boyden

24 h Actin 0,4 chamber assay. Fold changes of invasion were calculated relative to SK-Mel-147.ZIP-Scr (set as 1). E2F1 expression was detected by immunoblotting. Actin served as a loading control. ZEB2 0,2 Data information: Bar graphs are represented as means SD, n = 3 (A) or n = 5 (B, G), *P ≤ 0.05, two-sided Student’s t-test. Æ normalized expression level ZIP-Scr ZIP-452 Actin 0 E

n-fold invasion of primary melanomas, when they colonize the underlying dermis by inhibiting TXNIP, which contributes to skin cancer progression using expression data from a melanoma study (Xu et al 2008) (Fig 7E). This finding hints toward a putative therapeutic approach 0,2 available at OncomineTM database (Compendia Bioscience, Ann to avoid EMT of tumor cells either by knockdown of miR-224/452 Arbor, MI) [48] and own human samples of different Breslow depth. or addition of TXNIP. 0 II As shown in Fig 7A, E2F1 levels positively correlate with Breslow 24 h I depth of primary tumors (> 4 mm, high E2F1 expression versus F-actin < 1 mm, low E2F1 expression, P < 0.0053). Conversely, TXNIP Discussion expression and Breslow depth of invasion show an inverse I II correlation with low TXNIP levels at > 4 mm versus high levels at E2F1 is a critical factor for metastasis in melanoma. Since the exact F < 1 mm depth of invasion (P < 0.0039). This was confirmed by mechanisms are incompletely understood, we investigated the role RT–PCR (Fig 7B) and immunohistochemical analysis (Fig 7C) in a of miRs in E2F1-induced tumor progression. There are several

F-actin statistically significant number of own primary patient tumors with reports on miRs regulating this transcription factor, but except some low versus high Breslow Index. These data, clearly indicating a hints, it was less known about E2F1-induced miRs. Here, we uncov- statistically significant association between high E2F1 versus low ered the importance of E2F1-induced miRs for its oncogenic activity G TXNIP expression and the highest invasion depth of primary mela- in melanoma progression. We could show that the miR-224/miR- * * noma predictive for progression and severity of the disease, again 452 cluster is upregulated in invasive/metastatic melanoma and underscore the relevance of the uncovered E2F1-miR-224/452- controls as part of a gene regulatory module the stage of melanoma 2,5 * TXNIP axis and support that the components of this module can progression where an EMT-like switch takes place. The results 2 serve as novel prognostic markers for melanoma patients at an early demonstrate that expression of miR-224/miR-452 is necessary and 1,5 sh.control stage of metastasis initiation. sufficient to elicit an EMT phenotype and that this process is 1 sh.E2F1 induced by the E2F1 TF via direct activation of the GABRE host gene 0,5 E2F1 and TXNIP contribute to cancer progression in a regulatory and its intronic miRs. Increased miR-224/452 promotes oncogenic n-fold invasion feedback loop transition into a metastatic state by repressing the metastasis 0 Cell line i.v. Metastasis suppressor TXNIP. Interestingly, this factor in turn controls E2F1 E2F1 miRZip-Scr 4/5 (80%) Previously, TXNIP has been implicated in growth suppression in activity in a negative regulatory loop. Actin association with an increase of p16 expression and reduction of reti- As yet, there are only few studies on the whole cluster describing miRZip-224/452 0/5 (0%) noblastoma (RB) phosphorylation [49]. Since hypophosphorylated a correlation in miR-224/452 and GABRE expression, such as two RB leads to the inhibition of E2F1 function, we suspected a mutual recent publications on concomitant regulation of miR-224/452 and regulation of TXNIP and E2F1 to avoid uncontrolled E2F1 activity GABRE by epigenetic mechanisms, resulting in either high expres- by building a negative feedback loop. Immunoblotting demonstrated sion in HCC patients or low expression in prostate cancer [50,51]. Figure 3. a reduced expression of E2F1 in SK-Mel-103/-147 cells after TXNIP These findings already reflect the ability of both miRs to act as overexpression (Fig 7D). To identify the possible mechanism for the tumor suppressor or oncogene in a cell-context-dependent manner.

6 EMBO reports ª 2014 The Authors ª 2014 The Authors EMBO reports 7 EMBO reports E2F1-miR-224/452-TXNIP axis promotes EMT Susanne Knoll et al Susanne Knoll et al E2F1-miR-224/452-TXNIP axis promotes EMT EMBO reports

0h ZIP-Scr ZIP-224/452 E2F1-induced decrease of E-cadherin and increase of mesenchymal TXNIP-induced decrease of E2F1, we tested its influence on different markers was to some extend reversed through re-expression of regulators of E2F1 activity. In accordance with the observations TXNIP (Fig 6C, left). Furthermore, gain of epithelial and loss of made by Nishinaka et al, p16 increased and phospho-RB decreased mesenchymal markers in SK-Mel-147 cells ablated for E2F1 was to through TXNIP also in melanoma cells. The CDK inhibitor p16 B 1,2 * a significant part recovered by shTXNIP treatment (Fig 6C, right). inhibits different cyclin-dependent kinases, which are able to phos- 24 h E-cadherin Hence, loss of TXNIP in melanoma cells due to aberrant E2F1-miR- phorylate RB. We found that CDK4 was strongly downregulated 1 224/452 expression is crucial for the transcription factor to induce under these circumstances (Fig 7D). This argues that TXNIP indeed cancer progression. provides a negative feedback loop that prevents excessive E2F1 0,8 ZIP-Scr ZIP-224/452 Based on these findings, we compared the expression levels of activity. In line with elevated levels of CDK4, hyperphosphorylation 0,6 E2F1 and TXNIP in vivo at the stages of melanoma progression of RB and consecutive stimulation of E2F1 in both aggressive mela- where an EMT-like switch takes place, namely at the invasive front noma cell lines, miR-224/452 potentiates the accumulation of E2F1 Vimentin 0,4 0h n-fold invasion of primary melanomas, when they colonize the underlying dermis by inhibiting TXNIP, which contributes to skin cancer progression using expression data from a melanoma study (Xu et al 2008) (Fig 7E). This finding hints toward a putative therapeutic approach 0,2 available at OncomineTM database (Compendia Bioscience, Ann to avoid EMT of tumor cells either by knockdown of miR-224/452 Arbor, MI) [48] and own human samples of different Breslow depth. or addition of TXNIP. 0 II As shown in Fig 7A, E2F1 levels positively correlate with Breslow 24 h I depth of primary tumors (> 4 mm, high E2F1 expression versus F-actin < 1 mm, low E2F1 expression, P < 0.0053). Conversely, TXNIP Discussion expression and Breslow depth of invasion show an inverse I II correlation with low TXNIP levels at > 4 mm versus high levels at E2F1 is a critical factor for metastasis in melanoma. Since the exact F < 1 mm depth of invasion (P < 0.0039). This was confirmed by mechanisms are incompletely understood, we investigated the role RT–PCR (Fig 7B) and immunohistochemical analysis (Fig 7C) in a of miRs in E2F1-induced tumor progression. There are several

A B SK-Mel-29 SK-Mel-147 A B Figure 5. TXNIP repression is required for the oncogenic switch by SK-Mel-147 SK-Mel-147 miR-224/452. A, B Overexpression of TXNIP in aggressive melanoma cells with low * * 1,2 1,4 endogenous levels reduces migration and invasion. C Effect of ectopic TXNIP on EMT markers in SK-Mel-147 and SK-Mel-103 TXNIP 1 1,2 cells as indicated by Western blot. D MiR-224/452-induced invasion is inhibited by TXNIP. TXNIP expression 1 was determined by immunoblot using actin as a control. Fold changes Actin 0,8 were calculated relative to SK-Mel-29.miR-Scr (set as 1). 0,8 Data information: Bar graphs are represented as means SD, n = 3 (A) or 0,6 Æ 0,6 n = 5 (B, D), *P ≤ 0.05, two-sided Student’s t-test.

0,4 n-fold invasion 0,4

TXNIP 0,2 0,2 In the present study, functional analyses on miR-224/452 clearly

relative quantity of wound healing revealed tumorigenic actions of both effectors which are character- Actin 0 0 ized by the stimulation of migratory and invasive properties of less TXNIP TXNIP aggressive melanoma cells and a decrease of motility and invasion when miR-224/452 was depleted in aggressive cell lines as evident Actin Actin C D SK-Mel-29 SK-Mel-147 from the complete lack of lung metastases in mice. As re-introduction * of these miRs after E2F1 ablation in invasive/metastatic melanoma cells could recover their oncogenic effects, they are able to act SK-Mel-147 SK-Mel-29. * * ER-E2F1 1,00E+08 6,00E+08 independently of the transcription factor. The epithelial-mesenchymal transition is an important prerequi- 5,00E+08 8,00E+07 C SK-Mel-147 SK-Mel-103 site for metastatic cancer. As a process of epithelial plasticity, 4,00E+08 6,00E+07 it includes dissolution of epithelial cell–cell adhesions, actin 3,00E+08 n.s. cytoskeleton reorganization, as well as an increase in cell–matrix 4,00E+07 TXNIP TXNIP 2,00E+08 contacts, leading to enhanced migration and invasion [42]. It is well RLU/mg protein E2F1 E2F1 2,00E+07 RLU/mg protein 1,00E+08 ZEB1 ZEB1 known that miRs are able to influence EMT [52]. One of the first GAPDH GAPDH and well-described examples is the regulation of ZEB proteins by 0,00E+00 0,00E+00 E-cadherin E-cadherin miR-205 as well as the miR-200 family. Inhibition of the E-cadherin Vimentin Vimentin repressors ZEB1 and ZEB2 by these miRs results in the stabilization of an epithelial phenotype of cancer cells [53–55]. Concerning a TXNIP TXNIP putative role of miR-224/452 on EMT, there are only limited and controversial reports by Zhang et al [13] who found an increase of E Slug Slug E-cadherin in miR-224-depleted Huh-7 cells, whereas miR-224 has TXNIP-3'UTR 224-mut TXNIP-3'UTR 452-mut been shown to correlate with high E-cadherin expression in normal E2F1 ER-E2F1 * * Actin Actin 2,50E+08 breast epithelium [56]. In addition, miR-452 is known to be highly * 2,50E+08 * TXNIP TXNIP expressed in neural crest cells having an influence on an epithelial- Survivin Survivin 2,00E+08 2,00E+08 D SK-Mel-29 mesenchymal signaling pathway in the first pharyngeal arch [36]. In Actin Actin 1,50E+08 1,50E+08 our study, ectopic miR-224/452 mediates the EMT-like phenotype in * 12 non-invasive melanoma cells through increased expression of the 1,00E+08 1,00E+08 10 transcription factors Slug and ZEB1, which are known repressors of RLU/mg protein 5,00E+07 RLU/mg protein 5,00E+07 epithelial E-cadherin [57]. In line with E-cadherin decrease, the 8 0,00E+00 0,00E+00 intermediate filament protein vimentin as mesenchymal marker that CTRL 6 promotes cell migration and invasion is upregulated [58]. The TXNIP reverse effect on these EMT markers was observed when both 4 miRs were knocked down in metastatic tumor cells. Consistent n-fold invasion 2 with previous studies indicating that remodeling of actin filaments is essential for EMT as it promotes cell migration and metastatic 0 Figure 4. Identification of TXNIP as miR-224/452 target. spread from primary tumors [42], we have shown that E2F1-miR- A Scheme of target prediction for miR-224/452. TXNIP 224/452-induced EMT involves cytoskeletal changes toward an B MiR-224/452 regulates TXNIP. Overexpression of miR-224/452 in SK-Mel-29 resulted in less TXNIP expression, while inhibition of the endogenous miRNAs in SK-Mel-147 induced TXNIP. Transcript and protein levels were determined using actin as a control. aggressive cancer phenotype. Actin C Knockdown of E2F1 in SK-Mel-147 increases and E2F1 activation by addition of 4-OHT in SK-Mel-29.ER-E2F1 reduces TXNIP expression on RNA and protein level. We uncovered TXNIP as an important target of both miRs. GAPDH and actin served as loading controls. TXNIP was initially identified as thioredoxin-binding protein that D Luciferase reporter assay revealed a direct regulation of TXNIP by miR-224 and miR-452. Co-transfection of pMiR-Report-30 UTR(TXNIP) and miR-224/452 plasmids in inhibits thioredoxin (TRX), thereby contributing to redox homeo- SK-Mel-29 results in less luciferase activity (in comparison with miR-Scr). Promoter activity is upregulated in miRZip-transfected SK-Mel-147. stasis [59]. Since TRX also promotes tumor progression by angio- E Mutation of miR-224/452 binding sites in the TXNIP-30 UTR completely abolishes their repressive effects. Luciferase activity was measured after co-expression of genesis induction [60] and apoptosis inhibition [61,62], TXNIP, as pMiR-Report-30 UTR(TXNIP-224-mut) or -(TXNIP-452-mut) with miR-224/452 in comparison with miR-Scr in SK-Mel-29. Data information: Bar graphs are represented as means SD, n = 3,*P ≤ 0.05, two-sided Student’s t-test. its inhibitor, has great potential as a tumor suppressor gene. � Several studies on reduced expression of TXNIP in different forms

8 EMBO reports ª 2014 The Authors ª 2014 The Authors EMBO reports 9 EMBO reports E2F1-miR-224/452-TXNIP axis promotes EMT Susanne Knoll et al Susanne Knoll et al E2F1-miR-224/452-TXNIP axis promotes EMT EMBO reports

0,4 n-fold invasion 0,4

TXNIP 0,2 0,2 In the present study, functional analyses on miR-224/452 clearly

A SK-Mel-29 B SK-Mel-29 SK-Mel-147 C SK-Mel-29 SK-Mel-147 A B 6 * 30 * 1,2 *

5 25 1

4 20 0,8 ZEB1

3 15 0,6 E-cadherin

2 10 0,4 n-fold invasion n-fold invasion n-fold invasion Vimentin 1 5 0,2 TXNIP 0 0 0 E2F1 E2F1 E2F1 E2F1 Slug TXNIP TXNIP TXNIP C Actin Actin Actin Actin ZIP-Scr ZIP-Scr ZIP-224 ZIP-452 ZIP-224 ZIP-452 ZIP-224/452 ZIP-224/452

control E2F1

of aggressive cancers such as acute myeloid leukemia (AML), effects to an incomplete but considerable extend, resulting in hepatocellular carcinoma (HCC), and breast and bladder cancer mesenchymal-epithelial transition and inhibition of invasion. In emphasize its relevance for tumor prevention, since less TXNIP contrast, knockdown of E2F1 in highly aggressive melanoma cells D E expression clearly contributes to cancer malignancy [63–67]. Inter- leads to the induction of TXNIP and a more epithelial phenotype estingly, TXNIP is also described as a metastasis suppressor, which (gain of E-cadherin and loss of Slug, ZEB1 and vimentin). This leads to less lung metastases in mice after overexpression in process is partially reversible after ablation of the metastasis suppres- aggressive melanoma cells [47]. In accordance with these findings, sor and shows that the removal of TXNIP during E2F1-induced mela- knockdown of miR-224/452 results in enhanced TXNIP expression, noma progression is absolutely critical for the transcription factor to reduced migration and invasion, and the absence of metastases promote an EMT-like phenotype. The new E2F1 activity is also formation in vivo. Goldberg et al [47] detected low TXNIP expres- reflected by an inverse correlation between high E2F1 and low sion in metastatic melanoma cells as a reason of Chr6 deletion. TXNIP expression in primary melanoma samples from patients with TXNIP itself is not located on Chr6; however, relevant regulators more than 4 mm Breslow depth of invasion compared to non- of this protein originate from Chr6, like CRSP3. In this regard, our invasive tumor states. A previous study by Masaki et al [68] data provide another way of TXNIP inhibition via E2F1-mediated reported that loss of TXNIP enhances TGF-b-induced EMT, and this miR-224/452 expression that is clearly important in melanoma was associated with loss of E-cadherin and a gain of Slug and cells without chromosome 6 deletion and supports the impact of vimentin. In contrast, impaired TGF-b-mediated EMT has been silencing TXNIP to facilitate malignant progression. This is the detected in TXNIP-ablated cells in the context of diabetic nephropathy case for SK-Mel-28/-29/-103 and -147 used in this study, which [69]. These contradictory observations may result from the differ- have apparently no Chr6 deletion as TXNIP is still expressed in ent cellular systems, which means breast and lung cancer cell lines SK-Mel-29 cells and can be recovered by inhibition of the E2F1- versus proximal tubular cells derived from normal kidney. Thus, miR-224/452 module in SK-Mel-147 cells. In addition, we detected E2F1-induced oncogenic EMT bears some similarity with the CRSP3 as a gene, which is located on Chr6 in these melanoma cell signaling of the well-known EMT inducer TGF-b in cancer cells. As Figure 7. TXNIP and E2F1 contribute to malignant progression in a regulatory loop. lines (data not shown). However, miR-224/452 induction by E2F1 an inhibitor of the antioxidant protein TRX, overexpression of A–C Inverse correlation of E2F1 and TXNIP in primary melanoma of different Breslow depth obtained by Oncomine data analysis (A), real-time PCR (B) and is also essential in cells having such a modification, since TXNIP TXNIP in vitro increases the production of reactive oxygen species immunohistochemistry (C). is still expressed and might be active at a certain threshold [47]. (ROS) and induces oxidative stress [70]. Although ROS have D TXNIP inhibits E2F1 via the p16-CDK4-RB pathway. Levels of phospho-RB, CDK4,p16,E2F1 and TXNIP are indicated. The E2F1 target survivin was used as a positive control and actin for loading. Thus, the E2F1-miR-224/452-TXNIP axis represents a key regula- controversial roles in tumorigenesis by inducing DNA mutations, E Schematic illustration of the E2F1-miR-224/452-TXNIP axis in early- and late-stage primary melanoma. High TXNIP levels in early primary tumors restrict E2F1 tory mechanism to force tumor progression. genomic instability and aberrant pro-tumorigenic signaling on the activity through p16/CDK4-induced RB hypophosphorylation. During malignant progression, the miR-224/452 cluster is induced by increasing E2F1 levels, which According to the role of miR-224/452 in an E2F1-induced EMT- one hand, as well as being toxic to cancer cells through the induc- leads to TXNIP suppression at the tumor invasion front associated with an EMT-like phenotype, enhanced migration/invasion and metastatic dissemination. like program, downregulation of its target TXNIP in non-invasive tion of cell death [71], oxidative stress apparently plays no role in Data information: Boxes indicate the 25 and 75% quartile surrounding the median, *P ≤ 0.05 (n = 4), Mann–Whitney U-test; **P ≤ 0.01 (n = 9), two-sided Student’s SK-Mel-29 cells upon E2F1 induction showed identical effects with mediating miR-224/452-induced tumor cell EMT and invasion. t-test. The lines represent the minimum and maximum transcript levels, and circles represent outliers. an increase of mesenchymal and decrease of epithelial markers as TXNIP has also critical functions in regulating glucose homeostasis, described above. However, re-introduction of TXNIP rescued these linking loss of TXNIP to the Warburg cancer phenotype. However,

10 EMBO reports ª 2014 The Authors ª 2014 The Authors EMBO reports 11 EMBO reports E2F1-miR-224/452-TXNIP axis promotes EMT Susanne Knoll et al Susanne Knoll et al E2F1-miR-224/452-TXNIP axis promotes EMT EMBO reports

A SK-Mel-29 B SK-Mel-29 SK-Mel-147 C SK-Mel-29 SK-Mel-147 A B 6 * 30 * 1,2 *

5 25 1

4 20 0,8 ZEB1

3 15 0,6 E-cadherin

Figure 6. Loss of TXNIP due to aberrant E2F1-miR-224/452 expression is crucial for transcription factor-mediated cancer progression. A–CE2F1-induced EMT and invasion requires downregulation of TXNIP by miR-224/452. Invasion was measured by Boyden chamber, and fold changes were calculated relative to the controls (set as 1). Expression of indicated proteins is shown by Western blot. Data information: Bar graphs are represented as means SD, n = 3 (A) or n = 5 (B), *P ≤ 0.05, two-sided Student’s t-test. Æ of aggressive cancers such as acute myeloid leukemia (AML), effects to an incomplete but considerable extend, resulting in hepatocellular carcinoma (HCC), and breast and bladder cancer mesenchymal-epithelial transition and inhibition of invasion. In emphasize its relevance for tumor prevention, since less TXNIP contrast, knockdown of E2F1 in highly aggressive melanoma cells D E expression clearly contributes to cancer malignancy [63–67]. Inter- leads to the induction of TXNIP and a more epithelial phenotype estingly, TXNIP is also described as a metastasis suppressor, which (gain of E-cadherin and loss of Slug, ZEB1 and vimentin). This leads to less lung metastases in mice after overexpression in process is partially reversible after ablation of the metastasis suppres- aggressive melanoma cells [47]. In accordance with these findings, sor and shows that the removal of TXNIP during E2F1-induced mela- knockdown of miR-224/452 results in enhanced TXNIP expression, noma progression is absolutely critical for the transcription factor to reduced migration and invasion, and the absence of metastases promote an EMT-like phenotype. The new E2F1 activity is also formation in vivo. Goldberg et al [47] detected low TXNIP expres- reflected by an inverse correlation between high E2F1 and low sion in metastatic melanoma cells as a reason of Chr6 deletion. TXNIP expression in primary melanoma samples from patients with TXNIP itself is not located on Chr6; however, relevant regulators more than 4 mm Breslow depth of invasion compared to non- of this protein originate from Chr6, like CRSP3. In this regard, our invasive tumor states. A previous study by Masaki et al [68] data provide another way of TXNIP inhibition via E2F1-mediated reported that loss of TXNIP enhances TGF-b-induced EMT, and this miR-224/452 expression that is clearly important in melanoma was associated with loss of E-cadherin and a gain of Slug and cells without chromosome 6 deletion and supports the impact of vimentin. In contrast, impaired TGF-b-mediated EMT has been silencing TXNIP to facilitate malignant progression. This is the detected in TXNIP-ablated cells in the context of diabetic nephropathy case for SK-Mel-28/-29/-103 and -147 used in this study, which [69]. These contradictory observations may result from the differ- have apparently no Chr6 deletion as TXNIP is still expressed in ent cellular systems, which means breast and lung cancer cell lines SK-Mel-29 cells and can be recovered by inhibition of the E2F1- versus proximal tubular cells derived from normal kidney. Thus, miR-224/452 module in SK-Mel-147 cells. In addition, we detected E2F1-induced oncogenic EMT bears some similarity with the CRSP3 as a gene, which is located on Chr6 in these melanoma cell signaling of the well-known EMT inducer TGF-b in cancer cells. As Figure 7. TXNIP and E2F1 contribute to malignant progression in a regulatory loop. lines (data not shown). However, miR-224/452 induction by E2F1 an inhibitor of the antioxidant protein TRX, overexpression of A–C Inverse correlation of E2F1 and TXNIP in primary melanoma of different Breslow depth obtained by Oncomine data analysis (A), real-time PCR (B) and is also essential in cells having such a modification, since TXNIP TXNIP in vitro increases the production of reactive oxygen species immunohistochemistry (C). is still expressed and might be active at a certain threshold [47]. (ROS) and induces oxidative stress [70]. Although ROS have D TXNIP inhibits E2F1 via the p16-CDK4-RB pathway. Levels of phospho-RB, CDK4,p16,E2F1 and TXNIP are indicated. The E2F1 target survivin was used as a positive control and actin for loading. Thus, the E2F1-miR-224/452-TXNIP axis represents a key regula- controversial roles in tumorigenesis by inducing DNA mutations, E Schematic illustration of the E2F1-miR-224/452-TXNIP axis in early- and late-stage primary melanoma. High TXNIP levels in early primary tumors restrict E2F1 tory mechanism to force tumor progression. genomic instability and aberrant pro-tumorigenic signaling on the activity through p16/CDK4-induced RB hypophosphorylation. During malignant progression, the miR-224/452 cluster is induced by increasing E2F1 levels, which According to the role of miR-224/452 in an E2F1-induced EMT- one hand, as well as being toxic to cancer cells through the induc- leads to TXNIP suppression at the tumor invasion front associated with an EMT-like phenotype, enhanced migration/invasion and metastatic dissemination. like program, downregulation of its target TXNIP in non-invasive tion of cell death [71], oxidative stress apparently plays no role in Data information: Boxes indicate the 25 and 75% quartile surrounding the median, *P ≤ 0.05 (n = 4), Mann–Whitney U-test; **P ≤ 0.01 (n = 9), two-sided Student’s SK-Mel-29 cells upon E2F1 induction showed identical effects with mediating miR-224/452-induced tumor cell EMT and invasion. t-test. The lines represent the minimum and maximum transcript levels, and circles represent outliers. an increase of mesenchymal and decrease of epithelial markers as TXNIP has also critical functions in regulating glucose homeostasis, described above. However, re-introduction of TXNIP rescued these linking loss of TXNIP to the Warburg cancer phenotype. However,

is that metastasis suppression by TXNIP similar to its metabolic comparative CT method using actin for normalization. For primer Institutional Animal Care and Use Committee. function does not occur via TRX/ROS regulation. This view is also sequences, see Supplementary Table S2. Supplementary information for this article is available online: underscored by the finding that TXNIP depletion in the context of Boyden chamber and scratch assays http://embor.embopress.org TGF-b-induced EMT did not result in altered TRX levels [68]. Western blotting and immunofluorescence Strikingly, we observed that high levels of TXNIP cause a For Boyden chamber assay, cells were seeded on a 8-lm PET Acknowledgements decrease of E2F1 expression. In order to elucidate the mechanism Protein analysis was performed as described previously [74]. Briefly, membrane (BD BioCoatTM BD MatrigelTM Invasion Chamber, 6-well) We thank Anja Stoll for excellent technical assistance and Ilona Klamfuß for behind it, we investigated different upstream regulators of E2F1. In cell lysis was carried out using RIPA buffer, containing PhosSTOP covered with BD MatrigelTM Basement Membrane Matrix (BD help with animal experiments. Work related to this paper was supported by accordance with Nishinaka and colleagues who showed that TXNIP Phosphatase Inhibitor Cocktail (Roche), and protein concentration Bioscience). A concentration gradient between media in cell culture grants from Erich und Gertrud Roggenbuck-Stiftung, Federal Ministry of induces p16 resulting in reduced phospho-RB levels in HTLV-I posi- was determined by Bradford assay (Bio-Rad). The same quantity of inserts and surrounding well caused cell invasion through the pores Education and Research (BMBF) as part of the project eBio:SysMet, and FORUN tive T cells [49], the same effect on p16 and RB together with different protein samples was separated by SDS–PAGE and of the membrane. Finally, cells of the upper membrane surface were program of Rostock University Medical Faculty. SK received a fellowship of the decreasing expression of cyclin-dependent kinase 4 was seen in transferred to nitrocellulose membranes (Amersham Biosciences) detached, whereas the ones of the lower surface were stained with Landesgraduiertenförderung des Landes Mecklenburg-Vorpommern. melanoma cells. Hence, TXNIP is able to inhibit E2F1 via the p16/ [74]. Selected proteins were detected by the use of specific antibodies DAPI and documented by fluorescence microscopy. Wound healing CDK4/RB axis. This interesting finding gives strong support that for E2F1 (KH-95, BD Biosciences), E-cadherin (Cell Signaling, BD), assay (scratch assay) was performed with Ibidi’s culture inserts that Author contributions E2F1 and TXNIP are integral part of a regulatory loop in which vimentin (V9, Santa Cruz), ZEB1 (H-102, Santa Cruz), Slug (H-140, allow creating a defined gap of 500 lm between two areas with the SK and BMP conceived the idea and designed the experiments; SK, KF and BK increasingly high levels of E2F1 during cancer progression promote Santa Cruz), GABRE (H-110, Santa Cruz), CDK4 (C-22, Santa Cruz), same amount of confluent cells. Cell migration was determined by executed the experiments; SK and SM analyzed the data; US performed target an EMT-like switch by miR-224/452-mediated repression of TXNIP, RB-P (Cell Signaling), survivin (Abcam), E2F1 (BD), TXNIP (K0205- the extent of the gap closure, which was documented by microscopy prediction and OW provided the general framework for bioinformatic study; whereas high TXNIP expression in early primary tumors initially 3, MBL), p16 (M-156, Santa Cruz), ZEB2 (Sigma) and actin (Sigma) recordings at certain times. HM analyzed tissue sections; SK, US and BMP wrote the manuscript. restricts malignant progression by inhibiting E2F1 (Fig 7E). and their corresponding HRP-conjugated secondary antibodies. ECL In sum, our results demonstrate that E2F1 is a new EMT- Plus Western Blotting Detection Reagents (GE Healthcare) allowed Human melanoma tissues and immunohistochemistry Conflict of interest promoting transcription factor. The miR-224/452 cluster represents detection of HRP activity with X-ray films. The authors declare that they have no conflict of interest. a key mediator of E2F1-induced tumor progression, since it inhibits For immunofluorescence, cells were grown on coverslips, fixed For molecular analysis, primary melanoma and melanoma metasta- the metastasis suppressor TXNIP and contributes to EMT, invasion/ with 4% paraformaldehyde, permeabilized with Triton X-100 and ses were investigated by real-time RT–PCR and immunohistochem- migration and metastatic spread of melanoma cells. The uncovered blocked with bovine serum albumin (BSA). Incubation with primary istry. This study was approved by the Ethics Committees of the References mutual regulation between E2F1 and TXNIP points toward new antibodies was performed overnight. For visualization with the Universities of Rostock and Kiel, Germany, and informed consent prognostic and therapeutic options by using components of the inverted confocal laser scanning microscope (Zeiss, ELYRA PS.1), was obtained from all subjects. 1. Jemal A, Siegel R, Xu J, Ward E (2010) Cancer statistics, 2010. CA Cancer E2F1-miR-224/452-TXNIP axis as potential key targets to restrict fluorescence-labeled secondary antibodies (Cy3, Cy5; Molecular Formaldehyde-fixed paraffin-embedded (FFPE) primary melano- J Clin 60: 277 – 300 cancer progression. Probes) were used [74]. Cytoskeleton staining was determined by mas with different Breslow Index were immunohistologically 2. Soengas MS, Lowe SW (2003) Apoptosis and melanoma chemoresistance. Phalloidin/TRITC staining. analyzed. 4-lm-thick slices were cut and for heat-induced epitope Oncogene 22: 3138 – 3151 retrieval pretreated with Tris/EDTA buffer pH 9.0. Tissue slides were 3. Alla V, Engelmann D, Niemetz A, Pahnke J, Schmidt A, Kunz M, Emmrich Materials and Methods Chromatin immunoprecipitation (ChIP) blocked with BSA and incubated overnight using the following S, Steder M, Koczan D, Putzer BM (2010)E2F1 in melanoma progression antibodies and dilutions: TXNIP (1:50, K0205-3, MBL) and E2F1 and metastasis. J Natl Cancer Inst 102: 127 – 133 Cell culture and lentiviral transduction ChIP assay was carried out essentially as described [75] using SK- (1:50, KH-95, Santa Cruz). Finally, slides were washed and developed 4. Chen HZ, Tsai SY, Leone G (2009) Emerging roles of E2Fs in cancer: an Mel-29.ER-E2F1 cells. using LSAB+ System-HRP (Dako) and microscopically analyzed. exit from cell cycle control. Nat Rev Cancer 9: 785 – 797 Melanoma cell lines SK-Mel-28, SK-Mel-29, SK-Mel-103 and SK-Mel- 5. Engelmann D, Putzer BM (2010) Translating DNA damage into cancer

expressing either miRNAs or anti-microRNAs were obtained by 30 UTR were amplified by PCR using genomic DNA as template and target interaction sites that match computationally predicted target WJ, Chu IS (2010) Expression signature of E2F1 and its associated genes lentiviral transduction. Virus particle production required co-trans- cloned into the pWPXL-, pGL3-basic- and pMiR-Report-vector. TXNIP sites of miR-224 or miR-452. Those targets in which binding sites predict superficial to invasive progression of bladder tumors. J Clin Oncol fection of the expression plasmids, psPAX2 (‘packaging’) and was amplified from cDNA and cloned into pWPXL-expression vector. for both miRNAs exist where considered for further analysis. 28: 2660 – 2667 pMD2.G vector (‘envelope’) in HEK 293T cells [73]. For primer sequences, see Supplementary Table S2. For luciferase Furthermore, we retrieved predicted targets that are shared by 7. Tuve S, Wagner SN, Schittek B, Putzer BM (2004) Alterations of Delta- assays, cells were transfected using Turbofect (Thermo Scientific). both miRNAs from the miRror Suite [44]. From these targets, we TA-p 73 splice transcripts during melanoma development and progres- TaqMan®MicroRNA single assays and PCR MiR knockdown was achieved with commercially available considered only those with approved AGO binding sites based on sion. Int J Cancer 108: 162 – 166 miRZipTM vectors (SBI System Biosciences). Luciferase activity was starBase. In total, we received a set of 20 target genes. From this 8. Knoll S, Emmrich S, Putzer BM (2013) The E2F1-miRNA cancer progres- In general, large and small RNA was extracted using the NucleoSpin measured 36 h after transfection using the Luciferase Reporter Assay set, we selected those genes with relevance in cancer based on sion network. Adv Exp Med Biol 774: 135 – 147 miRNA kit (MACHEREY-NAGEL). MicroRNA expression levels were System (Promega) and normalized to total protein concentration in their associated Gene Ontology terms [45] (Fig 4A; Supplementary 9. Esquela-Kerscher A, Slack FJ (2006) Oncomirs - microRNAs with a role in measured using TaqManMicroRNA single assays and the 7900HT cell extract [73]. Table S1). cancer. Nat Rev Cancer 6: 259 – 269 Fast Real-Time PCR System (Applied Biosystems). For expression 10. Hauptman N, Glavac D (2013) MicroRNAs and long non-coding RNAs:

analysis, the comparative CT method was used with RNU6B as Animal studies and histological analysis Statistical analysis prospects in diagnostics and therapy of cancer. Radiol Oncol 47: endogenous control [73]. For semiquantitative PCR, 1 lg of RNA 311 – 318 was reverse transcribed using First Strand cDNA Synthesis Kit 2 × 106 tumor cells stably expressing miRZIP-Scr or miRZIP-224/ P-values were calculated using the Student’s t-test or Wilcoxon– 11. Emmrich S, Putzer BM (2010) Checks and balances: E2F-microRNA cros- (Thermo Scientific). cDNA was added to Thermo Scientific PCR 452 were injected into the tail vein of athymic NMRI nude mice at Mann–Whitney (Mann–Whitney) U-test as indicated. All bar graphs stalk in cancer control. Cell Cycle 9: 2555 – 2567

12 EMBO reports ª 2014 The Authors ª 2014 The Authors EMBO reports 13 EMBO reports E2F1-miR-224/452-TXNIP axis promotes EMT Susanne Knoll et al Susanne Knoll et al E2F1-miR-224/452-TXNIP axis promotes EMT EMBO reports

there was no evidence that supported a connection between TXNIP Master Mix (2×) and amplified with specific primers in a MyCycler the age of 6 to 8 weeks (Charles River). Mice were monitored for are plotted as mean SD. Box-whisker plots indicate the 25 and Æ and metabolism through increased redox stress, suggesting that Thermal Cycler (Bio-Rad). Actin was used as a loading control. For lung metastases over several weeks. Lung tissue was surgically 75% quartile surrounding the median and the minimum and maxi- TXNIP does not act primarily as an inhibitor of thioredoxin [72]. qRT-PCR, cDNA was added to iQTM SYBR Green Supermix excised, fixed in 4% paraformaldehyde, paraffin-embedded and mum as well as outliers. Statistical significance: n.s., no signifi- Given that miR-224/452-mediated TXNIP repression, which should and analyzed using iQ5 Multicolor Real-Time PCR Detection processed for histological analysis with hematoxylin and eosin cance, *P ≤ 0.05; **P ≤ 0.01. Statistical significance was calculated reduce oxidative stress, drives EMT, the obvious implication here System (Bio-Rad). Relative gene expression was calculated by the staining. All animal experiments were performed according to the using the R software package (http://www.r-project.org). is that metastasis suppression by TXNIP similar to its metabolic comparative CT method using actin for normalization. For primer Institutional Animal Care and Use Committee. function does not occur via TRX/ROS regulation. This view is also sequences, see Supplementary Table S2. Supplementary information for this article is available online: underscored by the finding that TXNIP depletion in the context of Boyden chamber and scratch assays http://embor.embopress.org TGF-b-induced EMT did not result in altered TRX levels [68]. Western blotting and immunofluorescence Strikingly, we observed that high levels of TXNIP cause a For Boyden chamber assay, cells were seeded on a 8-lm PET Acknowledgements decrease of E2F1 expression. In order to elucidate the mechanism Protein analysis was performed as described previously [74]. Briefly, membrane (BD BioCoatTM BD MatrigelTM Invasion Chamber, 6-well) We thank Anja Stoll for excellent technical assistance and Ilona Klamfuß for behind it, we investigated different upstream regulators of E2F1. In cell lysis was carried out using RIPA buffer, containing PhosSTOP covered with BD MatrigelTM Basement Membrane Matrix (BD help with animal experiments. Work related to this paper was supported by accordance with Nishinaka and colleagues who showed that TXNIP Phosphatase Inhibitor Cocktail (Roche), and protein concentration Bioscience). A concentration gradient between media in cell culture grants from Erich und Gertrud Roggenbuck-Stiftung, Federal Ministry of induces p16 resulting in reduced phospho-RB levels in HTLV-I posi- was determined by Bradford assay (Bio-Rad). The same quantity of inserts and surrounding well caused cell invasion through the pores Education and Research (BMBF) as part of the project eBio:SysMet, and FORUN tive T cells [49], the same effect on p16 and RB together with different protein samples was separated by SDS–PAGE and of the membrane. Finally, cells of the upper membrane surface were program of Rostock University Medical Faculty. SK received a fellowship of the decreasing expression of cyclin-dependent kinase 4 was seen in transferred to nitrocellulose membranes (Amersham Biosciences) detached, whereas the ones of the lower surface were stained with Landesgraduiertenförderung des Landes Mecklenburg-Vorpommern. melanoma cells. Hence, TXNIP is able to inhibit E2F1 via the p16/ [74]. Selected proteins were detected by the use of specific antibodies DAPI and documented by fluorescence microscopy. Wound healing CDK4/RB axis. This interesting finding gives strong support that for E2F1 (KH-95, BD Biosciences), E-cadherin (Cell Signaling, BD), assay (scratch assay) was performed with Ibidi’s culture inserts that Author contributions E2F1 and TXNIP are integral part of a regulatory loop in which vimentin (V9, Santa Cruz), ZEB1 (H-102, Santa Cruz), Slug (H-140, allow creating a defined gap of 500 lm between two areas with the SK and BMP conceived the idea and designed the experiments; SK, KF and BK increasingly high levels of E2F1 during cancer progression promote Santa Cruz), GABRE (H-110, Santa Cruz), CDK4 (C-22, Santa Cruz), same amount of confluent cells. Cell migration was determined by executed the experiments; SK and SM analyzed the data; US performed target an EMT-like switch by miR-224/452-mediated repression of TXNIP, RB-P (Cell Signaling), survivin (Abcam), E2F1 (BD), TXNIP (K0205- the extent of the gap closure, which was documented by microscopy prediction and OW provided the general framework for bioinformatic study; whereas high TXNIP expression in early primary tumors initially 3, MBL), p16 (M-156, Santa Cruz), ZEB2 (Sigma) and actin (Sigma) recordings at certain times. HM analyzed tissue sections; SK, US and BMP wrote the manuscript. restricts malignant progression by inhibiting E2F1 (Fig 7E). and their corresponding HRP-conjugated secondary antibodies. ECL In sum, our results demonstrate that E2F1 is a new EMT- Plus Western Blotting Detection Reagents (GE Healthcare) allowed Human melanoma tissues and immunohistochemistry Conflict of interest promoting transcription factor. The miR-224/452 cluster represents detection of HRP activity with X-ray films. The authors declare that they have no conflict of interest. a key mediator of E2F1-induced tumor progression, since it inhibits For immunofluorescence, cells were grown on coverslips, fixed For molecular analysis, primary melanoma and melanoma metasta- the metastasis suppressor TXNIP and contributes to EMT, invasion/ with 4% paraformaldehyde, permeabilized with Triton X-100 and ses were investigated by real-time RT–PCR and immunohistochem- migration and metastatic spread of melanoma cells. The uncovered blocked with bovine serum albumin (BSA). Incubation with primary istry. This study was approved by the Ethics Committees of the References mutual regulation between E2F1 and TXNIP points toward new antibodies was performed overnight. For visualization with the Universities of Rostock and Kiel, Germany, and informed consent prognostic and therapeutic options by using components of the inverted confocal laser scanning microscope (Zeiss, ELYRA PS.1), was obtained from all subjects. 1. Jemal A, Siegel R, Xu J, Ward E (2010) Cancer statistics, 2010. CA Cancer E2F1-miR-224/452-TXNIP axis as potential key targets to restrict fluorescence-labeled secondary antibodies (Cy3, Cy5; Molecular Formaldehyde-fixed paraffin-embedded (FFPE) primary melano- J Clin 60: 277 – 300 cancer progression. Probes) were used [74]. Cytoskeleton staining was determined by mas with different Breslow Index were immunohistologically 2. Soengas MS, Lowe SW (2003) Apoptosis and melanoma chemoresistance. Phalloidin/TRITC staining. analyzed. 4-lm-thick slices were cut and for heat-induced epitope Oncogene 22: 3138 – 3151 retrieval pretreated with Tris/EDTA buffer pH 9.0. Tissue slides were 3. Alla V, Engelmann D, Niemetz A, Pahnke J, Schmidt A, Kunz M, Emmrich Materials and Methods Chromatin immunoprecipitation (ChIP) blocked with BSA and incubated overnight using the following S, Steder M, Koczan D, Putzer BM (2010)E2F1 in melanoma progression antibodies and dilutions: TXNIP (1:50, K0205-3, MBL) and E2F1 and metastasis. J Natl Cancer Inst 102: 127 – 133 Cell culture and lentiviral transduction ChIP assay was carried out essentially as described [75] using SK- (1:50, KH-95, Santa Cruz). Finally, slides were washed and developed 4. Chen HZ, Tsai SY, Leone G (2009) Emerging roles of E2Fs in cancer: an Mel-29.ER-E2F1 cells. using LSAB+ System-HRP (Dako) and microscopically analyzed. exit from cell cycle control. Nat Rev Cancer 9: 785 – 797 Melanoma cell lines SK-Mel-28, SK-Mel-29, SK-Mel-103 and SK-Mel- 5. Engelmann D, Putzer BM (2010) Translating DNA damage into cancer

147 were maintained in Dulbecco’s modified Eagle medium (high Cloning of expression plasmids and 30 UTRs constructs and Identification of miR-224/452 candidate targets using cell death-A roadmap for E2F1 apoptotic signalling and opportunities glucose, 4.5 g/l) containing sodium pyruvate as previously luciferase reporter assay bioinformatic tools for new drug combinations to overcome chemoresistance. Drug Resist described [3]. For E2F1 induction in SK-Mel-29.ER-E2F1 cells, 1 lM Updat 13: 119 – 131 of 4-hydroxytamoxifen (4-OHT) was used [3]. Stable cell lines MiRNAs hsa-miR-224, hsa-miR-452, GABRE promoter and TXNIP We extracted from the starBase database (v1.0) [43] Argonaute- 6. Lee JS, Leem SH, Lee SY, Kim SC, Park ES, Kim SB, Kim SK, Kim YJ, Kim expressing either miRNAs or anti-microRNAs were obtained by 30 UTR were amplified by PCR using genomic DNA as template and target interaction sites that match computationally predicted target WJ, Chu IS (2010) Expression signature of E2F1 and its associated genes lentiviral transduction. Virus particle production required co-trans- cloned into the pWPXL-, pGL3-basic- and pMiR-Report-vector. TXNIP sites of miR-224 or miR-452. Those targets in which binding sites predict superficial to invasive progression of bladder tumors. J Clin Oncol fection of the expression plasmids, psPAX2 (‘packaging’) and was amplified from cDNA and cloned into pWPXL-expression vector. for both miRNAs exist where considered for further analysis. 28: 2660 – 2667 pMD2.G vector (‘envelope’) in HEK 293T cells [73]. For primer sequences, see Supplementary Table S2. For luciferase Furthermore, we retrieved predicted targets that are shared by 7. Tuve S, Wagner SN, Schittek B, Putzer BM (2004) Alterations of Delta- assays, cells were transfected using Turbofect (Thermo Scientific). both miRNAs from the miRror Suite [44]. From these targets, we TA-p 73 splice transcripts during melanoma development and progres- TaqMan®MicroRNA single assays and PCR MiR knockdown was achieved with commercially available considered only those with approved AGO binding sites based on sion. Int J Cancer 108: 162 – 166 miRZipTM vectors (SBI System Biosciences). Luciferase activity was starBase. In total, we received a set of 20 target genes. From this 8. Knoll S, Emmrich S, Putzer BM (2013) The E2F1-miRNA cancer progres- In general, large and small RNA was extracted using the NucleoSpin measured 36 h after transfection using the Luciferase Reporter Assay set, we selected those genes with relevance in cancer based on sion network. Adv Exp Med Biol 774: 135 – 147 miRNA kit (MACHEREY-NAGEL). MicroRNA expression levels were System (Promega) and normalized to total protein concentration in their associated Gene Ontology terms [45] (Fig 4A; Supplementary 9. Esquela-Kerscher A, Slack FJ (2006) Oncomirs - microRNAs with a role in measured using TaqManMicroRNA single assays and the 7900HT cell extract [73]. Table S1). cancer. Nat Rev Cancer 6: 259 – 269 Fast Real-Time PCR System (Applied Biosystems). For expression 10. Hauptman N, Glavac D (2013) MicroRNAs and long non-coding RNAs: analysis, the comparative CT method was used with RNU6B as Animal studies and histological analysis Statistical analysis prospects in diagnostics and therapy of cancer. Radiol Oncol 47: endogenous control [73]. For semiquantitative PCR, 1 lg of RNA 311 – 318 was reverse transcribed using First Strand cDNA Synthesis Kit 2 × 106 tumor cells stably expressing miRZIP-Scr or miRZIP-224/ P-values were calculated using the Student’s t-test or Wilcoxon– 11. Emmrich S, Putzer BM (2010) Checks and balances: E2F-microRNA cros- (Thermo Scientific). cDNA was added to Thermo Scientific PCR 452 were injected into the tail vein of athymic NMRI nude mice at Mann–Whitney (Mann–Whitney) U-test as indicated. All bar graphs stalk in cancer control. Cell Cycle 9: 2555 – 2567

12. van Rooij E, Purcell AL, Levin AA (2012) Developing microRNA therapeu- 28. Graff JW, Powers LS, Dickson AM, Kim J, Reisetter AC, Hassan IH, 44. Friedman Y, Naamati G, Linial M (2010) MiRror: a combinatorial analy- 60. Welsh SJ, Bellamy WT, Briehl MM, Powis G (2002) The redox protein tics. Circ Res 110: 496 – 507 Kremens K, Gross TJ, Wilson ME, Monick MM (2012) Cigarette smoking sis web tool for ensembles of microRNAs and their targets. Bioinformat- thioredoxin-1 (Trx-1) increases hypoxia-inducible factor 1alpha protein 13. Zhang Y, Takahashi S, Tasaka A, Yoshima T, Ochi H, Chayama K (2013) decreases global microRNA expression in human alveolar macrophages. ics 26: 1920 – 1921 expression: Trx-1 overexpression results in increased vascular endothe- Involvement of microRNA-224 in cell proliferation, migration, invasion, PLoS ONE 7:e44066 45. Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, Davis AP, lial growth factor production and enhanced tumor angiogenesis. Cancer and anti-apoptosis in hepatocellular carcinoma. J Gastroenterol Hepatol 29. Dmitriev P, Barat A, Polesskaya A, O’Connell MJ, Robert T, Dessen P, Dolinski K, Dwight SS, Eppig JT et al (2000) Gene ontology: tool for the Res 62: 5089 – 5095 28: 565 – 575 Walsh TA, Lazar V, Turki A, Carnac G et al (2013) Simultaneous miRNA unification of biology. The Gene Ontology Consortium. Nat Genet 25: 61. Saitoh M, Nishitoh H, Fujii M, Takeda K, Tobiume K, Sawada Y, Kawabata 14. Chen PJ, Yeh SH, Liu WH, Lin CC, Huang HC, Chen CL, Chen DS, Chen PJ and mRNA transcriptome profiling of human myoblasts reveals a novel 25 – 29 M, Miyazono K, Ichijo H (1998) Mammalian thioredoxin is a direct (2012) Androgen pathway stimulates microRNA-216a transcription to set of myogenic differentiation-associated miRNAs and their target 46. Zhou J, Chng WJ (2013) Roles of thioredoxin binding protein (TXNIP) in inhibitor of apoptosis signal-regulating kinase (ASK) 1. EMBO J 17: suppress the tumor suppressor in lung cancer-1 gene in early hepato- genes. BMC Genomics 14: 265 oxidative stress, apoptosis and cancer. Mitochondrion 13: 163 – 169 2596 – 2606 carcinogenesis. Hepatology 56: 632 – 643 30. Fang X, Yoon JG, Li L, Yu W, Shao J, Hua D, Zheng S, Hood L, Goodlett 47. Goldberg SF, Miele ME, Hatta N, Takata M, Paquette-Straub C, Freed- 62. Liu Y, Min W (2002) Thioredoxin promotes ASK1 ubiquitination and 15. Gao P, Wong CC, Tung EK, Lee JM, Wong CM, Ng IO (2011) Deregulation of DR, Foltz G et al (2011) The SOX2 response program in glioblastoma man LP, Welch DR (2003) Melanoma metastasis suppression by chromo- degradation to inhibit ASK1-mediated apoptosis in a redox activity-inde- microRNA expression occurs early and accumulates in early stages of multiforme: an integrated ChIP-seq, expression microarray, and microR- some 6: evidence for a pathway regulated by CRSP3 and TXNIP. Cancer pendent manner. Circ Res 90: 1259 – 1266 HBV-associated multistep hepatocarcinogenesis. J Hepatol 54: 1177 – 1184 NA analysis. BMC Genomics 12: 11 Res 63: 432 – 440 63. Zhou J, Bi C, Cheong LL, Mahara S, Liu SC, Tay KG, Koh TL, Yu Q, Chng 16. Scisciani C, Vossio S, Guerrieri F, Schinzari V, De Iaco R, D’Onorio de Meo 31. van Schooneveld E, Wouters MC, Van der Auwera I, Peeters DJ, Wildiers H, 48. Rhodes DR, Yu J, Shanker K, Deshpande N, Varambally R, Ghosh D, WJ (2011) The histone methyltransferase inhibitor, DZNep, up-regulates P, Cervello M, Montalto G, Pollicino T, Raimondo G et al (2012) Tran- Van Dam PA, Vergote I, Vermeulen PB, Dirix LY, Van Laere SJ (2012) Expres- Barrette T, Pandey A, Chinnaiyan AM (2004) ONCOMINE: a cancer micro- TXNIP, increases ROS production, and targets leukemia cells in AML. scriptional regulation of miR-224 upregulated in human HCCs by sion profiling of cancerous and normal breast tissues identifies microRNAs array database and integrated data-mining platform. Neoplasia 6: 1 – 6 Blood 118: 2830 – 2839 NFkappaB inflammatory pathways. J Hepatol 56: 855 – 861 that are differentially expressed in serum from patients with (metastatic) 49. Nishinaka Y, Nishiyama A, Masutani H, Oka S, Ahsan KM, Nakayama Y, 64. Kwon HJ, Won YS, Suh HW, Jeon JH, Shao Y, Yoon SR, Chung JW, Kim 17. Boguslawska J, Wojcicka A, Piekielko-Witkowska A, Master A, Nauman A breast cancer and healthy volunteers. Breast Cancer Res 14:R34 Ishii Y, Nakamura H, Maeda M, Yodoi J (2004) Loss of thioredoxin-bind- TD, Kim HM, Nam KH et al (2010) Vitamin D3 upregulated protein 1

(2011) MiR-224 targets the 30UTR of type 150-iodothyronine deiodinase 32. Veerla S, Lindgren D, Kvist A, Frigyesi A, Staaf J, Persson H, Liedberg F, ing protein-2/vitamin D3 up-regulated protein 1 in human T-cell leuke- suppresses TNF-alpha-induced NF-kappaB activation in hepatocarcino- possibly contributing to tissue hypothyroidism in renal cancer. PLoS ONE Chebil G, Gudjonsson S, Borg A et al (2009) MiRNA expression in urothe- mia virus type I-dependent T-cell transformation: implications for adult genesis. J Immunol 185: 3980 – 3989 6:e24541 lial carcinomas: important roles of miR-10a, miR-222, miR-125b, miR-7 T-cell leukemia leukemogenesis. Cancer Res 64: 1287 – 1292 65. Sheth SS, Bodnar JS, Ghazalpour A, Thipphavong CK, Tsutsumi S, Tward 18. Liu H, Brannon AR, Reddy AR, Alexe G, Seiler MW, Arreola A, Oza JH, Yao and miR-452 for tumor stage and metastasis, and frequent homozygous 50. Wang Y, Toh HC, Chow P, Chung AY, Meyers DJ, Cole PA, Ooi LL, Lee CG AD, Demant P, Kodama T, Aburatani H, Lusis AJ (2006) Hepatocellular M, Juan D, Liou LS et al (2010) Identifying mRNA targets of microRNA losses of miR-31. Int J Cancer 124: 2236 – 2242 (2012) MicroRNA-224 is up-regulated in hepatocellular carcinoma carcinoma in Txnip-deficient mice. Oncogene 25: 3528 – 3536 dysregulated in cancer: with application to clear cell Renal Cell Carci- 33. Puerta-Gil P, Garcia-Baquero R, Jia AY, Ocana S, Alvarez-Mugica M, through epigenetic mechanisms. FASEB J 26: 3032 – 3041 66. Butler LM, Zhou X, Xu WS, Scher HI, Rifkind RA, Marks PA, Richon VM noma. BMC Syst Biol 4: 51 Alvarez-Ossorio JL, Cordon-Cardo C, Cava F, Sanchez-Carbayo M (2012) 51. Kristensen H, Haldrup C, Strand S, Mundbjerg K, Mortensen MM, Thor- (2002) The histone deacetylase inhibitor SAHA arrests cancer cell 19. Arndt GM, Dossey L, Cullen LM, Lai A, Druker R, Eisbacher M, Zhang C, miR-143, miR-222, and miR-452 are useful as tumor stratification and sen K, Ostenfeld MS, Wild PJ, Arsov C, Goering W et al (2014) Hyperme- growth, up-regulates thioredoxin-binding protein-2, and down-regulates Tran N, Fan H, Retzlaff K et al (2009) Characterization of global microR- noninvasive diagnostic biomarkers for bladder cancer. Am J Pathol 180: thylation of the GABRE~miR-452~miR-224 promoter in prostate cancer thioredoxin. Proc Natl Acad Sci USA 99: 11700 – 11705 NA expression reveals oncogenic potential of miR-145 in metastatic 1808 – 1815 predicts biochemical recurrence after radical prostatectomy. Clin Cancer 67. Nishizawa K, Nishiyama H, Matsui Y, Kobayashi T, Saito R, Kotani H, colorectal cancer. BMC Cancer 9: 374 34. Liu SG, Qin XG, Zhao BS, Qi B, Yao WJ, Wang TY, Li HC, Wu XN (2013) Res 20: 2169 – 2181 Masutani H, Oishi S, Toda Y, Fujii N et al (2011) Thioredoxin-interacting 20. Fu J, Tang W, Du P, Wang G, Chen W, Li J, Zhu Y, Gao J, Cui L (2012) Differential expression of miRNAs in esophageal cancer tissue. Oncol Lett 52. Diaz-Lopez A, Moreno-Bueno G, Cano A (2014) Role of microRNA in protein suppresses bladder carcinogenesis. Carcinogenesis 32: Identifying microRNA-mRNA regulatory network in colorectal cancer by 5: 1639 – 1642 epithelial to mesenchymal transition and metastasis and clinical 1459 – 1466 a combination of expression profile and bioinformatics analysis. BMC 35. Liu C, Kelnar K, Vlassov AV, Brown D, Wang J, Tang DG (2012) Distinct perspectives. Cancer Manag Res 6: 205 – 216 68. Masaki S, Masutani H, Yoshihara E, Yodoi J (2012) Deficiency of Syst Biol 6: 68 microRNA expression profiles in prostate cancer stem/progenitor cells 53. Park SM, Gaur AB, Lengyel E, Peter ME (2008) The miR-200 family deter- thioredoxin binding protein-2 (TBP-2) enhances TGF-beta signaling 21. Wang YX, Zhang XY, Zhang BF, Yang CQ, Chen XM, Gao HJ (2010) Initial and tumor-suppressive functions of let-7. Cancer Res 72: 3393 – 3404 mines the epithelial phenotype of cancer cells by targeting the E-cadh- and promotes epithelial to mesenchymal transition. PLoS ONE 7: study of microRNA expression profiles of colonic cancer without lymph 36. Sheehy NT, Cordes KR, White MP, Ivey KN, Srivastava D (2010) The erin repressors ZEB1 and ZEB2. Genes Dev 22: 894 – 907 e39900 node metastasis. J Dig Dis 11: 50 – 54 neural crest-enriched microRNA miR-452 regulates epithelial-mesenchy- 54. Gregory PA, Bert AG, Paterson EL, Barry SC, Tsykin A, Farshid G, Vadas 69. Wei J, Shi Y, Hou Y, Ren Y, Du C, Zhang L, Li Y, Duan H (2013) Knock- 22. Lu S, Wang S, Geng S, Ma S, Liang Z, Jiao B (2013) Upregulation of micr- mal signaling in the first pharyngeal arch. Development 137: 4307 – 4316 MA, Khew-Goodall Y, Goodall GJ (2008) The miR-200 family and down of thioredoxin-interacting protein ameliorates high glucose-in- oRNA-224 confers a poor prognosis in glioma patients. Clin Transl Oncol 37. Najafi-Shoushtari SH, Kristo F, Li Y, Shioda T, Cohen DE, Gerszten RE, miR-205 regulate epithelial to mesenchymal transition by targeting duced epithelial to mesenchymal transition in renal tubular epithelial 15: 569 – 574 Naar AM (2010) MicroRNA-33 and the SREBP host genes cooperate to ZEB1 and SIP1. Nat Cell Biol 10: 593 – 601 cells. Cell Signal 25: 2788 – 2796 23. Mian C, Pennelli G, Fassan M, Balistreri M, Barollo S, Cavedon E, control cholesterol homeostasis. Science 328: 1566 – 1569 55. Korpal M, Lee ES, Hu G, Kang Y (2008) The miR-200 family inhibits 70. Kaimul AM, Nakamura H, Masutani H, Yodoi J (2007) Thioredoxin and Galuppini F, Pizzi M, Vianello F, Pelizzo MR et al (2012) MicroRNA 38. Bullock MD, Sayan AE, Packham GK, Mirnezami AH (2012) MicroRNAs: epithelial-mesenchymal transition and cancer cell migration by direct thioredoxin-binding protein-2 in cancer and metabolic syndrome. Free profiles in familial and sporadic medullary thyroid carcinoma: preli- critical regulators of epithelial to mesenchymal (EMT) and mesenchymal targeting of E-cadherin transcriptional repressors ZEB1 and ZEB2. J Biol Radic Biol Med 43: 861 – 868 minary relationships with RET status and outcome. Thyroid 22: 890 – 896 to epithelial transition (MET) in cancer progression. Biol Cell 104: 3 – 12 Chem 283: 14910 – 14914 71. Glasauer A, Chandel NS (2014) Targeting antioxidants for cancer ther- 24. Gokhale A, Kunder R, Goel A, Sarin R, Moiyadi A, Shenoy A, Mamidipally 39. Sreekumar R, Sayan BS, Mirnezami AH, Sayan AE (2011) MicroRNA 56. Giricz O, Reynolds PA, Ramnauth A, Liu C, Wang T, Stead L, Childs G, apy. Biochem Pharmacol pii: S0006-2952: 00423 – 00427 doi: 10.1016/ C, Noronha S, Kannan S, Shirsat NV (2010) Distinctive microRNA signa- control of invasion and metastasis pathways. Front Genet 2: 58 Rohan T, Shapiro N, Fineberg S et al (2012) Hsa-miR-375 is differentially j.bcp.2014.07.017. [Epub ahead of print] ture of medulloblastomas associated with the WNT signaling pathway. J 40. Tam WL, Weinberg RA (2013 ) The epigenetics of epithelial-mesenchymal expressed during breast lobular neoplasia and promotes loss of 72. Patwari P, Lee RT (2012) An expanded family of arrestins regulate Cancer Res Ther 6: 521 – 529 plasticity in cancer. Nat Med 19: 1438 – 1449 mammary acinar polarity. J Pathol 226: 108 – 119 metabolism. Trends Endocrinol Metab 23: 216 – 222 25. Scapoli L, Palmieri A, Lo Muzio L, Pezzetti F, Rubini C, Girardi A, Farinella 41. Denecker G, Vandamme N, Akay O, Koludrovic D, Taminau J, Lemeire K, 57. Bolos V, Peinado H, Perez-Moreno MA, Fraga MF, Esteller M, Cano A 73. Alla V, Kowtharapu BS, Engelmann D, Emmrich S, Schmitz U, Steder M, F, Mazzotta M, Carinci F (2010) MicroRNA expression profiling of oral Gheldof A, De Craene B, Van Gele M, Brochez L et al (2014) Identifica- (2003) The transcription factor Slug represses E-cadherin expression and Putzer BM (2012)E2F1 confers anticancer drug resistance by targeting carcinoma identifies new markers of tumor progression. Int J Immunopa- tion of a ZEB2-MITF-ZEB1 transcriptional network that controls mela- induces epithelial to mesenchymal transitions: a comparison with Snail ABC transporter family members and Bcl-2 via the p73/DNp73-miR-205 thol Pharmacol 23: 1229 – 1234 nogenesis and melanoma progression. Cell Death Differ 21: 1250 – 1261 and E47 repressors. J Cell Sci 116: 499 – 511 circuitry. Cell Cycle 11: 3067 – 3078 26. Kim YK, Yu J, Han TS, Park SY, Namkoong B, Kim DH, Hur K, Yoo MW, 42. Haynes J, Srivastava J, Madson N, Wittmann T, Barber DL (2011) 58. Satelli A, Li S (2011) Vimentin in cancer and its potential as a molecular 74. Knoll S, Furst K, Thomas S, Villanueva Baselga S, Stoll A, Schaefer S, Lee HJ, Yang HK et al (2009) Functional links between clustered microR- Dynamic actin remodeling during epithelial-mesenchymal transition target for cancer therapy. Cell Mol Life Sci 68: 3033 – 3046 Putzer BM (2011) Dissection of cell context-dependent interactions NAs: suppression of cell-cycle inhibitors by microRNA clusters in gastric depends on increased moesin expression. Mol Biol Cell 22: 4750 – 4764 59. Nishiyama A, Matsui M, Iwata S, Hirota K, Masutani H, Nakamura H, between HBx and p53 family members in regulation of apoptosis: a role cancer. Nucleic Acids Res 37: 1672 – 1681 43. Yang JH, Li JH, Shao P, Zhou H, Chen YQ, Qu LH (2011) starBase: a Takagi Y, Sono H, Gon Y, Yodoi J (1999) Identification of thioredox- for HBV-induced HCC. Cell Cycle 10: 3554 – 3565 27. Madhyastha R, Madhyastha H, Nakajima Y, Omura S, Maruyama M database for exploring microRNA-mRNA interaction maps from Argona- in-binding protein-2/vitamin D(3) up-regulated protein 1 as a negative 75. Racek T, Buhlmann S, Rust F, Knoll S, Alla V, Putzer BM (2008) Transcrip- (2012) MicroRNA signature in diabetic wound healing: promotive role of ute CLIP-Seq and Degradome-Seq data. Nucleic Acids Res 39: regulator of thioredoxin function and expression. J Biol Chem 274: tional repression of the prosurvival endoplasmic reticulum chaperone miR-21 in fibroblast migration. Int Wound J 9: 355 – 361 D202 – D209 21645 – 21650 GRP78/BIP by E2F1. J Biol Chem 283: 34305 – 34314

14 EMBO reports ª 2014 The Authors ª 2014 The Authors EMBO reports 15 EMBO reports E2F1-miR-224/452-TXNIP axis promotes EMT Susanne Knoll et al Susanne Knoll et al E2F1-miR-224/452-TXNIP axis promotes EMT EMBO reports

14 EMBO reports ª 2014 The Authors ª 2014 The Authors EMBO reports 15 EMBO Molecular Medicine A generic signature to quantify EMT phenotype Tuan Zea Tan et al Research Article

transcriptional regulators such as SNAIL, ZEB and TWIST (Thiery immunohistochemistry for known Epi markers, CDH1 and CK19, et al, 2009; Jordan et al, 2011; Lee & Nelson, 2012; Frisch et al, was significantly enriched in Luminal cell lines with a low EMT 2013; Tam & Weinberg, 2013). These diverse mechanisms nonethe- score (P = 0.035 and P = 0.005, respectively). Cell lines with an Epithelial-mesenchymal transition spectrum less converge and generate similar EMTed phenotypic endpoints intermediate EMT score were of a mixed Basal–Luminal phenotype, (Thiery et al, 2009; Tam & Weinberg, 2013), and this convergence with enriched expression of CK5, a myoepithelial or basal marker likely reflects a series of molecular features common to all cancers (P = 0.0002). Basal cell lines had an intermediate-to-high EMT quantification and its efficacy in deciphering undergoing EMT (Jordan et al, 2011). Thus, we sought to establish score, whereas Luminal cell lines had a lower EMT score (P = 1.6E-7; a generic EMT signature to capture a set of universal molecular Fig 1C). The bladder cancer-specific EMT signature was validated survival and drug responses of cancer patients features exhibited by a broad spectrum of cancers during EMT. (Supplementary Text, Supplementary Fig S1), whereas the ovarian Here, we developed an approach to quantitatively estimate the EMT cancer-specific EMT signature was already assessed in a previous Tuan Zea Tan1, Qing Hao Miow2, Yoshio Miki3, Tetsuo Noda3, Seiichi Mori3, Ruby Yun-Ju Huang1,4,† & status amongst clinical samples and cell lines using transcriptomics. study (Miow et al, 2014). These results corroborate the cancer- We first established bladder, breast, colorectal, gastric, lung and specific EMT signature scoring, which forms the basis of the generic 1,2,5,*,† Jean Paul Thiery ovarian cancer-specific EMT signatures and, from these, derived a EMT signature. generic EMT signature. We posit that this generic EMT signature exemplifies the common molecular features of EMT in tumours and Generic EMT signature cell lines of different origins and believe that this signature will be Abstract cancer-related events, including cancer invasion, metastasis, resis- important in the future objective and systematic study of the role To quantitatively score any cancer for its EMT status, we derived a tance to cell death, refractory responses to chemotherapy and EMT and its dynamic nature in cancer progression, treatment generic EMT signature for tumours and cell lines based on the Epithelial-mesenchymal transition (EMT) is a reversible and immunotherapy, immunosuppression and the acquisition of stem response and survival. weighted sum of the significance analysis of microarray (SAM) and dynamic process hypothesized to be co-opted by carcinoma during cell-like properties (Lee et al, 2006; Onder et al, 2008; Thiery et al, receiver operating characteristic (ROC) results from each of the invasion and metastasis. Yet, there is still no quantitative measure 2009; Jordan et al, 2011; Huang et al, 2012; Lee & Nelson, 2012; cancer-specific EMT signatures (Fig 2A; see Material and Methods). to assess the interplay between EMT and cancer progression. Here, Frisch et al, 2013; Tam & Weinberg, 2013). In EMT, polarized Results Genes that were present in all six of the cancer-specific EMT signa- we derived a method for universal EMT scoring from cancer- epithelial (Epi) cells progressively alter their junctional and polarity tures with a high z-transformed weighted sum (P < 0.001) were specific transcriptomic EMT signatures of ovarian, breast, bladder, complexes to acquire morphological and biochemical characteristics Cancer-specific EMT signature included in the generic EMT signature (Fig 2A). As illustrated by the lung, colorectal and gastric cancers. We show that EMT scoring typical of mesenchymal (Mes) cells (Thiery et al, 2009). EMT was interconnecting links in the heatmap, we noted a high overlap of exhibits good correlation with previously published, cancer-specific first described as a mechanism driving critical morphogenetic steps We first generated EMT signatures specific to bladder, breast, colo- genes amongst the cancer-specific EMT signatures. A total of 315 EMT signatures. This universal and quantitative EMT scoring was (for example, gastrulation) in the development of most metazoans rectal, gastric, lung and ovarian cancer according to the six-step genes (Epi: 145, Mes: 170) and 218 genes (Epi: 170, Mes: 48) were used to establish an EMT spectrum across various cancers, with (Jordan et al, 2011; Lim & Thiery, 2012) and, more recently, in scheme depicted in Fig 1A (see Materials and Methods). First, we selected for tumour and cell line generic EMT signatures, respec- good correlation noted between cell lines and tumours. We show wound-healing and carcinoma progression (Thiery et al, 2009). curated published EMT signatures (Subramanian et al, 2005; Lee tively (Supplementary Table S1A and B). Amongst these, 88 Epi and correlations between EMT and poorer disease-free survival in ovar- However, despite its potential involvement in invasion and metasta- et al, 2006; Carretero et al, 2010) and applied single-sample gene 30 Mes genes were up-regulated in both signatures (Supplementary ian and colorectal, but not breast, carcinomas, despite previous sis, the role of EMT in human tumours is still inadequately docu- set enrichment analysis (ssGSEA) (Verhaak et al, 2013) to provide a Table S1A and B). Known EMT transcripts—CDH1, EPCAM, GRHL2, notions. Importantly, we found distinct responses between epithe- mented (Wang et al, 2004; Chaffer & Weinberg, 2011; Kong et al, gross assessment of the EMT phenotype of each cell line or tumour. KRT19, RAB25, CDH2, VIM, ZEB1, ZEB2, SNAI2 and TWIST1 lial- and mesenchymal-like ovarian cancers to therapeutic regimes 2011). This is so even after the identification of a transitioned or An EMT signature that correlated best with known EMT transcripts (Thiery et al, 2009; Cieply et al, 2012; Huang et al, 2012; Zhang administered with or without paclitaxel in vivo and demonstrated ‘EMTed’ phenotype—either partially or completely—in circulating was next established, and the most Epi and most Mes cell lines or et al, 2013)—were consistently selected in the generic EMT signa- that mesenchymal-like tumours do not always show resistance to tumour cells (CTCs) (Jordan et al, 2011; Valastyan & Weinberg, tumours were chosen to build the EMT signature using BinReg ture; this successful identification of genes relevant to EMT lends chemotherapy. EMT scoring is thus a promising, versatile tool for 2011; Yu et al, 2013). Initially believed to be a binary process, EMT (Gatza et al, 2010). This BinReg EMT signature was then used to support to the validity of our strategy. Furthermore, the expression the objective and systematic investigation of EMT roles and is now well documented to be a dynamic course, with the existence predict the EMT phenotype in cell lines and tumours. The most Epi of miRNAs reported to suppress EMT, such as those from the dynamics in cancer progression, treatment response and survival. of intermediate states (Jordan et al, 2011; Kong et al, 2011; Huang and most Mes cell lines or tumours were again selected to generate miR-200 (miR-200a, miR-200b, miR-200c, miR-141, miR-429) and et al, 2013; Tam & Weinberg, 2013). Cells stuck or transitioning in the final EMT signature. Finally, we computed an EMT score of a miR-34 (miR-34a, miR-34b, miR-34c) families (Zhang & Ma, 2012; Keywords drug response; epithelial-mesenchymal transition; gene expression these intermediate or ‘metastable’ states of EMT (Jordan et al, 2011) given sample using a two-sample Kolmogorov–Smirnov test (2KS). Hao et al, 2014), was significantly and consistently anti-correlated signature; microarray; prognosis —often called ‘fused cells’ (Kong et al, 2011)—have attributes of Samples with a positive (high) EMT score were more Mes, whereas with the generic EMT score (Supplementary Text, Supplementary Subject Categories Biomarkers & Diagnostic Imaging; Cancer both Epi and Mes phenotypes and exhibit stem cell-like properties. those with a negative (low) score were more Epi. We developed a Fig S2). This suggests the potential to incorporate miRNAs in the DOI 10.15252/emmm.201404208 | Received 28 April 2014 | Revised 7 August They also display high plasticity between the Epi and Mes states, cancer-specific EMT signature for tumours and cell lines separately, generic EMT signature. 2014 | Accepted 8 August 2014 | Published online 11 September 2014 which is critical for metastasis, and hence, it is becoming increas- acknowledging the limitations that cell lines mimic only certain Functional annotation analyses on gene ontology and KEGG EMBO Mol Med (2014) 6: 1279–1293 ingly clear that these intermediate phenotypes must also be quanti- aspects of cancer biology, do not propagate in a stromal microenvi- pathway (Huang da et al, 2009) for all 315 genes in the generic EMT tatively assessed and considered in the design of new therapeutic ronment, and often accumulate additional mutations to survive in signature revealed a significant enrichment in EMT-related biologi- strategies (Chaffer & Weinberg, 2011; Valastyan & Weinberg, 2011). artificial culture systems (Borrell, 2010; Gillet et al, 2013). cal processes, for example, cell adhesion (FDR = 1.2E-9) and cell Introduction Numerous signalling pathways initiate and execute the biochemi- To first ensure the validity of these cancer-specific EMT signa- migration (FDR = 6.0E-4; Supplementary Table S2). The generic cal programs that lead to EMT in a context-dependent manner, tures, we verified our breast cancer-specific EMT signature on the EMT signature was then compared with published cancer-specific Accumulating evidence indicates that epithelial-mesenchymal including those associated with surface tyrosine or serine/threonine GSE16795 breast cancer cell line data set (Hollestelle et al, 2010). EMT signatures (Supplementary Table S1C). By comparing the transition (EMT) is of paramount importance in a plethora of kinases, WNT signalling, cytokine receptors and downstream EMT scores for breast cancer cell lines with a spindle-like enrichment score from ssGSEA, the generic EMT signature was morphology were significantly higher than those for cell lines with- found to strongly correlate with the six cancer-specific EMT signa- out a spindle-like morphology (Fig 1B; P = 1.4E-6); this is consistent tures that were used for its derivation (Rho [+0.73, +0.97] and 2 1 Cancer Science Institute of Singapore, National University of Singapore, Singapore with the reported spindle-shaped morphology of Mes cells (Lee & with the majority of published cancer-specific EMT signatures (Rho 2 Institute of Molecular and Cell Biology, A*STAR, Singapore 3 Cancer Institute of Japanese Foundation for Cancer Research, Kyoto, Japan Nelson, 2012). In addition, cell lines with a high EMT score [+0.32, +0.84]; Supplementary Table S1C) for each respective 2 4 Department of Obstetrics and Gynaecology, National University Health System, Singapore displayed a significantly higher positive staining for VIM and CDH2, cancer type despite the small overlap in the signature genes. Surpris- 5 Department of Biochemistry, Yong Loo Lin School of Medicine, National University of Singapore, Singapore *Corresponding author. Tel: +65 6516 3242; Fax: +65 6516 1453; E-mail: [email protected] known markers of an EMTed phenotype (Thiery et al, 2009) ingly, EMT scores computed from the generic EMT signatures of †Co-senior authors. (P = 2.1E-5 and P = 9.1E-6, respectively; Fig 1C). Conversely, tumour and cell lines were strongly correlated (Rho >+0.89), even

ª 2014 The Authors. Published under the terms of the CC BY 4.0 license EMBO Molecular Medicine Vol 6 | No 10 | 2014 1279 1280 EMBO Molecular Medicine Vol 6 | No 10 | 2014 ª 2014 The Authors EMBO Molecular Medicine A generic signature to quantify EMT phenotype Tuan Zea Tan et al Research Article

ª 2014 The Authors. Published under the terms of the CC BY 4.0 license EMBO Molecular Medicine Vol 6 | No 10 | 2014 1279 1280 EMBO Molecular Medicine Vol 6 | No 10 | 2014 ª 2014 The Authors Tuan Zea Tan et al A generic signature to quantify EMT phenotype EMBO Molecular Medicine EMBO Molecular Medicine A generic signature to quantify EMT phenotype Tuan Zea Tan et al

AB AB

Figure 1. Derivation and application of cancer-specific epithelial-mesenchymal transition (EMT) signature. A A six-step scheme illustrating the generation of a cancer-specific EMT signature. Note that tumours and cell lines have their own cancer-specific EMT signatures. (Top right panel) Red and green bars on sample enrichment score (ES) bar chart indicate, respectively, mesenchymal-like (Mes) and epithelial-like (Epi) samples selected for building the BinReg EMT signature. (Middle right panel) Heatmap of the EMT signature from Significance Analysis of Microarray (SAM)/Receiver Operating Characteristics (ROC) analysis. The colour bar shows the EMT phenotype probability of cell line or tumour samples, sorted from most Epi to most Mes. Red and green bars indicate Mes and Epi samples selected for SAM/ROC analysis. (Bottom right panel) Plots of empirical cumulative distribution function of Mes (red) and Epi (green) gene sets. B Dot plot of EMT score (mean SEM) for breast cancer cell lines (n = 34) with spindle- and non-spindle-like morphologies. Mann–Whitney U-test P-value is shown. Figure 2. Derivation and application of generic epithelial-mesenchymal transition (EMT) signature. Æ C Immunohistochemistry staining heatmap of Oestrogen Receptor (ER), Progesterone Receptor (PR), and Epi (CDH1, ERBB2, CK19) as well as Mes (CK5, VIM, CDH2) A Circos plot illustrating the generic EMT signature: the overlap of ovarian (blue), breast (purple), lung (green), colorectal (yellow), bladder (red) and gastric (orange) markers (black = low, red = high, white = no data). Breast cancer cell lines (n = 39) are aligned from the most Epi to most Mes based on the EMT score, as shown by cancer-specific EMT signatures is shown. Links indicate overlapping genes (red = mesenchymal, green = epithelial). Heatmap on the inner ring indicates weight the bar chart. Dot plot is the -log10 P-value of two-sample Kolmogorov–Smirnov test. Arbitrary threshold of P < 0.001 was used to define Epi, intermediate and Mes computed based on Significance Analysis of Microarray (SAM) fold-change, false discovery rate, Receiver Operating Characteristics (ROC) and number of samples of a cell lines. Breast cancer cell line microarrays and subtype are from GSE16795 (Hollestelle et al, 2010). Subtype colour code: blue, Luminal; maroon, Basal. gene in each cancer-specific EMT signature (red = high, blue = low weight). On the outermost ring, genes are represented by ticks and aligned from the highest SAM fold-change to the lowest for each cancer type. Selected genes are labelled. B EMT score (mean SEM; y-axis) of breast cancer molecular subtypes as predicted using ssGSEA and signature from Prat et al (2010) in non-laser-capture micro- Æ dissected (non-LCM) cohort (n = 3,992; upper panel) and LCM cohort (n = 417; lower panel). The Mann–Whitney U-test P-value of binary comparison for each though the cell line generic EMT signature does not include stromal Claudin-Low breast cancers were more Mes (P = 1.98E-40 and subtype is given. Colour code: maroon, Basal; yellow, Claudin-low; light blue, Luminal-A; dark blue, Luminal-B; orange, ERBB2+; green, Normal-like. N.A, not applicable. components. This indicates that stroma-related genes have a limited P = 2.47E-68, respectively) in both non-LCM and LCM cohorts. Of Note that no P-value is available for Claudin-low and Normal-like subtypes in lower panel because n < 3. influence on the generic EMT score of tumours. We noted, however, note, the high similarity between the EMT profiles of breast cancer that the generic EMT signature had a marginal or no correlation subtypes in LCM and non-LCM cohorts indicates that EMT scoring with four of the published EMT signatures, probably due to the is able to capture an overall EMT status of a sample, even in the phenotype of a cell line under different interventions, which is in Finally, with the aim of developing a smaller, more cost-effective small number of genes in these signatures, or because the signature presence of stroma. To further ensure the validity of the generic full agreement with previous EMT studies (Onder et al, 2008; EMT signature, we explored the possibility of reducing the number was derived from non-malignant cells. Overall, these results demon- EMT signature, we computed the EMT scores for a panel of in vitro Hellner et al, 2009; Malizia et al, 2009; Yanagawa et al, 2009; of genes in our generic EMT signature (Supplementary Text, Supple- strate the consistency of the generic EMT signature with previously functional studies across various cancers (Supplementary Fig S3, Maupin et al, 2010; Taube et al, 2010; Ohashi et al, 2011; Cieply mentary Fig S5). We identified a 40–50% smaller generic EMT reported EMT-related genes and cancer-specific EMT signatures. Supplementary Table S3). In each functional study, the generic EMT et al, 2012; D’Amato et al, 2012; Cai et al, 2013; Deshiere et al, signature that has an overall correlation of 0.85–0.88 with the full Furthermore, the generic EMT signature is both versatile for the score accurately reflected the EMT phenotype regardless of the 2013), thus again validating our generic EMT scoring method. generic EMT signature and has good concordance (75.08–95.8%) in quantitation of EMT in all cancer types and not strikingly sensitive cancer type (Supplementary Fig S3). For example, consistently Pancreatic cancer was not included in our original derivation of estimating EMT status (Supplementary Text, Supplementary Tables to the presence of stroma, two important advantages for this system higher EMT scores were found for cell lines with CDH1 or NOTCH3 the EMT signature. As EMT has been implicated in pancreatic S3 and S4A). However, the following analyses continue to use the of classification. knockdown, cell lines treated with TGFb, and cell lines constitu- cancers, it is important that this generic EMT signature can also full generic EMT signature. To assess the utility of this generic EMT signature, we computed tively expressing EMT inducers, TWIST1, SNAIL, GSC, as compared accurately estimate the EMT status in pancreatic cancers. We found the EMT scores for laser-capture-micro-dissected (LCM) and non- with control cell lines (Supplementary Fig S3; P < 0.05). Conversely, that the generic EMT score correlates positively with the immuno- Application of the generic EMT signature LCM breast carcinoma (Fig 2B). Consistent with previous reports cell lines with over-expressed GRHL2—a transcription factor fluorescence staining of EMT markers such as ZEB1, VIM and meta- (Blick et al, 2008; Taube et al, 2010), we observed that Luminal-A, commonly under-expressed in EMTed cells (Cieply et al, 2012)— static ability in various pancreatic cancer cell lines (Supplementary A spectrum of EMT is found in multiple cancers Luminal-B and ERBB2+ breast cancers were more Epi (P = 0.0496, displayed a lower EMT score, indicating a more Epi phenotype. Fig S4, Supplementary Text). The data thus validate the generic We next performed generic EMT scoring on multiple clinical P = 3.34E-79 and P = 2.48E-6, respectively), whereas Basal and Thus, the EMT score could routinely identify the Epi or Mes EMT signature in pancreatic cancers. samples and cell lines (Fig 3, Supplementary Fig S6, Supplementary

ª 2014 The Authors EMBO Molecular Medicine Vol 6 | No 10 | 2014 1281 1282 EMBO Molecular Medicine Vol 6 | No 10 | 2014 ª 2014 The Authors Tuan Zea Tan et al A generic signature to quantify EMT phenotype EMBO Molecular Medicine EMBO Molecular Medicine A generic signature to quantify EMT phenotype Tuan Zea Tan et al

AB AB

Table S4A–D). A wide range of EMT scores was observed in blad- for survival where the composition of histotype or molecular der, breast, gastric, lung, ovarian and prostate cancers. Surprisingly, subtype may play a role; this suggests the requirement for stratifica- haematopoietic and lymphoid malignancies, such as lymphoma, tion of cancers in addition to deciphering the EMT status. This is acute myeloid leukaemia and multiple myeloma, also displayed a exemplified by the stratification of breast cancer molecular subtypes spectrum of EMT scoring, albeit over a narrower range. Colorectal (Prat & Perou, 2011), where there is a correlation for better DFS for cancer was predominantly Epi (P < 1E-50), whereas renal carci- patients with Epi breast cancers that are of a Basal and Claudin-Low noma exhibited strong Mes features (P = 2.47E-53), perhaps reflect- subtypes, but no correlation for other subtypes (Supplementary Fig ing that kidney epithelium derives from the condensation of S7). However, this correlation of EMT and DFS in Basal and mesodermal Mes cells. Interestingly, although hepatocytes originate Claudin-Low subtypes was not coherent in all breast cancer cohorts from the primitive Epi endoderm, liver carcinoma displayed an probably due to small sample sizes. extensive range in EMT score. Other tumours that were primarily Mes included germ cell tumours (P = 1.9E-22), malignant mela- EMT status does not necessarily translate to noma (P = 1.38E-42), sarcoma (P = 1.7E-34), and glioblastoma and chemotherapeutic resistance neuroblastoma (P < 1E-50). A similar mean and dispersion of the To investigate the association between EMT and chemotherapeutic EMT score was seen in cell lines (Fig 3), with a wide spectrum resistance, we compared the clinical outcomes of patients using the noted for cell lines derived from bladder, breast, gastric, liver, lung response evaluation criteria in solid tumours (RECIST) available for and prostate carcinoma. Colorectal carcinoma cell lines were breast (Horak et al, 2013) and ovarian cancer (The Cancer Genome predominantly Epi (P = 2.61E-17), whereas renal carcinoma Atlas Research, 2011) cohorts (Fig 5A). In these cohorts, patients (P = 7.92E-5), malignant melanoma (P = 8.17E-9), sarcoma with breast cancer had been treated with sequential neoadjuvant (P = 1.51E-7) and glioblastoma (P = 5.67E-19) cell lines were gener- therapy (doxorubicin and cyclophosphamide), whereas patients ally Mes, mimicking the observations in tumours. In concordance with ovarian cancer had undergone primarily platinum-based ther- with clinical samples, germ cell tumour cell lines showed a tendency apy. Without considering the treatment regimen, there was no to be Mes (P = 0.58); the lack of significance was presumably significant difference between the RECIST groups in terms of EMT because of the limited number of cell lines. Note that the tumours score. Thus, we categorized tumours into Epi, intermediate and and cell lines in Fig 3 were not paired. As a result, the composition Mes based on 2KS (P < 0.05) and observed an enrichment of Mes of histology, grade, stage of tumours and cell lines are different and breast cancers in the progressive disease (PD) category that leads to the difference in EMT score distribution, such as is the (P = 0.3303). Analyses with another 11 breast cancer cohorts case in prostate cancer. These results show that each cancer type (GSE48905, GSE33658, GSE23428, GSE22226, GSE18864, has a characteristic EMT spectrum. GSE28796, GSE16646, GSE22513, GSE4779, GSE18728 and GSE50948) (Farmer et al, 2009; Bauer et al, 2010; Korde et al, EMT status does not necessarily correlate with poorer survival 2010; Silver et al, 2010; Lehmann et al, 2011; Massarweh et al, To investigate if an EMTed status universally correlates with poor 2011; Carey et al, 2012; Esserman et al, 2012; Evans et al, 2012; survival, we performed Kaplan–Meier analyses by cohort and by Knudsen et al, 2014; Prat et al, 2014), within which patients had Figure 3. Epithelial-mesenchymal transition (EMT) scores in different cancers types. cancer type comparing Epi and Mes tumours (Fig 4). Intriguingly, a been administered with different neoadjuvant treatment regimens, Scatter plot of EMT scores (mean SEM; y-axis) of various cancers in clinical samples (upper panel) and cell lines (lower panel) sorted by cancer type and mean EMT score. � transitioned status did not universally correlate with overall survival including fulvestrant, anastrazole, carboplatin, doxorubicin and EMT score nearer to +1.0 is more mesenchymal-like (Mes), whereas EMT score nearer to 1.0 is more epithelial-like (Epi). EMT scores of overlapping cell lines in Cancer Cell � (OS) or disease-free survival (DFS), as shown in the hazard ratio other drugs (Supplementary Fig S8A), showed a similar distribution Line Encyclopedia (CCLE) (Barretina et al, 2012) and SANGER/COSMIC (Garnett et al, 2012) collections were averaged. (HR) plots (Fig 4). In order to include as many data as possible, we of Epi, intermediate and Mes breast cancers in each clinical adopted a broad definition of DFS which encompasses progression- response group. Notably, the worst response group (PD or residual free, (local) recurrence-free, and distant recurrence/metastasis-free disease) comprised mostly patients with Mes breast cancers. Thus, responders tended to have lower EMT score in predominantly Mes samples for certain drugs. Surprisingly, the EMT status did not survival. In general, patients with Epi ovarian cancer (cohort mean there was a trend towards either an increasing proportion of Mes or melanoma (Supplementary Fig S8B). These data suggest that EMT systematically translate to cellular chemotherapeutic resistance

atic cancer (lHR = 0.6006) and glioblastoma (lHR = 0.81) showed resistant patients. We also noted a trend towards a decrease in the results remain inconclusive at this time. The lack of a conclusive between cellular phenotype and attaining resistance (Witta et al, better OS. There was no correlation between EMT status and OS for Epi proportion amongst patients with ovarian cancer and a change result might be because the majority of these patients had been 2006; Arumugam et al, 2009; Hrstka et al, 2010; Sethi et al, 2010; patients with acute myeloid leukaemia, colorectal or lung cancer. from complete response (CR) to PD (50–42%), albeit there was no treated with more than one chemotherapeutic compound, which Marchini et al, 2013). Regardless of the cancer type, Mes and Epi

Surprisingly, patients with Mes breast cancer (lHR = 1.48; significant enrichment in PD for patients with Mes ovarian cancers may confound the role of EMT and chemotherapeutic resistance in cell lines were preferentially sensitive to certain compounds. Mes

P = 0.006) and malignant melanoma (lHR = 1.48) had better OS (P = 0.556). these patients. Furthermore, as these data are from relatively small cell lines were more resistant to Afatinib and Gefinitib (against (Fig 4A), which is in stark contrast with previous reports (Thiery Since the distribution of the EMT score did not allow us to segre- cohorts, further study is required to validate the current observa- EGFR), but were more sensitive to the PDK1 kinase inhibitor, et al, 2009; Hrstka et al, 2010; Loboda et al, 2011; Cieply et al, gate certain other cancers into Epi, intermediate and Mes groups, tions. BX-795 and the HSP-90 inhibitor, Elesclomol (Fig 5B). Intriguingly, 2012; Huang et al, 2012; Byers et al, 2013; Frisch et al, 2013). we next investigated the EMT score profiles of responders and non- Thus, to search for an association between EMT and chemother- Epi cell lines were resistant to 64 compounds, whereas Mes cell Equally intriguing results were observed for DFS (Fig 4B), with responders in these cancers (Supplementary Fig S8B). This was apeutics, and to explore the potential therapeutic options for Epi lines were resistant to only 7, albeit the correlation was weak (Rho poorer DFS for patients with ovarian and colorectal cancers performed using cohorts of predominantly Epi colorectal cancers and Mes cancers, we analysed drug sensitivity data from the [ 0.35, +0.37]). When stratified by cancer type, we observed a 2 � (lHR = 0.5165, P < 0.001; and lHR = 0.7669, P = 0.002, respec- (GSE19862, GSE35452, GSE46862) (Gim et al, 2014), a cohort of SANGER/COSMIC (Garnett et al, 2012) database (April 16, 2013) in similar pattern of preferential sensitivity to certain compounds. tively), and a marginal correlation noted for patients with bladder head and neck cancers (GSE32877) (Tomkiewicz et al, 2012) and a cell line models. We correlated the EMT score with the half-maxi- Notably, Mes pancreatic cancer, malignant melanoma, renal cancer

carcinoma (lHR = 0.8473). For liver and renal carcinoma, patients cohort of predominantly Mes melanoma (GSE22968) (Beasley et al, mal inhibitory concentration (IC50) of 138 compounds (Fig 5B, and liver cancer cell lines were more sensitive to compounds with Mes tumours had better DFS than their Epi counterparts 2011). There was no significant difference between responders and Supplementary Fig S9, Supplementary Table S5) using the Spear- targeting microtubule dynamics, such as Vinblastine and Docetaxel.

(lHR = 1.238 and 3.948, respectively). The result for DFS in patients non-responders in terms of EMT score, albeit there was a slight man’s correlation coefficient test, as it measures the overall trend Comparatively, Mes breast, lung and uterine cancer cell lines were with breast cancer was unclear (HR = 0.4432–2.622; P = 0.252). trend towards responders tending to have a higher EMT score in and requires no definition of sensitive or resistant categories. We more resistant to Afatinib and Gefinitib (Supplementary Fig S9). Overall, the EMT status is unlikely to be the sole prognostic factor predominantly Epi colorectal cancer, and a slight trend that employed a less stringent threshold (P < 0.1) because of the limited Previous observations reported that EMT is associated with EGFR

ª 2014 The Authors EMBO Molecular Medicine Vol 6 | No 10 | 2014 1283 1284 EMBO Molecular Medicine Vol 6 | No 10 | 2014 ª 2014 The Authors Tuan Zea Tan et al A generic signature to quantify EMT phenotype EMBO Molecular Medicine EMBO Molecular Medicine A generic signature to quantify EMT phenotype Tuan Zea Tan et al

HR [lHR] = 0.68, P = 0.018), gastric cancer, (lHR = 0.7013), pancre- a decreasing proportion of Epi breast cancers amongst chemo- may correlate with chemotherapeutic resistance; however, the (Fig 5B, Supplementary Fig S9), contradicting previous associations atic cancer (lHR = 0.6006) and glioblastoma (lHR = 0.81) showed resistant patients. We also noted a trend towards a decrease in the results remain inconclusive at this time. The lack of a conclusive between cellular phenotype and attaining resistance (Witta et al, better OS. There was no correlation between EMT status and OS for Epi proportion amongst patients with ovarian cancer and a change result might be because the majority of these patients had been 2006; Arumugam et al, 2009; Hrstka et al, 2010; Sethi et al, 2010; patients with acute myeloid leukaemia, colorectal or lung cancer. from complete response (CR) to PD (50–42%), albeit there was no treated with more than one chemotherapeutic compound, which Marchini et al, 2013). Regardless of the cancer type, Mes and Epi

AB AB

Figure 5. Generic epithelial-mesenchymal transition (EMT) and drug sensitivity. Figure 4. Correlation of Epithelial-mesenchymal transition (EMT) scores and survival. A Bar plots of breast (n = 270; left panel) and ovarian (n = 328; right panel) cancers stratified by EMT status and clinical response based on response evaluation criteria A, B Plot of log hazard ratio (HR; mean 95% confidence interval) comparing (A) overall survival (OS) and (B) disease-free survival (DFS) of Epi and Mes tumours in in solid tumours (RECIST). Regimen was neoadjuvant doxorubicin and cyclophosphamide for breast cancer, and platinum-based adjuvant/progression/recurrence 2 Æ different cancers and cohorts. DFS includes progression-free, recurrence-free and distant metastasis-free survival (cohorts inclusive of distant metastasis-free chemotherapy for ovarian cancer. Percentage distribution of EMT status is given in each clinical response group. Abbreviation: CR, complete response; PR, partial survival were indicated with *). Corresponding P-values from the log-rank test are given next to each cohort, and those with significant differences (P < 0.05) are response; SD, stable disease; PD, progressive disease. Green, epithelial-like (Epi); orange, intermediate; red, mesenchymal-like (Mes). marked red. Log HR < 0.0 indicates Epi tumours with survival benefit, whereas log HR > 0.0 indicates Mes tumours with survival benefit. Meta-analysis P-value B Volcano plot of EMT correlation with drug sensitivity regardless of cancer type. Rho [ 1.0,+1.0](x-axis) and -log P-value (y-axis) were computed by Spearman’s 2 2 2 � 10 for effect or heterogeneity was computed using DerSimonian–Laird binary random effect (for overall) or Peto fixed effect method (for individual cancer). correlation coefficient test. Dashed line of P-value = 0.1 is plotted. Red and green indicate higher drug resistance in Mes tumours (Rho [0,+1.0]) and Epi tumours 2 (Rho [ 1.0, 0]), respectively. 2 � C Kaplan–Meier analysis comparing overall survival (left panel) and disease-free survival (right panel) of Epi (green) and Mes (red) ovarian cancer patients who underwent a treatment regimen with (dark colour) or without (light colour) paclitaxel. P-value reported was computed by log-rank test. Abbreviation: HR = hazard inhibitor resistance in non-small cell lung cancer (NSCLC) (Byers responses to certain compounds. In addition, we show that Epi and ratio. et al, 2013). Using our generic EMT score, we observed that Epi Mes cell lines also have preferential responses to certain cell lines not limited to NSCLC were more sensitive to inhibitors of compounds and that EMT is not the only mechanism driving resis- EGFR or both EGFR and ERBB2 (Erlotinib, Lapatinib, BIBW2992 tance in all chemo- or targeted therapies. patients (Fig 5C); this was performed using the 2KS criteria tumours is also observed in glioma (Desmedt et al, 2009) and and Gefitinib) in the SANGER/COSMIC (Garnett et al, 2012) and These intriguing preferential drug sensitivities of Epi and Mes described for Fig 5A. Because few patients were treated without multiple myeloma (Erdem-Eraslan et al, 2013) (Supplementary Fig

Cancer Cell Line Encyclopedia (CCLE) (Barretina et al, 2012) data- cancers in the correlation analysis of EMT score and the IC50 of cis-/carboplatin (n < 10), we focused our analysis on the effect of S10, Supplementary Information). Even though glioma patients bases (Supplementary Table S5). Cell lines with sensitizing EGFR 138 compounds (EMT score–IC50; Supplementary Table S5) paclitaxel. Surprisingly, patients with Epi and Mes ovarian receiving radiotherapy and chemotherapy generally have better activating mutations (L861Q, G719S, exon 19 in-frame deletion) prompted us to investigate the relevance of these findings, as cell cancers, who had received a regimen with paclitaxel, had signifi- OS, the benefit is greater in patients with Mes glioma (Carr et al, 2004) exhibited significantly lower EMT scores lines do not fully exemplify the behaviours of primary tumours. cantly different OS and DFS outcomes as compared with those (P = 0.0117). In contrast, patients with Epi multiple myeloma compared with wild-type cell lines (P = 0.0056). On the other We selected ovarian cancer as a model for this pilot study who had received a regimen without paclitaxel (Fig 5C). Epi have better DFS rates when administered with bortezomib instead hand, cell lines harbouring the secondary gatekeeper EGFR-T790M because the first-line treatment for ovarian cancer is primarily patients receiving a regimen containing paclitaxel showed poorer of dexamethasone (P = 0.0349). However, we observed no differ- mutation, which confers resistance to EGFR inhibitors (Gazdar, cisplatin/carboplatin and paclitaxel, which could provide a less DFS (P = 0.0039), whereas Mes patients treated with paclitaxel ence in patients with ER+ breast cancers (Mulligan et al, 2007) 2009), were more Mes (EMT score =+0.23). Hence, a higher preva- convoluted mechanism of interaction between EMT and the drugs. showed better DFS (P = 0.039) and OS (P = 0.0006); these results who were administered with letrozole or tamoxifen in terms of lence of sensitizing EGFR mutations could account for the higher Provocatively, in the EMT score–IC50 correlation analysis, we for the Mes patients mirrored those garnered from the EMT EMT stratification. Overall, these results provide in vivo evidence response rate of Epi cancers to EGFR inhibitors. Although it is too found that Mes ovarian cancers have preferential sensitivity to score–IC50 analysis, indicating that Mes is more sensitive to pacli- for the findings of the EMT score–IC50 correlation analysis and preliminary to conclude if Epi or Mes is resistant to certain cisplatin (Rho = 0.37) and paclitaxel (Rho = 1.0; Supplemen- taxel. We found no significant difference in DFS or OS outcome show the preferential drug sensitivity in patients with Epi and À À compounds (due to the modest correlation and P-values), these tary Table S5). A Kaplan–Meier analysis was then performed to for ovarian cancer patients who exhibited an intermediate EMTed Mes tumours as well as their differential responses to particular results suggest that Epi and Mes cell lines have differential stratify the treatment regimens into Epi versus Mes ovarian cancer phenotype. Such differential therapeutic response in Epi and Mes chemotherapeutic regimens.

ª 2014 The Authors EMBO Molecular Medicine Vol 6 | No 10 | 2014 1285 1286 EMBO Molecular Medicine Vol 6 | No 10 | 2014 ª 2014 The Authors Tuan Zea Tan et al A generic signature to quantify EMT phenotype EMBO Molecular Medicine EMBO Molecular Medicine A generic signature to quantify EMT phenotype Tuan Zea Tan et al

AB AB

Discussion by changes in the tumour microenvironment (Valastyan & Weinberg, expression, this may generate a high EMT score in some Epi highly Mes, germ cell tumour, is extraordinarily sensitive to 2011; Lee & Nelson, 2012; Tam & Weinberg, 2013; Van den Eynden tumours. The EMT scores of LCM and non-LCM breast cancer cisplatin (Masters & Koberle, 2003; Eckstein, 2011). Furthermore, Increasing evidence points to the role of EMT in cancer progression, et al, 2013) or the influence of drug treatment or cytotoxic stress cohorts (Fig 2B) showed a marginally lower EMT score for Epi, even though EMT is often linked with the acquisition of stem cell- metastasis and drug resistance. However, the difficulty in making (Frisch et al, 2013; Marchini et al, 2013), amongst other factors. Luminal-A subtype, suggesting that stromal contribution may, to like features, Prrx1 uncouples EMT and stemness, resulting in a an adequate assessment of EMT in tumours has caused dispute as to This is exemplified in pancreatic carcinoma, which derives from the some extent, obscure a precise assessment of the EMT score. drug-resistant, metastatic colonization (Ocana et al, 2012). Thus, whether EMT exists in cancer (Jordan et al, 2011). To address this same endodermal anlage as the colon yet exhibits a relatively Mes However, assessing the stromal contribution is non-trivial given the we postulate that it is not solely the acquisition of EMT but the issue, we developed a generic EMT signature to quantitatively esti- phenotype as compared with colon carcinoma. Although pancreatic RNA instability and labour-intensive procedure of segregating stro- EMT stem cell-like phenotype that engenders drug resistance mate the extent of EMT in tumours and cell lines. To the best of our carcinoma comprises a large fraction of stromal cells (Beatty et al, mal from cancer cells (Park et al, 2011). It is therefore difficult to (Brabletz, 2012). Frisch et al (2013) proposed a similar concept, knowledge, this is the first time a generic EMT signature has been 2011), pancreatic carcinoma cell lines exhibit the same EMT quantify the influence of stroma in our EMT scoring. Nevertheless, suggesting that EMT is acquired by triggering EMT inducers to sought in order to capture the universal features of EMT in tumours spectrum as colon cells (Fig 3), supporting the notion that the EMT in addition to the minute EMT score differences in LCM and non- repress cell polarity and that stem cell-like features are acquired by or in cells. Previous reports indicate that intermediate states of EMT score still arises from the contributions of the pancreatic carcinoma LCM breast cancer cohorts, we have also shown a strong correlation engaging additional programs such as the WNT and Hippo path- display the highest plasticity (Jordan et al, 2011; Huang et al, 2013) cells not just the stromal cells. In a similar way, liver carcinoma of the generic EMT score computed using tumour- and cell line- ways. It is likely that the present generic EMT signature estimates and thus represent an appropriate stage within which to induce or shows a wide spectrum of EMT scores. As the liver also derives specific signatures (Supplementary Table S1C). This result indicates the degree of EMT but cannot estimate the degree or behaviour of reverse EMT. The change in EMT score captures and reflects this from the primitive endoderm, it would be expected that liver that whereas stroma may obscure a precise assessment of EMT by a cancer stem cell-like phenotype. This distinction is evident in the phenotypic transition in the cell or tissue; this method is judiciously carcinoma would exhibit an Epi phenotype. In this case, in addition transcriptome, the influence is not overwhelmingly striking. Thus, minute differences between control and HMGA2-knockdown—a illustrated in a previous application where Epi MCF7 breast cancer to the role of the stroma, it is intriguing to consider that the cell of we believe the EMT scoring is relatively independent of stromal gene implicated in stemness (Copley et al, 2013)—MDA-MB-231 cells displayed a shift in the EMT spectrum when transfected with origin may have undergone an E- to N-cadherin switch. influence, but likely not of clonal heterogeneity. Others have shown breast cancer cells (Supplementary Fig S3) and may explain the SNAI1 (Akalay et al, 2013). Having the capacity to monitor such a Although many reports have associated the EMT status with that, although there is a higher proportion of EMT carcinoma cells limited correlation between generic EMT score and therapeutic transition would be instrumental for assessing the effectiveness of survival (Witta et al, 2006; Arumugam et al, 2009; Hrstka et al, in basal-like tumours, such cells are also seen in luminal breast resistance, as cancer stem cells may have an impact on tumour EMT reversion therapy and for identifying an intermediate state of 2010; Sethi et al, 2010; Loboda et al, 2011; Cieply et al, 2012; Byers tumours (Sarrio et al, 2008). It would thus be useful to analyse the progression in breast (Sarrio et al, 2008) and colon (Brabletz, EMT that would have an improved chemotherapeutic response. It is et al, 2013; Marchini et al, 2013), our EMT scoring does not wholly phenotype and clonogenicity of these EMTed cells and of CTCs in 2012) Epi tumours. In our preliminary analysis, although there are important to note that full reversion of EMT may not be desirable, support these findings. We show that the EMT status is linked to OS addition to EMT scoring (Thiery & Lim, 2013). On-going studies on some moderate correlations between stemness and generic EMT as Mes micro-metastases must re-acquire an Epi phenotype to prolif- in ovarian cancer, gastric cancer and glioblastoma, but not in other CTCs in our laboratory are exploring whether the EMT score reflects score, the correlation was not consistent across cancer types, erate at the metastatic site (Thiery, 2002); in agreement was the carcinoma types. In terms of DFS, patients with Epi ovarian and the propensity of a cancer to disseminate and become refractory to which may suggest that different cancers enrol distinct programs recent demonstration that reversing EMT may promote metastatic colorectal cancers have a better prognosis. The discrepancy in the therapy. CTCs exhibit a wide spectrum of EMT phenotypes, irre- to acquire stemness (Supplementary Text, Supplementary Fig S11). colonization (Tsai et al, 2012). The main challenge, therefore, is reported correlations between EMT status and survival is intriguing, spective of the primary tumour (Valastyan & Weinberg, 2011; In addition, the existence of different types of stem cells within a that we do not know the precise intermediate states and under what as EMT was posited to be involved in cancer progression, metastasis Thiery & Lim, 2013; Yu et al, 2013). Thus, the capacity of a primary cancer—as shown in breast cancers—has to be taken into account conditions or context cancer cells in the primary tumour or at the and drug resistance, all of which are strongly connected with poorer tumour to metastasize may reside in a small subset of cells, the when considering the correlation of stemness and EMT (Liu et al, distant metastatic sites can exit dormancy and resume growth or survival. It is noteworthy that most breast carcinoma of the lobular phenotype of which is not known and cannot be assessed by an 2014). Finally, the lower sensitivity of Mes cell lines to various become chemo-resistant. Here, we validated the efficacy of our histotype, which are notoriously known for not expressing E-cadherin, EMT scoring method. compounds (EGFR inhibitors) may be due to a lower prevalence of generic EMT scoring system to reflect the EMT transition in a panel are not more aggressive than E-cadherin-positive invasive ductal Our findings are apparently discrepant with previous connections the targeted mutations in these cell lines. However, it is still of functional studies across multiple cancer types. The spectrum of carcinoma (Ferlicot et al, 2004). Here, we also showed that patients between EMT status and drug resistance (Witta et al, 2006; Arumugam unknown whether an EGFR mutation is the main driver in these EMT identified across the various tumours implies a causal role of with Mes breast cancers appear to have better OS and DFS than et al, 2009; Hrstka et al, 2010; Sethi et al, 2010; Marchini et al, cancers and whether these mutations—acquired or inherent—play the EMT status in the differential characteristics of these cancers those with Epi breast cancers, seemingly in opposition to what has 2013). Whilst we acknowledge the limitations of cell lines and IC50 a role in initiating or regulating EMT. and in their responses to treatment. Remarkably, the EMT spectrum been previously reported (Hrstka et al, 2010; Taube et al, 2010; Cai as a drug assay (Haibe-Kains et al, 2013), we believe our results Overall, we demonstrate the feasibility of applying a generic was highly similar between cell lines and tumours in a given cancer et al, 2013). On closer look, this difference likely arises from the give a bird’s-eye view of EMT and drug resistance. By assessing OS EMT score for the examination of the EMT spectrum in different type, which verifies the capacity of the EMT score to capture the distribution of patients with luminal and triple-negative breast and DFS outcomes of ovarian cancer patients through EMT status cancers, as well as its correlation with survival and chemotherapeu- EMT phenotype rather than the influence of the stroma. cancers in the cohort. As shown in Fig 4B, a breast cancer cohort and treatment regimen, we found that Epi ovarian cancers are more tic resistance. We believe the proposed generic EMT score is a The type of cancer is generally considered to be a good indicator (GSE25066) with a lower percentage of Luminal-B and ERBB2+ resistant to paclitaxel, whereas Mes ovarian cancers have a prefer- promising, general-purpose tool with which to estimate EMT pheno- of the EMT status (Fig 3). For example, colorectal carcinoma is breast cancers would show a better DFS for Epi breast cancers. Even ential sensitivity to paclitaxel. This shows that cancers with different types, regardless of cancer type, to systematically investigate EMT primarily Epi, whereas glioblastoma, neuroblastoma, osteosarcoma, though Luminal-B and ERBB2+ breast cancer subtypes are of the Epi degrees of EMT respond distinctly to particular compounds—in and to more objectively assess the impact of EMT effectors or drugs malignant melanoma and germ cell tumours are primarily Mes. It is type (Blick et al, 2008) (Fig 2B), they have poor OS and DFS, similar accordance with our previous work in ovarian cancer (Miow et al, on phenotype changes. It also offers a more objective EMT scoring unclear, however, whether these phenotypic traits are inherent or to that of the Mes type, triple-negative breast cancers (Prat & Perou, 2014)—and is supportive of the utility of the EMT score-IC50 correla- in vitro as opposed to estimations by visual inspection or marker acquired. Inherent EMT traits could be a reflection of the cell of 2011; Ishitobi et al, 2013). Consequently, the more prevalent Epi tion analysis in cell lines. More importantly, these results identify assessment. origin or the lineage of the cancer. Indeed, melanoma and neuro- Luminal-B and ERBB2+ breast cancers give rise to poorer survival that patients with Mes, but not Epi, ovarian cancer would benefit blastoma are derived from transformed melanocytes and sympa- curves in Epi breast cancer cohorts, suggesting that heterogeneity from therapeutic regimens that contain paclitaxel. In line with this, thetic neural progenitor cells, respectively (Nakaya & Sheng, 2013), within a cancer type could mask and perplex the role of EMT. Thus, the Japanese Gynecologic Oncology Group demonstrated a survival Materials and Methods which originate from the neural crest and delaminate through an stratification by molecular subtypes may be required to study the advantage for a weekly administration of paclitaxel compared with EMT before colonizing different embryonic sites where they role of EMT. Indeed, when stratified by breast cancer molecular a once in 3 week administration of paclitaxel in combination with Data pre-processing for Affymetrix microarray expression data undergo differentiation into melanocytes, glial cells and neurons of subtype (Prat & Perou, 2011), patients with Epi breast cancers show carboplatin in relapsed patients with ovarian cancer (Baird et al, the peripheral nervous system. Thus, these neural crest cell deriva- better DFS if their cancers are of the Basal and Claudin-Low 2010), where relapsed ovarian cancer is shown to be enriched for Pre-processing and quality checks were performed as described (Tan tives maintain an intrinsic Mes phenotype. Another example is subtypes, but not the other subtypes (Supplementary Fig S7). Mes (Tan et al, 2013). Similar to our findings, gastric cancer et al, 2013) (Supplementary Materials and Methods). Data sets on found in breast cancer. The most EMTed breast cancers belong to The crosstalk between stromal and cancer cells plays a major patients, who have an enrichment for Mes, respond differently to the Affymetrix U133A or U133Plus2 platforms for bladder (n = 132), the Claudin-Low subtype and are likely derived from the highly role in metastasis (Park et al, 2011) and hence may influence the chemotherapy from the subtype of patients not enriched for Mes breast (n = 3992), colorectal (n = 1820), gastric (n = 231) and ovar- plastic cells of the basal layer of the mammary gland; the less plastic results of EMT scoring. In breast cancer, mammary Epi cells can and are more sensitive to cisplatin (Tan et al, 2011). Overall, our ian (n = 1538) cancers, as well as NSCLC and lung adenocarcinoma luminal cells, in contrast, are thought to give rise to the Basal adopt a stromal gene expression pattern indistinguishable from reac- data indicate that not all Mes carcinoma are resistant to chemother- (n = 481) were downloaded from Gene Expression Omnibus (GEO), subtype (Taddei et al, 2008; Lim et al, 2009; Molyneux et al, 2010) tive stroma when undergoing EMT (Farmer et al, 2009). As there is apy and that the EMT status does not necessarily translate to a Array Express, Expression Project for Oncology (ExpO) and The (Fig 2B). In the case of acquired EMT, the process may be triggered no distinction between reactive stroma and EMT-induced stromal propensity towards drug resistance. Indeed, testicular carcinoma, a Cancer Genome Atlas (TCGA) (Supplementary Table S6). An LCM

ª 2014 The Authors EMBO Molecular Medicine Vol 6 | No 10 | 2014 1287 1288 EMBO Molecular Medicine Vol 6 | No 10 | 2014 ª 2014 The Authors Tuan Zea Tan et al A generic signature to quantify EMT phenotype EMBO Molecular Medicine EMBO Molecular Medicine A generic signature to quantify EMT phenotype Tuan Zea Tan et al

breast cancer cohort (n = 417) was provided by the Japanese The final cancer-specific EMT signature (generated by SAM/ Molecular Signature Database (MSigDB): http://www.broadinstitute.org/gsea/ The paper explained Foundation for Cancer Research (http://www.jfcr.or.jp/english; ROC) is a refinement of the initial EMT signature (generated by msigdb/index.jsp Supplementary Materials and Methods; GSE54002). Normalization BinReg). Although it seems redundant to have an initial followed Cancer Cell Line Encyclopedia (CCLE): http://www.broadinstitute.org/ccle/home Problem was performed independently on each cohort using R version 3.01, by a refined final EMT signature, the benefit of this approach is During epithelial-mesenchymal transition (EMT), epithelial cells lose SANGER COSMIC cell line: http://cancer.sanger.ac.uk/cancergenome/projects/ Bioconductor Affy Package 1.38.1, Robust Multichip Average threefold. First, since some of the collected, published EMT signa- polarity and acquire migratory properties reminiscent of mesenchymal cell_lines/ (Gautier et al, 2004), and ComBat (Johnson et al, 2007) was applied tures are derived from different cell types and from a relatively cells. EMT is a dynamic process, not a binary process, with intermedi- Compute Generic EMT score: http://www.csi.nus.edu.sg/bioinfo/index.php for batch adjustment on the compiled, normalized data sets sepa- smaller number of cell lines, these published EMT signatures may ary states, and is thus still not easily ascertained in cultures or in vivo. BinReg/Profiler software: http://dig.genome.duke.edu/software.html Consequently, its role in cancer remains controversial. rately. Normal tissues were removed from the batch-adjusted data. not be applicable universally, as they may be cell line-specific or DAVID functional annotation tool: http://david.abcc.ncifcrf.gov/summary.jsp Cell line collections (Supplementary Table S7), including SANGER/ cancer-specific. In this study, we used a large panel of cell lines to Results COSMIC (Garnett et al, 2012), CCLE (Barretina et al, 2012) data sets derive an EMT signature specific to each cancer type, and hence, it We used gene expression to establish an EMT scoring method and and validation data sets (Supplementary Table S8), were subjected is less likely that the derived signature contains features unique to quantitatively estimated the degree of EMT ( 1.0 to +1.0) in a large References À to the same normalization procedure. a single cell line. Second, to ensure accuracy of the final EMT collection of cell lines and tumours reflecting epithelial and mesen- signature, we validated the initial EMT signature on two indepen- chymal states as well as the potential intermediate states that occur Akalay I, Janji B, Hasmim M, Noman MZ, Andre F, De Cremoux P, Bertheau P, during transition. Predictive modelling and validation by BinReg dent functional EMT studies. Third, regenerating the EMT signa- Badoual C, Vielh P, Larsen AK et al (2013) Epithelial-to-mesenchymal ture by SAM/ROC from the most Epi or most Mes tumours transition and autophagy induction in breast carcinoma promote escape Impact from T-cell-mediated lysis. Cancer Res 73: 2418 – 2427 Expression data analysis based on a binary regression model using ensured the additional changes sometimes acquired in cell lines We applied EMT scoring to ascertain its efficacy in correlating EMT the BinReg v2.0 (Profiler, http://dig.genome.duke.edu/software. would not be included and distort the EMT signature for tumours status with patient survival rates and responses to treatment. Such Arumugam T, Ramachandran V, Fournier KF, Wang H, Marquis L, Abbruzzese html) was described previously (Gatza et al, 2010; Tan et al, 2013). in general. versatile EMT scoring may enable the objective and systematic inves- JL, Gallick GE, Logsdon CD, McConkey DJ, Choi W (2009) Epithelial to Details are given in the Supplementary Materials and Methods. tigation of EMT across many parameters of cancer progression, mesenchymal transition contributes to drug resistance in pancreatic Derivation of generic EMT signature survival and throughout the clinical response to therapy. cancer. Cancer Res 69: 5820 – 5828 Generation of cancer-specific EMT signature Baird RD, Tan DS, Kaye SB (2010) Weekly paclitaxel in the treatment of We derived a generic EMT signature from the overlap between recurrent ovarian cancer. Nat Rev Clin Oncol 7: 575 – 582 Aside from an ovarian- and breast cancer-specific EMT signatures, specific EMT signatures generated for bladder, breast, colorectal, Barretina J, Caponigro G, Stransky N, Venkatesan K, Margolin AA, Kim S, which we derived previously from CDH2 and CDH1 immunofluores- gastric, lung and ovarian cancer types. We weighted the genes that Statistical analysis Wilson CJ, Lehar J, Kryukov GV, Sonkin D et al (2012) The Cancer Cell Line cence staining (Akalay et al, 2013; Miow et al, 2014), we devised a were selected in six cancer-specific EMT signatures using the Encyclopedia enables predictive modelling of anticancer drug sensitivity. strategy to generate cancer-specific EMT signatures for the other formula: for gene g, the weight of the gene is given by: DerSimonian–Laird binary random or Peto fixed effect meta-analysis Nature 483: 603 – 607 types of cancer (bladder, colorectal, gastric and lung), as depicted was conducted using OpenMeta[Analyst] software with the default Bauer JA, Chakravarthy AB, Rosenbluth JM, Mi D, Seeley EH, De Matos by the six-step scheme in Fig 1A: D settings. The log-rank test in the Kaplan–Meier analyses was Granja-Ingram N, Olivares MG, Kelley MC, Mayer IA, Meszoely IM et al 2:0 nd weight g log 2 fcgd ROCgd 0:5 (2010) Identification of markers of taxane sensitivity using proteomic and 1 Published EMT gene sets from Molecular Signature database q 1:0 D computed by GraphPad Prism version 5.0 (GraphPad Software, La ð Þ¼d 1 ð Þ� gd �ð � Þ� i 1 ni v4.0 (Subramanian et al, 2005) and previous literature (Lee X¼ ð þ Þ ¼ Jolla, CA). Mann–Whitney, Fisher’s exact and Spearman’s correla- genomic analyses of breast tumors from patients receiving neoadjuvant et al, 2006; Carretero et al, 2010) (Supplementary Table S9) P tion coefficient tests were computed by Matlab R2012a, statistics paclitaxel and radiation. Clin Cancer Res 16: 681 – 690

were collated. where D is the total number of diseases (D = 6 in this case), fcgd toolbox version 8.0 (MathWorks, Natick, MA). Beasley GM, Riboh JC, Augustine CK, Zager JS, Hochwald SN, Grobmyer SR,

2 ssGSEA score (Verhaak et al, 2013) was computed for EMT gene and qgd are the fold-change and q-value% of the gene, g, of Peterson B, Royal R, Ross MI, Tyler DS (2011) Prospective multicenter

sets on cancer cell lines and correlated with gene expression of disease, d, as computed by SAM. ROCgd is the ROC value of gene, Supplementary information for this article is available online: phase II trial of systemic ADH-1 in combination with melphalan via

known Mes and Epi markers (TWIST1, SNAI1, SNAI2, VIM, g, of disease, d, and nd is the number of samples in disease, d. http://embomolmed.embopress.org isolated limb infusion in patients with advanced extremity melanoma. J CDH2, ZEB1 and CDH1, DDR1, ERBB2, ERBB3, KRT19) (Thiery The formula will give higher weights to genes that have a large Clin Oncol 29: 1210 – 1215 et al, 2009). fold-change, a small q-value%, a large ROC value and a large Acknowledgements Beatty GL, Chiorean EG, Fishman MP, Saboury B, Teitelbaum UR, Sun W, 3 The gene set that best correlated with the enrichment score was number of samples. We ranked and selected the genes with a We thank Dr. R. Jackson for her careful English editing. We thank the finan- Huhn RD, Song W, Li D, Sharp LL et al (2011) CD40 agonists alter tumor chosen to rank the cell lines. The 10–20 most Mes and most Epi z-transformed weight > 3.09 or P < 0.001 (Supplementary Table S1A cial support from the Cancer Science Institute of Singapore, National stroma and show efficacy against pancreatic carcinoma in mice and cell lines were selected for BinReg modelling. Two data sets, and B). Research Foundation Prime Minister’s Office Singapore, National Medical humans. Science 331: 1612 – 1616 GSE9691 (Onder et al, 2008) and GSE24202 (Taube et al, 2010), Research Council/National University Cancer Institute of Singapore Center Blick T, Widodo E, Hugo H, Waltham M, Lenburg ME, Neve RM, Thompson were used for BinReg parameter settings and to ensure validity Computation of EMT score Grant/EMT Theme CG12Aug12, Department of Biochemistry of the National EW (2008) Epithelial mesenchymal transition traits in human breast of the derived EMT signature. (Note: Steps 1–3 can be recursive University of Singapore, and Institute of Molecular and Cell Biology at cancer cell lines. Clin Exp Metastasis 25: 629 – 642 to identify an initial BinReg EMT signature of sufficient accuracy To compute the EMT score of a sample, we adopted a similar A*STAR, Singapore. Borrell B (2010) How accurate are cancer cell lines? Nature 463: 858 in predicting the EMT status.) approach to that used in ssGSEA (Verhaak et al, 2013). The empirical Brabletz T (2012) To differentiate or not–routes towards metastasis. Nat Rev 4 The BinReg EMT signature was then used to predict the EMT cumulative distribution function (ECDF) was estimated for Epi and Author contributions Cancer 12: 425 – 436 status of cell lines and tumour samples specific to a particular Mes gene sets. The 2KS test was employed to compute the difference JPT and RYH conceived the idea; JPT and RYH devised the project and obtained Byers LA, Diao L, Wang J, Saintigny P, Girard L, Peyton M, Shen L, Fan Y, Giri

cancer type. between Mes ECDF (ECDFMes) and Epi ECDF (ECDFEpi). The 2KS funding; TZT, QHM, JPT and RYH wrote the paper; TZT performed bioinformat- U, Tumula PK et al (2013) An epithelial-mesenchymal transition gene 5 The extreme 25% of the most Mes and Epi cell lines or the score was then taken as the EMT score. A sample with a positive EMT ics analyses; YM, MM, TN and SM provided on laser-capture micro-dissected signature predicts resistance to EGFR and PI3K inhibitors and identifies extreme 100 Mes and Epi tumours were chosen to generate the score exhibits a more Mes phenotype, whereas a negative EMT score breast cancer samples. Axl as a therapeutic target for overcoming EGFR inhibitor resistance. Clin EMT signatures for cell lines and tumours, respectively; this reflects a more Epi phenotype. Note that the 2KS test allows segrega- Cancer Res 19: 279 – 290

prevented the signature from over-fitting the training data. EMT tion of samples into Epi (2KS score ECDFEpi > ECDFMes; P < 0.05), Conflict of interest Cai J, Guan H, Fang L, Yang Y, Zhu X, Yuan J, Wu J, Li M (2013)

signatures were generated using SAM/ROC (Tusher et al, 2001; intermediate Epi (2KS score ECDFEpi > ECDFMes; P ≥ 0.05), intermedi- The authors declare that they have no conflict of interest. MicroRNA-374a activates Wnt/beta-catenin signaling to promote breast

Verhaak et al, 2013), with applied thresholds of: SAM q%=0, ate Mes (2KS score ECDFEpi < ECDFMes, P ≥ 0.05) and Mes (2KS score cancer metastasis. J Clin Invest 123: 566 – 579

and ROC > 0.8–0.85 or < 0.15–0.2. ECDFEpi < ECDFMes, P < 0.05). The EMT signature is given in For more information Carey LA, Rugo HS, Marcom PK, Mayer EL, Esteva FJ, Ma CX, Liu MC, Storniolo 6 Using this SAM/ROC-derived EMT signature, we then computed Supplementary Table S1A and B. The Matlab R2012a script for Gene Expression Omnibus (GEO): http://www.ncbi.nlm.nih.gov/gds AM, Rimawi MF, Forero-Torres A et al (2012) TBCRC 001: randomized the EMT score of a sample using a two-sample Kolmogorov– computing the EMT score and computation of the EMT score can be ArrayExpress: http://www.ebi.ac.uk/arrayexpress/ phase II study of cetuximab in combination with carboplatin in stage IV Smirnov test (2KS). requested through http://www.csi.nus.edu.sg/bioinfo/index.php. The Cancer Genome Atlas (TCGA): https://tcga-data.nci.nih.gov/tcga/ triple-negative breast cancer. J Clin Oncol 30: 2615 – 2623

ª 2014 The Authors EMBO Molecular Medicine Vol 6 | No 10 | 2014 1289 1290 EMBO Molecular Medicine Vol 6 | No 10 | 2014 ª 2014 The Authors Tuan Zea Tan et al A generic signature to quantify EMT phenotype EMBO Molecular Medicine EMBO Molecular Medicine A generic signature to quantify EMT phenotype Tuan Zea Tan et al

2 ssGSEA score (Verhaak et al, 2013) was computed for EMT gene and qgd are the fold-change and q-value% of the gene, g, of Peterson B, Royal R, Ross MI, Tyler DS (2011) Prospective multicenter sets on cancer cell lines and correlated with gene expression of disease, d, as computed by SAM. ROCgd is the ROC value of gene, Supplementary information for this article is available online: phase II trial of systemic ADH-1 in combination with melphalan via known Mes and Epi markers (TWIST1, SNAI1, SNAI2, VIM, g, of disease, d, and nd is the number of samples in disease, d. http://embomolmed.embopress.org isolated limb infusion in patients with advanced extremity melanoma. J CDH2, ZEB1 and CDH1, DDR1, ERBB2, ERBB3, KRT19) (Thiery The formula will give higher weights to genes that have a large Clin Oncol 29: 1210 – 1215 et al, 2009). fold-change, a small q-value%, a large ROC value and a large Acknowledgements Beatty GL, Chiorean EG, Fishman MP, Saboury B, Teitelbaum UR, Sun W, 3 The gene set that best correlated with the enrichment score was number of samples. We ranked and selected the genes with a We thank Dr. R. Jackson for her careful English editing. We thank the finan- Huhn RD, Song W, Li D, Sharp LL et al (2011) CD40 agonists alter tumor chosen to rank the cell lines. The 10–20 most Mes and most Epi z-transformed weight > 3.09 or P < 0.001 (Supplementary Table S1A cial support from the Cancer Science Institute of Singapore, National stroma and show efficacy against pancreatic carcinoma in mice and cell lines were selected for BinReg modelling. Two data sets, and B). Research Foundation Prime Minister’s Office Singapore, National Medical humans. Science 331: 1612 – 1616 GSE9691 (Onder et al, 2008) and GSE24202 (Taube et al, 2010), Research Council/National University Cancer Institute of Singapore Center Blick T, Widodo E, Hugo H, Waltham M, Lenburg ME, Neve RM, Thompson were used for BinReg parameter settings and to ensure validity Computation of EMT score Grant/EMT Theme CG12Aug12, Department of Biochemistry of the National EW (2008) Epithelial mesenchymal transition traits in human breast of the derived EMT signature. (Note: Steps 1–3 can be recursive University of Singapore, and Institute of Molecular and Cell Biology at cancer cell lines. Clin Exp Metastasis 25: 629 – 642 to identify an initial BinReg EMT signature of sufficient accuracy To compute the EMT score of a sample, we adopted a similar A*STAR, Singapore. Borrell B (2010) How accurate are cancer cell lines? Nature 463: 858 in predicting the EMT status.) approach to that used in ssGSEA (Verhaak et al, 2013). The empirical Brabletz T (2012) To differentiate or not–routes towards metastasis. Nat Rev 4 The BinReg EMT signature was then used to predict the EMT cumulative distribution function (ECDF) was estimated for Epi and Author contributions Cancer 12: 425 – 436 status of cell lines and tumour samples specific to a particular Mes gene sets. The 2KS test was employed to compute the difference JPT and RYH conceived the idea; JPT and RYH devised the project and obtained Byers LA, Diao L, Wang J, Saintigny P, Girard L, Peyton M, Shen L, Fan Y, Giri cancer type. between Mes ECDF (ECDFMes) and Epi ECDF (ECDFEpi). The 2KS funding; TZT, QHM, JPT and RYH wrote the paper; TZT performed bioinformat- U, Tumula PK et al (2013) An epithelial-mesenchymal transition gene 5 The extreme 25% of the most Mes and Epi cell lines or the score was then taken as the EMT score. A sample with a positive EMT ics analyses; YM, MM, TN and SM provided on laser-capture micro-dissected signature predicts resistance to EGFR and PI3K inhibitors and identifies extreme 100 Mes and Epi tumours were chosen to generate the score exhibits a more Mes phenotype, whereas a negative EMT score breast cancer samples. Axl as a therapeutic target for overcoming EGFR inhibitor resistance. Clin EMT signatures for cell lines and tumours, respectively; this reflects a more Epi phenotype. Note that the 2KS test allows segrega- Cancer Res 19: 279 – 290 prevented the signature from over-fitting the training data. EMT tion of samples into Epi (2KS score ECDFEpi > ECDFMes; P < 0.05), Conflict of interest Cai J, Guan H, Fang L, Yang Y, Zhu X, Yuan J, Wu J, Li M (2013) signatures were generated using SAM/ROC (Tusher et al, 2001; intermediate Epi (2KS score ECDFEpi > ECDFMes; P ≥ 0.05), intermedi- The authors declare that they have no conflict of interest. MicroRNA-374a activates Wnt/beta-catenin signaling to promote breast

Verhaak et al, 2013), with applied thresholds of: SAM q%=0, ate Mes (2KS score ECDFEpi < ECDFMes, P ≥ 0.05) and Mes (2KS score cancer metastasis. J Clin Invest 123: 566 – 579 and ROC > 0.8–0.85 or < 0.15–0.2. ECDFEpi < ECDFMes, P < 0.05). The EMT signature is given in For more information Carey LA, Rugo HS, Marcom PK, Mayer EL, Esteva FJ, Ma CX, Liu MC, Storniolo 6 Using this SAM/ROC-derived EMT signature, we then computed Supplementary Table S1A and B. The Matlab R2012a script for Gene Expression Omnibus (GEO): http://www.ncbi.nlm.nih.gov/gds AM, Rimawi MF, Forero-Torres A et al (2012) TBCRC 001: randomized the EMT score of a sample using a two-sample Kolmogorov– computing the EMT score and computation of the EMT score can be ArrayExpress: http://www.ebi.ac.uk/arrayexpress/ phase II study of cetuximab in combination with carboplatin in stage IV Smirnov test (2KS). requested through http://www.csi.nus.edu.sg/bioinfo/index.php. The Cancer Genome Atlas (TCGA): https://tcga-data.nci.nih.gov/tcga/ triple-negative breast cancer. J Clin Oncol 30: 2615 – 2623

Carr EA, Mead J, Vershon AK (2004) Alpha1-induced DNA bending is required Garnett MJ, Edelman EJ, Heidorn SJ, Greenman CD, Dastur A, Lau KW, Knudsen S, Jensen T, Hansen A, Mazin W, Lindemann J, Kuter I, Laing N, breast cancers originate from luminal epithelial progenitors and not from for transcriptional activation by the Mcm1-alpha1 complex. Nucleic Acids Greninger P, Thompson IR, Luo X, Soares J et al (2012) Systematic Anderson E (2014) Development and validation of a gene expression score basal stem cells. Cell Stem Cell 7: 403 – 417 Res 32: 2298 – 2305 identification of genomic markers of drug sensitivity in cancer cells. that predicts response to fulvestrant in breast cancer patients. PLoS ONE Mulligan G, Mitsiades C, Bryant B, Zhan F, Chng WJ, Roels S, Koenig E, Fergus Carretero J, Shimamura T, Rikova K, Jackson AL, Wilkerson MD, Borgman CL, Nature 483: 570 – 575 9:e87415 A, Huang Y, Richardson P et al (2007) Gene expression profiling and Buttarazzi MS, Sanofsky BA, McNamara KL, Brandstetter KA et al (2010) Gatza ML, Lucas JE, Barry WT, Kim JW, Wang Q, Crawford MD, Datto MB, Kelley Kong D, Li Y, Wang Z, Sarkar FH (2011) Cancer stem cells and correlation with outcome in clinical trials of the proteasome inhibitor Integrative genomic and proteomic analyses identify targets for M, Mathey-Prevot B, Potti A et al (2010) A pathway-based classification of epithelial-to-mesenchymal transition (EMT)-phenotypic cells: are they bortezomib. Blood 109: 3177 – 3188 Lkb1-deficient metastatic lung tumors. Cancer Cell 17: 547 – 559 human breast cancer. Proc Natl Acad Sci U S A 107: 6994 – 6999 cousins or twins? Cancers (Basel) 3: 716 – 729 Nakaya Y, Sheng G (2013) EMT in developmental morphogenesis. Cancer Lett Chaffer CL, Weinberg RA (2011) A perspective on cancer cell metastasis. Gautier L, Cope L, Bolstad BM, Irizarry RA (2004) affy—analysis of Affymetrix Korde LA, Lusa L, McShane L, Lebowitz PF, Lukes L, Camphausen K, Parker JS, 341: 9 – 15 Science 331: 1559 – 1564 GeneChip data at the probe level. Bioinformatics 20: 307 – 315 Swain SM, Hunter K, Zujewski JA (2010) Gene expression pathway analysis Ocana OH, Corcoles R, Fabra A, Moreno-Bueno G, Acloque H, Vega S, Cieply B, Riley P, Pifer PM, Widmeyer J, Addison JB, Ivanov AV, Denvir J, Frisch Gazdar AF (2009) Activating and resistance mutations of EGFR in to predict response to neoadjuvant docetaxel and capecitabine for breast Barrallo-Gimeno A, Cano A, Nieto MA (2012) Metastatic colonization SM (2012) Suppression of the epithelial-mesenchymal transition by non-small-cell lung cancer: role in clinical response to EGFR tyrosine cancer. Breast Cancer Res Treat 119: 685 – 699 requires the repression of the epithelial-mesenchymal transition inducer Grainyhead-like-2. Cancer Res 72: 2440 – 2453 kinase inhibitors. Oncogene 28(Suppl 1): S24 – S31 Lee JM, Dedhar S, Kalluri R, Thompson EW (2006) The Prrx1. Cancer Cell 22: 709 – 724 Copley MR, Babovic S, Benz C, Knapp DJ, Beer PA, Kent DG, Wohrer S, Treloar Gillet JP, Varma S, Gottesman MM (2013) The clinical relevance of cancer cell epithelial-mesenchymal transition: new insights in signaling, development, Ohashi S, Natsuizaka M, Naganuma S, Kagawa S, Kimura S, Itoh H, Kalman DQ, Day C, Rowe K et al (2013) The Lin28b-let-7-Hmga2 axis determines lines. J Natl Cancer Inst 105: 452 – 458 and disease. J Cell Biol 172: 973 – 981 RA, Nakagawa M, Darling DS, Basu D et al (2011) A NOTCH3-mediated the higher self-renewal potential of fetal haematopoietic stem cells. Nat Gim JCY, Hong SH, Kim HC, Chun HK, Park T, Park WY, Lee WY (2014) Lee K, Nelson CM (2012) New insights into the regulation of squamous cell differentiation program limits expansion of EMT-competent Cell Biol 15: 916 – 925 Predicting multi-class response to preoperative chemoradiotherapy in epithelial-mesenchymal transition and tissue fibrosis. Int Rev Cell Mol Biol cells that express the ZEB transcription factors. Cancer Res 71: 6836 – 6847 D’Amato NC, Ostrander JH, Bowie ML, Sistrunk C, Borowsky A, Cardiff RD, Bell rectal cancer patients, manuscript in preparation 294: 171 – 221 Onder TT, Gupta PB, Mani SA, Yang J, Lander ES, Weinberg RA (2008) Loss of K, Young LJ, Simin K, Bachelder RE et al (2012) Evidence for phenotypic Haibe-Kains B, El-Hachem N, Birkbak NJ, Jin AC, Beck AH, Aerts HJ, Lehmann BD, Bauer JA, Chen X, Sanders ME, Chakravarthy AB, Shyr Y, E-cadherin promotes metastasis via multiple downstream transcriptional plasticity in aggressive triple-negative breast cancer: human biology is Quackenbush J (2013) Inconsistency in large pharmacogenomic studies. Pietenpol JA (2011) Identification of human triple-negative breast cancer pathways. Cancer Res 68: 3645 – 3654 recapitulated by a novel model system. PLoS ONE 7:e45684 Nature 50: 389 – 393 subtypes and preclinical models for selection of targeted therapies. J Clin Prat A, Parker JS, Karginova O, Fan C, Livasy C, Herschkowitz JI, He X, Perou Deshiere A, Duchemin-Pelletier E, Spreux E, Ciais D, Combes F, Vandenbrouck Hao J, Zhang Y, Deng M, Ye R, Zhao S, Wang Y, Li J, Zhao Z (2014) MicroRNA Invest 121: 2750 – 2767 CM (2010) Phenotypic and molecular characterization of the claudin-low Y, Coute Y, Mikaelian I, Giusiano S, Charpin C et al (2013) Unbalanced control of epithelial-mesenchymal transition in cancer stem cells. Int J Lim E, Vaillant F, Wu D, Forrest NC, Pal B, Hart AH, Asselin-Labat ML, Gyorki intrinsic subtype of breast cancer. Breast Cancer Res 12:R68 expression of CK2 kinase subunits is sufficient to drive Cancer 135: 1019 – 1027 DE, Ward T, Partanen A et al (2009) Aberrant luminal progenitors as the Park ES, Kim SJ, Kim SW, Yoon SL, Leem SH, Kim SB, Kim SM, Park YY, Cheong epithelial-to-mesenchymal transition by Snail1 induction. Oncogene 32: Hellner K, Mar J, Fang F, Quackenbush J, Munger K (2009) HPV16 E7 oncogene candidate target population for basal tumor development in BRCA1 JH, Woo HG et al (2011) Cross-species hybridization of microarrays for 1373 – 1383 expression in normal human epithelial cells causes molecular changes mutation carriers. Nat Med 15: 907 – 913 studying tumor transcriptome of brain metastasis. Proc Natl Acad Sci U S Desmedt C, Giobbie-Hurder A, Neven P, Paridaens R, Christiaens MR, Smeets indicative of an epithelial to mesenchymal transition. Virology 391: 57 – 63 Lim J, Thiery JP (2012) Epithelial-mesenchymal transitions: insights from A 108: 17456 – 17461 A, Lallemand F, Haibe-Kains B, Viale G, Gelber RD et al (2009) The Gene Hollestelle A, Nagel JH, Smid M, Lam S, Elstrodt F, Wasielewski M, Ng SS, development. Development 139: 3471 – 3486 Prat A, Bianchini G, Thomas M, Belousov A, Cheang MC, Koehler A, Gomez P, expression Grade Index: a potential predictor of relapse for French PJ, Peeters JK, Rozendaal MJ et al (2010) Distinct gene mutation Liu S, Cong Y, Wang D, Sun Y, Deng L, Liu Y, Martin-Trevino R, Shang L, Semiglazov V, Eiermann W, Tjulandin S et al (2014) Research-based endocrine-treated breast cancer patients in the BIG 1-98 trial. BMC Med profiles among luminal-type and basal-type breast cancer cell lines. Breast McDermott SP, Landis MD et al (2014) Breast cancer stem cells transition PAM50 subtype predictor identifies higher responses and improved Genomics 2: 40 Cancer Res Treat 121: 53 – 64 between epithelial and mesenchymal states reflective of their normal survival outcomes in HER2-positive breast cancer in the NOAH study. Clin Eckstein N (2011) Platinum resistance in breast and ovarian cancer cell lines. Horak CE, Pusztai L, Xing G, Trifan OC, Saura C, Tseng LM, Chan S, Welcher R, counterparts. Stem Cell Rep 2: 78 – 91 Cancer Res 20: 511 – 521 J Exp Clin Cancer Res 30: 91 Liu D (2013) Biomarker analysis of neoadjuvant doxorubicin/ Loboda A, Nebozhyn MV, Watters JW, Buser CA, Shaw PM, Huang PS, Van’t Prat A, Perou CM (2011) Deconstructing the molecular portraits of breast Erdem-Eraslan L, Gravendeel LA, de Rooi J, Eilers PH, Idbaih A, Spliet WG, den cyclophosphamide followed by ixabepilone or Paclitaxel in early-stage Veer L, Tollenaar RA, Jackson DB, Agrawal D et al (2011) EMT is the cancer. Mol Oncol 5: 5 – 23 Dunnen WF, Teepen JL, Wesseling P, Sillevis Smitt PA et al (2013) Intrinsic breast cancer. Clin Cancer Res 19: 1587 – 1595 dominant program in human colon cancer. BMC Med Genomics 4: 9 Sarrio D, Rodriguez-Pinilla SM, Hardisson D, Cano A, Moreno-Bueno G, molecular subtypes of glioma are prognostic and predict benefit from Hrstka R, Nenutil R, Fourtouna A, Maslon MM, Naughton C, Langdon S, Malizia AP, Lacey N, Walls D, Egan JJ, Doran PP (2009) CUX1/Wnt signaling Palacios J (2008) Epithelial-mesenchymal transition in breast cancer adjuvant procarbazine, lomustine, and vincristine chemotherapy in Murray E, Larionov A, Petrakova K, Muller P et al (2010) The regulates epithelial mesenchymal transition in EBV infected epithelial relates to the basal-like phenotype. Cancer Res 68: 989 – 997 combination with other prognostic factors in anaplastic oligodendroglial pro-metastatic protein anterior gradient-2 predicts poor prognosis in cells. Exp Cell Res 315: 1819 – 1831 Sethi S, Macoska J, Chen W, Sarkar FH (2010) Molecular signature of brain tumors: a report from EORTC study 26951. J Clin Oncol 31: 328 – 336 tamoxifen-treated breast cancers. Oncogene 29: 4838 – 4847 Marchini S, Fruscio R, Clivio L, Beltrame L, Porcu L, Nerini IF, Cavalieri D, epithelial-mesenchymal transition (EMT) in human prostate cancer bone Esserman LJ, Berry DA, Cheang MC, Yau C, Perou CM, Carey L, DeMichele A, Huang RY, Chung VY, Thiery JP (2012) Targeting pathways contributing to Chiorino G, Cattoretti G, Mangioni C et al (2013) Resistance to metastasis. Am J Transl Res 3: 90 – 99 Gray JW, Conway-Dorsey K, Lenburg ME et al (2012) Chemotherapy epithelial-mesenchymal transition (EMT) in epithelial ovarian cancer. Curr platinum-based chemotherapy is associated with epithelial to Silver DP, Richardson AL, Eklund AC, Wang ZC, Szallasi Z, Li Q, Juul N, Leong response and recurrence-free survival in neoadjuvant breast cancer Drug Targets 13: 1649 – 1653 mesenchymal transition in epithelial ovarian cancer. Eur J Cancer 49: CO, Calogrias D, Buraimoh A et al (2010) Efficacy of neoadjuvant Cisplatin depends on biomarker profiles: results from the I-SPY 1 TRIAL Huang RY, Wong MK, Tan TZ, Kuay KT, Ng AHC, Chung VY, Chu Y-S, 520 – 530 in triple-negative breast cancer. J Clin Oncol 28: 1145 – 1153 (CALGB 150007/150012; ACRIN 6657). Breast Cancer Res Treat 132: Matsumura N, Lai H-C, Lee YF et al (2013) An EMT Spectrum defines an Massarweh S, Tham YL, Huang J, Sexton K, Weiss H, Tsimelzon A, Beyer A, Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, 1049 – 1062 anoikis resistant and spheroidogenic Intermediate Mesenchymal state that Rimawi M, Cai WY, Hilsenbeck S et al (2011) A phase II neoadjuvant trial Paulovich A, Pomeroy SL, Golub TR, Lander ES et al (2005) Gene set Evans AL, Faial T, Gilchrist MJ, Down T, Vallier L, Pedersen RA, Wardle FC, is sensitive to E-cadherin restoration by a Src-kinase inhibitor, Saracatinib of anastrozole, fulvestrant, and gefitinib in patients with newly diagnosed enrichment analysis: a knowledge-based approach for interpreting Smith JC (2012) Genomic targets of Brachyury (T) in differentiating mouse (AZD0530). Cell Death Dis 4:e915 estrogen receptor positive breast cancer. Breast Cancer Res Treat 129: genome-wide expression profiles. Proc Natl Acad Sci U S A 102: 15545 – 15550 embryonic stem cells. PLoS ONE 7:e33346 Huang da W, Sherman BT, Lempicki RA (2009) Systematic and integrative 819 – 827 Taddei I, Deugnier MA, Faraldo MM, Petit V, Bouvard D, Medina D, Fassler R, Farmer P, Bonnefoi H, Anderle P, Cameron D, Wirapati P, Becette V, Andre S, analysis of large gene lists using DAVID bioinformatics resources. Nat Masters JR, Koberle B (2003) Curing metastatic cancer: lessons from testicular Thiery JP, Glukhova MA (2008) Beta1 integrin deletion from the basal Piccart M, Campone M, Brain E et al (2009) A stroma-related gene Protoc 4: 44 – 57 germ-cell tumours. Nat Rev Cancer 3: 517 – 525 compartment of the mammary epithelium affects stem cells. Nat Cell Biol signature predicts resistance to neoadjuvant chemotherapy in breast Ishitobi M, Okumura Y, Arima N, Yoshida A, Nakatsukasa K, Iwase T, Shien T, Maupin KA, Sinha A, Eugster E, Miller J, Ross J, Paulino V, Keshamouni VG, 10: 716 – 722 cancer. Nat Med 15: 68 – 74 Masuda N, Tanaka S, Tanabe M et al (2013) Breast cancer subtype and Tran N, Berens M, Webb C et al (2010) Glycogene expression alterations Tam WL, Weinberg RA (2013) The epigenetics of epithelial-mesenchymal Ferlicot S, Vincent-Salomon A, Medioni J, Genin P, Rosty C, Sigal-Zafrani B, distant recurrence after ipsilateral breast tumor recurrence. Ann Surg associated with pancreatic cancer epithelial-mesenchymal transition in plasticity in cancer. Nat Med 19: 1438 – 1449 Freneaux P, Jouve M, Thiery JP, Sastre-Garau X (2004) Wide metastatic Oncol 20: 1886 – 1892 complementary model systems. PLoS ONE 5:e13002 Tan IB, Ivanova T, Lim KH, Ong CW, Deng N, Lee J, Tan SH, Wu J, Lee MH, Ooi spreading in infiltrating lobular carcinoma of the breast. Eur J Cancer 40: Johnson WE, Li C, Rabinovic A (2007) Adjusting batch effects in microarray Miow QH, Tan TZ, Ye J, Lau JA, Yokomizo T, Thiery J-P, Mori S (2014) CH et al (2011) Intrinsic subtypes of gastric cancer, based on gene 336 – 341 expression data using empirical Bayes methods. Biostatistics 8: 118 – 127 Epithelial-mesenchymal status renders differential responses to cisplatin expression pattern, predict survival and respond differently to Frisch SM, Schaller M, Cieply B (2013) Mechanisms that link the oncogenic Jordan NV, Johnson GL, Abell AN (2011) Tracking the intermediate stages of in ovarian cancer. Oncogene doi: 10.1038/onc.2014.136 chemotherapy. Gastroenterology 141: 476 – 485 epithelial-mesenchymal transition to suppression of anoikis. J Cell Sci 126: epithelial-mesenchymal transition in epithelial stem cells and cancer. Cell Molyneux G, Geyer FC, Magnay FA, McCarthy A, Kendrick H, Natrajan R, Tan TZ, Miow QH, Huang RY, Wong MK, Ye J, Lau JA, Wu MC, Bin Abdul 21 – 29 Cycle 10: 2865 – 2873 Mackay A, Grigoriadis A, Tutt A, Ashworth A et al (2010) BRCA1 basal-like Hadi LH, Soong R, Choolani M et al (2013) Functional genomics

ª 2014 The Authors EMBO Molecular Medicine Vol 6 | No 10 | 2014 1291 1292 EMBO Molecular Medicine Vol 6 | No 10 | 2014 ª 2014 The Authors Tuan Zea Tan et al A generic signature to quantify EMT phenotype EMBO Molecular Medicine EMBO Molecular Medicine A generic signature to quantify EMT phenotype Tuan Zea Tan et al

identifies five distinct molecular subtypes with clinical relevance and Verhaak RG, Tamayo P, Yang JY, Hubbard D, Zhang H, Creighton CJ, Fereday pathways for growth control in epithelial ovarian cancer. EMBO Mol Med S, Lawrence M, Carter SL, Mermel CH et al (2013) Prognostically relevant 5: 983 – 998 gene signatures of high-grade serous ovarian carcinoma. J Clin Invest 123: Taube JH, Herschkowitz JI, Komurov K, Zhou AY, Gupta S, Yang J, Hartwell K, 517 – 525 Onder TT, Gupta PB, Evans KW et al (2010) Core Wang W, Goswami S, Lapidus K, Wells AL, Wyckoff JB, Sahai E, Singer RH, epithelial-to-mesenchymal transition interactome gene-expression Segall JE, Condeelis JS (2004) Identification and testing of a gene signature is associated with claudin-low and metaplastic breast cancer expression signature of invasive carcinoma cells within primary mammary subtypes. Proc Natl Acad Sci U S A 107: 15449 – 15454 tumors. Cancer Res 64: 8585 – 8594 The Cancer Genome Atlas (2011) Integrated genomic analyses of ovarian Witta SE, Gemmill RM, Hirsch FR, Coldren CD, Hedman K, Ravdel L, Helfrich carcinoma. Nature 474: 609 – 615 B, Dziadziuszko R, Chan DC, Sugita M et al (2006) Restoring E-cadherin Thiery JP (2002) Epithelial-mesenchymal transitions in tumour progression. expression increases sensitivity to epidermal growth factor receptor Nat Rev Cancer 2: 442 – 454 inhibitors in lung cancer cell lines. Cancer Res 66: 944 – 950 Thiery JP, Acloque H, Huang RY, Nieto MA (2009) Epithelial-mesenchymal Yanagawa J, Walser TC, Zhu LX, Hong L, Fishbein MC, Mah V, Chia D, transitions in development and disease. Cell 139: 871 – 890 Goodglick L, Elashoff DA, Luo J et al (2009) Snail promotes CXCR2 Thiery JP, Lim CT (2013) Tumor dissemination: an EMT affair. Cancer Cell 23: ligand-dependent tumor progression in non-small cell lung carcinoma. 272 – 273 Clin Cancer Res 15: 6820 – 6829 Tomkiewicz C, Hans S, Mucchielli MH, Agier N, Delacroix H, Marisa L, Yu M, Bardia A, Wittner BS, Stott SL, Smas ME, Ting DT, Isakoff SJ, Ciciliano Brasnu D, Aggerbeck LP, Badoual C, Barouki R et al (2012) A head and JC, Wells MN, Shah AM et al (2013) Circulating breast tumor cells exhibit neck cancer tumor response-specific gene signature for cisplatin, dynamic changes in epithelial and mesenchymal composition. Science 339: 5-fluorouracil induction chemotherapy fails with added taxanes. PLoS 580 – 584 ONE 7:e47170 Zhang J, Ma L (2012) MicroRNA control of epithelial-mesenchymal transition Tsai JH, Donaher JL, Murphy DA, Chau S, Yang J (2012) Spatiotemporal and metastasis. Cancer Metastasis Rev 31: 653 – 662 regulation of epithelial-mesenchymal transition is essential for squamous Zhang JX, Wei J, Lu J, Tong Z, Liao B, Yu B, Zheng F, Huang X, Chen Z, Fang Y cell carcinoma metastasis. Cancer Cell 22: 725 – 736 et al (2013) Overexpression of Rab25 contributes to metastasis of bladder Tusher VG, Tibshirani R, Chu G (2001) Significance analysis of microarrays cancer through induction of epithelial-mesenchymal transition applied to the ionizing radiation response. Proc Natl Acad Sci U S A 98: and activation of Akt/GSK-3beta/Snail signaling. Carcinogenesis 34: 5116 – 5121 2401 – 2408 Valastyan S, Weinberg RA (2011) Tumor metastasis: molecular insights and evolving paradigms. Cell 147: 275 – 292 License: This is an open access article under the Van den Eynden GG, Majeed AW, Illemann M, Vermeulen PB, Bird NC, terms of the Creative Commons Attribution 4.0 Hoyer-Hansen G, Eefsen RL, Reynolds AR, Brodt P (2013) The multifaceted License, which permits use, distribution and reproduc- role of the microenvironment in liver metastasis: biology and clinical tion in any medium, provided the original work is implications. Cancer Res 73: 2031 – 2043 properly cited.

ª 2014 The Authors EMBO Molecular Medicine Vol 6 | No 10 | 2014 1293 Article

A computational study of the Warburg effect identifies metabolic targets inhibiting cancer migration

Keren Yizhak1,*,†, Sylvia E Le Dévédec2,†, Vasiliki Maria Rogkoti2, Franziska Baenke3, Vincent C de Boer4, Christian Frezza5, Almut Schulze3, Bob van de Water2,‡ & Eytan Ruppin1,6,‡,**

Abstract Subject Categories Genome-Scale & Integrative Biology; Metabolism; Computational Biology Over the last decade, the field of cancer metabolism has mainly DOI 10.15252/msb.20134993 | Received 18 November 2013 | Revised 6 July focused on studying the role of tumorigenic metabolic rewiring 2014 | Accepted 7 July 2014 in supporting cancer proliferation. Here, we perform the first Mol Syst Biol. (2014) 10: 744 genome-scale computational study of the metabolic underpin- nings of cancer migration. We build genome-scale metabolic models of the NCI-60 cell lines that capture the Warburg effect Introduction (aerobic glycolysis) typically occurring in cancer cells. The extent of the Warburg effect in each of these cell line models is quan- Altered tumor metabolism has become a generally regarded hall- tified by the ratio of glycolytic to oxidative ATP flux (AFR), mark of cancer (Hanahan & Weinberg, 2011). The initial recognition which is found to be highly positively associated with cancer that metabolism is altered in cancer can be traced back to Otto cell migration. We hence predicted that targeting genes that Warburg’s early studies, showing that transformed cells consume mitigate the Warburg effect by reducing the AFR may specifi- glucose at an abnormally high rate and largely reduce it to lactate, cally inhibit cancer migration. By testing the anti-migratory even in the presence of oxygen (Warburg, 1956). Over the last effects of silencing such 17 top predicted genes in four breast decade, much of the field of cancer metabolism has focused on the and lung cancer cell lines, we find that up to 13 of these novel role of the Warburg effect in supporting cancer proliferation (Vander predictions significantly attenuate cell migration either in all or Heiden et al, 2009). However, the role of this process in supporting one cell line only, while having almost no effect on cell prolifer- other fundamental cancer phenotypes such as cellular migration has ation. Furthermore, in accordance with the predictions, a signifi- received far less attention. cant reduction is observed in the ratio between experimentally Contemporary cytotoxic cancer treatment has been mainly measured ECAR and OCR levels following these perturbations. based on drugs that kill proliferating cells generally unselectively Inhibiting anti-migratory targets is a promising future avenue in and are therefore accompanied by many undesirable side effects. treating cancer since it may decrease cytotoxic-related side Drug targets that can inhibit migration but leave cellular prolifer- effects that plague current anti-proliferative treatments. ation relatively spared may be able to avoid such side effects. Furthermore, it may reduce cytotoxic-related clonal selection of Such targets may have the additional benefit of reducing the more aggressive cancer cells and the likelihood of emerging selection for more resistant clones that occurs due to the elimi- resistance. nation of treatment-sensitive cells. The growing availability of high-throughput measurements for a range of cancer cells Keywords cancer cell migration; cellular metabolism; genome-scale presents an opportunity to study a wider scope of dysregulated metabolic modeling metabolism across many different cancers. Here, we aim to

1 The Blavatnik School of Computer Science, Tel-Aviv University, Tel-Aviv, Israel 2 Division of Toxicology, Leiden Academic Centre for Drug Research, Leiden University, Leiden, The Netherlands 3 Gene Expression Analysis Laboratory, Cancer Research UK, London Research Institute, London, UK 4 Laboratory Genetic Metabolic Diseases, Academic Medical Center, Amsterdam, The Netherlands 5 MRC Cancer Unit, Hutchison/MRC Research Centre, University of Cambridge, Cambridge, UK 6 The Sackler School of Medicine, Tel-Aviv University, Tel-Aviv, Israel *Corresponding author. Tel: +972 3 6405378; E-mail: [email protected] **Corresponding author. Tel: +972 3 6406528; E-mail: [email protected] †These authors contributed equally to this study ‡These authors contributed equally to this study

ª 2014 The Authors. Published under the terms of the CC BY 4.0 license Molecular Systems Biology 10: 744 | 2014 1 Molecular Systems Biology Identifying anti-migratory metabolic drug targets Keren Yizhak et al Keren Yizhak et al Identifying anti-migratory metabolic drug targets Molecular Systems Biology

A C integrate pertaining data with a genome-scale mechanistic model Results of human metabolism to study the role of the Warburg effect in tumor progression and its potential association with cellular Stoichiometric and flux capacity constraints successfully capture Glucose migration. the coupling of high cell proliferation rate to lactate secretion 3BrPA Genome-scale metabolic modeling is an increasingly widely across individual NCI-60 cancer models (HK2) used computational framework for studying metabolism. Given G6P the genome-scale metabolic model (GSMM) of a species along- As a starting point for this study, we developed a set of metabolic side contextual information such as growth media and ‘omics’ models specific for each of the NCI-60 cell lines. We built these 0 0.5 1 data, one can obtain a fairly accurate prediction of numerous models using a new algorithm we have recently developed termed metabolic phenotypes, including growth rates, nutrient uptake PRIME, for building individual models of cells from pertaining rates, gene essentiality, and more (Covert et al, 2004). GSMMs omics data (Yizhak et al, submitted, Supplementary Information IodoacetateG3P have been used for various applications (Oberhardt et al, 2009; and Supplementary Fig S1). PRIME uses the generic human model (G3PDH) Chandrasekaran & Price, 2010; Jensen & Papin, 2010; Szappanos as a scaffold and sets maximal flux capacity constraints over a et al, 2011; Wessely et al, 2011; Lerman et al, 2012; Nogales subset of its growth-associated reactions according to the expression 1,3BPG 0 0.5 1 et al, 2012; Schuetz et al, 2012) including drug discovery levels of their corresponding catalyzing enzymes in each of the (Trawick & Schilling, 2006; Oberhardt et al, 2013; Yizhak et al, target cell lines. B 2013) and metabolic engineering (Burgard et al, 2003; Pharkya An important hallmark of cancerous cells is the production of 2PG et al, 2004). Over the last few years, GSMMs have been success- lactate through the Warburg effect (Warburg, 1956). As a first step Fluoride fully used for modeling human metabolism as well (Duarte et al, in validating the basic function of our NCI-60 models, we assessed (Enolase) 2007; Ma et al, 2007; Shlomi et al, 2008; Gille et al, 2010; Lewis whether maximizing biomass forces production of lactate, which PEP et al, 2010; Mardinoglu et al, 2013). Specifically, GSMM models would signify proper coupling of biomass production with lactate 0 0.5 1 of cancer cells have been reconstructed and applied for predict- output as seen in cancer cells. We found that the models indeed ing selective drug targets, as well as for studying the role of must secrete lactate under biomass maximization (Supplementary Pyruvate tumor suppressors and oxidative stress (Folger et al, 2011; Frezza Information and Supplementary Fig S2). Hence, in contrast to the et al, 2011; Agren et al, 2012, 2014; Jerby et al, 2012; Goldstein original generic model of human metabolism, they enable us to Oxamate et al, 2013; Gatto et al, 2014). In the context of studying the systematically assess the extent of lactate secretion and study the (LDH) Lactate Warburg effect, the original human metabolic model does not Warburg effect across a wide range of cancer cell lines without predict forced lactate secretion under maximal biomass produc- needing to add (mostly unknown) solvent capacity constraints, thus 0 0.5 1 tion rate, even when oxygen consumption rate equals zero. identifying its functional correlates on a genome scale. This renders it unsuitable for studying the Warburg effect as is, as already noted by (Shlomi et al, 2011). While the addition of Comparing predicted versus experimentally measured solvent capacity constraints has been shown to overcome this bioenergetics capacity hurdle in principle (Shlomi et al, 2011), this addition requires enzymatic kinetic data which are still largely absent on a We compared the predicted lactate secretion rates across all cell Figure 1. A comparison between experimental and predicted in silico measurements of lactate secretion (or ECAR) and OCR across different cancer cell lines. genome-scale. lines to those measured experimentally by Jain et al (Jain et al, A Measured versus predicted lactate secretion rates across the 59 cell lines available at Jain et al (2012). In this study, we utilize individual genome-scale metabolic 2012), obtaining a moderate but significant correlation (Spearman B Measured versus predicted lactate secretion rates in hypoxic (red) and normoxic (blue) conditions for four breast cancer cell lines: T47D, MCF7, BT549, and Hs578T. models tailored separately to each of the NCI-60 cancer cell lines correlation R = 0.36, P-value = 5.7e 3, Fig 1A, Materials and Meth- Bars represent the measured lactate secretion rates and the line represents the corresponding predicted rates. Error bars represent SD; number of samples for À to study the role of the Warburg effect in supporting cancer cellu- ods). To further test the models’ performance under different envi- experimental data (bars) is n = 7; number of samples for predicted data (line) is n = 1000. C Predicted ECAR and OCR by the A549 and H460 cell line models following inhibitory perturbations in the glycolytic pathway. The models predictions show a decrease lar migratory capacity. We first test and validate the individual ronmental conditions, we measured lactate secretion rates in four in ECAR (red line) and an increase in OCR (blue line). As found experimentally, the predicted OCR increase in H460 cells is lower than that found for A549 cells. The models against both existing and novel bioenergetic experimental breast cancer cell lines, T47D, MCF7, BT549, and Hs578T (Supple- x-axes represent the level of inhibition imposed, starting from a zero to a maximal inhibition (Materials and Methods). The specific perturbations include 3BpRA that data. Then, we examine the extent of the Warburg effect occur- mentary Dataset S1), under both normoxic and hypoxic conditions inhibits the enzyme hexokinase 2; Iodoacetate that inhibits the enzyme glycerol-3-phosphate dehydrogenase; Fluoride that inhibits the enzyme enolase; and ring in a given cancer cell line, by quantifying the glycolytic to (see Materials and Methods). Utilizing the corresponding cell line Oxamate that inhibits the enzyme lactate dehydrogenase. oxidative ATP flux ratio (AFR). We find that the AFR is highly models from the NCI-60 set, we found a high correlation between positively correlated with cancer cell migration, emphasizing the measured and predicted lactate secretion levels across both condi- role of glycolytic flux in supporting the more aggressive meta- tions (Spearman correlation R = 0.95, P-value = 1.1e 3, Fig 1B). observed after all glycolysis inhibitors was lower than the corre- state-of-the-art ECAR/OCR ratio (EOR) (Materials and Methods and À static stages of tumor development. To determine whether a The ratio of glycolytic versus oxidative capacity in a cell can be sponding increase in A549 cells (Fig 1C). Supplementary Dataset S2) showed a significant correlation across causal relation exists between AFR levels and cell migration, we quantified using its extracellular acidification rate (ECAR, a proxy of the NCI-60 models (Spearman correlation R = 0.66, P-value = 2e 8). À predict gene silencing that reduce this ratio. These potential lactate secretion) and its oxygen consumption rate (OCR). To Quantifying the Warburg effect and its relation to proliferation Testing both measures using a genome-wide NCI-60 drug response targets are then filtered further to exclude those predicted to further examine how well our cell line models capture measured and migration across the NCI-60 cell lines dataset (Scherf et al, 2000), we find that the model-predicted wild- result in cell lethality. Reassuringly, the predicted targets are Warburg-related activity in response to genetic perturbations, we type AFR levels across all cell line models are significantly corre- found to be significantly more highly expressed in metastatic and utilized measured ECAR and OCR levels in response to perturba- While ECAR and OCR are the commonly used measures for experi- lated (Spearman P-value < 0.05; FDR corrected with a = 0.05) with high-grade breast cancer tumors. Experimental investigation of tions in two NCI-60 lung cancer cell lines (A549 and H460), and mentally quantifying the bioenergetic capacity of the cell and thus Gi50 values of 30% of the compounds across these cell lines the top predicted targets via siRNA-mediated knockdown shows compared the results to predictions from our models (Materials and the Warburg effect, the genome-wide scope of GSMMs enables us to (empiric P-value < 9.9e 4), whereas the model-predicted EOR À that a significant portion of them truly attenuate cancer cell Methods) (Wu et al, 2007). Qualitatively similar ECAR and OCR examine other putative measures as well. One promising such measure accomplish this task for only 19% of the compounds migration without inducing a lethal effect. Furthermore, in accor- changes are found in response to various enzymatic perturbations measure we examined is the ratio between the ATP flux rate in the (Materials and Methods). Interestingly, we find that out of the 30% dance with the predictions, a significant reduction is observed in along the glycolytic pathway. Specifically, increased glycolytic inhi- glycolysis versus its flux rate in OXPHOS (AFR). Clearly, higher AFR AFR-Gi50-correlated compounds, 97% are positively correlated, the ratio between ECAR and OCR levels following these genes bition resulted in reduced ECAR and elevated OCR levels in both values denote more ‘Warburgian’ cell lines and vice versa. A suggesting that the more ‘Warburgian’ cell lines are less responsive silencing perturbations. cells, while the maximum cellular respiration increase in H460 cells comparison of our new AFR metric versus the aforementioned and therefore require higher dosage of compound to suppress their

2 Molecular Systems Biology 10: 744 | 2014 ª 2014 The Authors ª 2014 The Authors Molecular Systems Biology 10: 744 | 2014 3 Molecular Systems Biology Identifying anti-migratory metabolic drug targets Keren Yizhak et al Keren Yizhak et al Identifying anti-migratory metabolic drug targets Molecular Systems Biology

growth. The effect of most of these compounds is also negatively the more aggressive mesenchymal cell lines exhibiting larger Fig 2D and Supplementary Table S1), it correlates even more Wilcoxon P-value < 0.05, Fig 3C). Finally, lower expression of nine correlated with the cells’ growth rates, suggesting that slowly Warburg effect (Sarrio et al, 2008), Fig 2A). Once again, the AFR strongly in the positive direction with cancer cell migration of the predicted targets is significantly associated with improved proliferating cells are more resistant to treatment (similar results were was more predictive of this experimental observation than the EOR (Spearman correlation of R = 0.88, P-value = 0.03, Fig 2D and long-term survival (Curtis et al, 2012) (log-rank P-value < 0.05, previously shown for compounds targeting cell growth (Penault- (Supplementary Dataset S2). Supplementary Table S1). Controlling for the cell lines’ measured Fig 3D), testifying for their potential role as therapeutic targets. All Llorca et al, 2009; Vincent-Salomon et al, 2004)). Interestingly, the We next turned to our primary objective of examining the rela- growth rates, this correlation becomes even more significant (partial P-values are corrected for multiple hypothesis using FDR with a = 0.05. response to many compounds in this dataset shows a significant tion between the Warburg effect and tumor proliferation and migra- Spearman correlation of R = 0.96, P-value = 7e 3, Supplementary À association with the AFR measure while having no association with tion. To this end, we experimentally measured the migration speed Table S1). Overall, this finding suggests that glycolytic flux correlates siRNA-mediated gene knockdown experiments testing the the cells’ growth rate. 133 such compounds were identified (Supple- of six NCI-60 breast cancer cell lines (Fig 2B and C, Materials and with migration rather than with growth, while OXPHOS flux exhibits predicted targets mentary Dataset S3), possibly suggesting that their mechanism Methods, Supplementary Fig S3, and Supplementary Dataset S2) the opposite behavior. A similar association between lactate secretion might be related to the Warburg level of the cells rather than to their and utilized publically available measured growth rates for these and growth rate has been recently found in an experimental study To experimentally test our predictions we silenced the 17 predicted proliferation. Finally, predicted AFR values correctly separate cell lines. While the AFR correlates markedly negatively with cell by Jain et al (Jain et al, 2012) across the entire NCI-60 collection AFR-reducing genes and examined their phenotypic effects in the between epithelial and mesenchymal breast cancer cell lines (with growth rate (Spearman correlation of R = 0.55, P-value = 4.53e 6, (Spearman correlation of R = 0.22, P = 0.09). Furthermore, MDA-MB-231, MDA-MB-435, BT549, and A549 cell lines. Knock- À À À previous studies have shown that high concentrations of lactate down experiments were performed with SmartPools from Dharma- A C correlate with a high incidence of distant metastasis (Hirschhaeuser con using a live cell migration and fixed proliferation assays et al, 2011). The overall picture portrayed by these correlations is that (Materials and Methods). 8–13 out of the 17 enzymes (8–10 out of while glycolytic carbon diverted to biosynthetic pathways may 12 metabolic reactions) were found to significantly attenuate migra- support cell proliferation, non-diverted glycolytic carbon supports cell tion speed in each cell line (two-sided t-test P-value < 0.05, FDR migration and metastasis (Supplementary Fig S4). corrected with a = 0.05, Fig 4, Materials and Methods and Supple- mentary Dataset S4). This result is highly significant as only 17% of Predicting drug targets that revert the AFR and hence may the metabolic genes were found to impair cell migration in a siRNA inhibit cancer migration screen of 190 metabolic genes (Fokkelman M, Rogkoti VM et al, unpublished data, Bernoulli P-value in the range of 3.9e 3 and À The congruence between AFR levels and disease severity led us to 1.18e 7). Of note, the association between the gene expression of À ask if we could build upon this association to identify potential new the predicted targets and the measured migration speed is insignifi- drug targets. We searched for drug targets predicted to reduce the cant for all targets but one, testifying for the inherent value of our AFR ratio by simulating the knockout of each metabolic reaction model-based prediction analysis (Supplementary Table S3). It across the NCI-60 models, and examining the effects of the knock- should also be noted that the knockdown of the three splices of the outs on biomass production, lactate secretion, and the AFR. As enolase gene have almost no significant effect on these cells’ migra- lactate secretion is a basic indicator of the Warburg effect, we first tion speed, possibly because of isoenzymes backup mechanisms. identified a set of 113 reactions whose knockout is predicted to Importantly, most of the gene knockdown experiments do not abolish lactate secretion rate in all cancer cell lines under biomass manifest any significant effects on cell proliferation (Fig 4). In maximization. Interestingly, the set of enzymes catalyzing these accordance with the findings of Simpson et al (Simpson et al, B D reactions is significantly more highly expressed in the NCI-60 cell 2008), we found that the correlation between the reduction in * lines than the background metabolic genes (one-sided Wilcoxon migration speed and reduction in proliferation rate is mostly P-value < 1.6e 8), indicating the potential oncogenic nature of insignificant (Supplementary Dataset S4), suggesting that the À these genes. reduced migration observed is not simply a consequence of To avoid selecting for drug-resistant clones it would be advanta- common mechanisms hindering proliferation, but rather that it geous to develop drugs that reduce the virulence of cancer cells but occurs due to the disruption of distinct migratory-associated avoid killing them. The knockout of 12 of 113 lactate-reducing reac- metabolic pathways. Luminal tions reduces the AFR but relatively spares biomass production * * (Materials and Methods and Supplementary Table S2). Importantly, ECAR and OCR levels following selected gene silencing the knockout of these 12 reactions according to models of healthy lymphoblast cells built by PRIME (Choy et al, 2008) also spares To further study the association between reduced AFR levels and their biomass production (Materials and Methods). Moreover, we impaired cell migration we used the Seahorse XF96 extracellular found that none of the lymphoblast cell lines show the forced lactate flux analyzer to measure both ECAR and OCR fluxes in the MDA-

Basal secretion that is observed in cancer cells. While the Warburg effect MB-231 cell line, following knockdown of a selected group of targets is sometimes referred in the literature as occurring in highly prolifer- (Materials and Methods and Supplementary Fig S6). As the AFR ating cells in general, our analysis finds that this phenomenon is measure is very difficult to measure experimentally, we tested the apparently more prominent in cancer cells, at least with regard to conventionally measured EOR (ECAR/OCR) as its proxy. We the lymphoblastoid cell population studied here. focused on a subset of seven genes (Fig 5) whose knockdown is Figure 2. Association between AFR levels and cell proliferation and migration. The final list of predicted gene targets includes 17 metabolic predicted to have the highest effect on cell migration and span all A The 20 cell lines that are predicted to exhibit the Warburg effect to the greatest/least extent according to the AFR measure. The x-axis and y-axis represent the mean enzymes that are associated with the final 12 reactions, spanning three predicted metabolic pathways. As shown in Fig 5, a significant and SD of the normalized ATP flux rate in glycolysis and OXPHOS, respectively (Materials and Methods). The AFR measure correctly separates between mesenchymal glycolysis, serine, and methionine metabolism (Fig 3A). 10 of the EOR reduction versus the control is found for all seven examined (orange) and epithelial cell lines (green), showing that the former (which are known to be more aggressive) have higher AFR levels. B We analyzed a panel of six breast cancer cell lines for their migration capacity using live cell imaging. Differential Interference Contrast (DIC) images of the six cell predicted targets have significantly higher expression levels in meta- genes (two-sided t-test P-value < 0.05, FDR corrected with a = 0.05, lines in the order of their respective migration speed (from low to high), scale bar is 100 lm (Materials and Methods). static versus non-metastatic breast cancer patients (Chang et al, Materials and Methods and Supplementary Table S4). The silencing C The average migration speed of cells followed for 12 h in complete medium. Error bars represent SEM; the number of samples is between n = 100 and n = 200. 2005) (one-sided Wilcoxon P-value < 0.05, Fig 3B). Moreover, 9 of of the four glycolytic genes (HK2, PGAM1, PGK2, and GAPDH) D The correlation of predicted model-based EOR and AFR measures to growth and migration rates measured experimentally. Both measures represent a negative the predicted targets exhibit higher expression levels in grade 3 results in both decreased ECAR and increased OCR levels, while correlation with growth and a positive correlation with migration rates. Significant results (P-value < 0.05) are marked with an asterisk. tumors than in grade 1 tumors (Miller et al, 2005) (one-sided the silencing of the serine- and methionine-associated genes

4 Molecular Systems Biology 10: 744 | 2014 ª 2014 The Authors ª 2014 The Authors Molecular Systems Biology 10: 744 | 2014 5 Molecular Systems Biology Identifying anti-migratory metabolic drug targets Keren Yizhak et al Keren Yizhak et al Identifying anti-migratory metabolic drug targets Molecular Systems Biology

A Glucose To Pentose C HK2 Phosphate Pathway P = 7.08e-13 G6P R5P P = 4.05e-4 Methionine P = 4.61e-7 P = 9.8e-3 F6P Glycine Metabolism Glycolysis THF MAT1/2 FBP L-Met adenosy P = 2.78e-8 l-Met mTHF DHAP G3P Serine TPI1 adenosy Serine GAPDH l-hcys 1,3BPG Biosynthesis P = 5.34e-5 P = 1.83e-4 AHCY/AHCYL P = 1.2-3 PGK -kg PHGDH Glu α adn + Hcys 3PG 3PHP P-Serine Serine PSAT1 PSPH P = 8.52e-5 PGAM NAD NADH 2PG ENO PEP PKM Ac-CoA Lactate Pyruvate Pyruvate TCA Gln Cycle

B D

P = 4.61e-8 P = 1.6e-2 P = 5.34e-4 P = 2.2e-6 P = 2.69e-5 P = 1.08e-7 P = 1.96e-6

P = 1.86e-5 P = 6.12e-5 P = 1.1e-4 P = 3.75e-4 P = 2.85e-4 P = 6.08e-4 P = 2.4e-3 P = 1.4e-2 P = 1.5e-3 P = 3.8e-5

P = 4.47e-5 P = 4.3e-4

Figure 3. Gene targets that are predicted to reduce the AFR and their association with prognostic markers of breast cancer patients. A A schematic representation of the 12 predicted gene targets, marked in red. B Ten predicted targets that show a significantly higher expression in metastatic versus non-metastatic tumor samples (n = 295). C Nine predicted targets that show a significantly higher expression in grade 3 versus grade 1 tumor samples (n = 236). D Nine predicted targets whose lower expression is significantly associated with improved long-term survival (n = 1568). Figure 4. Normalized to control mean speed per SmartPool gene silencing of the predicted targets. A–D The four different cell lines that were analyzed: MDA-MB-231, MDA-MB-435s, BT549, and A549. Significant results (two-sided t-test, P-value < 0.05 after correcting for multiple hypothesis using FDR with a = 0.05) are marked with an asterisk. Two different controls are used: (1) non-targeting siRNA (= negative control); and (2) a positive control DNM2 which is known to block both migration and proliferation (Ezratty et al, 2005). Left panel shows migration speed and right panel shows (PSPH, AHCY, and PHGDH) results with decreased ECAR solely ratio were indeed found to attenuate cell migration, and result with nuclear count. Error bars represent SD; the number of samples is n = 3. (Fig 5A). Furthermore, a matching significant difference in a significant reduction in ECAR to OCR levels. Of note, our modeling experimentally measured EOR levels is found between the lowest and approach relies on gene expression differences between the cells highest AFR-reducing genes (one-sided Wilcoxon P-value = 0.05). and does not take into account specific uptake rates. It is therefore Overall, taken together our results testify that, as predicted, the more suited for capturing qualitative rather than exact quantitative and the b-catalytic subunit of ATP synthase forming the BEC index production in endothelial cells and that the silencing of the glyco- knockdown of the top-ranked genes results in attenuated cell differences between the cells, as demonstrated throughout the was found to have a prognostic value in assessing the clinical lytic regulator PFKFB3 impairs the cell migration capacity and inter- migration that is accompanied by reduced EOR and AFR levels. paper. Moreover, the lion share of our analysis is focused on the outcome of patients with early-stage colorectal carcinomas. The feres with vessel sprouting (De Bock et al, 2013). In addition, simulations of perturbations where specific uptake rates are not AFR measure and the BEC index (as computed by its corresponding silencing of PFKFB3 was shown to suppress cell proliferation in available. Nonetheless, utilizing such uptake measurements can RNA levels) are significantly correlated (Spearman R = 0.58, about 50% (De Bock et al, 2013). Overall, the results presented in Discussion significantly increase the correlation to the measured lactate rates P-value = 1.6e 6) across the NCI-60 cell lines, and the BEC index is this study, as well as findings reported by others (Simpson et al, À (Spearman correlation R = 0.67, P-value = 1.5e 8), suggesting that perfectly correlated with migration speed across the six breast 2008), suggest that proliferation and migration are not mutually À In this study we explored the role of the Warburg effect in support- uptake rates measurements under perturbation states can signifi- cancer cell lines (Spearman R = 1, P-value = 2.8e 3). However, the exclusive, and the effect of potential targets on both processes À ing tumor migration, going beyond recent investigations focusing on cantly increase the models’ prediction power. BEC index has inferior performance in predicting drug response should be carefully examined. its role in assisting cancer proliferation. A model-based investigation Our AFR measure is conceptually analogous to a bioenergetic (Supplementary Table S1). Some of our predicted targets have been previously studied in across cancer cell lines shows that the ratio between glycolytic and (BEC) index previously introduced by Cuezva et al (Cuezva et al, The finding that enhanced glycolytic activity plays a key role in the context of cell proliferation as well (Cheong et al, 2012). oxidative ATP flux rate is significantly associated with cancer migra- 2002). In that study, the ratio between the expression of the glyco- cancer cell migration is also in line with a very recent study by Possemato et al (Possemato et al, 2011) have showed that suppres- tory behavior. Gene silencing perturbations predicted to reduce this lytic enzyme glyceraldehyde-3-phosphate dehydrogenase (GAPDH) De Bock et al, showing that glycolysis is the major source of ATP sion of PHGDH in cell lines with elevated PHGDH expression, but not

6 Molecular Systems Biology 10: 744 | 2014 ª 2014 The Authors ª 2014 The Authors Molecular Systems Biology 10: 744 | 2014 7 Molecular Systems Biology Identifying anti-migratory metabolic drug targets Keren Yizhak et al Keren Yizhak et al Identifying anti-migratory metabolic drug targets Molecular Systems Biology

B D

P = 4.61e-8 P = 1.6e-2 P = 5.34e-4 P = 2.2e-6 P = 2.69e-5 P = 1.08e-7 P = 1.96e-6

P = 1.86e-5 P = 6.12e-5 P = 1.1e-4 P = 3.75e-4 P = 2.85e-4 P = 6.08e-4 P = 2.4e-3 P = 1.4e-2 P = 1.5e-3 P = 3.8e-5

P = 4.47e-5 P = 4.3e-4

corresponding metabolic reaction to zero. The biomass function Calculating the EOR and AFR measures for assessing the Warburg level utilized here is taken from (Folger et al, 2011). The media simu- of the cell lines and using them to predict drug response lated in all the analyses throughout the paper is the RPMI-1640 The EOR and AFR measures were calculated in a similar manner to media that was used to grow the cell lines experimentally (Lee that described above. Specifically, the EOR is calculated as the mean et al, 2007; Choy et al, 2008). over lactate secretion across all samples divided by the mean over oxygen consumption across all samples. Similarly, the AFR is calcu- Building cell-specific metabolic models and computing lactate secretion lated as the mean flux carried by the reactions producing ATP in Our method to reconstruct the NCI-60 cancer cell lines (see Supple- glycolysis versus the mean flux carried by the reaction producing mentary Material, based on the yet unpublished methods in Yizhak ATP in OXPHOS. To determine an empiric P-value in the drug et al, submitted) required several key inputs: (a) the generic human response analysis we randomly shuffled the drug response data model (Duarte et al, 2007), (b) gene expression data for each cancer 1,000 times, each time examining the resulting Wilcoxon P-value cell line from (Lee et al, 2007), and (c) growth rate measurements. over the original set of cell lines. The algorithm then reconstructs a specific metabolic model for each sample by modifying the upper bounds of growth-associated reac- Predicting the effect of reaction knockouts tions in accordance with their gene expression (Note: the growth Each metabolic reaction in each cell line model is perturbed by rates were used only to determine which reactions should be used constraining its flux to zero. Under each perturbation the minimal in constraining the models, in order to obtain models that were as lactate secretion (under maximal growth rate) and the maximal physiologically relevant as possible; they were not used to deter- growth rate is calculated. The set of reactions that eliminate forced mine reaction bounds). A similar procedure was used to reconstruct lactate secretion while maintaining a level of cell growth that is the lymphoblast metabolic models (Choy et al, 2008) for compari- > 10% of the wild-type growth prediction is further tested for the son against normal proliferating cells. A more detailed description is AFR level. The mean AFR level for each cell line under each of these found in the Supplementary Material. perturbations is calculated over 1,000 flux distribution samples as Simulations of the Warburg effect include the examination of described above. The final set of predicted reactions includes those Figure 5. ECAR and OCR levels of top predicted gene targets. minimal lactate production rate under different demands for whose knockout reduces the AFR to below 60% of its wild-type level. A Mean and SEM (normalized to nuclear count) ECAR and OCR levels after silencing of seven different genes (HK2, PGAM1, PGK2, GAPDH, PSPH, AHCY, and PHGDH) biomass production, glucose, glutamine, and oxygen uptake rates compared to the control. Silencing of the four glycolytic genes results in both a decrease in ECAR levels (x-axis) and an increase in OCR levels (y-axis), while the (Supplementary Material). We examined the minimal value of Datasets serine- and methionine-associated genes show only a decrease in ECAR levels. Error bars represent SEM. The number of samples is n = 18. lactate secretion as it testifies whether or not the cell is enforced to Growth rate measurements and drug response data were down- B Mean and SD of computed ECAR/OCR (EOR) levels for control and selected gene silencing (Materials and Methods). For all genes a significant reduction in EOR levels is observed. Error bars represent SD. The number of samples is n = 18. secrete lactate under a given condition (Supplementary Fig S1). All loaded from the NCI website. the correlations reported in the paper are Spearman rank correla- Growth rate: http://dtp.nci.nih.gov/docs/misc/common_files/ tions and their associated P-values are computed using the exact cell_list.html in those without, inhibits cell proliferation. Accordingly, as PHGDH Materials and Methods permutation distribution. Drug response: http://discover.nci.nih.gov/nature2000/naturein is not amplified in the cell line MDA-MB-231 which was examined tromain.jsp in both studies, its suppression is indeed non-lethal. However, we Computational methods Calculating wild-type and perturbed lactate secretion rates and show that its suppression significantly attenuates cell migration, OCR levels Experimentally measuring lactate secretion rates of breast suggesting that metabolic enzymes can promote different cancerous Genome-scale metabolic modeling (GSSM) For simulating lactate secretion under normoxic conditions (when cancer cell lines phenotypes in different cancer cells. A metabolic network consisting of m metabolites and n reactions comparing to Jain et al (Jain et al, 2012), Wu et al (Wu et al, 2007)

bolic correlates that are strongly associated with cell migration may model (i.e. the flux distribution). The exchange of metabolites with 50% of its normoxic state as described above. Under each of these Cells were cultured under normoxic (20% O2) and hypoxic (0.5%

lead to new anti-metastatic treatment opportunities. It is important the environment is represented as a set of exchange (transport) conditions, we sampled the solution space under maximal biomass O2) conditions for 72 h. Cells were starved of glucose and glutamine to note, however, that while the inhibition of migration alone might reactions, enabling a pre-defined set of metabolites to be either yield and obtained 1,000 feasible flux distributions (Bordel et al, for 1 h and full medium was added for 1 h. Lactate secretion was be a good strategy for avoiding the adverse side effects of cytotoxic taken up or secreted from the growth media. The steady-state 2010). The predicted lactate secretion rate is the average lactate determined from normoxic and hypoxic cells and normalized to treatment, cell migration is a crucial process also in normal physiol- assumption represented in equation (1) constrains the production secretion flux over these samples. For emulating the perturbation cell growth (increase in total protein during the 72 h incubation in ogy, for instance, in immune response and tissue repair (Fo¨rster rate of each metabolite to be equal to its consumption rate. Enzy- experiments in Wu et al we gradually lowered the bound of the normoxia). Lactate concentrations in media incubated with or et al, 1999; Ridley et al, 2003). Therefore, future anti-migratory matic directionality and flux capacity constraints define lower and corresponding compound target (from the maximal bound to 0) without cells were determined using lactate assay kits (BioVision). drugs may pose different drug selectivity challenges that should be upper bounds on the fluxes and are embedded in equation (2). and repeated the procedure described above for computing Total protein content determined by Sulforhodamine B assay was carefully addressed in the future studies. Irrespectively, they may In the following, flux vectors satisfying these conditions will the ECAR (lactate secretion) and the OCR, which in a similar used for normalization. Two experiments were performed with result in lesser clonal selection, and as a result, their usage may be be referred to as feasible steady-state flux distributions. Gene manner is defined as the average oxygen consumption flux across three or four biologically independent replicates (total of seven accompanied with lesser rate of emergence of drug-resistant clones. knockouts are simulated by constraining the flux through the all samples. replicates).

8 Molecular Systems Biology 10: 744 | 2014 ª 2014 The Authors ª 2014 The Authors Molecular Systems Biology 10: 744 | 2014 9 Molecular Systems Biology Identifying anti-migratory metabolic drug targets Keren Yizhak et al Keren Yizhak et al Identifying anti-migratory metabolic drug targets Molecular Systems Biology

Remarkably, analyzing the model-predicted flux rates has can be represented by a stoichiometric matrix S, where the entry Sij and the breast cancer data collected in this paper), oxygen maximal Cell Culture successfully uncovered a fundamental association between the AFR represents the stoichiometric coefficient of metabolite i in reaction j uptake rate was set to the highest value under which minimal The MCF7, T47D, Hs578T and BT549 breast cancer cell lines were and cancer migration, even given the relatively small set of cell lines (Price et al, 2004). A CBM model imposes mass balance, directional- lactate secretion is positive. Since metabolic models are designed to obtained from the American Type Culture Collection and London for which migration was measured. Our analysis has also revealed ity, and flux capacity constraints on the space of possible fluxes in maximize growth yield rather than growth rate, using an unlimited Research Institute Cell Services. Cells were cultured in DMEM/F12 other potential associations between individual fluxes and cell the metabolic network’s reactions through a set of linear equations: amount of oxygen in GSMM simulations will result in a state where (1:1), with 2 mM L-glutamine and penicillin/streptomycin. Medium migration (Supplementary Fig S4). However, future studies measur- the minimal lactate secretion rate equals zero. However, it’s impor- was supplemented with 10% FCS (GIBCO) for the cancer cell lines ing cellular migration data across a much wider array of cell lines Sv 0 (1) tant to note that even under the limited oxygen levels simulated and 5% horse serum, 20 ng/ml EGF, 5 lg/ml hydrocortisone, ¼ (of the order for which we already have proliferation data) are here, the generic human model doesn’t show lactate secretion (as 10 lg/ml insulin, and 100 ng/ml cholera toxin for the non- needed to determine the actual significance of these potential leads. v v v (2) opposed to the NCI-60 cancer cell line models described above). For malignant cell lines. min � � max As this study has shown, cellular proliferation and migration have simulating the hypoxic conditions measured here for the breast distinct underlying metabolite correlates; understanding the meta- where v stands for the flux vector for all of the reactions in the cancer cell lines, we lowered the oxygen maximal uptake rate by Lactate secretion measurements bolic correlates that are strongly associated with cell migration may model (i.e. the flux distribution). The exchange of metabolites with 50% of its normoxic state as described above. Under each of these Cells were cultured under normoxic (20% O2) and hypoxic (0.5% lead to new anti-metastatic treatment opportunities. It is important the environment is represented as a set of exchange (transport) conditions, we sampled the solution space under maximal biomass O2) conditions for 72 h. Cells were starved of glucose and glutamine to note, however, that while the inhibition of migration alone might reactions, enabling a pre-defined set of metabolites to be either yield and obtained 1,000 feasible flux distributions (Bordel et al, for 1 h and full medium was added for 1 h. Lactate secretion was be a good strategy for avoiding the adverse side effects of cytotoxic taken up or secreted from the growth media. The steady-state 2010). The predicted lactate secretion rate is the average lactate determined from normoxic and hypoxic cells and normalized to treatment, cell migration is a crucial process also in normal physiol- assumption represented in equation (1) constrains the production secretion flux over these samples. For emulating the perturbation cell growth (increase in total protein during the 72 h incubation in ogy, for instance, in immune response and tissue repair (Fo¨rster rate of each metabolite to be equal to its consumption rate. Enzy- experiments in Wu et al we gradually lowered the bound of the normoxia). Lactate concentrations in media incubated with or et al, 1999; Ridley et al, 2003). Therefore, future anti-migratory matic directionality and flux capacity constraints define lower and corresponding compound target (from the maximal bound to 0) without cells were determined using lactate assay kits (BioVision). drugs may pose different drug selectivity challenges that should be upper bounds on the fluxes and are embedded in equation (2). and repeated the procedure described above for computing Total protein content determined by Sulforhodamine B assay was carefully addressed in the future studies. Irrespectively, they may In the following, flux vectors satisfying these conditions will the ECAR (lactate secretion) and the OCR, which in a similar used for normalization. Two experiments were performed with result in lesser clonal selection, and as a result, their usage may be be referred to as feasible steady-state flux distributions. Gene manner is defined as the average oxygen consumption flux across three or four biologically independent replicates (total of seven accompanied with lesser rate of emergence of drug-resistant clones. knockouts are simulated by constraining the flux through the all samples. replicates).

Cell culture for live cell imaging and cell migration assays described earlier (Zhang et al, 2011). SRB data showed a complete computations. SLD, VMR and VCB performed the experimental procedures. KY, Ezratty EJ, Partridge MA, Gundersen GG (2005) Microtubule-induced focal overlap with the nuclear count so this measure is used in all SLD, BvW, and ER wrote the paper. adhesion disassembly is mediated by dynamin and focal adhesion kinase. T47D, MCF-7, MDA-MB-435, BT549, MDA-MB-231 and Hs578t were figures. Changes in proliferation rates upon knockdown when Nat Cell Biol 7: 581 – 590 cultured in RPMI (GIBCO, Life Technologies, Carlsbad, CA, USA) compared to control were evaluated in triplicate via a two-sided Conflict of interest Folger O, Jerby L, Frezza C, Gottlieb E, Ruppin E, Shlomi T (2011) Predicting supplemented with 10% FBS (PAA, Pashing Austria) and 100 t-test. The mean proliferation rate after knockdown between all The authors declare that they have no conflict of interest. selective drug targets in cancer through metabolic networks. Mol Syst Biol International Units/ml penicillin and 100 lg/ml streptomycin three replicates was calculated and normalized to the non-targeting 7: 501 (Invitrogen, Carlsbad, CA, USA). siRNA (= control). Genes achieving P-value < 0.05 after correcting Förster R, Schubel A, Breitfeld D, Kremmer E, Renner-Müller I, Wolf E, Lipp M for multiple hypothesis using FDR with a = 0.05 are considered as References (1999) CCR7 coordinates the primary immune response by establishing Gene silencing hits. functional microenvironments in secondary lymphoid organs. Cell 99: Agren R, Bordel S, Mardinoglu A, Pornputtapong N, Nookaew I, Nielsen J 23 – 33 Human siRNA SmartPools (a combination of four individual singles) Metabolic flux assay (2012) Reconstruction of genome-scale active metabolic networks for 69 Frezza C, Zheng L, Folger O, Rajagopalan KN, MacKenzie ED, Jerby L, Micaroni for the 17 predicted genes were purchased in siGENOME format human cell types and 16 cancer types using INIT. PLoS Comput Biol 8: M, Chaneton B, Adam J, Hedley A, Kalna G, Tomlinson IPM, Pollard PJ, from Dharmacon (Lafayette, CO, USA). Plates were diluted to 1 lM The bioenergetics flux of cells in response to gene silencing was e1002518 Watson DG, Deberardinis RJ, Shlomi T, Ruppin E, Gottlieb E (2011) Haem working concentration in complementary 1× siRNA buffer in a assessed using the Seahorse XF96 extracellular flux analyzer Agren R, Mardinoglu A, Asplund A, Kampf C, Uhlen M, Nielsen J (2014) oxygenase is synthetically lethal with the tumour suppressor fumarate 96-well plate format. A non-targeting siRNA was used as negative (Seahorse Bioscience). About 8,000 MDA-MB-231 cells per well Identification of anticancer drugs for hepatocellular carcinoma through hydratase. Nature 477: 225 – 228 control. A 50 nM reverse transfection was performed according to (Seahorse plate) were treated with siRNAs or control for 72 h. Each personalized genome-scale metabolic modeling. Mol Syst Biol 10: 721 Gatto F, Nookaew I, Nielsen J (2014) Chromosome 3p loss of heterozygosity is manufacturer’s guidelines. Complex time was 20 min and 5,000 gene (in total 7) was knockdown in six different wells and the Bordel S, Agren R, Nielsen J (2010) Sampling the solution space in associated with a unique metabolic network in clear cell renal carcinoma. cells were added. The plate was placed in the incubator overnight experiment was performed twice (so a total of six replicates per genome-scale metabolic networks reveals transcriptional regulation in key Proc Natl Acad Sci 111:E866 – E875 and the medium was refreshed the following morning. After plate and two plates). Prior to measurement, the medium was enzymes. PLoS Comput Biol 6:e1000859 Gille C, Bolling C, Hoppe A, Bulik S, Hoffmann S, Hubner K, Karlstadt A, 48–72 h cells were used for various assays. Cell migration and meta- replaced with unbuffered DMEM XF assay medium. The basal Burgard AP, Pharkya P, Maranas CD (2003) Optknock: a bilevel programming Ganeshan R, Konig M, Rother K, Weidlich M, Behre J, Holzhutter H-G bolic flux assay experiments were performed in duplicate while the oxygen consumption rate (OCR) and extracellular acidification rate framework for identifying gene knockout strategies for microbial strain (2010) HepatoNet1: a comprehensive metabolic reconstruction of the cell proliferation assay was performed in triplicate. (ECAR) were then determined using the XP96 plate reader with the optimization. Biotechnol Bioeng 84: 647 – 657 human hepatocyte for the analysis of liver physiology. Mol Syst Biol 6: 411 standard program as recommended by the manufacturer: three Chandrasekaran S, Price ND (2010) Probabilistic integrative modeling of Goldstein I, Yizhak K, Madar S, Goldfinger N, Ruppin E, Rotter V (2013)p53 Live cell imaging random cell migration assay measurements per well were done (so for each gene 18 measure- genome-scale metabolic and regulatory networks in Escherichia coli and promotes the expression of gluconeogenesis-related genes and enhances ments were obtained for both OCR and ECAR). After the measure- Mycobacterium tuberculosis. Proc Natl Acad Sci 107: 17845 – 17850 hepatic glucose production. Cancer Metab 1: 9 Glass bottom 96-well plates (Greiner Bio-one, Monroe, NC, USA) ments were completed, the plates were live stained with Hoechst Chang HY, Nuyten DSA, Sneddon JB, Hastie T, Tibshirani R, Sørlie T, Dai H, He Hanahan D, Weinberg RA (2011) Hallmarks of cancer: the next generation. were coated with 20 lg/ll collagen type I (isolated from rat tails) 33342 for 1 h and fixed with TCA allowing both a nuclear counting YD, van’t Veer LJ, Bartelink H, van de Rijn M, Brown PO, van de Vijver MJ Cell 144: 646 – 674 for 1 h at 37°C. 48 h after silencing, the MDA-MB-231 cells were re- and/or SRB readout. Whole wells were imaged using epi-fluores- (2005) Robustness, scalability, and integration of a wound-response gene Hirschhaeuser F, Sattler UGA, Mueller-Klieser W (2011) Lactate: a metabolic plated onto the collagen-coated glass bottom plate. 24 h after seed- cence and the number of nuclei was determined using a custom- expression signature in predicting breast cancer survival. Proc Natl Acad key player in cancer. Cancer Res 71: 6921 – 6925 ing, cells were pre-exposed for 45 min to 0.1 lg/µl Hoechst 33342 made ImagePro macro. Plates were further processed for SRB stain- Sci U S A 102: 3738 – 3743 Jain M, Nilsson R, Sharma S, Madhusudhan N, Kitami T, Souza AL, Kafri R, (Fisher Scientific, Hampton, NH, USA) to visualize nuclei. After ing as described earlier (Zhang et al, 2011). SRB data showed a Cheong H, Lu C, Lindsten T, Thompson CB (2012) Therapeutic targets in Kirschner MW, Clish CB, Mootha VK (2012) Metabolite profiling identifies a refreshing the medium, cells were placed on a Nikon Eclipse complete overlap with the nuclear count so this measure was used cancer cell metabolism and autophagy. Nat Biotechnol 30: 671 – 678 key role for glycine in rapid cancer cell proliferation. Science 336: TE2000-E microscope fitted with a 37°C incubation chamber, 20× for normalization. All values are normalized to nuclear count. EOR Choy E, Yelensky R, Bonakdar S, Plenge RM, Saxena R, De Jager PL, Shaw SY, 1040 – 1044 objective (0.75 NA, 1.00 WD) automated stage and perfect focus for control and each gene knockdown is computed by dividing the Wolfish CS, Slavik JM, Cotsapas C, Rivas M, Dermitzakis ET, Jensen PA, Papin JA (2010) Functional integration of a metabolic network system. Three positions per well were automatically defined, and corresponding ECAR and OCR values. A two-sided t-test is applied Cahir-McFarland E, Kieff E, Hafler D, Daly MJ, Altshuler D (2008) Genetic model and expression data without arbitrary thresholding. Bioinformatics the Differential Interference Contrast (DIC) and Hoechst signals to examine significant changes between control and knockdown- analysis of human traits in vitro: drug response and gene expression in 27: 541 – 547 were acquired with a CCD camera (Pixel size: 0.64 lm) every induced EOR. lymphoblastoid cell lines. PLoS Genet 4:e1000287 Jerby L, Wolf L, Denkert C, Stein GY, Hilvo M, Oresic M, Geiger T, Ruppin E 20 min for a total imaging period of 12 h using NIS software Covert MW, Knight EM, Reed JL, Herrgard MJ, Palsson BO (2004) Integrating (2012) Metabolic associations of reduced proliferation and oxidative stress (Nikon). All data were converted and analyzed using custom-made Supplementary information for this article is available online: high-throughput and computational data elucidates bacterial networks. in advanced breast cancer. Cancer Res 72: 5712 – 5720 ImagePro Plus macros (Roosmalen et al, 2011). Cell migration was http://msb.embopress.org Nature 429: 92 – 96 Lee JK, Havaleshko DM, Cho H, Weinstein JN, Kaldjian EP, Karpovich J, quantified by tracking nuclei in time. Changes in migration speed Cuezva JM, Krajewska M, de Heredia ML, Krajewski S, Santamaría G, Kim H, Grimshaw A, Theodorescu D (2007) A strategy for predicting the per knockdown were evaluated via a two-sided t-test comparing the Acknowledgements Zapata JM, Marusawa H, Chamorro M, Reed JC (2002) The bioenergetic chemosensitivity of human cancers and its application to drug discovery. speed for every individual cell followed overtime for 16 h and the We would like to thank Hans de Bont and Michiel Fokkelman for their technical signature of cancer: a marker of tumor progression. Cancer Res 62: Proc Natl Acad Sci 104: 13086 – 13091 corresponding control values. Data shown are normalized to control support, Yoav Teboulle, Matthew Oberhardt, Edoardo Gaude, Gideon Y. Stein 6674 – 6681 Lerman JA, Hyduke DR, Latif H, Portnoy VA, Lewis NE, Orth JD, and represent only one replicate. Of note, for all four cell lines both and Tami Geiger for their helpful comments on the manuscript. KY is partially Curtis C, Shah SP, Chin S-F, Turashvili G, Rueda OM, Dunning MJ, Speed D, Schrimpe-Rutledge AC, Smith RD, Adkins JN, Zengler K, Palsson BO (2012) replicates showed a R2 of reproducibility above 0.75. Genes achiev- supported by a fellowship from the Edmond J. Safra Bioinformatics center at Lynch AG, Samarajiwa S, Yuan Y, Graf S, Ha G, Haffari G, Bashashati A, In silico method for modelling metabolism and gene product expression ing P-value < 0.05 after correcting for multiple hypothesis using Tel-Aviv University and is grateful to the Azrieli Foundation for the award of an Russell R, McKinney S, Langerod A, Green A, Provenzano E, Wishart G et al at genome scale. Nat Commun 3: 929 FDR with a = 0.05 are considered as hits. Azrieli Fellowship; SLD is supported by the Netherlands Consortium for Systems (2012) The genomic and transcriptomic architecture of 2,000 breast Lewis NE, Schramm G, Bordbar A, Schellenberger J, Andersen MP, Cheng JK, Biology and the EU FP7 Systems Microscopy NoE project (258068) and BvdW tumours reveals novel subgroups. Nature 486: 346 – 352 Patel N, Yee A, Lewis RA, Eils R, Konig R, Palsson BO (2010) Large-scale in Proliferation assay from the Netherlands Genomics Initiative. ER acknowledges the generous De Bock K, Georgiadou M, Schoors S, Kuchnio A, Wong Brian W, Cantelmo silico modeling of metabolic interactions between cell types in the human support of grants from the Israeli Science Foundation (ISF), the Israeli Cancer Anna R, Quaegebeur A, Ghesquière B, Cauwenberghs S, Eelen G, Phng L-K, brain. Nat Biotechnol 28: 1279 – 1285 Cells were directly transfected and plated onto micro-clear 96-well Research Fund (ICRF) and the I-CORE Program of the Planning and Budgeting Betz I, Tembuyser B, Brepoels K, Welti J, Geudens I, Segura I, Cruys B, Ma H, Sorokin A, Mazein A, Selkov A, Selkov E, Demin O, Goryanin I (2007) plates (Greiner Bio-one). After 5 days of incubation, the cells were Committee and The Israel Science Foundation (grant No 41/11). Bifari F, Decimo I et al (2013) Role of PFKFB3-driven glycolysis in vessel The Edinburgh human metabolic network reconstruction and its stained with Hoechst 33342 and fixed with TCA (Trichloroacetic sprouting. Cell 154: 651 – 663 functional analysis. Mol Syst Biol 3: 135 acid) allowing both a nuclear counting and/or Sulforodamine B Author contributions Duarte NC, Becker SA, Jamshidi N, Thiele I, Mo ML, Vo TD, Srivas R, Mardinoglu A, Agren R, Kampf C, Asplund A, Nookaew I, Jacobson P, Walley (SRB) readout. Whole wells were imaged using epi-fluorescence KY and ER conceived and designed the research. SLD, VCB, CF, and BvW Palsson BO (2007) Global reconstruction of the human metabolic AJ, Froguel P, Carlsson LM, Uhlen M, Nielsen J (2013) Integration of clinical and the number of nuclei was determined using a custom-made designed the experimental procedures. FB and AS contributed the lactate network based on genomic and bibliomic data. Proc Natl Acad Sci data with a genome-scale metabolic model of the human adipocyte. Mol ImagePro macro. Plates were further processed for SRB staining as secretion data. KY performed the computational analysis and the statistical USA104: 1777 – 1782 Syst Biol 9: 649

10 Molecular Systems Biology 10: 744 | 2014 ª 2014 The Authors ª 2014 The Authors Molecular Systems Biology 10: 744 | 2014 11 Molecular Systems Biology Identifying anti-migratory metabolic drug targets Keren Yizhak et al Keren Yizhak et al Identifying anti-migratory metabolic drug targets Molecular Systems Biology

Miller LD, Smeds J, George J, Vega VB, Vergara L, Ploner A, Pawitan Y, Hall P, Shlomi T, Cabili MN, Herrgård MJ, Palsson BØ, Ruppin E (2008) Klaar S, Liu ET, Bergh J (2005) An expression signature for p53 status in Network-based prediction of human tissue-specific metabolism. Nat human breast cancer predicts mutation status, transcriptional effects, and Biotechnol 26: 1003 – 1010 patient survival. Proc Natl Acad Sci U S A 102: 13550 – 13555 Simpson KJ, Selfors LM, Bui J, Reynolds A, Leake D, Khvorova A, Brugge JS Nogales J, Gudmundsson S, Knight EM, Palsson BO, Thiele I (2012) Detailing (2008) Identification of genes that regulate epithelial cell migration using the optimality of photosynthesis in cyanobacteria through systems an siRNA screening approach. Nat Cell Biol 10: 1027 – 1038 biology analysis. Proc Natl Acad Sci 109: 2678 – 2683 Szappanos B, Kovacs K, Szamecz B, Honti F, Costanzo M, Baryshnikova A, Oberhardt MA, Palsson BO, Papin JA (2009) Applications of genome-scale Gelius-Dietrich G, Lercher MJ, Jelasity M, Myers CL, Andrews BJ, Boone C, metabolic reconstructions. Mol Syst Biol 5: 320 Oliver SG, Pal C, Papp B (2011) An integrated approach to characterize Oberhardt MA, Yizhak K, Ruppin E (2013) Metabolically re-modeling the drug genetic interaction networks in yeast metabolism. Nat Genet 43: pipeline. Curr Opin Pharmacol 13: 778 – 785 656 – 662 Penault-Llorca F, André F, Sagan C, Lacroix-Triki M, Denoux Y, Verriele V, Trawick JD, Schilling CH (2006) Use of constraint-based modeling for the Jacquemier J, Baranzelli MC, Bibeau F, Antoine M, Lagarde N, Martin A-L, prediction and validation of antimicrobial targets. Biochem Pharmacol 71: Asselain B, Roché H (2009) Ki67 expression and docetaxel efficacy in 1026 – 1035 patients with estrogen receptor-positive breast cancer. J Clin Oncol 27: Vander Heiden MG, Cantley LC, Thompson CB (2009) Understanding the 2809 – 2815 Warburg effect: the metabolic requirements of cell proliferation. Science Pharkya P, Burgard AP, Maranas CD (2004) OptStrain: a computational 324: 1029 – 1033 framework for redesign of microbial production systems. Genome Res 14: Vincent-Salomon A, Rousseau A, Jouve M, Beuzeboc P, Sigal-Zafrani B, 2367 – 2376 Fréneaux P, Rosty C, Nos C, Campana F, Klijanienko J, Al Ghuzlan A, Possemato R, Marks KM, Shaul YD, Pacold ME, Kim D, Birsoy K, Sastre-Garau X (2004) Proliferation markers predictive of the pathological Sethumadhavan S, Woo H-K, Jang HG, Jha AK, Chen WW, Barrett FG, response and disease outcome of patients with breast carcinomas treated Stransky N, Tsun Z-Y, Cowley GS, Barretina J, Kalaany NY, Hsu PP, Ottina by anthracycline-based preoperative chemotherapy. Eur J Cancer 40: K, Chan AM et al (2011) Functional genomics reveal that the serine 1502 – 1508 synthesis pathway is essential in breast cancer. Nature 476: 346 – 350 Warburg O (1956) On the origin of cancer cells. Science 123: 309 – 314 Price ND, Reed JL, Palsson BO (2004) Genome-scale models of microbial cells: Wessely F, Bartl M, Guthke R, Li P, Schuster S, Kaleta C (2011) Optimal evaluating the consequences of constraints. Nat Rev Microbiol 2: 886 – 897 regulatory strategies for metabolic pathways in Escherichia coli depending Ridley AJ, Schwartz MA, Burridge K, Firtel RA, Ginsberg MH, Borisy G, Parsons on protein costs. Mol Syst Biol 7: 515 JT, Horwitz AR (2003) Cell migration: integrating signals from front to Wu M, Neilson A, Swift AL, Moran R, Tamagnine J, Parslow D, Armistead S, back. Science 302: 1704 – 1709 Lemire K, Orrell J, Teich J, Chomicz S, Ferrick DA (2007) Multiparameter van Roosmalen W, Le Dévédec SE, Zovko S, de Bont H, van de Water B (2011) metabolic analysis reveals a close link between attenuated mitochondrial Functional screening with a live cell imaging-based random cell migration bioenergetic function and enhanced glycolysis dependency in human assay. Methods Mol Biol 769: 435 – 448 tumor cells. Am J Physiol Cell Physiol 292:C125 – C136 Sarrio D, Rodriguez-Pinilla SM, Hardisson D, Cano A, Moreno-Bueno G, Yizhak K, Gabay O, Cohen H, Ruppin E (2013) Model-based identification of Palacios J (2008) Epithelial-mesenchymal transition in breast cancer drug targets that revert disrupted metabolism and its application to relates to the basal-like phenotype. Cancer Res 68: 989 – 997 aging. Nat Commun 4: 2632 Scherf U, Ross DT, Waltham M, Smith LH, Lee JK, Tanabe L, Kohn KW, Reinhold Zhang Y, Moerkens M, Ramaiahgari S, de Bont H, Price L, Meerman J, van de WC, Myers TG, Andrews DT, Scudiero DA, Eisen MB, Sausville EA, Pommier Y, Water B (2011) Elevated insulin-like growth factor 1 receptor signaling Botstein D, Brown PO, Weinstein JN (2000) A gene expression database for induces antiestrogen resistance through the MAPK/ERK and PI3K/Akt the molecular pharmacology of cancer. Nat Genet 24: 236 – 244 signaling routes. Breast Cancer Res 13:R52 Schuetz R, Zamboni N, Zampieri M, Heinemann M, Sauer U (2012) Multidimensional optimality of microbial metabolism. Science 336: License: This is an open access article under the 601 – 604 terms of the Creative Commons Attribution 4.0 Shlomi T, Benyamini T, Gottlieb E, Sharan R, Ruppin E (2011) Genome-scale License, which permits use, distribution and reproduc- metabolic modeling elucidates the role of proliferative adaptation in tion in any medium, provided the original work is causing the Warburg effect. PLoS Comput Biol 7:e1002018 properly cited.

12 Molecular Systems Biology 10: 744 | 2014 ª 2014 The Authors further reading

The EMBO Journal EMBO Molecular Medicine The Huntington disease protein accelerates breast tumour Foxm1 transcription factor is Prolyl-isomerase Pin1 controls development and metastasis required for lung fibrosis and normal and cancer stem cells of through ErbB2/HER2 signalling. epithelial-to-mesenchymal the breast. Moreira Sousa C, McGuire JR, Thion MS, transition. Rustighi A, Zannini A, Tiberi L, Gentien D, de la Grange P, Tezenas du Balli D, Ustiyan V, Zhang Y, Wang Sommaggio R, Piazza S, Sorrentino G, Montcel S, Vincent-Salomon A, Durr A, IC, Masino AJ, Ren X, Whitsett JA, Nuzzo S, Tuscano A, Eterno V, Benvenuti Humbert S. Kalinichenko VV, Kalin TV. F, Santarpia L, Aifantis I, Rosato A, DOI: 10.1002/emmm.201201546 | Bicciato S, Zambelli A, Del Sal G. DOI: 10.1038/emboj.2012.336 Published 09.01.2013 Published 04.01.2013 DOI: 10.1002/emmm.201302909 | Published 12.12.2013 embomolmed.embopress.org emboj.embopress.org Targeting the androgen receptor with siRNA promotes Molecular Systems Biology EMBO Reports prostate cancer metastasis through enhanced macrophage Metabolic shifts toward Kindlin 2 forms a transcriptional recruitment via CCL2/CCR2- glutamine regulate tumor complex with β-catenin and induced STAT3 activation. growth, invasion and TCF4 to enhance Wnt signalling. Izumi K, Fang LY, Mizokami A, Namiki M, bioenergetics in ovarian cancer. Yu Y, Wu J, Wang Y, Zhao T, Ma B, Liu Y, Li L, Lin WJ, Chang C. Yang L, Moss T, Mangala LS, Marini J, Fang W, Zhu WG, Zhang H. DOI: 10.1002/emmm.201202367 | Zhao H, Wahlig S, Armaiz-Pena G, Jiang DOI: 10.1038/embor.2012.88 Published 27.08.2013 D, Achreja A, Win J, Roopaimoole R, Published 15.06.2012 Rodriguez-Aguayo C, Mercado-Uribe I, Lopez-Berestein G, Liu J, Tsukamoto T, Tie1 deficiency induces Inhibition of endotrophin, a Sood AK, Ram PT, Nagrath D. endothelial-mesenchymal cleavage product of collagen VI, DOI: 10.15252/msb.20134993 | transition. confers cisplatin sensitivity to Published 05.05.2014 Garcia J, Sandi MJ, Cordelier P, Binétruy B, tumours. Pouysségur J, Iovanna JL, Tournaire R. Park J, Morley TS, Scherer PE. msb.embopress.org DOI: 10.1038/embor.2012.29 DOI: 10.1002/emmm.201202006 | Published 16.03.2012 Published 30.04.2013

embor.embopress.org EMBO Molecular Medicine

embor.embopress.org | emboj.embopress.org | embomolmed.embopress.org | msb.embopress.org

58403 / MITM027479