A cluster of noncoding RNAs activates the ESR1 locus during breast cancer adaptation

Estrogen receptor-α (ER)-positive breast cancer cells undergo hormone-independent proliferation after deprivation of oestrogen, leading to endocrine therapy resistance. Up-regulation of the ER gene (ESR1) is critical for this process, but the underlying mechanisms remain unclear. Here we show that the combination of transcriptome and fluorescence in situ hybridization analyses revealed that oestrogen deprivation induced a cluster of noncoding RNAs that defined a large chromatin domain containing the ESR1 locus. We termed these RNAs as Eleanors (ESR1 locus enhancing and activating noncoding RNAs). Eleanors were present in ER-positive breast cancer tissues and localized at the transcriptionally active ESR1 locus to form RNA foci. Depletion of one Eleanor, upstream (u)-Eleanor, impaired cell growth and transcription of intragenic Eleanors and ESR1 mRNA, indicating that Eleanors cis-activate the ESR1 gene. Eleanor-mediated gene activation represents a new type of locus control mechanism and plays an essential role in the adaptation of breast cancer cells.

C ancer cells adapt to the surrounding environment and maintain their proliferation, resulting in malignant transformation and resistance to anticancer treatments 1 . Breast cancers expressing estrogen receptor-a (ER) depend on oestrogen for cellular growth and survival. ER functions as a nuclear receptor-type transcription factor upon binding to oestrogen and regulates the expression of various target genes. Endocrine therapies, such as the use of an aromatase inhibitor (AI) that blocks oestrogen production, are the most effective for ER-positive breast cancers 2 . However, these treatments are frequently followed by disease recurrence because most breast tumours, which are initially responsive to these therapies, develop resistances through unknown mechanisms [2][3][4] .
MCF7 human breast cancer cells are ER-positive and acquire oestrogen-independent proliferation when they are cultured under an oestrogen-depleted condition for a prolonged period of time (long-term oestrogen deprivation; LTED) 5,6 . LTED adaptation is a well-established cellular model that recapitulates acquisition of AI resistance or postmenopausal tumorigenesis [5][6][7][8][9][10] . Previous studies have reported that the gene-encoding ER (ESR1) is up-regulated during LTED adaptation 8 , which is found in ERpositive human breast cancers. Understanding the molecular mechanism of this gene activation is critical because overproduction of ER may lead to an enhanced response to low concentrations of oestrogen, which is responsible for the LTEDadapted phenotype 7,9,11 . Paradoxically, administration of oestrogen is an effective treatment for AI-resistant breast cancers [12][13][14] , and the LTED cell model may be used to gain the mechanistic evidence for such therapeutic efficacy.
Gene expression patterns are reprogrammed in response to environmental changes or during development and linked to the conversion of cellular phenotypes. Several events that occur in chromatin include recruitment of transcriptional activators/ repressors, changes in histone/DNA modifications, RNA polymerase II (RNA Pol II) binding, long-range chromosomal interactions and chromatin domain formation [15][16][17][18] . It was classically shown that the b-globin locus forms a distinct open chromatin domain during erythropoiesis [16][17][18] . To date, various types of chromatin domains have been characterized as B10 kb to a few Mb in length by genome-wide chromosome conformation, histone modification patterns, association with specific nuclear architectures and nuclease sensitivities [19][20][21][22] . These data suggest that interphase chromosomes are organized by hierarchical folding through which transcription can be regulated through chromatin domain formation.
Recent studies have revealed that noncoding RNAs (ncRNAs) are also involved in transcriptional regulation through diverse functions 23 . The mammalian transcriptome includes thousands of long noncoding RNAs (lncRNAs) that are longer than 200 nucleotides and devoid of protein-coding potential 24 . Some lncRNAs show unique expression under specific conditions such as X chromosome inactivation, genomic imprinting and maintenance or differentiation of stem cells [25][26][27] . LncRNAs are encoded at virtually any site of the genome, including enhancer, promoter, intron and intergenic regions, which regulate genes both in cis and trans. Currently, the potential roles of lncRNAs in cancer cell adaptation are unknown.
In the present study, we found that up-regulation of ESR1 was important for LTED cell adaptation, which was maintained by novel ncRNAs produced from a large chromatin domain of the ESR1 gene. Fluorescence in situ hybridization (FISH) analyses showed that these ncRNAs, termed Eleanors (ESR1 locus enhancing and activating noncoding RNAs), were localized at the site of active transcription, resulting in the formation of distinct RNA foci in the nucleus. One of the Eleanors, upstream-Eleanor (u-Eleanor), originated from an enhancer-like sequence upstream of the ESR1 gene, which was necessary for enhanced expression of both ESR1 mRNA and intragenic Eleanors in LTED cells. Our genome-wide transcriptome analyses revealed that coordinated expression of ncRNA and mRNA, exemplified by the ESR1 gene, was conserved in a set of long genes. These findings uncover the molecular basis for endocrine therapyresistant breast cancer, which involves a new type of ncRNAmediated regulation of a chromatin domain and protein-coding genes.

Results
ESR1 up-regulation is accompanied by Eleanor expression. To understand the mechanism of hormonal adaptation and the action of resveratrol in ER-positive breast cancers, we used a cell model system in which MCF7 cells were cultured under three different conditions: normal (MCF7), oestrogen deprivation for 2-4 months (LTED) and further treatment with 100 mM resveratrol for 24 h (LTED-RES, Fig. 1a). Resveratrol is structurally similar to oestrogen, binds to ER in vitro and exerts oestrogenic effects on breast cancer cells 28,29 . Quantitative PCR with reverse transcription (qRT-PCR) and immunofluorescence analyses showed that ESR1 expression was significantly increased in LTED cells and dramatically suppressed by resveratrol (Fig. 1b,c). Notably, knockdown of ER significantly reduced LTED cell proliferation at 96 h after transfection of the small interfering RNA (siRNA) (Fig. 1d). This result suggests that the up-regulation of ER plays a role in acquisition of oestrogenindependent cancer cell growth.
To further investigate activation of the ESR1 gene, we performed mRNA-Seq and RNA-Seq analyses of cells under the three conditions. We prepared poly (A) þ RNA for mRNA-Seq, and total RNA that was devoid of ribosomal RNA for RNA-Seq, respectively (see Methods for details). Gene tracks representing mRNA-Seq and RNA-Seq data are shown in Fig. 1e. The human ESR1 locus resides on chromosome 6, consists of eight exons and is B300 kb in length. As expected, mRNA-Seq data showed up-regulation of ESR1 exons in LTED cells and repression in LTED-RES cells (Fig. 1e, top three tracks). Interestingly, RNA-Seq analyses detected a significant amount of intragenic transcripts in LTED cells, which extended along the entire ESR1 locus including introns and upstream noncoding regions, but not to the neighbouring silent gene, SYNE1 (Fig. 1e, fifth track). We named the noncoding RNAs produced from inside and around the ESR1 locus (6q25.1; 152083078-152424447) as 'Eleanors'. Production of Eleanors was well synchronized with production of ESR1 mRNA, suggesting that Eleanors participate in regulation of the ESR1 gene. Eleanors were distinct from previously reported types of ncRNAs, such as enhancer/promoter-RNAs and miRNAs, because Eleanors were expressed from a much larger chromatin region.
Eleanors are localized at the site of ESR1 transcription. To confirm the presence of Eleanors, we assessed transcripts from ESR1-intron 2 by qRT-PCR (Fig. 2a). Generally, introns are rapidly processed to undetectable levels after transcription as shown for intron 11 of ERBB2. However, transcripts from intron 2 of ESR1 were unusually stable in LTED cells. To further examine the presence of ncRNAs derived from the broad region of the ESR1 locus, we performed FISH analyses using bacterial artificial chromosome (BAC) probes that covered most of the ESR1 locus (ESR1-BAC) (Figs 1e and 2b). First, cells were processed for DNA FISH using the BAC clone for the centromeric region of chromosome 6 (CEN6-BAC) as a control (Fig. 2b, top panels). We detected three or four CEN6 signals (red) in the nucleus because of the standard karyotypes of MCF7 cells 30 . As expected, ESR1 signals (green) were detected close to each CEN6 signal. However, it was surprising that the ESR1 signals were significantly larger in LTED cells (B2-fold higher ESR1/CEN6 area ratio compared with that in MCF7 cells) and obviously smaller in LTED-RES cells (Fig. 2b, bottom right panel).
The enlarged FISH signals may suggest homogeneously staining regions that are cytogenetic hallmarks of genomic amplification in cancer 31 . However, such a notion was not the case for the ESR1 locus in LTED cells, because the enlarged foci were promptly reduced to small dots by resveratrol treatment. Indeed, we found no ESR1 gene amplification in copy number variation analysis ( Supplementary Fig. 1a,b).
Further analyses showed that most of the enlarged FISH signals consisted of RNA molecules, because they were significantly diminished by treatment with RNase, but not DNase ( Supplementary Fig. 1c). To clarify the origin of the RNAs accumulating in the enlarged foci, we performed FISH under non-denaturing conditions (RNA FISH) using the ESR1-BAC probe (covering noncoding regions as well as exons), ESR1-cDNA probe (covering exons exclusively) and ESR1-intron 2 probe (Fig. 2b lower three panels). The enlarged signals in LTED cells were detected with the BAC and intron 2 probes, but not the ESR1-cDNA probe, indicating that ncRNAs derived from the qRT-PCR results under the MCF7 condition were set to 1. Primers were designed to cover the exon-exon junction. Values are the means±s.d.; n ¼ 3. P-values were calculated using Student's t-test (*Po0.05, ***Po0.001). (c) Immunofluorescence of ER showing enhanced expression in LTED cells and its suppression by resveratrol treatment (LTED-RES). Scale bar, 10 mm. (d) ESR1 knockdown inhibits LTED cell proliferation. LTED cells were treated with siRNA targeting ESR1 for the indicated periods. Cell growth is shown as fold changes. Values are the means ± s.d.; n ¼ 3. P-values were calculated using Student's ttest (*Po0.05, ***Po0.001). (e) Gene tracks representing mRNA-Seq and RNA-Seq data of the human ESR1 locus. Novel ncRNAs, termed Eleanors, were abundantly expressed in LTED cells from the entire ESR1 locus, which were detected as read signals in non-exonic regions. Eleanors were suppressed in LTED-RES cells. The structures of ESR1 and downstream SYNE1 genes are shown below. Green bars indicate the FISH probes used in this study. Regions highlighted in Figs 2a and 4a are denoted by # and ##, respectively. 2M, two months; 4M, four months. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms7966 ARTICLE intragenic region (Eleanors) were the major components of the enlarged FISH signals.
We investigated the spatial positioning of Eleanors relative to the ESR1 locus in the LTED nucleus by sequential hybridizations. First, Eleanors were hybridized with the ESR1-BAC2 probe (shown in Fig. 1e), followed by fixation of the signals and subsequent RNase treatment, and then the ESR1 gene was hybridized with the ESR1-BAC probe. The results showed that Eleanors and the ESR1 locus were co-localized in the nucleus (Fig. 2c). Collectively, these data demonstrate that Eleanors are produced from the intragenic region of the transcriptionally active ESR1 locus in LTED cells and remain associated with the site of transcription, resulting in the formation of distinct RNA foci. Our stranded RNA-Seq results indicated that    Eleanors were transcribed in the same direction as that of ESR1 mRNA.
To investigate whether ncRNA production in LTED cells is specific to the ESR1 locus, we examined the ERBB2 gene that plays a role in a subset of breast cancers. Gene tracks representing mRNA-Seq and RNA-Seq data of the ERBB2 locus are shown in Fig. 2d. ERBB2 gene expression was not accompanied by ncRNA production in LTED cells. However, similar to ESR1 expression, ERBB2 expression was activated by more than 3-fold (Fig. 2e). Consistently, FISH signals for the ERBB2 locus (green) showed no changes, while signals for the ESR1 locus (red) were large in the same nucleus (Fig. 2f,g). Two other genes, APP and ERGIC, also showed similar levels of up-regulation in LTED cells without significant RNA expression from noncoding regions ( Supplementary Fig. 2).
Adaptation to hormone deprivation is conserved among several breast cancer cells. Another ER-positive breast cancer cell line, HCC1428 acquires oestrogen-independent proliferation with up-regulation of ER after a long period of oestrogen deprivation (HCC1428 LTED cells) 32 . Using an ESR1-BAC probe, we detected the enlarged FISH signals in HCC1428 LTED cells, which were suppressed by resveratrol treatment (HCC1428 LTED-RES cells), similar to LTED and LTED-RES cells ( Supplementary Fig. 1d).
Together, the Eleanor-containing foci were characteristic of the ESR1 locus in LTED cells, suggesting a new type of gene activation mechanism for the specific gene locus.
Eleanors are present in ER-positive breast cancer tissues. Naturally occurring human breast cancers are grouped into at least three subtypes: a luminal type that is ER-positive, an ERBB2 type that highly expresses ERBB2 with genomic amplifications, and a triple-negative type that is negative for ER, ERBB2 and the progesterone receptor 33,34 . To confirm the appearance of Eleanor-containing foci in vivo, we performed a combination of immunohistochemistry (IHC) and FISH analyses using serial sections of various breast cancer tissues (Fig. 3). We detected the enlarged FISH signals for Eleanors (green) in some luminal-type breast cancers, which were well correlated to ER overexpression found in IHC. On the other hand, these FISH signals were absent from normal breast tissue as well as ERBB2-type and triplenegative-type cancer tissues, all of which were ER negative. These results suggest significant implications of Eleanors in ER-positive breast cancer cells.
u-Eleanor enhances intragenic Eleanor and ESR1 mRNA. Our RNA-Seq data showed that the region producing ncRNAs extended further upstream of the ESR1 gene (Fig. 4a). One of the peaks on the gene track was positioned at site c that was B40 kb upstream of the canonical promoter A (site f) in MCF7 cells 35,36 . qRT-PCR analyses showed local transcription from site c (Fig. 4b). Stranded RNA-Seq data indicated that u-Eleanor was transcribed in the same direction as that of intragenic Eleanors and ESR1 mRNA. In agarose gel electrophoresis of RT-PCR products, we detected a transcript of at least 1,200 nucleotides in length ( Supplementary Fig. 3a,b). We termed this lncRNA as u-Eleanor.
Alignment with the chromatin immunoprecipitationsequencing (ChIP-Seq) data revealed that the u-Eleanor gene region (site c) was bound by RNA Pol II, CTCF and transactivators (GATA3, CEBPB and p300) in MCF7 cells (Supplementary Table 1). Consistently, this site was enriched with active histone marks (trimethyl-H3K4 and acetyl-H3K27) and devoid of repressive marks (trimethyl-H3K9 and trimethyl-H3K27). These active chromatin features at the u-Eleanor gene region were characteristic of cell types of breast origin. Similar to enhancer RNAs (eRNAs), these data suggest that site c functions as an upstream regulatory element that is actively transcribed into ncRNAs 37 .
Alternatively, u-Eleanor could be an unannotated upstream promoter of ESR1. Although promoter A is most frequently used in MCF7 cells 36 , transcription of ESR1 can be complex under certain circumstances because of the nature of differential promoter usage 35,36 . Gene annotations in UCSC and GENCODE genome browsers 38,39 revealed a variety of cDNAs and ESTs, including independent transcripts, which terminate before the ESR1 gene, as well as a transcript that spans from the upstream region to the complete body of ESR1 ( Supplementary Fig. 3a). However, u-Eleanor did not correspond to any of the previously described alternative promoters 35,36 . u-Eleanor was unlikely to be contiguous with the downstream Eleanors because we did not detect any transcripts from sites d or e ( Fig. 4b and Supplementary Fig. 3a,b). In addition, RT-PCR using several primer sets starting in u-Eleanor and ending in ESR1 exon 1 failed to detect any transcripts (cg in Supplementary Fig. 3b). Furthermore, no protein-coding possibility was predicted in the u-Eleanor gene region (Supplementary Table 5). Taken together, we conclude that u-Eleanor is a transcript independent of intragenic Eleanors and ESR1 mRNA.
Transcription of u-Eleanor may enhance ESR1 mRNA expression, because both RNAs were up-regulated in LTED cells and repressed in LTED-RES cells (Figs 1b and 4b). We suspected that u-Eleanor might function as an eRNA, which was in agreement with our ChIP-qPCR data showing that the u-Eleanor region and promoter A of ESR1 were bound by the active form of RNA Pol II (phosphorylated at serine 5) in LTED cells (Fig. 4c). In addition, u-Eleanor chromatin was enriched with monomethylation of H3K4 (H3K4me1) rather than tri-methylation (H3K4me3) ( Supplementary Fig. 3c,d), suggesting that this region functions as an enhancer 40,41 . On the other hand, promoter A was high in H3K4me3 relative to H3K4me1. The aligned ChIP-Seq data showed that the u-Eleanor region was also bound by CTCF, an organizer for the three-dimensional structure of the genome. Chromosome conformation capture experiments suggested that the u-Eleanor region as well as exon 1 and intron 1 of the ESR1 gene were close together in LTED cells compared with that in MCF7 and LTED-RES cells ( Supplementary Fig. 4). These results suggest that the upstream chromatin structure of this locus is significantly altered during LTED adaptation.
We tested whether u-Eleanor is involved in transcriptional activation of the ESR1 locus in LTED cells. Upon reduction of u-Eleanor to 50-70% by two independent siRNAs, ESR1 mRNA expression was decreased to B60% (Po0.01, P-values; Student's t-test ) without affecting ERBB2 mRNA (Fig. 4d). Consistently, the ER protein level was decreased with the reduction of u-Eleanor (Fig. 4e). Furthermore, FISH analyses of LTED cells were performed using ESR1-BAC and ESR1-BAC2 (shown in Fig. 1e) as independent probes (Fig. 4f). u-Eleanor knockdown efficiently diminished the enlarged FISH signals, indicating that u-Eleanor maintains downstream intragenic Eleanor in LTED cells.
Next, we examined the effect of Eleanors on the proliferative activity of LTED cells. As a result, the cell number was significantly reduced at 96 h after knockdown of u-Eleanor by siRNA (Fig. 4g). Thus, u-Eleanor has an essential role in the enhanced transcription of Eleanors and mRNA from the ESR1 locus, as well as the cell proliferation and viability during LTED adaptation.
Resveratrol exerts a repressive effect on Eleanor via ER. As described above, addition of resveratrol to LTED cells dramatically co-suppressed the expression of both Eleanors and ESR1 mRNA (Figs 1b,e and 4b, red bars). As a result, resveratrol inhibited the proliferative activity of LTED cells in a dose-and time-dependent manner (Fig. 5a). Because oestrogen-loaded ER negatively regulates the ESR1 gene 42 , we expected that resveratrol would repress the ESR1 locus through ER. To test this hypothesis, we depleted ER with specific siRNAs in LTED cells followed by resveratrol treatment (RES, Fig. 5b). FISH analyses using the ESR1-BAC probe showed that the Eleanor-containing foci in control cells (siGL3) became smaller after resveratrol treatment, but they remained large in ER-depleted cells (siESR1, Fig. 5c,d). These results indicated that suppression of Eleanors by resveratrol is dependent on ER. To confirm this result, we used a specific ER antagonist, ICI 182,780, which induces degradation of ER through the ubiquitin-mediated pathway 43 . After ICI 182,780 treatment of LTED cells for 48 h, ER was absent (Fig. 5e). Under this condition, the enlarged Eleanor FISH signals became insensitive to the suppressive action of resveratrol (Fig. 5f). Moreover, we found that ER degradation by ICI 182,780 in LTED-RES cells resulted in de-repression of u-Eleanor transcripts (Fig. 4b, blue bars), indicating that resveratrol inhibits u-Eleanor expression via oestrogenic effects on ER. This result may be explained by the presence of multiple oestrogen-response elements in the u-Eleanor region ( Supplementary Fig. 5, marked with yellow). Because u-Eleanor is required for enhanced expression of intragenic Eleanor in LTED cells (Fig. 4f), it is possible that resveratrol represses u-Eleanor through ER, leading to a subsequent reduction of intragenic Eleanor. In addition, repression of u-Eleanor and Eleanors by resveratrol was abrogated by ICI 182,780, whereas ESR1 mRNA remained repressed under the same condition (Fig. 5g). Thus, Eleanors were expressed even under repression of the ESR1 gene, indicating that Eleanors do not simply represent nascent transcripts or by-products of ESR1 mRNA.
Resveratrol has been reported to activate SIRT1, a member of the sirtuin family of NAD þ -dependent deacetylases. Therefore, we examined whether SIRT1 is involved in the repressive effect of resveratrol on Eleanors. We depleted SIRT1 by siRNAs in LTED cells ( Supplementary Fig. 6a,b) Fig. 6c).
In addition, overexpression of SIRT1 in LTED cells did not change the FISH signals ( Supplementary Fig. 6d). Thus, the effect of resveratrol on the ESR1 locus is unlikely to depend on SIRT1.
A large chromatin domain is defined by a cluster of ncRNAs. An overview of the RNA-Seq data around the ESR1 gene revealed that the region associated with ncRNAs spanned B700 kb (6q25.1; 151,720,000-152,424,447; Fig. 6a). In addition to ESR1, this region includes three previously annotated genes, C6orf96, C6orf211 and C6orf97, all of which are co-regulated in breast cancer cells 44 . Alignment with the published ChIP-Seq data revealed that the region was heavily bound by RNA Pol II and enriched with an active histone mark, trimethyl-H3K36, in breast cancer cell lines, but not other cell types such as HeLa (Fig. 6a). The length of the region was well correlated with the recently proposed size for a single unit of a chromatin domain [19][20][21] .
To visualize characteristic RNAs originating from this large region, we performed RNA-FISH scanning using a series of BAC probes for subregions along the chromatin domain (Fig. 6a,b). We detected large RNA foci with probes 108N8, 404G5 and 130E4, which were similar to those detected with ESR1-BAC and ESR1-BAC2 probes (Figs 6b,c and 2b). It should be noted that 404G5, which corresponded to a region completely devoid of any protein-coding sequences, was able to detect the large RNA foci. In contrast, RNAs were not produced from outside of the domain (403M6 and 445H2). Using the combination of RNA-Seq and RNA-FISH, we found a novel chromatin domain including four co-regulated genes that were defined by induction of a ncRNA cluster during hormone deprivation (see Fig. 6d).
A set of long genes exhibit co-regulation of ncRNA and mRNA. Our genome-wide transcriptome analyses revealed changes in the mRNA expression of 2,918 genes under MCF7, LTED and LTED-RES conditions (interquartile range 410). They were classified into 14 distinct clusters on the basis of expression patterns ( Supplementary Fig. 7a). Among them, genes in cluster 4 (199 genes including ESR1) showed induction in LTED cells and repression in LTED-RES cells (Fig. 7a). Gene ontology analysis of cluster 4 showed enrichment of genes for the cellular metabolic process, transcriptional regulation, apoptosis and cell death, and cellular response to oestrogen stimulus (Fig. 7b).
We further characterized ncRNAs that were co-regulated with their neighbouring protein-coding genes. RNA-Seq analyses showed that B168,000 gene regions were significantly up-or down-regulated during LTED adaptation (MCF7 to LTED) and resveratrol treatment (LTED to LTED-RES, Fig. 7c). Interestingly,  67,68 . The other 13 genes (double asterisks) also showed up-or down-regulation of mRNAs in LTED and LTED-RES cells, respectively, while ncRNAs showed no co-regulation. The genes in these groups and their gene lengths are listed in Table 1. over 60% of these gene regions corresponded to potential ncRNAs, including exon-intron fusion, intron and intergenic sequences. We then extracted gene regions in which both mRNAs and ncRNAs were up-regulated during LTED adaptation and down-regulated by resveratrol treatment (Fig. 7d). Thirteen gene regions (indicated with a single asterisk in Fig. 7d and Table 1), including the ESR1 locus, exhibited such coordinate expression in LTED and LTED-RES cells (Fig. 7d and Supplementary Fig. 7b). We found that the genes in this group were generally long. Their average length was close to 300 kb (280,142 bp), while the average for genes in the overall genome is B60 kb (ref. 45). In the other 13 gene regions (indicated with double asterisks in Fig. 7d and Table 1), similar to the ESR1 locus, mRNAs were up-and downregulated in LTED and LTED-RES cells, respectively. However, they lacked coordinated regulation with neighbouring ncRNAs (Fig. 7d and Supplementary Fig. 7c). Their average length was B17 kb (16,839 bp).
Thus, our comprehensive transcriptome analyses revealed that coordinated regulation of ncRNA and mRNA, as exemplified by Eleanors and ESR1 mRNA, is common in a set of long genes. On the basis of recent studies [45][46][47] , ncRNA production from the entire gene locus may overcome physical transcription complications and maintain open chromatin.

Discussion
In this study, we found a novel type of ncRNA-mediated gene locus control in breast cancer adaptation. While cells undergo hormone-independent proliferation, the ESR1 gene is upregulated and ncRNAs are produced from a broad chromatin domain of B700 kb including ESR1 and other co-regulated genes. Eleanors originate from and around the ESR1 gene, and maintain the transcriptionally active locus. Up-regulation of ESR1 is important for the hormone-independent cell growth, which is suppressed by inhibition of Eleanors with either u-Eleanor knockdown or resveratrol. Eleanors are overexpressed in ER-positive breast cancers, suggesting that resveratrol and an inhibitor of u-Eleanor may be potential therapeutic agents for endocrine therapy resistance.
On the basis of our data, we propose a mechanistic model as illustrated in Fig. 6d. ESR1 mRNA is expressed at the basal level in MCF7 cells as they are ER positive (Fig. 6d, MCF7). During LTED adaptation, because of the loss of negative control by oestrogen-bound ER, there is significant induction of u-Eleanor expression, leading to coordinate up-regulation of intragenic Eleanor and ESR1 mRNA (Fig. 6d, LTED). Characteristic Eleanor-containing RNA foci are formed in the nuclei of LTED cells and subsequent resveratrol treatment represses both u-Eleanor and intragenic Eleanor by the oestrogenic actions on ER (Fig. 6d, LTED-RES).
The mechanism for induction of u-Eleanor in LTED cells is intriguing, and one possible mechanism is de-repression. In MCF7 cells, oestrogen-bound ER mildly inhibits the ESR1 locus by negative feedback 48 . On removal of oestrogen, ER becomes unliganded and fails to repress the ESR1 locus. By the addition of resveratrol to LTED cells, resveratrol-bound ER strongly represses the ESR1 locus, possibly because of its high concentration or structural properties.
Cancer cells can survive during various environmental changes by adjusting their global gene expression to acquire suitable phenotypes. Indeed, our transcriptome analyses of MCF7 and LTED cells showed that 2918 mRNAs were significantly up-or down-regulated under oestrogen deprivation ( Supplementary  Fig. 7a). These transcriptional changes were not limited to protein-coding regions (Fig. 7c, complete exon) and rather prevalent (B2-fold more frequent) in noncoding regions (Fig. 7c, exon-intron, intron and intergene). These results are in good agreement with the fact that o2% of the mammalian genome encodes proteins, whereas 75% of the genome is transcribed to produce ncRNAs that may modulate chromatin structure and gene expression 26,49 .
For the following reasons, Eleanors identified in this study are unique. First, transcription of Eleanors is inducible. They accumulate during oestrogen deprivation, before ESR1 mRNA up-regulation, and are abruptly suppressed by resveratrol. Furthermore, genome-wide analyses revealed that the ESR1 locus is one of the 13 genetic loci where ncRNAs and mRNAs are co-regulated during the hormonal changes.
Second, intragenic Eleanors are transcribed from a broad region covering the entire ESR1 gene body, which spans up to 300 kb. In addition, u-Eleanor is derived from an enhancer-like element at B40 kb upstream of the ESR1 gene. u-Eleanor is responsible for up-regulation of downstream transcripts including ESR1 mRNA and intragenic Eleanors (Fig. 4d,f). To date, there are few reports of simultaneous transcription of ncRNAs from upstream and the gene body. One related but distinct example is the b-globin locus where ncRNAs are produced from LCR, a central upstream regulatory element, and downstream intergenic regions in the locus 50 .
Third, Eleanors cover their own transcription sites to regulate gene expression and form a so-called 'RNA cloud' in the nucleus (Fig. 2c). This property may be shared with other chromosomal RNAs including XACT, Air, C 0 T-1 repeat RNAs and snoRNAs [51][52][53][54] . XIST RNA also forms a large nuclear domain by coating the entire inactive X chromosome and plays a role in X chromosome inactivation 55 . Eleanors may be functional introns that are stably maintained in the nucleus [56][57][58] or represent pervasive transcription of the genome 59 , which are enhanced under oestrogen deprivation. Eleanors were unusually stable because of resistance to the denaturation procedure that is normally performed only in DNA FISH and degrades most RNAs. Because the Eleanor FISH signals were clearer with denaturation (Fig. 2b, top panels), Eleanors may be tethered to the sites of transcription by RNA-DNA hybrid formations.
We found that coordinate expression of ncRNA and mRNA is conserved in a set of long genes (Fig. 7, Table 1). Accumulation of ncRNA at the site of transcription may counteract length-  Fig. 7d). w Genes that showed up-and down-regulation of mRNAs in LTED and LTED-RES cells, respectively, while ncRNAs showed no co-regulation (denoted with double asterisks in Fig. 7d).
NATURE COMMUNICATIONS | DOI: 10.1038/ncomms7966 ARTICLE dependent impairment of gene transcription that was reported recently 45,47 . We and others have previously reported that a population of ER-positive breast cancer cells has abnormally large ESR1 FISH signals [60][61][62] . Considering that Eleanors are extremely stable, it is possible that some of the previously detected large FISH signals in breast cancer patients include Eleanors as shown in Fig. 3. In addition, several molecular events have been reported during LTED adaptation, including up-regulation of ER, ERBB2, c-Myb, c-Myc and MAP kinases, and activation of the PI-3 kinase pathway, NOTCH pathway, growth factor pathways related to mTOR and EGFR/ERBB/AKT, as well as changes in the phosphorylation pattern of ER 7,8,10,31,63 . It would be interesting to determine how Eleanors are integrated in these events in LTED cells and ER-positive breast cancer cells.
In summary, Eleanors, a novel type of ncRNA, are actively involved in the epigenetic adaptation of ER-positive breast cancer cells by activating transcription of the ESR1 gene locus. These findings highlight ncRNA-mediated mechanisms in cancer cell adaptation, which may be diagnostic and therapeutic targets for endocrine therapy-resistant breast cancer.

Methods
Cell culture. MCF7 cells (ATCC) were cultured in RPMI 1640 (Sigma) supplemented with 10% fetal bovine serum (FBS). For LTED, MCF7 cells were grown in phenol red-free RPMI 1640 (Wako) containing 4% dextran-coated charcoalstripped FBS for 2-4 months. For LTED-RES, LTED cells were treated with 50 or 100 mM resveratrol (Sigma Aldrich, R5010) for 24 h. For ICI 782,780 treatment, cells were cultured with 100 nM ICI 782,780 (Tocris, 1047) for 48 h. Human mammary epithelial cells 64 (Lonza) were cultured in mammary epithelial growth media (Lonza) according to manufacturer's protocol at 37°C with 5% CO 2 . Primary invasive breast carcinoma specimens were obtained by surgical excision from patients at the Department of Breast and Endocrine Surgery, Kumamoto University Hospital (Kumamoto, Japan).Informed written consent was obtained from all the patients before surgery. The study protocol was approved by the Ethics Committee of Kumamoto University Graduate School of Medicine (Kumamoto, Japan). Preparation of mRNA-Seq libraries. Total RNA was extracted from cultured cells with an RNeasy Mini Kit (Qiagen). mRNA-Seq libraries were generated using an mRNA-Seq Sample Preparation Kit (Illumina) according to manufacturer's protocol with minor modifications. Poly(A) þ RNA was enriched from 1 mg total RNA by two successive rounds of oligo(dT) selection, fragmented and then used for first-strand cDNA synthesis by random hexamer priming. After second-strand cDNA synthesis, double-stranded DNA was repaired using T4 DNA polymerase, Klenow enzyme and T4 polynucleotide kinase (New England Biolabs), and then treated with Klenow exo-to add an adenine to the 3 0 end. After ligation of the Index PE Adaptor oligo mix (Illumina) using Takara ligation mix (Takara), the adaptor-ligated DNA was amplified using primers, InPE2.0/1.0 and index (Illumina), by 18 cycles of PCR. The amplified libraries were isolated from an agarose gel. The DNA samples were purified using a QIAquick MinElute Kit (Qiagen) at each preparation step.
Preparation of RNA-Seq libraries. Total RNA was extracted from cultured cells with an RNeasy Mini Kit and then ribosomal RNA was removed by a Ribo-Zero Gold Kit (Illumina). RNA-Seq libraries were generated using a ScriptSeq v2 RNA-Seq Library Preparation Kit (Illumina), according to manufacturer's instructions.
mRNA-Seq and RNA-Seq analyses. An Illumina Genome Analyzer IIx was used to generate 41-base single-end reads for mRNA-Seq and an Illumina HiSeq 1000 was used to generate 51 base (stranded) single-end reads for RNA-Seq. The sequence reads were aligned with the human reference genome (UCSC hg19, http://genome.ucsc.edu/) using Tophat (v1. 4 Clustering of sequence reads and gene ontology. Data for mRNA-Seq experiments were analysed using CLC Genomics Workbench ver 5.51 (CLC Bio). Results obtained from mRNA-Seq experiments were selected on the basis of the following parameters: degree of variance in gene expression (inter quantile range 410) in two samples among MCF7, LTED and LTED-RES cells, and sufficient abundance of mRNA (FPKM 45). A total of 2,918 genes were chosen and used for downstream analyses including k-means clustering (Mev (http://www/tm4.org/men/)). In addition, cluster 4 (199 genes) obtained by k-means clustering was used for gene ontology analysis that was performed using the Genomatix Pathway System in the Genomatix Genome Analyzer. The distribution of reads in the RNA-Seq data set was determined by counting the number of reads in a 100-bp sliding window by the NGS Analyzer of the Genomatix Genome Analyzer and selected on the basis of the following parameters: differentially expressed transcripts in two different cell states (MCF7 and LTED, or LTED and LTED-RES) by fold-change ranking (log2 fold-change o À 1.0, down-regulation; 41.0, up-regulation) together with P-values (Po0.01) computing by method described in chapter 'Testing for differential expression' in DEseq algorithm version 1.0.6 (ref. 67) in the NGS Analyzer of the Genomatix Genome Analyzer. This analysis identified 168,389 and 168,749 genetic regions that were differentially expressed in LTED cells compared with that in MCF7 cells, and LTED-RES cells compared with that in LTED cells, respectively. Furthermore, each region was subdivided by ElDorado database genome annotation version 08-2011.
Combined expression analysis of mRNA and ncRNA. Genetic regions corresponding to mapped read clusters in exon-intron junctions, introns and intergenes were detected using the NGS Analyzer and defined as ncRNA candidates (Fig. 7c). Moreover, read clusters with different levels between LTED and MCF7 cells, and LTED-RES and LTED cells were extracted using edgeR1.6.15 in the Genomatix Genome Analyzer (Fig. 7d). Genetic regions that were up-regulated in LTED cells compared with that in MCF7 cells were selected on the basis of fold-change ranking (log2 fold-change o À 1.0) together with P-values (Po0.01), which identified 388 mRNAs and 10,808 ncRNAs (Genomatix Software GmbH). P-values were calculated by edgeR algorithm version 1.6.15 (ref. 68). Similar selections were performed for down-regulated mRNAs and ncRNAs in LTED-RES cells compared with that in LTED cells, resulted in 359 mRNAs and 933 ncRNAs. Overlapping among the four groups was analysed and summarized to show the number of shared genes in a four-way Venn diagram (Genomatix Software GmbH).
Nonspecific enrichments detected in each sample were subtracted from the local enrichments (clusters) representing genomic regions bound in comparative samples.
ChIP-Seq data used in this study. ChIP-Seq data of MCF7 cells were obtained from ENCODE Consortium through UCSC genome browser (University of California, Santa Cruz), and compared with our mRNA-Seq and RNA-Seq data (Figs 4a and 6a and Supplementary Figs 3a and 4a). The files used in the analyses are listed in Supplementary Table 1.
Fluorescence in situ hybridization. Cells grown on coverslips were fixed with 4% formaldehyde and 0.5% Triton X-100 in phosphate-buffered saline (PBS) for 15 min, and then permeabilized with 0.5% saponin and 0.5% Triton X-100 in PBS for 20 min. Samples were immersed in 20% glycerol in PBS for 30 min, and then subjected to four cycles of freeze-thawing by freezing the cells in liquid nitrogen for 30 s each time. The cells were then treated with 0.1 N HCl for 15 min. For denaturation and hybridization, the cells were incubated in hybridization mixtures (2 Â SSC, 50% formamide, 10% dextran sulfate, 1 mg ml À 1 tRNA and 5-10 mg ml À 1 probe DNA) at 75°C for 4-10 min and then 37°C for 48-72 h. BAC and plasmid probes were labelled with biotin or digoxigenin in a nick translation mixture (Roche) according to the manufacturer's protocol. After hybridization, the cells were washed with 2 Â SSC and 50% formamide at 37°C for 5 min, followed by 2 Â SSC at 37°C for 5 min. FISH signals were detected with FITC-anti-digoxigenin (Roche; dilution 1:250) or Cy3-streptavidin (Jackson ImmunoResearch; dilution 1:1,000). For dual-colour FISH, a SPEC ESR1/CEN6 Dual Color Probe kit (ZytoVision) was used. For nuclease treatment, cells were pre-treated with 0.5 U ml À 1 DNase I (Promega) or 1 mg ml À 1 RNase A (Roche) for 30 min before hybridizations.
For FISH with tissue microarray (BioChain, Z7020005 and Biomax, BR1504), sections were processed using ZytoLight SPEC ESR1/CEN 6 Dual Color Probe Kit (Zytovision, Bremerhaven, Z-2070-20) according to manufacturer's protocol. Briefly, slides were heated for 10 min at 70°C, treated with xylene twice for 10 min at room temperature, rehydrated in a series of graded ethanol solutions, washed with dH 2 O, incubated in Pretreatment Solution Citric at 98°C for 15 min and then washed again with dH 2 O. The slides were then treated with pepsin solution for 10 min at 37°C, and washed with 2 Â SSC solution for 5 min and with dH2O for 1 min at room temperature. The slides were dehydrated, air-dried and incubated with a probe at 76°C for 10 min for denaturation, and at 37°C overnight for hybridization. DNA was counterstained with 5,6-diamidino-2-phenylindole (DAPI).
Immunofluorescence and immunohistochemistry. For immunofluorescence, cells were fixed with 4% paraformaldehyde in PBS for 15 min at room temperature. The cells were washed three times with PBS for 5 min, then permeabilized with PBS containing 0.2% Triton X-100 and 0.5% normal goat serum (GIBCO) for 5 min on ice. The cells were incubated at room temperature with PBS with 0.5% normal goat serum (GIBCO) three times for 5 min, then with the primary antibodies for 60 min, followed by the Cy3-or Alexa488-conjugated secondary antibodies for 60 min. The cells were washed with PBS three times for 10 min each. DNA was counterstained with 1 mg ml À 1 DAPI.
For immunohistochemical staining, sections in a tissue microarray (Biomax, BR1504) were processed with automated IHC staining system, BenchMark XT (Ventana). The sections were de-paraffinized in EZ prep (Ventana) at 72°C for 4 min and then incubated in Immunoblock (DS Pharma Biomedicals) at 37°C for 12 min to block endogenous peroxidase. Antigens were retrieved by incubating 95°C for 8 min, then treated with rabbit monoclonal anti-human ERa clone SP1 (Ventana, 790-4325, used without dilution) at 37°C for 36 min. The antigens were visualized by avidin-biotin-peroxidase complex method using iVIEW DAB detection kit (Ventana). The slides were then counterstained with haematoxylin at 37°C for 8 min.
Microscopy and image analysis. Images were obtained with an IX-71 microscope (Olympus) equipped with a Â 60 NA1.0 Plan Apo objective lens, a cooled CCD camera (Hamamatsu) and image acquisition software (Lumina Vision Version 2.4; Mitani Corporation). For imaging cytometry analyses of FISH, image stacks of three-dimensional data sets were collected at 0.5-1.0 mm intervals through the z axis, subjected to projections and used for automatic detection and counting of FISH signals with CELAVIEW RS100 software (Olympus) and Cellomics CellInsight (Thermo Fisher Scientific). Images in Figs 1c and 2c and Supplementary  Fig. 1c were obtained with a confocal laser-scanning microscope (LSM 780, Carl Zeiss) equipped with Â 63/1.4 Plan-Apochrome objective lens and a cooled CCD camera (Carl Zeiss). Image acquisition was done using LSM software (Carl Zeiss).
Immunoblotting. To prepare total cell lysate, cells were dissolved in SDS sample buffer containing benzonase (Sigma). Proteins were separated by SDSpolyacrylamide gel electrophoresis and then transferred to a nitrocellulose membrane, Amersham Hybond ECL (GE Healthcare). The membrane was blocked for 1 h with PBS containing 10% nonfat dry milk and then incubated with primary antibodies in PBS containing 0.03% Tween 20 for 1 h. The membrane was washed with PBS containing 0.3% Tween 20 three times for 10 min each, and incubated with horseradish peroxidase-conjugated secondary antibodies for 40 min. After the membrane was washed with PBS containing 0.3% Tween 20 three times for 10 min each, signals were visualized with Western Lightning Plus-ECL (PerkinElmer). Uncropped immunoblot images for Figs 4e and 5b, and Supplementary Fig. 6a are provided in Supplementary Figs 8 and 9, respectively. PCR with reverse transcription. Total RNA was isolated from cultured cells with TRIzol (Invitrogen) and then treated with DNase I (Roche) before cDNA synthesis. Reverse transcription was carried out with a High Capacity cDNA reverse transcription kit (Applied Biosystems). qRT-PCR was performed with SYBR green fluorescence on an ABI Prism 7300 system (Applied Biosystems). Values were normalized to b-actin or glyceraldehyde-3-phosphate dehydrogenase (GAPDH) gene expression before calculating relative fold changes. The amplification efficiency for each primer set was measured using genomic DNA and used for normalization in Figs 2a and 4b. To indicate absolute expression levels of each RNA in Figs 2a,e and 4b, DC t values are summarized in Supplementary Table 6. Primer sequences are listed in Supplementary Table 3. For RT-PCR, amplified products were separated by electrophoresis on 2% agarose gels and stained with ethidium bromide. All PCR amplifications were performed within a quantitative range by adjusting the cycle numbers. For RT-PCR in Supplementary Fig. 3b, PCR-amplified products were run on agarose gel and stained with ethidium bromide. Corresponding uncropped gel images are provided in Supplementary  Fig. 9.
Chromatin immunoprecipitation-qPCR. MCF7 and LTED cells were crosslinked with 1% formaldehyde at 37°C for 10 min. Crude cell lysates were sonicated to generate DNA fragments of 200-500 bp using a Bioruptor USD-250 (Cosmo-Bio; 15 sonications of 30 s with 30 s intervals at 250 W). Chromatin was precipitated with antibodies at 4°C overnight, washed and de-crosslinked for 5 h. DNA enrichment in ChIP samples was determined by qPCR with SYBR green fluorescence on an ABI Prism 7300 system. The threshold was set to cross a point where PCR amplification was linear, and the cycle number required to reach the threshold was recorded and analysed using Microsoft Excel. PCR was performed using precipitated DNA and the input DNA. Primers used in ChIP-qPCR are listed in Supplementary Table 3. RNA interference. Cells were transfected with specific siRNAs (Nippon EGT) using RNAiMAX (Invitrogen). Target sequences for each siRNA are listed in Supplementary Table 3. The cells were analysed at 48 and 72 h after transfection.
Cell counting. The number of proliferating cells was counted using an automatic cell imaging counter (CYTORECON; GE Healthcare) or image cytometer (Cellomics CellInsight; Thermo Fisher Scientific).
3C assay. Formaldehyde-crosslinked chromatin from MCF7 and LTED cells was digested with BglII restriction enzyme overnight, followed by ligation with T4 DNA ligase at 16°C for 4 h. To prepare control templates for standard curves, a BAC clone covering the ESR1 locus (RP11-450E24 and RP3-404G5) was digested with BglII, followed by random religation. After reversing the cross-links, genomic DNA was purified by phenol extraction and ethanol precipitation. The ligated products were assessed by qPCR with the ABI Prism 7300 and Thunderbird SYBR qPCR Mix (Toyobo). The efficiency of BglII digestion was evaluated by qPCR using primers that only amplified undigested DNA fragments containing the BglII site. More than 80% of the individual restriction sites were digested under the experimental condition. 3C-qPCR data were normalized to a loading control that was obtained with primers that amplified a genomic fragment in the ESR1 locus (intron 3), which represented the amount of template DNA. Primers used in the 3C assay are listed in Supplementary Table 3.
Overexpression of SIRT1 in LTED cells. Full-length cDNA for human SIRT1 was amplified by PCR using primers described in Supplementary Table 3. The amplified fragments were digested with EcoRV and XbaI, and then cloned into pcDNA3-myc (pcDNA3-myc-SIRT1). LTED cells (1 Â 10 5 cells) were transfected with the plasmid (2 mg) using 5 ml FuGene HD (Roche Applied Science) in a sixwell plate for 48 h, and then subjected to immunofluorescence or immuno-FISH.
Statistical analysis. Comparisons between groups were analysed using the two-tailed Student's t-test. A value of Po0.05 was considered statistically significant. For RNA-and mRNA-Seq experiments, P-values were calculated with DESeq and edgeR 67,68 .