Differentiation of mammary tumors and reduction in metastasis upon Malat1 lncRNA loss

Mammary tumor progression is impaired upon genetic loss of Malat1 lncRNA

Human luminal B breast cancer is one of the most prevalent subtypes of breast cancer, affecting nearly 15%–20% of individuals diagnosed with this disease (for review, see Schnitt 2010; Ades et al. 2014). In order to investigate the potential in vivo role of Malat1 in mammary tumor progression, we used the MMTV-PyMT mouse mammary tumor model of human luminal B breast cancer (Guy et al. 1992; Lin et al. 2003; Herschkowitz et al V体育官网. 2007). In this model, the polyoma middle-T antigen is under the control of the MMTV promoter, resulting in highly penetrant mammary tumors and lung metastasis (Lin et al. 2003).

Malat1 homozygous knockout (Malat1^−/−) female mice previously generated in our laboratory (Zhang et al. 2012) were crossed to MMTV-PyMT males (C57Bl/6) to obtain MMTV-PyMT;Malat1^+/+, PyMT;Malat1^+/−, and PyMT;Malat1^−/− mice. Individuals of all three genotypes were born in expected Mendelian ratios with a normal gestation period and postnatal development. In accord with published data (Lin et al. 2003), tumor onset in MMTV-PyMT mice began soon after puberty, and palpable tumors developed within 12–16 wk of birth in >90% of the mice. Tumor onset did not differ significantly in the presence or absence of Malat1 (Fig. 1A). Prior to 5 mo of age, nearly 90% of the mice developed palpable multifocal mammary gland lesions, demonstrating that loss of Malat1 does not affect the initiation of PyMT-induced mammary gland tumors. Although the tumors that developed in the Malat1^−/− background initially exhibited a relatively slower tumor growth rate compared with control mice (Fig. 1B), this difference was not found to be significant, especially at the later time points. Surprisingly, the MMTV-PyMT;Malat1^−/− mice developed highly cystic liquid-filled tumors (Fig. 1C, right panel), in contrast to the control mice, which developed poorly differentiated solid carcinomas (Fig. 1C, left panel). Hematoxylin and eosin (H&E) staining of tumor sections from MMTV-PyMT;Malat1^−/− mice demonstrated that nearly 90% of the tumors that were >1 cm in diameter exhibited a cystic phenotype, with >75% of the tumor area showing encapsulated well-differentiated cystic histopathology (Fig. 1D, bottom panels), in stark contrast to the poorly differentiated solid carcinomas observed in the control tumors (Fig. 1D, top panels).

In the MMTV-PyMT model, nearly 80% of the mice develop lung micro- and macrometastases. Therefore, we examined the impact of Malat1 loss on lung metastases. MMTV-PyMT;Malat1^−/− mice exhibited a significant reduction in lung macrometastases, where only three of 12 mice had macro lesions as compared with seven of eight mice in control MMTV-PyMT;Malat1^+/+ mice (Fig. 1E,F). The lungs were further examined for the incidence of micrometastases (Fig. 1G), and the majority of MMTV-PyMT;Malat1^−/− animals did not show any micrometastasis (nine of 12). Next, we calculated the total metastatic burden as a percentage of lung volume (Fig. 1H) and found that the MMTV-PyMT;Malat1^−/− animals showed a significantly decreased metastatic burden (3%–5%) as compared with the control MMTV-PyMT;Malat1^+/+ mice, for which the metastatic burden reached up to 40% of the total lung volume. These results clearly demonstrate a role for the lncRNA Malat1 in mammary tumor progression.

MALAT1 is up-regulated in human breast cancer metastases

To understand the significance of MALAT1 in human breast cancer, we evaluated its expression in human breast tumors by RNA-FISH on matched primary tumor and metastases tissue sections from the same individuals. MALAT1 was generally expressed at a low level in the primary lesions (Fig. 2A, top panels) of breast tumors irrespective of the ER/PR status. However, very high levels of MALAT1 were observed in the lung metastases from the same individuals (Fig. 2A, bottom panels). Interestingly, higher levels of MALAT1 expression were confined only to the tumor cells, while the adjacent stromal cells showed a very weak signal for MALAT1 in both the primary tumor and the metastatic lesions (Supplemental Fig. S1A). Some primary tumors showed elevated MALAT1 expression in only a small fraction of the cancer cells or cell clusters, while neighboring cancer cells remained very low in MALAT1 expression, suggesting that these high MALAT1-expressing cancer cells may escape the primary tumors preferentially to establish metastasis or that potentially microenvironmental cues may influence MALAT1 expression levels (Supplemental Fig. S1B,C). To determine the extent of occurrence of high MALAT1 expression in human samples, we obtained a breast cancer tissue array and lung metastasis tissue array and evaluated MALAT1 expression level using RNA-FISH. MALAT1 transcript levels were increased between 1.5-fold and threefold in >60% of the lung metastases compared with the primary tumor sections (Fig. 2B). The lung metastasis array was generated from primary breast cancer patients. While only some of the primary tumor samples and lung metastases were matched, we could detect high MALAT1 expression in the majority of the lung metastasis samples, confirming our matched sample data. High MALAT1 expression was also observed in sections of human breast cancer brain metastatic lesions (Supplemental Fig. S1D). Interestingly, the MMTV-PyMT tumors also displayed a similar expression pattern for Malat1, where higher Malat1 expression was confined to a small fraction of cells (Supplemental Fig. S2A–E). Furthermore, more higher Malat1-expressing cells were found in advanced carcinomas compared with adenomas or hyperplasia (Supplemental Fig. S2A–E). This observation suggests that MALAT1 likely plays a crucial role in human breast cancer pathogenesis, prompting us to further evaluate its potential as a therapeutic target.

Knockdown of V体育ios版 - Malat1 results in mammary tumor differentiation

Next, we were interested in determining whether we could recapitulate the phenotype observed by genetic knockout upon systemically knocking down Malat1 levels in established tumors in a more therapeutic setting. Since Malat1 is a nuclear retained lncRNA, we used ASOs to achieve knockdown. Malat1-specific ASOs (16mers) were synthesized using a next-generation design comprised of phosphorothioate-modified short S-cEt (S-2′-O-Et-2′,4′-bridged nucleic acid) gapmer chemistry (Teplova et al. 1999). We used two independent Malat1-specific ASOs targeting two different regions of the Malat1 transcript as well as a control scrambled ASO (ScASO).

Three-month-old to 4-mo-old mice were divided into three cohorts (seven to 12 mice each) and randomized based on their age, tumor size, and number of tumors. Each mouse in the cohort received either ScASO or Malat1-specific ASO1 or ASO2 via subcutaneous injections of 25 mg/kg per day (125 mg/kg per week) for 5 d with a rest period of 2 d (Fig. 2C). The injections were carried out for a period of 8 wk, upon which at least one tumor from most of the control mice reached 2 cm in size. During the course of the treatment, tumors were measured twice per week. At the end of the treatment period, the animals were euthanized, and the primary tumors and lungs were collected. An average Malat1 knockdown of ∼60% was achieved in the mammary tumor tissue in mice treated with either ASO1 or ASO2 as compared with ScASO-treated control mice (Fig. 2D). Knockdown efficiency was also confirmed by RNA-FISH on tumor sections (Fig. 2E). Different regions of the treated tumors showed varying knockdown efficiencies ranging from 20% to 80% knockdown (data not shown).

Interestingly, when tumor growth was monitored over the course of treatment, the Malat1 ASO-treated group showed a 50% slower tumor growth rate than tumors treated with ScASO (Fig. 2F). The decrease in growth rate was observed as early as week 3 into the treatment period, and, by the end of the treatment, the tumor volume was reduced by 50% in Malat1 ASO-treated mice compared with the ScASO-treated group. Consistent with the slower tumor growth rate, Ki-67 labeling of the tumor sections showed a marked decrease in Ki-67-positive nuclei. The proliferative index of Malat1 ASO1- or ASO2-treated tumors, calculated as a measure of Ki-67 positivity, was found to be 22% and 26%, respectively, compared with 51% in the control tumors treated with ScASO (Fig. 2G). Although this was in contrast to the knockout experiment, where there was no significant difference in primary tumor growth rate, we reason that this was due to the fact that the Malat1 knockout tumors were extensively cystic, and, when we measured the tumor volume, it is likely that the cyst growth contributed significantly to the total volume rather than actual tumor mass.

The primary tumors from each group were analyzed for histopathological changes. Notably, the Malat1 ASO-treated tumors were cystic, liquid-filled, encapsulated, well-differentiated primary tumors closely resembling those from the MMTV-PyMT;Malat1^−/− animals, whereas the tumors in the ScASO-treated control animals remained poorly differentiated, as in tumors from MMTV-PyMT;Malat1^+/+ mice (Fig. 3A). Also, similar to the knockout phenotype, a single layer of epithelial cells lined the enlarged ducts in the Malat1 ASO-treated mice. There was a significant increase in cystic/ductular tumors (31% in ASO1-treated and 22% in ASO2-treated) (Fig. 3B), whereas only 4% of the ScASO-treated mice developed cystic tumors similar in morphology to the Malat1 ASO-treated groups. Furthermore, there was a twofold decrease in the number of solid carcinomas, characterized by poorly differentiated cells, in Malat1 ASO-treated mice. Quantitation of the cystic area of Malat1 ASO-treated tumors showed that, on average, 40% of the total tumor area contained cystic/ductular histopathology (Fig. 3C).

Given that Malat1 ASO-treated tumors display a remarkable histological differentiation phenotype, we assayed for differentiation markers by immunofluorescence on sections of primary tumors. E-cadherin, a major cell adhesion protein present on the cell membrane of differentiated epithelial cells, exhibited increased labeling on the cell membrane of Malat1 ASO-treated tumors (Fig. 3D, bottom panels), whereas the control tumors showed lower levels to undetectable levels of E-cadherin in several regions within the tumors of the control mice (Fig. 3D, top panels). Surprisingly, intense β-casein localization was observed in the enlarged ducts of Malat1 ASO-treated tumors (Fig. 3E, bottom panels), consistent with the differentiation phenotype. In contrast, β-casein was absent in the control tumor sections (Fig. 3E, top panels).

Malat1 ASO treatment significantly reduces lung metastasis

Given the reduction of metastasis seen in the MMTV-PyMT;Malat1^−/− animals (Fig. 1E–H), it was important to determine whether Malat1 knockdown could also result in a reduction in the incidence of lung metastasis in this therapeutic setting. Indeed, Malat1 ASO treatment resulted in a significant reduction in lung metastatic incidence (Fig. 4A). A small fraction of Malat1 ASO-treated animals (six of 18) developed macrometastatic nodules as compared with the control mice (10 of 12) (Fig. 4B). Furthermore, the lungs were serial-sectioned to determine the incidence of micrometastases and obtain the total metastatic burden for a given lung volume. Consistent with data obtained from genetic loss of Malat1 (Fig. 1E–H), the Malat1 ASO-treated group showed a significantly reduced metastatic burden (<10%) (Fig. 4C,D) compared with the larger metastatic burden (>30%) observed in the ScASO-treated group. Micrometastases were also not observed in lungs that were free of macrometastatic nodules in mice treated with Malat1 ASOs. The livers and brains from the Malat1 ASO-treated groups and control groups were also analyzed for metastatic incidence, and no evidence of macrometastatic lesions was found in those organs (data not shown). Thus, we conclude that Malat1 loss impairs the metastatic progression of the disease. As such Malat1 ASOs represent a potent therapeutic approach for metastatic breast cancer.

Malat1 loss affects branching morphogenesis in tumor-derived organoids

Three-dimensional (3D) organotypic culture systems allow studies of physiologically relevant cellular processes as well as disease states under defined culture conditions (Ewald et al. 2008; for review, see Sato and Clevers 2013; Sachs and Clevers 2014). Mammary gland organoids preserve the gland characteristics and undergo various morphogenetic changes upon growth factor stimulus that recapitulate normal physiological processes such as branching morphogenesis (Ewald et al. 2008; Cheung et al. 2013). To investigate the role of Malat1 in these processes, organoids were prepared from MMTV-PyMT tumors <1 cm in size. These tumor organoids were grown on Matrigel for up to 6 d in the presence or absence of Malat1 ASOs, allowing us to track growth, dynamics, and branching morphogenesis of the organoids (Fig. 5A).

The Malat1 ASOs were taken up “freely” by the organoids without the use of transfection reagents, and an ∼90% knockdown of the Malat1 transcript was achieved with either Malat1 ASO relative to the ScASO control (Fig. 5B). After 6 d in culture, mock-treated organoids or ScASO-treated organoids underwent extensive branching morphogenesis (Fig. 5C [top panels], D). In contrast, organoids treated with Malat1 ASO1 or ASO2 failed to undergo branching morphogenesis (Fig. 5C [bottom panels], D) and remained as spherical acini that were smaller than the branched control organoids. The loss of branching was found to be specific for ASOs targeting Malat1, as knockdown of Neat1, another nuclear lncRNA encoded in the same chromosomal region as Malat1, resulted in organoids with branching morphogenesis similar to control PyMT tumor organoids (Supplemental Fig. S3A). Furthermore, detailed time-lapse differential interference contrast (DIC) imaging showed that control organoids underwent active remodeling throughout the imaging window (4 d), displaying extensive movement within the Matrigel (Supplemental Movie S1). The mock-treated or ScASO-treated organoids expanded rapidly, forming lumens, and, as early as 72 h, initiated the branching process (Supplemental Movie S1). Numerous cells migrated within the organoids, and these cells could be followed using organoids developed from MMTV-PyMT;CAG::H2B-GFP bitransgenic mice treated with ScASO (Supplemental Movie S3). In contrast, Malat1 ASO1- or ASO2-treated organoids remained in the same position over the 84-h time course of imaging and did not show collective cell migration within the Matrigel (Supplemental Movie S2). Most importantly, cells within individual Malat1 ASO-treated organoids also lacked cellular motility, and the organoids appeared highly compact (Supplemental Movie S4). Interestingly, organoids treated with Malat1 ASOs showed increased apoptotic cell death and a significant reduction in the number of dividing cells (Supplemental Fig. S3B).

To further characterize the distinct morphological change induced upon Malat1 ASO treatment, transmission electron microscopic analysis was performed on the organoids to evaluate ultrastructural differences (Fig. 5E; Supplemental Fig. S3C). Consistent with the differentiation phenotype observed in the tumors treated with Malat1 ASOs, the organoids also showed well-differentiated subcellular structures, such as the presence of secretory lipid droplets and dense core granules that were absent in the control organoids treated with ScASO or PBS. Importantly, Malat1 ASO-treated organoids exhibited tight cell–cell contacts with little to no intercellular spaces. An increase in cell–cell junctional complexes such as desmosomes was observed in Malat1 ASO-treated organoids compared with the control branched organoids (Supplemental Fig. S3C). Consistent with this observation, desmosomes have previously been shown to negatively regulate branching morphogenesis (Basham et al. 2013). The cells within the Malat1 ASO-treated organoids were aligned adjacent to each other in a polarized manner with intercellular ducts that are present in normal mammary glands; these ducts are one of the distinct features of differentiated epithelial cells (Zhan et al. 2008; Watson and Campbell 2009). The ScASO- or mock-treated PyMT organoids lacked polarity, and the vast majority of the organoids contained Matrix proteins within the organoids, which reflected the active extracellular matrix (ECM) remodeling necessary for migration (Supplemental Fig. S3C). Last, organoids derived from MMTV-PyMT;Malat1^−/− tumors did not undergo branching morphogenesis and remained as spherical tightly packed acini (Fig. 5F), whereas the MMTV-PyMT;Malat1^+/+ organoids were extensively branched. The loss of branching phenotype observed in the MMTV-PyMT;Malat1^−/− organoids could be rescued by full-length mouse Malat1 that was nucleofected into those organoids (Supplemental Fig. S4A–D). We observed ∼30% branching in MMTV-PyMT;Malat1^−/− organoids that were nucleofected with full-length mouse Malat1. In contrast, 5% of organoids displayed branching in mock transfected or GFP transfected organoids (Supplemental Fig. S4D). The nucleofection efficiency ranged between 35% and 45% (data not shown). The effect of Malat1 knockdown was also studied in organoids derived from MMTV-cre;Flox-neo-neuNT mice, a model that phenocopies the HER2-amplified subtype of human breast cancer. Here, the first exon of endogenous neu (mouse homolog of HER2/ERBB2) has been replaced with Cre-inducible activated neu-cDNA, resulting in constitutively active Neu protein transcribed from its endogenous promoter in mammary epithelial cells (Andrechek et al. 2000). Control organoids derived from this model underwent branching morphogenesis, while the Malat1 ASO-treated organoids failed to undergo branching morphogenesis, resulting in the formation of tightly packed spherical acinar structures (Fig. 5G–I).

Malat1 knockdown in tumors disrupts protumorigenic signaling pathways

Next, we sought to investigate the gene expression changes that are triggered upon Malat1 knockdown in both tumor organoids and primary tumors by RNA-seq. RNA was isolated from day 6 organoids that were either untreated mock control or treated with ScASO or Malat1 ASO1 or ASO2. We used tumors from four MMTV-PyMT mice to prepare organoids that were subjected to different ASO intervention. The differentially expressed genes in Malat1 ASO1- or ASO2-treated organoids or tumor sections were compared with ScASO- or mock-treated controls. When we subjected the entire organoid data set to pathway analysis using the Kyoto Encyclopedia of Genes and Genomes (KEGG), several tumorigenic signaling pathways were enriched. Interestingly, enrichment of genes belonging to the ECM receptor, focal adhesion, and cell adhesion pathways was observed (false discovery rate [FDR] < 0.1) (Supplemental Table S1). These pathway changes are highly consistent with the phenotypic changes observed in the organoids upon Malat1 ASO treatment; i.e., loss of cellular migration and increased adhesion observed by the loss of branching morphogenesis.

RNA-seq analysis was also carried out from tumors derived from Malat1 ASO1- or ASO2-treated mice that were compared with tumors from ScASO-treated mice, and 52 genes were affected upon Malat1 ASO treatment, with a significance threshold of FDR < 0.1 (Supplemental Fig. S5A). Additionally, 478 genes were found to be differentially expressed; these genes were filtered based on fold change cutoff from both ASO1- and ASO2-treated tumors and exhibited a significant change in one or the other treated tumors (Supplemental Table S2). The sample correlation plot also reflected the heterogeneity between the tumors. Nevertheless, the top 10 up-regulated and down-regulated genes from the tumors were indeed very interesting candidates (Fig. 6A,B). Of the up-regulated genes, the casein gene family was significantly enriched. This included Csn2 that encodes for the β-casein protein, the major milk protein, and Csn1s2a, a member of the α-casein family. Additionally, Wap (whey acidic protein), Lao, and Cel, which are also constituents of mammalian milk, were also up-regulated in many of the tumors. In light of the significant differentiation phenotype that was observed in the Malat1 ASO-treated tumors, it was reassuring to observe that these genes were also up-regulated >20-fold in these tumors. Additionally, up-regulation of the Pip gene is a confirmation of the cystic phenotype, as Pip (also known as gross mammary gland cystic fluid) is a protein that is secreted in cystic mammary gland lesions (Collette et al. 1986). Many of the lactation-related genes were validated by RT-qPCR from the ASO-treated tumors as well as Malat1 knockout tumors (Supplemental Fig. S5B). Interestingly, none of the genes genomically adjacent to Malat1 showed significant expression change except Cdc42ep2, which showed 1.3-fold up-regulation in the ASO-treated tumors. We subjected the entire data set to gene set enrichment analysis (GSEA) (Subramanian et al. 2005; Luo et al. 2009), and, interestingly, significant enrichment of several gene sets involved in mammary tumor progression was observed (Fig. 6C), indicating that loss of Malat1 inhibits several gene expression programs involved in tumor progression. Detailed pathway analysis using KEGG also resulted in enrichment of genes belonging to the ECM signaling, focal adhesion, and cell adhesion pathways (Fig. 6D). Nearly 90% of the tumor RNA-seq enriched pathways (eight of nine) were also enriched in the organoid RNA-seq data analysis. One of the most significantly affected pathways was integrin signaling, with many of the protumorigenic integrins (Itga2b and Itga5b3) being down-regulated in Malat1-depleted tumors (Fig. 6E). To identify the transcription factors potentially cooperating with Malat1, we performed both known and de novo motif analysis of the promoter region (−900, +100 in relation to the TSS) of the differentially expressed genes, and they showed significant enrichment in the Sox5-, Tcfcp2l1-, and E2A-binding motifs in their promoter regions (Fig. 6F). In addition, we took a sequence-independent approach by searching for overlaps between Malat1 putative targets and >359 ChIP-seq (chromatin immunoprecipitation [ChIP] combined with deep sequencing) and ChIP–ChIP experiments annotated in the ChIP Enrichment Analysis (ChEA) database. From a total of 140 transcription factors, Sox9 was identified as a top scorer (P = 0.0002). Sox9 is a HMG protein that shares the similar DNA-binding element that was predicted for Sox5 in our motif analysis (data not shown). Interestingly, when comparing differentially expressed genes upon Malat1 knockdown to CHART-seq (West et al. 2014) performed in MCF7 cells, we found that 11% of our differentially expressed genes possessed MALAT1 peaks (Supplemental Table S3), suggesting that Malat1 could directly regulate gene expression by physically binding to regulatory regions of those genes. Interestingly, one of the up-regulated genes, Tnxb (Tenascin Xb), an ECM protein that has been shown to have anti-metastatic properties (Matsumoto et al. 2001), had five Malat1-binding peaks with high peak scores (Supplemental Table S3).

Malat1 knockdown affects multiple splicing events

As Malat1 has been implicated in alternative pre-mRNA splicing regulation in several cell lines (Tripathi et al. 2010) and in modulating the recruitment of splicing factors to an actively transcribing transgene reporter (Bernard et al. 2010), we investigated whether Malat1 knockdown also resulted in changes in the in vivo splicing pattern of genes involved in tumorigenesis. The SpliceTrap/SpliceDuo pipeline (Wu et al. 2011) was applied to the RNA-seq data sets generated from the tumors and organoids. SpliceTrap performs probabilistic inference of the extent to which an exon is included in the mature transcript, is skipped, or is subjected to size variations due to alternative 3′/5′ splice sites or intron retention. SpliceDuo implements a noise-based methodology to identify statistically significant alternative splicing events across two conditions (see Supplemental Fig. S6A for splicing analysis pipeline). Interestingly, numerous alternative splicing changes were observed in both the Malat1 ASO-treated tumors and organoids when compared with the ScASO-treated samples (Fig. 7A,B; Supplemental Table S4). We detected a total of 1351 splicing changes spanning the four major classes of alternative splicing patterns in at least two biological replicates, resulting in a 60% overlap between ASO1 and ASO2 treatment (Fig. 7B). This finding suggests an involvement of Malat1 in alternative splicing of many important genes. Interestingly, Esr1 coding for estrogen receptor α (ERα) was found to be alternatively spliced in its 5′ untranslated region (UTR) in Malat1 ASO-treated samples compared with ScASO-treated tumors (Fig. 7C; Supplemental Table S4). When we looked at ERα protein expression using immunoblot analysis (Supplemental Fig. S6B), we observed that the control tumors showed lower ERα expression compared with Malat1 ASO-treated tumors. It is known that the MMTV-PyMT tumors lose ER expression in the advanced carcinoma stage. Since most of the Malat1 ASO-treated tumors did not progress to advanced carcinomas, it is possible that they retained ERα expression. It remains to be further investigated whether the switch in the ER isoform confers stability to the protein.

Of particular note, Sox5 was alternatively spliced, giving rise to an isoform that lacked the exon that encodes for a loop domain of this HMG transcription factor (Supplemental Fig. S6C,D). This finding was exciting, as motif analysis of differentially regulated genes upon Malat1 knockdown also showed enrichment of Sox5-binding motifs in their promoters (Fig. 6F). Furthermore, many members of ECM signaling were also alternatively spliced, and Itga2b (Fig. 7D; Supplemental Fig. S6E) belonging to the integrin signaling pathway showed alternate 3′ splice site utilization that resulted in retention of part of an intron. This transcript was shown to be down-regulated in our gene expression analysis as well by qRT–PCR analysis.

Overall, Malat1 knockdown affects both gene expression and alternative splicing of numerous protumorigenic signaling molecules and differentiation-related genes, resulting in a dramatic shift in mammary tumors to a highly differentiated less aggressive state.

VSports最新版本 - Malat1 level increases upon cancer progression

Consistent with our findings that Malat1 plays a role in tumor progression, RNA-FISH analysis of paired human breast tumor samples and metastases indicated that while the level of Malat1 is higher in the primary tumor than in the surrounding stroma, its level is further increased in the lung metastases (Fig. 7E). Of particular interest is the fact that some cells in the primary tumor exhibit a higher level of Malat1, and we suggest that it may be these cells that escape from the primary tumor and metastasize to the lungs. The up-regulation of Malat1 may be triggered by physiological stress and/or oncogene activation. Interestingly, several previous studies in other cancers have found a correlation of Malat1 levels with EMT transcription factors such as Zeb1 and Snail. Although we did not observe changes in the expression of EMT transcription factors in our data sets, it is possible that Malat1 expression could increase in cells that are undergoing EMT, and that may correlate well with those cells preferentially metastasizing. In this context, it has been shown previously that hypoxia-inducible factor HIF1A regulates MALAT1 expression (Choudhry et al. 2014; Michalik et al. 2014). Hif1a is a major transcription factor that governs oxidative stress and regulates tumor progression and has also been shown to play a role in EMT (Semenza et al. 2003). In a number of cell culture studies, Malat1 was found to be up-regulated upon stress such as serum starvation, hypoxia, and genotoxic stress, all of which are linked to tumor progression (Yang et al. 2011; Mizutani et al. 2012; Choudhry et al. 2014). Consistent with this, our findings demonstrated that Malat1 loss in dedifferentiated primary tumors results in extensive alteration in gene expression, activating pathways that are involved in morphogenesis and differentiation.

Mammary gland differentiation has been shown to be regulated by GATA3 (Kouros-Mehr et al. 2008) and Stat signaling (Liu et al. 1997). Although we found that the Jak–Stat signaling pathway is altered in tumor organoids upon loss of Malat1, this change was not captured in the primary tumors, and Gata3 was not differentially expressed in our data sets. Thus, additional factors likely regulate mammary tumor differentiation upon Malat1 loss. Consistent with this possibility, ECM proteins and integrin signaling are altered upon loss of Malat1, and differentiation markers of the mammary gland, such as casein genes, are up-regulated several-fold, resulting in secretion of these proteins within the cysts and thereby restoring a functional activity of the mammary gland. Interestingly, ECM proteins and integrin signaling have been previously shown to play a crucial role in the regulation of casein genes (Streuli et al. 1991, 1995; Pontier and Muller 2009), and prior studies have indicated that lactation confers cancer protection to the mammary gland due to differentiation of epithelial cells (Kotsopoulos et al. 2012).

Poorly differentiated primary tumors are often associated with a poor outcome and higher incidence of metastasis (for review, see Medema 2013). Therefore, the ability to differentiate primary tumors by loss or knockdown of Malat1 significantly reduces the possibility that they will undergo EMT-like processes and become invasive. Multiple lines of evidence converge on this aspect of Malat1 function, as its loss has been previously shown to reduce motility in cell lines and the homing behavior of lung cells in xenograft mouse models (Tano et al. 2010; Gutschner et al. 2013). Our findings add significantly to these prior observations by showing that loss of Malat1 in tumor-bearing mice results in a 70% reduction in lung metastasis. We also found that Malat1 loss inhibited branching morphogenesis in tumor organoids, consistent with its requirement for migration and local ECM remodeling of the organoids in Matrigel, a process that is similar to invasion and matrix degradation observed in vivo in malignant tumors (Ewald et al. 2008; Cheung et al. 2013). In addition, Malat1 loss in both MMTV-PyMT and Her2/neu-amplified MMTV-cre;Flox-neo-neuNT tumor organoids resulted in increased epithelial cell adhesion and loss of migration of cells, resulting in tightly packed spherical acinar structures. Given that the branching phenotype could be restored by addition of mouse full-length Malat1, Malat1 is essential for this process, and the phenotype observed is specific for Malat1. Our data add significant support to the concept that organoids can provide a rapid means of assessing therapeutic potential in a patient-specific manner (Boj et al. 2015).

Malat1 functions as a cotranscriptional splicing scaffold

Although Malat1 knockout mice exhibit no phenotype (Eissmann et al. 2012; Nakagawa et al. 2012; Zhang et al. 2012), upon Malat1 knockout or knockdown in the MMTV-PyMT mammary cancer model, we observed changes in gene expression and pre-mRNA splicing of numerous cancer-relevant genes. Lack of significant overlap of differentially expressed genes upon Malat1 loss among various studies supports a role for Malat1 in a context-dependent manner. Based on cross-linking studies, it has been shown that Malat1 is bound to either chromatin (West et al. 2014) or pre-mRNAs (Engreitz et al. 2014). Since it is well documented that transcription and pre-mRNA splicing are coupled processes (for review, see Bentley 2014), our data suggest that Malat1 may act as a scaffold to coordinate these processes in a gene- and context-specific manner (Fig. 7F). The presence of Malat1 in nuclear speckles, domains that are commonly found near actively transcribing genes and are also enriched in splicing and transcription factors (for review, see Spector and Lamond 2011), could facilitate the function of Malat1 at sites of transcription. Interestingly, based on comparisons between a CHART-seq data set from MCF7 cells that identified direct MALAT1-binding sites across the genome and our gene expression data set, we identified ∼11% of our differentially expressed genes to be potential direct targets of Malat1, which is consistent with the possibility of Malat1 coordinating expression of these genes. In line with this, we previously showed that loss of Malat1 impairs the recruitment of SR splicing factors to a stably integrated transgene array (Bernard et al. 2010). It has also been previously shown that numerous splicing changes occur during tumor progression, giving rise to novel oncogenic splice variants (Venables et al. 2008; Eswaran et al. 2013). In this regard, SRSF1, a SR protein-splicing factor, can act as an oncoprotein that promotes breast cancer (Anczukow et al. 2012). During tumor progression, Malat1 may facilitate efficient transcription and splicing of protumorigenic genes such as integrins, ECM proteins, and genes involved in migration and metastasis, leading to a more favorable outcome for the tumor. Thus, loss of Malat1 results in altered transcription and pre-mRNA splicing events more consistent with the differentiated phenotype of the impacted tissue. Given the importance of estrogen signaling in luminal breast cancer and our observation that the estrogen receptor (Esr1) transcript was alternatively spliced in its 5′ UTR upon Malat1 loss, it will be important in follow-up studies to determine whether Esr1 isoform switching plays a role in the observed differentiation phenotype.

It is presently unclear whether Malat1 binds to certain proteins or transcription factors to elicit its regulatory role. Previously published CLIP-seq (cross-linking immunoprecipitation [CLIP] combined with deep sequencing) and RNA pull-down studies from several groups have shown that Malat1 binds to splicing factors such as SRSF1, TDP43, and PRC2 components, including EZH2 (Tripathi et al. 2010; Tollervey et al. 2011; Yang et al. 2011; Guil et al. 2012). Detailed motif analysis of the promoters of the differentially expressed genes in our data sets show them to be enriched for TCFCP2L1, SOX5, and E2A transcription factor-binding sites, which have established roles in tissue differentiation and pluripotency (Chen et al. 2008; Kamachi and Kondoh 2013; Ye et al. 2013). Therefore, we suggest that the specificity of the Malat1-responsive genes may be context-dependent and arise from the transcription factors that primarily regulate those genes. For example, TCFCP2L1 has been shown to cooperate with the estrogen receptor family of transcription factors such as ESRRB (Boroviak et al. 2014) and has been shown to be associated with breast cancer metastasis to the lungs (Landemaine et al. 2008).

Malat1 ASOs are a promising therapeutic

Of potential clinical impact, we employed antisense-mediated knockdown using Malat1-specific ASOs in a mouse model of metastatic mammary cancer. ASOs are currently in clinical trials for many diseases, including spinal muscular atrophy and Duchenne muscular dystrophy (Aartsma-Rus et al. 2004; Rigo et al. 2012). The in vivo efficacies of Malat1 ASOs have been tested in primates, where >80% knockdown efficiency was achieved (Koller et al. 2011; Hung et al. 2013). Here we show that Malat1 knockdown results in slower tumor proliferation and differentiation and significant reduction in metastasis, all recapitulating the phenotype of the MMTV-PyMT;Malat1^−/− genetic model, thus confirming the specificity of these ASOs. Future studies using patient-derived xenograft models as well as patient-derived tumor organoids will enable us to further study the efficacy of Malat1 ASOs as a therapeutic to induce differentiation of primary human breast tumors and significantly reduce tumor progression.

Animals

Animal experiments were carried out in the Cold Spring Harbor Laboratory Animal Shared Resource in accordance with Institutional Animal Care and Use Committee-approved procedures. Malat1 knockout (C57BL/6) mice were generated as described previously (Zhang et al. 2012) and bred with MMTV-PyMT (C57BL/6) mice (Guy et al. 1992). CAG::H2B-EGFP mice (Hadjantonakis and Papaioannou 2004) were obtained from Jackson Laboratory and bred with MMTV-PyMT mice. MMTV-Cre and Fl-neo-neu-NT mice were obtained from Dr. William Muller (McGill University, Montreal) and were bred to generate MMTV-Cre;Fl-neo-neu-NT.

RNA in situ hybridization

RNA in situ hybridization was performed on formalin/PFA-fixed paraffin-embedded tissue sections using specific probes against human or mouse Malat1 purchased from Affymetrix. The human tissue sections were obtained from Dr. Edi Brogi (Memorial Sloan Kettering Cancer Center). Both matched primary tumor sections/metastasis and unmatched array samples were scored for MALAT1 level. The in situ hybridization protocol using ViewRNA technology was performed according to the manufacturer's instructions (Affymetrix). Slides were scanned using an Aperio scanner, and the images were analyzed using Leica Scanscope software. Alternatively, nick-translated Malat1 probes labeled with fluorescently conjugated dUTPs were used for RNA-FISH as described previously (Zhang et al. 2012). Images were captured using a DeltaVision microscope (GE). For all of the quantitative analysis, the fluorescent intensity profiles were obtained from at least five different random fields. When array spots were used for intensity measurement, the total intensity of the spot with ROI markup was used to estimate the signal intensity. The intensity values were binned to obtain the range, which was then used for plotting.

"VSports在线直播" Organoid culture

Organoids from MMTV-PyMT and MMTV-Cre;Flox neo-neu NT tumors were prepared and cultured as described previously (Ewald 2013). Organoids were mixed with Matrigel and plated in 24-well Mat-Tek dishes. For ASO treatment, 500 nM Malat1 ASO or scASO was added to 1 mL of organoid culture medium. Images were acquired using a Zeiss Axio-Observer light microscope. For live-cell imaging, organoids from MMTV-PyMT or MMTV-PyMT;CAG::H2B-EGFP mice were treated with Malat1 or ScASOs. At least five organoids per well were imaged in three dimensions over a period of 4 d at 15- or 30-min intervals using a Perkin-Elmer spinning-disc confocal microscope. Time-lapse images were processed using Volocity (PerkinElmer) and ImageJ (National Institutes of Health) software.

Electron microscopy

Organoids grown in Matrigel in 12-well culture dishes were fixed overnight with 2% glutaraldehyde and 2% paraformaldehyde in PBS. While in the multiwell dish, the samples were rinsed with distilled water and then post-fixed with 1% osmium tetroxide in 1.5% potassium ferrocyanide for 1 h. After a distilled water rinse, discs of Matrigel containing embedded organoids were lifted out of the culture dishes, cut into 2-mm × 10-mm × 4-mm pieces, and transferred into 20-mL glass scintillation vials. Samples were dehydrated in a graded series of ethanol and then rinsed in 100% acetone. Resin embedding was done with continuous agitation in 50% epon-araldite resin in acetone for 1 h and overnight incubation in 100% epon-araldite with agitation. Samples were then placed into BEEM capsules (placed flat into the cap end of a conical capsule made by cutting off the conical end), and the resin was polymerized overnight at 60°C in a vented oven. Samples were sectioned at 100 nm, collected on butvar-coated nickel slot grids (1-mm × 2-mm opening; Electron Microscopy Sciences), and counterstained with lead citrate. Sections were examined in a Hitachi H-7000 transmission electron microscope operated at 75 kV. Representative images were recorded on Kodak 4489 film. Images were scanned at 2400 dpi with an Epson Perfection VP750 Pro.

ASO treatment and tumor studies (VSports在线直播)

The mice selected for ASO treatment were grouped based on age, initial tumor size, and separation of siblings into three cohorts. The ASOs against Malat1 or ScASO were reconstituted at a concentration of 5 mg/mL in PBS and injected subcutaneously every day at a concentration of 25 mg/kg per day in a volume of ∼125 μL for 5 d with a rest period of 2 d. Tumor growth was measured twice per week using calipers in a nonblinded fashion. Tumor volume was calculated based on the formula volume = width² × length/2. Mice were sacrificed after 8 wk of treatment, and tumors were removed and fixed in 4% PFA. Upon sacrifice, the lungs were also removed, and the macronodules were counted using a dissection microscope. Other internal organs, including the spleen, liver, kidney, and brain, were examined for metastasis or abnormalities using a dissection microscope.

Histology

Primary tumors and lungs were fixed in 4% PFA for up to 24 h and washed in PBS. Primary tumors were embedded in paraffin, sectioned, and stained with H&E. Slides were scanned using an Aperio slide scanner. The pathologist performed blind analyses to assess the histologic grade. The cystic area calculation was done using Aperio Scanscope software. After fixation, lungs were incubated in 20% sucrose overnight and embedded in OCT, and the lungs were cross-sectioned 2 mm apart. The lung sections were embedded horizontally to obtain serial sections of the entire lung, and H&E-stained sections were scanned with an Aperio image scope. The lung metastatic burden was calculated using Aperio Scanscope and ImageJ (National Institutes of Health) software.

Immunofluorescence

Immunofluorescence on the tumor cryosections was performed as described previously. Briefly, sections were washed in PBS and permeabilized for 5 min in 0.05% Triton X. The sections were blocked using 1% goat serum followed by incubation with primary antibody (Ki-67 from Vector Laboratories, E-cadherin from BD; β-casein was a gift from Dr. Mina Bissel) for 4 h to overnight at 4°C. Subsequently, the slides were washed with PBS + 0.01% Triton X. Finally, the sections were incubated in Alexa fluor-conjugated secondary antibodies (Jackson ImmunoResearch) followed by washes with PBS. The sections were stained with DAPI and mounted in 90% glycerol and 10% PBS plus PPD (paraphynelediamine). Images were acquired using a Zeiss LSM 710 or Zeiss Axio-Observer microscope. Ki-67 was quantified by counting the number of positive nuclei for every 100 nuclei from multiple fields on the slide to obtain the percentage of positive cells.

RNA-seq and analysis

Organoids generated from four MMTV-PyMT mice were cultured in 3D Matrigel domes in the presence of FGF2 for 6 d. The organoids were treated with 500 nM mock, ScASO, or Malat1 ASOs. In addition, tumors were removed from MMTV-PyMT mice that had undergone injections with ScASO (three mice) or Malat1 ASO1 (three mice) or ASO2 (four mice). Total RNA was extracted using TRIzol (Life Technologies) from both organoids and homogenized tumor tissue. Libraries for polyA⁺ RNA-seq were prepared from 1 μg of RNA per sample using TruSeq chemistry (Illumina), multiplexed, and sequenced to obtain paired-end 101-base-pair (bp) reads on an Illumina HiSeq 2000 platform, resulting in 20 million to 86 million reads per library.

The quality of the raw data was evaluated using FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc), and reads were mapped to mm9 using STAR (Dobin et al. 2013), resulting in an overall mapping efficiency of >90%. The Gencode mV1 annotation set was used as a reference, and the reads per gene record were counted using the HTSeq package (Anders et al. 2015). Replicate analysis was carried out using edgeR (Robinson et al. 2010). Differential gene expression was performed with DESeq2 (Love et al. 2014). Functional analysis of KEGG pathways and gene ontology (GO) terms was carried out with the R/Bioconductor packages GAGE (Luo et al. 2009) and Pathview (Luo and Brouwer 2013). Motif analyses were performed using Homer (http://homer.salk.edu/homer; Heinz et al. 2010).

Splicing analysis

Alternative splicing, including cassette exons, alternative 3′/5′ splice site, and intron retention events, were identified and quantified using the SpliceTrap/SpliceDuo pipeline (Wu et al. 2011). This tool combines RNA-seq data with prebuilt transcript models to quantify the level of inclusion of every exon in a transcript. The transcript models are exon trios composed of alternative exon candidates with their annotated flanking exons. These were derived from the mm9 TXdb database, which accounts for a total of >230,000 exon trio models describing the alternative splicing of ∼190,000 mouse exons in ∼44,000 transcripts. SpliceTrap uses the Bowtie read aligner to align reads against TXdb. Biological replicates were analyzed independently, and alternative splicing events that were reproduced in at least two samples were selected (SpliceDuo FDR < 0.1). Inconsistent splicing events showing opposite splicing directions (skipping versus inclusion) were further removed.

RT-qPCR (VSports)

One microgram of RNA was used to make cDNAs using oligo-dT primers. One-tenth the volume of the cDNA reaction mixture was used in the qPCR reaction containing 1× SYBR Green master mix (Applied Biosystems) and specific forward and reverse primers. The relative fold change was calculated using the ΔΔC_t method. The primers used were as follows: Malat1 F (AACCAGTTTCCCCAGCTTTT) and Malat1 R (CTACATTCCCACCCAGCACT).

Accession number

RNA-seq data presented in this study have been submitted to the NCBI Gene Expression Omnibus (GEO; http:// http://www.ncbi.nlm.nih.gov/geo) under accession number GSE67647