Hypoxia induces cancer cell-specific chromatin interactions and increases MALAT1 expression in breast cancer cells

Metastasis-associated lung adenocarcinoma transcript 1 (MALAT1) is a long noncoding RNA overexpressed in various cancers that promotes cell growth and metastasis. Although hypoxia has been shown to up-regulate MALAT1, only hypoxia-inducible factors (HIFs) have been implicated in activation of the MALAT1 promoter in specific cell types and other molecular mechanisms associated with hypoxia-mediated MALAT1 up-regulation remain largely unknown. Here, we demonstrate that hypoxia induces cancer cell-specific chromatin–chromatin interactions between newly identified enhancer-like cis-regulatory elements present at the MALAT1 locus. We show that hypoxia-mediated up-regulation of MALAT1 as well as its antisense strand TALAM1 occurs in breast cancer cells, but not in nontumorigenic mammary epithelial cells. Our analyses on the MALAT1 genomic locus discovered three novel putative enhancers that are located upstream and downstream of the MALAT1 gene body. We found that parts of these putative enhancers are epigenetically modified to a more open chromatin state under hypoxia in breast cancer cells. Furthermore, our chromosome conformation capture experiment demonstrated that noncancerous cells and breast cancer cells exhibit different interaction profiles under both normoxia and hypoxia, and only breast cancer cells gain specific chromatin interactions under hypoxia. Although the HIF-2α protein can enhance the interaction between the promoter and the putative 3′ enhancer, the gain of chromatin interactions associated with other upstream elements, such as putative −7 and −20 kb enhancers, were HIF-independent events. Collectively, our study demonstrates that cancer cell-specific chromatin–chromatin interactions are formed at the MALAT1 locus under hypoxia, implicating a novel mechanism of MALAT1 regulation in cancer.

noncoding regions of the genome that increase gene expression in select tissues or in response to various stresses (1,2). Enhancer sequences contain transcription factor-binding sites and the recruitment of these proteins to enhancers can up-regulate the target gene through close contact with the promoter (1). In addition, chromatin-chromatin interactions are formed within the three-dimensional chromatin structure, connecting enhancers present upstream, downstream, or at a long distance, to their target promoters (3)(4)(5). Gene expression can be further fine-tuned through regulation of histone modifications, such as methylation and acetylation (6).
Metastasis-associated lung adenocarcinoma transcript 1 (MALAT1) 2 is a long noncoding RNA, which is highly expressed in multiple tissues. Through transcriptional regulation and alternative splicing MALAT1 regulates expression of genes involved in proliferation and cellular motility (7)(8)(9). MALAT1 is widely overexpressed in solid tumors where it promotes proliferation, motility, angiogenesis, and metastasis (10). MALAT1 overexpression is a poor prognostic factor in cancer and is a strong indicator of metastatic potential. Although the functional role of MALAT1 is well-documented, the upstream factors involved in MALAT up-regulation are poorly understood. A handful of transcription factors, namely ␤-catenin/ TCF7L2 (11), c-Fos (12), NF-B (13), and Sp1/Sp3 (14,15) have been shown to activate MALAT1 transcription, whereas p53 represses transcription (16,17). However, involvement of these proteins in MALAT1 transcription have not been validated in more than one tumor type, and there is a lack of information regarding how MALAT1 is up-regulated in a broad range of cancer types. Moreover, besides the promoter regions, cis-regulatory elements involved in MALAT1 expression are completely unknown.
Hypoxia, caused by low oxygen levels in the tumor microenvironment, up-regulates a set of genes involved in angiogenesis, apoptosis, cell migration, glycolysis, and stemness (18,19). Many of these genes are targets of the hypoxia-inducible factor (HIF) proteins, a family of transcription factors stabilized by low-oxygen conditions (20,21). However, mechanisms that cause HIF-independent gene regulation are largely unexplored and demonstra-tion of these pathways and transcription factors relevant to cancer progression remain elusive. Several studies have shown that MALAT1 is up-regulated under hypoxia (9,(22)(23)(24)(25)(26). However, its underlying mechanisms remain poorly understood.
In this study, we demonstrate that the MALAT1 genomic locus contains potential distal enhancers upstream and downstream of the MALAT1 gene body and these regions are involved in longrange chromatin interactions in breast cancer. Importantly, hypoxia modifies histone markers resulting in more open chromatin and induces chromatin interactions in the MALAT1 locus in breast cancer cells. Although HIF facilitates the interaction between the MALAT1 promoter and the enhancer located downstream of the gene, other hypoxia-induced chromatin-chromatin interactions between the potential enhancer regions are HIF-independent, they only occur in breast cancer cell lines and they are not observed in nontumorigenic mammary lines. These findings suggest that the MALAT1 enhancers have a hypoxic-specific response in the cancer condition.

Hypoxia up-regulates MALAT1 in breast cancer cell lines
To determine the expression level of MALAT1 in different subtypes of breast cancer cells, we performed Northern blotting using RNA isolated from nine cancer lines encompassing each subtype: luminal A (MCF7, T47D), luminal B (BT474, MDA-MB-361, ZR-75-30), HER2-enriched (MDA-MB-453, SKBR3), and triple-negative (MDA-MB-157, MDA-MB-231). MCF10A was also included as a nontumorigenic mammary epithelial cell line. We observed high levels of MALAT1 expression in luminal A, luminal B, and HER2-enriched breast cancer cells compared with triple-negative breast cancer cell lines and MCF10A. All three cell lines of the luminal B subtype breast cancer cells had the highest expression levels of MALAT1. This result demonstrated that the MALAT1 expression level is indeed higher in breast cancer cell lines compared with that of a nontumorigenic line (Fig. 1A). To examine how hypoxia affects MALAT1 expression in these cells, we exposed nine of these cell lines to either normoxic (21% O 2 ) or hypoxic (1% O 2 ) conditions for 24 h prior to RNA isolation. Our quantitative PCR (qPCR) data demonstrated that under hypoxia all breast cancer lines increased expression of MALAT1 relative to normoxic levels (Fig. 1B). Hypoxia-mediated up-regulation of MALAT1 is more noticeable in the cell lines with low MALAT1 expression under normoxia (MDA-MB-157, MDA-MB-231, MCF7), compared with the cell lines with the highest MALAT1 expression in normoxia (BT474, MDA-MB-361). Surprisingly, MCF10A did not A, 5 g of total RNA was probed for MALAT1 expression by Northern blotting. Cell lines are colored according to receptor status and breast cancer subtype: gray, nontumorigenic; red, triple negative; blue, luminal A; green, luminal B; purple, HER2 positive. B, cells were grown in normoxia (21% O 2 ) or hypoxia (1% O 2 ) for 24 h and total RNA was isolated for RT-qPCR. Increase in MALAT1 expression for each cell line under hypoxia is normalized to its normoxic expression levels. Data are represented as mean Ϯ S.D., n ϭ 3-5, *, p Ͻ 0.05; **, p Ͻ 0.01. C, diagram of the MALAT1 genomic locus and the location of the antisense transcript TALAM1 gene. The locations indicated in kb are distances from the MALAT1 transcription start site. D, strand-specific qPCR assay design developed by Zong et al. (28). A fusion oligo consisting of a linker sequence and a strand-specific primer for reverse transcription (SSP-RT) is used for cDNA synthesis. The 5Ј portion of the linker is used for qPCR in conjunction with a SSP forward primer. E, MALAT1 and TALAM1 strand-specific qPCR for MCF10A and MCF7 exposed to normoxic or 24 hypoxic conditions for 24 h. Data are represented as mean Ϯ S.D., n ϭ 3, *, p Ͻ 0.05; **, p Ͻ 0.01.

Hypoxia up-regulates MALAT1 via chromatin looping
show any changes in MALAT1 expression under hypoxia (Fig.  1B). These results demonstrated that hypoxia strongly induces MALAT1 expression in breast cancer cells, but not in nontumorigenic mammary epithelial cells.

The antisense transcript TALAM1 is up-regulated under hypoxia
Increasing evidence has demonstrated that antisense transcripts are lowly expressed noncoding RNAs transcribed from the opposite strand of coding or noncoding RNAs which then, in many cases, regulate expression of the sense transcript (27). The MALAT1 locus contains an antisense transcript called TALAM1 (Fig. 1C), which promotes MALAT1 up-regulation via 3Ј end processing (28). Because MALAT1 was overexpressed under hypoxia and these two RNAs engage in a positive feedback loop with each other, we examined if hypoxia could also up-regulate TALAM1. We used a strand-specific primer (SSP) strategy devised by Zong et al. (28) to differentiate MALAT1 and TALAM1 (Fig. 1D), as regular reverse transcription and PCR cannot discriminate between sense and antisense transcripts. We used a linker sequence not found in the human genome fused to an antisense oligo specific for either MALAT1 or TALAM1 for cDNA synthesis. qPCR amplified an SSP forward primer with a reverse primer for the linker sequence. The result demonstrated no hypoxia-mediated up-regulation of MALAT1 in MCF10A and accordingly no change was found in TALAM1 expression under hypoxia (Fig. 1E). In contrast, when we examined expression of MALAT1 and TALAM1 in the MCF7 breast cancer cell line, both transcripts were significantly up-regulated under hypoxia (Fig. 1E). These expression profiles are consistent with increased MALAT1 transcription under hypoxia and provide support for the importance of this locus in cancer cells.

HIF-2␣ has a minor role in MALAT1 up-regulation
Having determined that hypoxia up-regulates MALAT1, we next sought to find a candidate factor that would function as a transcriptional activator in this process. We considered HIF-1␣ or HIF-2␣ as potential candidates, given their regulation of MALAT1 in various tissues under hypoxia (22,24,25). To examine whether HIFs are required for MALAT1 up-regulation under hypoxia, we used small interfering RNA (siRNA) to knockdown either HIF protein alone or in combination with each other. Knockdown of HIF-1␣ and HIF-2␣ was confirmed at both the mRNA and protein levels in MCF7 cells ( Fig. 2A and Fig. S1A). We next measured the effect of HIF knockdown on the well-known HIF target genes CAIX and PAI1 and found both mRNAs were reduced by HIF depletion (Fig. 2B). Under this condition, we observed a slight, yet significant, decrease of MALAT1 by HIF-2␣ knockdown alone (Fig. 2C). Consistent with the previous observation on HIF-1␣ function as a repressor of HIF-2␣ expression in MCF7 cells (29), our result showed A, siRNA targeting HIF-1␣ or HIF-2␣ was transfected into MCF7 cells for 48 h prior to transfer to hypoxia for 24 h, and Western blotting was performed to detect HIF-1␣ and HIF-2␣. B and C, MCF7 cells transfected with siRNA as described in A were used for RT-qPCR to measure the expression of the known HIF target genes CAIX and PAI1 (B) and MALAT1 (C). D and E, MCF7 cells were transfected with pcDNA-HA-HIF2␣-P405A/P531A (HIF2␣-CA; ϩ) or an empty vector (Ϫ) for 48 h prior to cell lysis and used for Western blots using the antibodies against HIF-2␣ and HA (tag) (D) and MALAT1 RT-qPCR (E). F, MCF7 cells were exposed to hypoxia or treated with 100 M CoCl 2 for 24 h prior to lysate preparation for an HIF-2␣ Western blotting. G, RT-qPCR was performed to measure the MALAT1 expression level following MCF7 treatment with 100 M CoCl 2 for 24 h. Actin was examined as a loading control in Western blots. In the bar graphs, data are represented as mean Ϯ S.D., n ϭ 3-5, *, p Ͻ 0.05; **, p Ͻ 0.01.

Hypoxia up-regulates MALAT1 via chromatin looping
that the combinational treatment of HIF-1␣ siRNA and HIF-2␣ siRNA abrogated the effect of HIF-2␣ siRNA on a marginal decrease of MALAT1.
We further examined the role of only HIF-2␣ in MALAT1 up-regulation by transfecting MCF7 cells with a plasmid encoding a constitutively active form of HIF-2␣ (HIF2␣-CA) with P405A and P531A point mutations (pcDNA-HA-HIF2␣-P405A/P531A), rendering the protein stable in the presence of oxygen (30). We confirmed HIF2␣-CA overexpression at the mRNA and protein levels and up-regulation of PAI1, a HIFtarget gene ( Fig. 2D and Fig. S1B). Under this condition, we found that MALAT1 was up-regulated about 1.3-fold (Fig. 2E).
To further verify the minor role of HIF-2␣ in MALAT1, we treated MCF7 cells with cobalt chloride (CoCl 2 ), which prevents the degradation of the ␣ subunits of HIF under normoxic conditions (31), resulting in increased levels of the HIF2␣ proteins (Fig. 2F). Upon CoCl 2 treatment, we observed only a 1.5fold increase in MALAT1 expression in MCF7 cells (Fig. 2G), indicating that CoCl 2 -mediated HIF-stabilization is not sufficient to fully recapitulate the effect of hypoxia. Therefore, it is likely that additional, HIF-independent changes are necessary to fully activate MALAT1 expression in breast cancer cells under hypoxia.

Putative enhancers are identified at the MALAT1 locus and active chromatin marks are increased in some of these enhancer regions under hypoxia
To address whether any cis-elements present near the MALAT1 gene are involved in hypoxia-induced MALAT1 expression, we extensively analyzed various chromatin marks and transcription factor binding patterns at the MALAT1 locus using ENCODE chromatin immunoprecipitation (ChIP)-seq data. Interestingly, besides the known promoter area, we found multiple sites containing the active enhancer/promoter marks histone-3 lysine-4 methylation (mono-methylation, H3K4me1; tri-methylation, H3K4me3) and histone-3 lysine-27 acetylation (H3K27Ac) as well as RNA polymerase II occupancy and transcription factor-binding sites in the sequence upstream of the MALAT1 transcription start site (TSS) and downstream of the transcription termination site (Fig. 3A). Based on enrichment levels of these chromatin features, we propose three putative enhancer regions for MALAT1. These enhancer regions are located between Ϫ21.4 and Ϫ17.6 kb and Ϫ11.1 and Ϫ2.7 kb upstream of the TSS of the MALAT1 gene, and between ϩ8.4 and ϩ11 kb, which is the region located immediately downstream of the 3Ј end of the MALAT1 gene. We named these putative enhancers Ϫ20 kb enhancer, Ϫ7 kb enhancer, and 3Ј enhancer, respectively (Fig. 3A). Examination of the mouse genome for orthologous sequence revealed each of these putative enhancer elements were conserved at the genetic and epigenetic levels (Fig. S2), supporting a potential functional role.
It is known that as a stress response to hypoxia, a selective group of genes are up-regulated, in part due to an increase of active epigenetic markers such as H3K4me3 at their promoters (32,33). Therefore, we questioned if hypoxia alters the chromatin status of the MALAT1 promoter or these potential enhancers regions. Nine sites throughout the promoter and putative enhancers were selected to determine whether histone modifi-cations occupancy changed under hypoxia by ChIP-qPCR (Fig.  3A). The P1 locus is found approximately Ϫ20.5 kb upstream of the MALAT1 TSS and is within the putative Ϫ20 kb enhancer. The P2 locus is at Ϫ15.9 kb upstream and is not contained within any putative enhancer. The putative Ϫ7 kb enhancer contains two primer sets at each end, P3 and P4, at Ϫ9.2 and Ϫ4.7 kb, respectively. The P5 locus is at Ϫ2.2 kb and is not contained within any putative enhancer. We designed three primer sets for the promoter regions; the P6 locus at Ϫ0.8 kb for the upstream promoter, and the P7 locus at ϩ0.8 kb and the P8 locus at ϩ1.4 kb for the internal promoter. Finally, the P9 locus is at ϩ9.2 kb and covers the putative 3Ј enhancer.
We performed ChIP-qPCR at the genomic loci P1-P9 under normoxia and hypoxia in the nontumorigenic MCF10A and tumorigenic MCF7 cell lines for the histone modifications H3K27Ac, H3K4me1, and H3K4me3, then calculated the logfold change of the ratio of hypoxic to normoxic enrichment of each histone modification. Our ChIP-qPCR results revealed distinct levels of histone modifications between these cell lines.
In MCF10A cells, the active enhancer/promoter marker H3K27Ac was enriched at P2, P5, and P9 and decreased at the other six loci under hypoxia (Fig. 3B). Active enhancer marker H3K4me1 levels showed fewer changes in MCF10A cells in response to hypoxia, only increasing at P8 and staying unchanged at P5, whereas decreasing at all other loci (Fig. 3C). Active promoter marker H3K4me3 levels were increased at P1, P7, and P8 but sharply decreased at all other loci, including the P6 locus covering the upstream promoter (Fig. 3D).
Interestingly, we found strikingly different patterns of histone modifications in response to hypoxia in MCF7 cells. Changes in H3K27Ac levels were overall similar to MCF10A, with enrichment at P4, P5, and P9 under hypoxia (Fig. 3B). Interestingly, we found significant increases of the H3K4me1 level under hypoxia compared with normoxia at seven of nine sites examined (except for the P6 and P7 sites) (Fig. 3C). In addition, we detected that the level of H3K4me3 increased at both the upstream and internal promoter loci (P6 and P7, respectively), a part of the putative Ϫ7 kb enhancer and its downstream region (P4 and P5) as well as the putative 3Ј enhancer (P9) under hypoxia in MCF7 cells (Fig. 3D).
In summary, all three histone markers showed enrichment at the loci of P4 (putative Ϫ7 kb enhancer), P5 (downstream of the Ϫ7kb enhancer), and P9 (putative 3Ј enhancer) in MCF7 under hypoxia, whereas only H3K27Ac was enriched at P5 and P9 in MCF10A. These data demonstrate that the Ϫ7 kb enhancer and its downstream region as well as the 3Ј enhancer are modified to open, active chromatin in breast cancer in response to hypoxia, but these modifications do not occur in nontumorigenic mammary cells.

Long-range chromatin interactions occur at the MALAT1 locus between enhancers and the MALAT1 promoter in a cell-type specific manner
Although H3K4me3 has traditionally been considered as an active promoter mark, a recent study demonstrated that active enhancers are also highly enriched with H3K4me3 modification (34). Because our ChIP-qPCR data demonstrated greater enrichment of H3K27Ac, H3K4me1, and H3K4me3 under Hypoxia up-regulates MALAT1 via chromatin looping hypoxia at multiple loci within the putative MALAT1 enhancers in breast cancer, we hypothesized that these regions may be in a more open chromatin state under hypoxia and become proximal to each other as well as with the MALAT1 promoter region to facilitate transcription. We tested this possibility using chromosome conformation capture (3C) (Fig. 4A). A dual-digest strategy with the restriction enzymes BamHI and BglII was selected given their complementary overhangs, which yielded 12 restriction sites throughout the MALAT1 genomic locus (Fig. 4B), stopping at Ϫ24 kb upstream due to the absence of histone modifications at this locus and further upstream (Fig.  3A). Fragments were named for their distance from the MALAT1 3Ј end (fragments A-J), with the fragment containing the MALAT1 gene body omitted (Fig. 4B).
Comparison of the fragments with the loci of our putative enhancers indicated that the putative Ϫ20 kb enhancer is con-tained within fragments H and I. The vast majority of the putative Ϫ7 kb enhancer is contained within fragments D and E. The 3Ј enhancer falls mostly into fragment A. The promoter region is contained within the B and C fragments (Fig. 4B). We selected the 3Ј enhancer as a 3C anchor point as normoxic epigenetic marker levels were low in this region but sharply increased under hypoxia in MCF7 and the very high density of transcription factors and RNA polymerase II bound at this locus (Fig. 4), which implicates its function in transcriptional regulation, potentially by mediating interaction with the promoter (35,36).
Our 3C and subsequent PCR using MCF10A and MCF7 cell lines demonstrated that the 3Ј enhancer (anchor/fragment A) indeed interacts with upstream fragments in varying degrees in these two cell lines (Fig. 4C and Fig. S3). Under the normoxic condition, the 3Ј enhancer demonstrates chromatin-

Hypoxia up-regulates MALAT1 via chromatin looping
chromatin interactions with the promoter (fragment B) in both cell lines, which demonstrates that chromatin looping between the 3Ј enhancer and the promoter regions occurs in both normal and breast cancer cells and this interaction may be important in expression of MALAT1 under normoxic condition. In sharp contrast, the interaction of the anchor with the G fragment was detected only in MCF7 cells, but not in MCF10A cells (Fig. 4C), suggesting this locus is important in cancer cells under normal conditions.

Hypoxia alters long-range chromatin interactions at the MALAT1 locus, resulting in extensive chromatin contacts in breast cancer cells
Because hypoxia altered chromatin modifications at multiple sites upstream and downstream of the MALAT1 gene (Fig. 3,  B-D), we next questioned whether hypoxia alters chromatin interactions at the MALAT1 locus. From our 3C-PCR, we found that MCF10A cells did not gain any significant interactions within the examined loci in response to hypoxia. The interaction between the anchor (3Ј enhancer) and fragment B (promoter) was even weakened under hypoxia in MCF10A (Fig.  4, C and D, and Fig. S3).
In contrast, in MCF7 cells, the interaction between the anchor (3Ј enhancer) and fragment B (promoter) is enhanced under hypoxia. The interaction between the anchor and fragment G was still observed under hypoxia. Interestingly, in addition to these interactions, several novel interactions formed in MCF7 cells under hypoxia; interactions between the anchor and fragments C (part of the upstream promoter), D (Ϫ7 kb enhancer), and H (Ϫ20 kb enhancer) (Fig. 4, C and E, and Fig.  S3). Specificity of these interactions were validated with negative control samples (genomic DNA and unligated samples; Fig.  S4). Furthermore, we observed no chromatin-chromatin interactions between the anchor and fragments E, F, I, or J in either cell line, which supports the specificity of the detected interactions. These data demonstrate that novel chromatinchromatin contacts are formed between MALAT1 enhancers at the MALAT1 locus under hypoxia in MCF7 breast cancer cells, but not in nontumorigenic MCF10A cells.

HIF proteins regulate chromatin-chromatin interactions between the MALAT1 promoter and 3 enhancer, but not the interactions between enhancers
Having observed a minor but measurable role for HIF-2␣ in MALAT1 up-regulation in MCF7 cells (Fig. 2), we questioned if this protein could regulate the observed hypoxic chromatinchromatin interactions. We either overexpressed HIF2␣-CA (HIF-2␣-P405A/P531A) or treated CoCl 2 in MCF7 cells under

Hypoxia up-regulates MALAT1 via chromatin looping
normoxic conditions and performed 3C-PCR. The results revealed that both HIF-2␣ overexpression or CoCl 2 -mediated HIF stabilization increased the chromatin interaction between anchor (3Ј enhancer) and fragment B (promoter) (Fig. 5, A, B, D, and E). In line with these results, knockdown of HIF-2␣ abrogated the effect of hypoxia-mediated chromatin interactions of 3Ј enhancer with the promoter region found in fragment B (Fig.  5, C and F). In contrast, HIF-2␣ does not enhance chromatinchromatin interactions between the anchor (3Ј enhancer) and fragment G (Fig. 5, A-F, and Fig. S5). These results are consistent with our results demonstrating constitutive interactions between these loci (Fig. 4C). Although we observed the interaction between the anchor and fragments C, D, and H under hypoxia in MCF7 cells (Fig. 4C), we did not observe any of those interactions under the condition of HIF-2␣ overexpression or CoCl 2 -mediated HIF stabilization (Fig. 5, A and B). These data demonstrate that HIF-2␣ plays a role in mediating chromatinchromatin interactions between the MALAT1 promoter and the 3Ј enhancer, but not in mediating interactions between other putative enhancers.
To support the notion that HIF-2␣ plays a role in mediating interaction between the MALAT1 promoter and the 3Ј enhancer, we examined whether HIF was recruited to either site under hypoxia. We identified two putative HIF-binding sites (hypoxic response elements; HREs) within the MALAT1 promoter and two HREs within the 3Ј enhancer (Fig. 5G). In MCF7 cells, our chromatin immunoprecipitation (ChIP) results revealed that, indeed, HIF-2␣ was enriched at all four HRE loci (Fig. 5H).
Taken together, our data demonstrate that whereas both nontumorigenic mammary cells and breast cancer cells form chromatin loops at the MALAT1 locus by the interaction

Hypoxia up-regulates MALAT1 via chromatin looping
between the promoter and 3Ј end of the gene, hypoxia triggers further gain of chromatin interactions only in breast cancer cells throughout the MALAT1 locus. This renders extensive chromatin contacts among the promoter and multiple enhancer regions, including the Ϫ7 kb enhancer and Ϫ20 kb enhancer (Fig. 6). Although the HIF-2␣ protein can enhance the interaction between the promoter and the 3Ј enhancer, the gain of chromatin interactions associated with other upstream enhancers is an HIF-independent event and specifically occurs in breast cancer cells. These findings reveal a mechanism inducing cancer cell-specific MALAT1 up-regulation under hypoxia.

Discussion
MALAT1 is an oncogenic lncRNA up-regulated in many solid tumors, which facilitates increased cellular proliferation rates, EMT, and metastasis. Few studies have examined the factors that up-regulate MALAT1 during tumor initiation. Several transcription factors have been shown to regulate MALAT1 transcription (11)(12)(13)(14)(15)(16)(17), but these are largely unvalidated in multiple tissues with the exception of p53 and Sp1/Sp3. Environmental stimuli that up-regulate MALAT1 are more wellestablished, such as transforming growth factor-␤ signaling in both normal and cancer cells (37)(38)(39)(40)(41)(42)(43). Similarly, hypoxia consistently up-regulates MALAT1 in a range of normal and cancerous tissues (9,(22)(23)(24)(25)(26); however, the downstream processes from hypoxia that actually result in MALAT1 up-regulation are elusive. In this study, we have demonstrated that hypoxia induces chromatin looping between previously unidentified enhancers located upstream and downstream of the MALAT1 gene in a HIF-independent manner and up-regulates MALAT1 as well as its antisense strand transcript TALAM1 (Fig. 6).
Large scale gene expression changes under hypoxia are regulated through various epigenetic modifying complexes. Interestingly, a large genome-wide increase is consistently noted in the active promoter mark H3K4me3 (32,44,45). Our results are consistent with these findings, as we demonstrate hypoxia causes increased H3K4me3 enrichment by ChIP at several enhancer sites throughout the MALAT1 locus in a cancerspecific manner. The active enhancer mark H3K4me1 also increased at a majority of these same loci in cancer cells, supporting the activity of these enhancers under hypoxia.
The most significant findings we present in this study are identification of novel enhancers at the MALAT1 locus and delineation of the chromatin interactions occurring at these potential enhancers under hypoxia. The Ϫ20 and Ϫ7 kb enhancers appear to be hypoxic-responsive for the cancer condition, as robust chromatin looping was observed between this locus and the 3Ј enhancer under hypoxia in breast cancer cells, but not in nontumorigenic mammary epithelial cells. A recent study also identified the downstream region from ϩ8.7 to ϩ17.2 kb relative to the MALAT1 TSS as a putative enhancer, with the 3Ј end of this region important for Oct4-mediated up-regulation of MALAT1 in lung cancer (46). This chromatin region largely overlaps our proposed 3Ј enhancer, providing support for the importance of this locus in MALAT1 up-regulation in cancer. It would be intriguing to determine whether the Ϫ20 and Ϫ7 kb enhancers are consistent in MALAT1 regulation in different types of cancer, as well.
The MALAT1 3Ј enhancer overlaps the putative promoter of the antisense strand transcript TALAM1, which up-regulates MALAT1 in a positive feedback loop (28). The same genomic regulatory elements are known to have dual functionality, where highly active enhancers can also function as weak promoters (47)(48)(49). This model fits well with endogenous expression of TALAM1, which has on average 300-fold lower expression than MALAT1 (28), and our finding of 3Ј enhancer and MALAT1 promoter interactions in both nontumorigenic and cancerous cells. This implicates the region downstream of MALAT1 as an important and active enhancer in regulation of this locus.
Although local and global changes to chromatin modifications under hypoxia are well-established, few studies have described changes in chromatin 3D structure and organization under hypoxia. At a genome-wide level, most enhancerpromoter interactions for hypoxic-regulated genes are present in normoxic conditions, with only a handful undergoing compartmental change to the active state upon oxygen deprivation, thereby gaining novel chromatin contacts (50,51). In a murine cardiac cell model, hypoxia resulted in global chromatin condensation through rapid histone deacetylation compacting nucleosomes, creating novel interactions but largely inactivating chromatin (52). However, these global, average views do not preclude the possibility of local chromatin looping for up-regulation of specific genes under hypoxia, which we observe for MALAT1. For instance, a hypoxic-specific interaction was observed between two enhancers of solute carrier family 2 member 3 (SLC2A3), which increased SLC2A3 gene expression (53). Together, these data indicate that whereas global chromatin structures of hypoxia responsive genes are poised under

Hypoxia up-regulates MALAT1 via chromatin looping
normoxia, individual gene loci undergo topological changes to facilitate their transcription for hypoxic stress response.
Growing evidence suggests chromatin organization and interactions are altered in cancer and other disease states compared with normal tissue. In breast cancer, recent work has established clear differences in chromatin-chromatin interaction profiles between normal and cancerous cell lines. An enhancer of insulin-like growth factor-binding protein 3 (IGFBP3) had a greater number of long-range chromatinchromatin interactions in MDA-MB-231 compared with human mammary epithelial cells, particularly with the EGFR gene (54). Our results provide another example of different chromatin interactions in cancer versus noncancerous lines, especially in response to hypoxia.
In this work, we have demonstrated that breast cancer cells undergo specific changes within their chromatin structure to facilitate up-regulation of the lncRNA MALAT1. Hypoxia causes relaxation and opening of chromatin at novel MALAT1 enhancers that specifically interact together in breast cancer to up-regulate MALAT1 in a HIF-independent manner. Our study provides a model for both investigation of genes up-regulated under hypoxia in a HIF-independent manner and changes to chromatin structure. It will be interesting to determine whether this looping occurs in other cancer types or if HIF regulates alternative looping in cell lines such as HeLa where MALAT1 up-regulation is HIF-dependent (23,25). We predict numerous looping events remain to be discovered between enhancers and promoters of various hypoxic stress response genes and preventing these loops from forming represents a novel therapeutic strategy for treatment of breast cancer.

Tissue culture and hypoxia
The nontumorigenic, immortalized mammary cell line MCF10A was obtained from the ATCC and was cultured in MEBM mammary epithelial cell growth basal medium (Lonza) supplemented with 10% FBS, 1% penicillin/streptomycin, 20 ng/ml of epidermal growth factor, 0.5 mg/ml of hydrocortisone, and 10 g/ml of insulin. Breast cancer cell lines BT474, MCF7, MDA-MB-231, MDA-MB-453, SKBR3, and T47D were obtained from the ATCC (Manassas, VA). MDA-MB-157 and MDA-MB-361 were gifts from Dr. Robert Sobol. All cancer cell lines were maintained in Dulbecco's modified Eagle's medium (Corning, NY) supplemented with 10% FBS ϩ 1% penicillin/ streptomycin. All cell lines were maintained in a 37°C incubator with 5% CO 2 . Cells were seeded and grown for 48 h under normoxic (21% O 2 ) conditions then either maintained at normoxia or transferred to BD GasPak TM EZ Gas Generating Systems and Supplies:Pouch System (Franklin Lakes, NJ) for hypoxic (ϳ1% O 2 ) conditions for an additional 24 h.

RNA isolation, cDNA synthesis, and qPCR
Total RNA was extracted using Total RNA Isolation (TRI) reagent, according to the manufacturer's instructions (Sigma). cDNA was generated using the SuperScript III Reverse Transcription Kit (Invitrogen), according to the manufacturer's instructions. The reaction was stopped by incubation at 85°C for 5 min, then 2 units of RNase H were added for a 20-min incubation at 37°C. qPCR was performed in triplicate reactions using 2ϫ iTaq Universal SYBR Green Supermix (Bio-Rad) on a Bio-Rad CFX Connect Real-Time System thermocycler. All qPCR were normalized to YWHAZ expression. Primers used are listed in Table S1.

Northern blotting
Five micrograms of total RNA was analyzed by gel electrophoresis in a 1% formaldehyde gel. RNA was transferred to a nylon membrane overnight by capillary action in a 20ϫ SSC buffer. RNA was UV cross-linked and prehybridized with Ultrahyb Hybridization Buffer (Invitrogen). Probes were 3Ј labeled with digoxigenin using the DIG Oligonucleotide Labeling Kit (Roche Applied Science), according to the manufacturer's protocol. Probes and RNA were hybridized overnight with gentle rocking at 42°C. Membranes were washed using the DIG Wash and Block Buffer Set (Roche), according to the manufacturer's instructions. RNA was detected with anti-digoxigen-AP, Fab fragments (Roche), and incubation with disodium 3-(4methoxyspiro {1,2-dioxetane-3,2Ј-(5Ј-chloro)tricyclo [3.3.1.13,7] decan}-4-yl)phenyl phosphate.

Transfection of siRNA and cDNA constructs
Small interfering RNA (siRNA) were purchased from Sigma targeting HIF-1␣ or HIF-2␣ and reconstituted in nucleasefree water. The plasmid HA-HIF2␣-P405A/P531A-pcDNA3 encoding a constitutively active mutant of HIF-2␣ was obtained through Addgene (Addgene number 18956) (30). Transfection reactions were prepared in the same way. siRNA was mixed with Opti-MEM (ThermoFisher), combined with Lipofectamine RNAiMax transfection reagent (Invitrogen) and added to the culture media. The HIF-2␣ construct was mixed with Opti-MEM, combined with Lipofectamine 3000 transfection reagent (Invitrogen), and added to culture media. Both siRNA and construct transfections were incubated at 37°C for ϳ8 h before media change. For HIF knockdown, cultures were grown for 48 h at normoxia, then maintained at 21% O 2 or moved to hypoxia for an additional 24 h prior to protein and RNA isolation and 3C analysis. For HIF-2␣ overexpression, cells were grown for 48 h prior to protein and RNA isolation. siRNA sequences are listed in Table S2.

Protein extraction and Western Blotting
Cells were lysed in an equal volume of 1% Nonidet P-40 buffer (10 mM Tris-HCl, pH 7.5, 100 mM NaCl, 1% Nonidet P-40) with ϫ20 PhosphoStop and ϫ50 protease inhibitor mixture (Roche). Protein was quantified by Pierce Coomassie Protein Assay Reagent (ThermoFisher Scientific), then boiled in a 1ϫ SDS solution. Protein was analyzed by 7-10% SDS-PAGE gel, then transferred to methanol-charged Immobilon PVDF Transfer Membrane (Millipore, Burlington, MA) and blocked in 5% skim milk in 1ϫ TBST prior to overnight incubation with primary antibodies (Table S3). Following washing and secondary antibody incubation at room temperature, protein bands were visualized using Clarity Western ECL Substrate (Bio-Rad).

Hypoxia up-regulates MALAT1 via chromatin looping
quenched with glycine, then lysed with lysis buffer (10 mM Tris-HCl, pH 8, 10 mM NaCl, 0.2% Nonidet P-40, protease inhibitor mixture). The lysate solution was homogenized by 15 strokes with a Dounce homogenizer and then centrifuged to collect nuclei. After washing, nuclei were counted and an equal number (0.5-1 ϫ 10 7 ) for normoxic and hypoxic conditions were collected by centrifugation. Nuclei were resuspended in 0.5 ml of 1.2ϫ New England Biolabs buffer 3.1 and proteins that are not cross-linked to the DNA were removed by adding 15 l of 10% SDS and incubated at 37°C for 1 h followed by quenching by adding 50 l of 20% Triton X-100. Fifty-five l was removed and set aside to determine digestion efficiency (no-digest control). Restriction enzymes BamHI (5,000 units) and BglII (1,000 units) were added and then incubated at 37°C overnight with 950 rpm shaking.
Fifty l of 1ϫ New England Biolabs buffer 3.1 and 100 l of 10% SDS were added and incubated at 65°C for 30 min with 950 rpm shaking. Sixty l was removed for no-ligation control and kept at Ϫ20°C. Chromatin was combined with 1.1ϫ T4 ligation buffer (550 l of 10ϫ T4 ligation buffer, 275 l of 20% Triton X-100, 3.5 ml of water), then incubated at 37°C for 1 h. For ligation, T4 ligase (800 units) was added and incubated at 16°C for 4 h. Proteinase K was added to a final concentration of 100 g/ml and the samples were incubated overnight at 65°C for reverse cross-linking. After RNase A (40 g/ml) treatment, DNA was purified using phenol/chloroform, precipitated by adding sodium acetate, pH 5.2, and washed with 70% ethanol.
The no-ligation control was purified by the Qiagen PCR Purification Kit according to the manufacturer's instructions. The concentration of dsDNA was determined by Qubit 3.0 Fluorometer (Invitrogen). Digestion efficiency was determined by PCR using primers listed in Table S1. Successfully digested 3C samples were centrifuged and washed in 70% ethanol, then air-dried and reconstituted in water. 3C sample interaction frequencies were conducted in triplicate by PCR and products were analyzed on a 2% agarose gel.

Statistical analysis
All experiments shown are the result of at least three biological replicates. Data are presented as the mean Ϯ S.D. with p values calculated by 2-tailed t test. All statistics were calculated using GraphPad Prism software version 7.02 (GraphPad).