Insights from Global Analyses of Long Noncoding RNAs in Breast Cancer

Abstract

Purpose of review

The goal of this review was to compare and contrast the results and implications from several recent transcriptomic studies that analyzed the expression of lncRNAs in breast cancer. How many lncRNAs are dysregulated in breast cancer? Do dysregulated lncRNAs contribute to breast cancer etiology? Are lncRNAs viable biomarkers in breast cancer?

Recent findings

Transcriptomic profiling of breast cancer tissues, mostly from The Cancer Genome Atlas, identified thousands of long noncoding RNAs that are expressed and dysregulated in breast cancer. The expression of lncRNAs alone can divide patients into molecular subtypes. Subsequent functional studies demonstrated that several of these lncRNAs have important roles in breast cancer cell biology.

Summary

Thousands of lncRNAs are dysregulated in breast cancer that can be developed as biomarkers for prognostic or therapeutic purposes. The reviewed reports provide a roadmap to guide functional studies to discover lncRNAs with critical biological functions relating to breast cancer development and progression.

I. Introduction

Breast cancer is the mostly commonly diagnosed cancer in women throughout the world. Nearly 1.7 million are diagnosed and more than 500,000 women die each year from the heterogeneous genetic disease [1]. There are two main histological subtypes—Invasive Ductal Carcinoma (IDC) and Invasive Lobular Carcinoma (ILC) and five main molecular subtypes—Luminal A, Luminal B, Basal-like, HER2-enriched, and normal-like [2–4]. The five molecular subtypes are defined by the PAM50 gene expression signature [5, 6] that largely correlates with the expression of pathological markers. Luminal A tumors typically express both the Estrogen (ER) and Progesterone Receptors (PR), Luminal B are generally ER+/PR-, HER2-enriched mostly contain amplified ERBB2/HER2 growth factor receptor, and basal-like largely do not express ER, PR, or HER2 (triple-negative or TNBC). Targeted therapies have dramatically improved the clinical outcomes of ER+ and HER2+ breast cancer patients, but resistance to these therapies and metastasis still remain deadly killers that are not fully understood. Targeted therapies do not exist for basal-like tumors, nor are there biomarkers to subdivide this group.

Long noncoding RNAs (lncRNAs) are RNAs longer than 200 nucleotides that have no apparent coding potential. This loose definition incorporates RNAs of varying genomic structures, expression patterns, and functions. LncRNAs perform a wide-variety of biological functions including regulation of transcription, translation, proliferation, apoptosis, etc. They perform these functions largely through their interactions—regulated by their structure—to DNA, RNA, and proteins by acting as guides, scaffolds, and decoys [7–9]. During the past three decades, the expressions, functions, and dysregulation of several lncRNAs were described for all phases of breast cancer initiation and progression. Comprehensive reviews detail the proposed roles of these lncRNAs [7, 10, 11]. It is not our intention to rehash those reviews. Instead, Table 1 and the following short section summarize very briefly and broadly the functions of selected lncRNAs in breast cancer biology that were identified through standard molecular biology and genetic approaches, before reviewing more carefully the insights from recent global analyses of lncRNAs in breast cancer. For comprehensive reviews on the functions of lncRNAs in breast cancer, please refer to [7, 10, 11].

Table 1.

Dysregulated lncRNAs in breast cancer

LncRNA	Expression in Breast Cancer	Biology
H19	Upregulated	Promotes anchorage-independent growth, expression correlates with tumor grade, may lead to drug resistance. [17–19, 104–110]
GAS5	Downregulated	Induces apoptosis, low expression associates with poor prognosis. [50–53, 111, 112]
LSINCT5	Upregulated	Enhances proliferation. [38]
HOTAIR	Upregulated	Enhances proliferation and metastasis, activates ligand-independent ER activation and possibly contributes to tamoxifen resistance. [45, 113–116]
XIST	Downregulated	Inhibits cell viability and regulates X chromosome inactivation. [13, 14, 22, 23]
NEAT1	Upregulated	Enhances Proliferation and Induces Apoptosis[35–37]
MALAT1	Upregulated	Promotes proliferation, invasion, migration, tumor progression, metastasis, and elevated expression associates with poor survival of ER-/lymph node negative patients and tamoxifen treated ER+ patients. [8, 39–44]
lncRNA-ARA	Upregulated in Adriamycin resistant cell lines	Induces proliferation, decreases apoptosis, expression inversely correlates with Adriamycin sensitivity[117]
BC200	Upregulated	Prognostic indicator of tumor progression and enhances cell growth [18, 118]
SOX2OT	Upregulated	Inhibits proliferation, enhances anchorage-independent growth, and may be involved in in maintaining a stem cell phenotype [119]
TreRNA	Upregulated	Enhances invasion and metastasis. [120]
UCA1	Upregulated	Enhances proliferation, inhibits apoptosis, and possibly involved in tamoxifen resistance. [121, 122]
NKILA	Downregulated	Low expression associated with breast cancer metastasis and poor patient prognosis. [123]
ZFAS1	Downregulated	Inhibits proliferation and differentiation. [55, 56]
lncRNA-ATB	Upregulated in trastuzumab resistant tumors and cell lines	Promotes trastuzumab resistance, enhances invasion, metastasis, and EMT. [124]
BCAR4	Upregulated	Promotes proliferation, tumorigenesis, metastasis, potentially involved in tamoxifen resistance, and expression associates with poor survival. [125–129]
lncRNA-HIT	Upregulated	Promotes migration, invasion, EMT, and metastasis. [130]
LINK-A	Upregulated in TNBC	Promotes tumorigenesis and glycolysis. [132]
JPX	Downregulated	Limits cell viability, activates XIST. [133]
PANDA	Upregulated	Prevents apoptosis. [134]
SRA	Upregulated	Enhances migration and tumorigenesis. [28–32]
LINC00511	Upregulated	Enriched in TNBC. [33]
91H	Upregulated	Promotes cell proliferation and tumorigenesis. [24–26]
LINC00520	Upregulated	Enhances migration and invasion[46]
Linc-ROR	Upregulated	Enriched in TNBC. Promotes EMT, invasion, metastasis, and multidrug resistance. [47–49]

II. lncRNAs in breast cancer – early evidence of dysregulation and function

Overview

In the early 1990s, the identification and functions of lncRNAs emerged through traditional molecular biology and genetic techniques. Two of the first described—H19 [12] and XIST [13, 14]—have roles in mammary gland development and breast cancer formation. H19 is a classically imprinted gene that is hormonally-regulated [15], expressed highly during pregnancy[16], overexpressed in breast cancer tumors, promotes anchorage-independent growth[17], significantly correlates with tumor grade [18], and may lead to drug resistance[19]. XIST induces X chromosome inactivation in female mammals by recruiting chromatin-modifying complexes to the targeted X chromosome resulting in histone modification and heterochromatin formation [15, 20, 21]. The expression of XIST is commonly reduced in breast cancer tumor tissues and cell lines with a corresponding increase in phosphorylated-AKT levels [22, 23], suggesting that XIST negatively regulates cell viability through inhibition of AKT. Over the past decade several other lncRNAs were shown to be dysregulated in breast cancer and have important biological functions. The next three subsections briefly discuss potential oncogenic and tumor suppressor lncRNAs involved in breast cancer formation or progression.

Upregulated lncRNAs in breast cancer promote proliferation and tumorigenesis

Several lncRNAs that are overexpressed in breast tumors regulate breast cancer biology (Table 1). The antisense compliment of H19—91H—is highly expressed in a subset of breast tumors. 91H promotes cell proliferation and tumorigenesis [24, 25] by preventing histone and DNA methylation on the maternal H19/IGF2 allele leading to increased expression of those two oncogenes [26]. The lncRNA SRA also regulates gene expression, but as a coactivator of ER and PR [27]. At least 470 genes are regulated by hormones in a SRA-dependent manner[28–30] and SRA is necessary for full ERα activity [31]. Overexpression of SRA in the mammary gland of mice leads to epithelial hyperplasia [32]. Interestingly, while it seems SRA functions in breast cancer as a coactivator to ER, it is necessary for full migration of a TNBC cell line, suggesting it also functions independently of ER. A different nuclear receptor, RORγ, regulates LINC00511, which is a highly expressed lncRNA in breast cancer. LINC00511 is enriched in TNBC cell lines and acts as an oncogene in lung cancer [33, 34]. It is unsurprising that lncRNAs involved in breast cancer biology interact with or are regulated by nuclear receptors and other breast cancer players. Another example is NEAT1, which is repressed by BRCA1. NEAT1 localizes to the paraspeckles in the nucleus, is upregulated by hypoxia, induces proliferation, inhibits apoptosis, and high expression correlates with poor survival in breast cancer tumors [35–37]. For other lncRNAs there is relatively little known, like LSINCT5, which is overexpressed in breast and ovarian cancer [38] and enhances proliferation through unknown mechanisms.

Upregulated lncRNAs in breast cancer promote migration, invasion, and tumor progression

MALAT1 is a conserved and alternatively spiced lncRNA [39] that regulates gene expression by altering transcription and post-transcriptional pre-mRNA processing of a large number of genes [40]. MALAT1 is upregulated in breast cancer and promotes invasion and migration of breast cancer cells [41]. The expression of MALAT1 is higher in ER+ and HER2+ tumors, but MALAT1 induces proliferation, tumor progression, and metastasis of TNBC cells, implying a role in all breast cancer subtypes. Further supporting this hypothesis, elevated expression of MALAT1 is associated with poor disease-specific survival of ER- and lymph node negative patients [40] and poor recurrence-free survival in tamoxifen treated ER+ patients [42]. Thus, MALAT1 might serve as an important biomarker or target of therapy for these groups. In fact, antisense oligonucleotides targeting MALAT1 subcutaneously injected in mice with established mammary tumors resulted in cystic and poorly metastasizing tumors implying targeting MALAT1 is a potential therapy to limit the spread of disease [43, 44]. HOTAIR is another lncRNA that enhances breast cancer metastasis. HOTAIR represses HOXD—a metastatic suppressor—by inducing H3K27me3 through the recruitment of PRC2 (25,26). Much like other breast cancer related lncRNAs, HOTAIR is repressed directly by ER. When HOTAIR is upregulated it promotes ligand-independent ER activation and contributes to tamoxifen resistance [45].

Other oncogenes also regulate lncRNAs in breast cancer cells. LINC00520 is controlled by SRC, PIK3CA, and STAT3 and enhances migration and invasion [46]. LINC00520 is highly elevated in breast cancer cells with highest expression in basal-like subtype. Likewise, Linc-ROR is a lncRNA upregulated in breast tumors that is enriched in TNBC. Linc-ROR promotes epithelial-mesenchymal transition (EMT), invasion, metastasis, and multidrug resistance in breast cancer in part by acting as a miRNA sponge [47, 48]. Additionally, estrogen metabolites downregulate NRF2 expression resulting in an upregulation of linc-ROR expression suggesting Linc-ROR may contribute to estrogen-mediated breast tumorigenesis as well as TNBC [49].

Downregulated lncRNAs in breast cancer regulate proliferation, apoptosis, and differentiation

In addition to XIST there are other tumor suppressor-like lncRNAs with downregulated expression in breast cancer (Table 1). Gas5 is a pro-apoptotic snoRNA host gene with reduced expression in breast cancer samples [50]. Gas5 induces apoptosis by acting as a mimic to the glucocorticoid receptor [51] and contains a stem-loop sequence that serves as hormone response element mimic in breast cancer [52]. Low expression of Gas5 is associated with poor prognosis in breast cancer likely due to reduced apoptosis stimulated by chemotherapeutic agents [53, 54]. Zfas1 is another snoRNA host gene that acts as a tumor suppressor. Zfas1 is highly expressed and regulated during mammary gland development of the mouse, and the human orthologue is down-regulated in breast tumors [55]. Reduction of Zfas1 expression enhanced proliferation and differentiation of the HC11 mammary epithelial cell line [55] potentially through interaction with ribosomes and by regulating RPS6 phosphorylation [56].

III. Global analyses reveal widespread dysregulation of lncRNAs in breast cancer tissues and cells

Overview

Early studies demonstrated that lncRNAs are critical to the biology of breast cancer cells and that other undiscovered lncRNAs likely play equally critical roles in breast cancer development and progression. The advent of global expression analyses, especially those applying next-generation sequencing, unleashed a plethora of information about the expression, dysregulation, and function of novel lncRNAs in breast cancer initiation and development. However, discrepancies in specific results exist between studies due to unstandardized analyses and annotations. RNAseq data can be analyzed to allow novel isoform/gene discovery or can simply quantify the expression of RNAs to predefined gene annotations. While there is overlap between annotations from different databases (RefSeq, GENCODE, BodyMap) there are numerous differences due to how they were generated and curation requirements. This section details the early evidence of pervasive transcription through recent studies utilizing novel and publically available datasets. We comment on the similarities and differences in the approaches and how they affect the discoveries of these studies. Table 2 summarizes key prognostic and functional lncRNAs identified through the global analyses.

Table 2.

Additional dysregulated lncRNAs in breast cancer identified through recent transcriptomic screens

lncRNAs	Notes
LINC00657, LINC00346, LINC00654, HCG11	High expression of all four predicted poorer overall survival. Knockdown of LINC00657 decreased proliferation.
LINC00705, LINC00310, LINC00704, LINC00574, FAM74A3, UMODL1-AS1, ARRDC1-AS1, HAR1A, LINC00323	High expression of all nine associated with poor disease free survival.
HOTAIRM1, LINC00340	Enriched in basal breast tumors.
HOTAIR	Enriched in HER2 subtype
TOPORS-AS1, RP11-35G9.3	High expression is associated with good patient outcome
BCAL8	Enhances proliferation, anchorage-independent growth, and tumor growth in nude mice.
CTD-2015G9.2, CTD2527121.15, LINC00393, LINC001198, RP11-10A14.5, RP11-19E11.1	Enriched in basal-like breast cancer.
LINC01297 RP11-303E16.2	In ER+ tumors, LINC01297 is highly expressed and RP11-303E16.2 is downregulated. Together their expression can divide patients into ER+ and ER-.
CYTOR	Correlates with poor survival and is critical for proliferation and migration
HIF1A-AS2, AK124454	Part of an mRNA-lncRNA signature that predicts recurrence-free survival and pathologic complete remission after taxane-based neoadjuvant chemotherapy. Both lncRNAs promote proliferation, invasion, and paclitaxel resistance.
BRCAT49	Breast cancer and lineage-associated lncRNA located near a SNP associated to breast cancer in several GWASs.
DCCAM-AS1	Highly expressed in breast cancer tissues. Strongly induced by estrogen. Enhances proliferation, invasion, anchorage-independent growth of cell lines and xenograft growth and metastasis in mice.
lncRNA152, lncRNA67	Estrogen regulated lncRNAs that are overexpressed in breast cancer tumors. Both are necessary and sufficient for proliferation of certain breast cancer cell lines.
LIMT	Prevents migration, invasion, and metastasis. Depleted in basal-like and HER2 positive tumors. Loss of LIMT is correlated with a decrease in survival.

Evidence of pervasive transcription

Commonly used gene expression microarrays do not contain probes for most lncRNAs—an obvious problem for using previously published data for lncRNA discovery. However, the use of genome tiling arrays provided early hints of pervasive transcription of the human genome. Bertone et al. constructed a series of 134 high-density tiling arrays that represented 1.5 Gb of non-repetitive sequence of the human genome and discovered 10,595 transcripts from poly-A selected liver RNA not detected by other methods [57]; early evidence that novel transcripts were awaiting discovery.

Next generation sequencing of total RNA (RNAseq) does not rely on predesigned probes and thus allows for easier discovery of novel RNAs. With this technology several groups demonstrated that over 70% of the genome is transcribed even though only 1.5% of the transcripts encode for proteins [58–61]. But the questions remained if the novel transcripts were real and functional or if they were the product of a leaky transcriptional system or imperfect detection and computational technologies. To address this, Guttman et. al used a multi-omic approach to define a conservative cohort of 1,675 large intergenic noncoding RNAs (lincRNAs) in mouse cells that are 1) expressed, 2) multi-exonic, 3) in areas of regulatory histone marks (H3K4me3 and H3K36me), and 4) do not overlap protein coding or miRNA genes [62]. Regulatory histone marks and splicing of transcripts imply that these lincRNAs are functional and are not merely transcriptional noise; however, it does not preclude single-exon RNAs or those without active histone marks from being functional. The ENCODE Consortium used RNAseq data from 15 different cell lines without the conservative requirements of Guttman et al. They reported that 74.7% of the human genome is covered by primary transcripts, but that no individual cell line expressed more than 56.7% of the known transcripts. Work from ENCODE is continuously updated and the latest version of Human GENCODE (v.25) as of December 2016 (https://www.gencodegenes.org/stats/current.html) [58] has 15,767 lncRNA genes and another 14,650 pseudogenes. This means the number of long noncoding genes is nearly double that of protein-coding genes, thus ushering in the excitement of new candidates to understand, diagnose, and treat human disease including breast cancer.

Insights from The Cancer Genome Atlas

Concurrent to ENCODE, The Cancer Genome Atlas (TCGA) was sequencing numerous cancers with the goal of understanding the molecular changes leading to tumor formation. TCGA is a collaborative that published multilevel-omic data on thousands of tumors in over 33 different cancer types. For breast cancer alone, there are publically available RNAseq expression data from 1098 tumors. The processed expression data that are publically available were calculated using MapSplice followed by RSEM with a generic annotation file to hg19 that is similar to the UCSC Gene standard track (December 2009 version) [63, 64]. This pipeline does not allow for novel gene discovery and includes the expression of 20,532 total genes and 73,671 transcripts. This data was mined and reanalyzed by several groups to determine the expression and function of lncRNAs in breast cancer biology and their potential utility as biomarkers, molecular classifiers, or therapeutic targets.

Through the cBioPortal [65] Liu et al.[66] examined the 2730 annotated lncRNAs in the fully processed TCGA data. They found that 577 of these lncRNAs have altered RNA expression, copy number variations, or mutations in at least 1% of the breast cancer tumors. The overexpression of four lncRNAs—LINC00657, LINC00346, LINC00654, and HCG11— predicts a poorer overall survival and the combined overexpression of another 9 lncRNAs— LINC00705, LINC00310, LINC00704, LINC00574, FAM74A3, UMODL1-AS1, ARRDC1-AS1, HAR1A, LINC00323— is associated with poorer disease free survival, suggesting that lncRNAs are viable biomarkers in breast cancer. However, because TCGA data is relatively new and the five-year relative survival rate of women diagnosed with localized breast cancer is 98.5%[67] associations with disease outcomes based on TCGA data may still be premature. To examine the functional significance of a lncRNA in the new signature, LINC00657 was knocked out with CRISPR/Cas9 in MCF7 and MDA-MB-231 breast cancer cell lines. The knock-out cell lines proliferated slowly and formed smaller and fewer colonies in a clonogenic assay. This indicates that LINC00657 is an oncogene that is overexpressed in a subset of breast cancer tumors.

Another comprehensive analysis of lncRNAs in breast cancer using TCGA data was performed on 658 non-normal-like IDC samples with defined molecular subtypes by PAM50 scores [68]. Using aligned data from TCGA, ‘raw’ sequencing reads were obtained by back converting alignment to FASTQ files. The reads were then realigned with MOSAIK[69] and transcripts were quantified with HTSeq and DESeq [70] using GENCODEv.15 that contained 13,159 annotated lncRNAs. After filtering for lncRNAs with expression near or below levels of detection (FPKM greater than or equal to 1 in at least 10% of samples), 1,623 lncRNAs were examined further. GREAT analysis [71] demonstrated that the expressed lncRNAs were genomically positioned near protein coding genes (PCGs) with established roles in breast cancer such as ESR1, GATA3, and FOXA1. Additionally, 937 of these lncRNAs have histone marks associated with enhancer regions (H3K37ac and H3K4me2) in human mammary epithelial cell lines, suggesting the lncRNAs are cis-acting regulators of expression of genes that drive breast carcinogenesis. Unsupervised hierarchical clustering of breast cancer samples by the expression of lncRNAs highly correlated with PAM50 scores and defined molecular subtypes of breast cancer. Notably, there is a clear distinction between basal and ER+ subtypes with 122 lncRNAs, including HOTAIRM1 and LINC00340, overexpressed (FC>=2; FDR<0.05) in the basal subgroup. HOTAIR and 57 other lncRNAs were overexpressed in the HER2-enriched subtype and 96 lncRNAs were upregulated in the luminal subtypes. In contrast to the previous study only 6 lncRNAs were associated with patient outcome (FDR < 0.05) including two— TOPORS-AS1 and RP11-35G9.3—that are associated with good outcome [68].

Yan et al. used a different computational pipeline to perform a comprehensive characterization of lncRNAs of 5,037 tumors across 13 cancer types—including breast cancer— in TCGA. They used Partek to calculate expression levels from the downloaded alignment (.bam) files to GENCODE V.18 that contains 13,562 lncRNA genes and 14,181 pseudogenes. All lncRNAs were considered to be expressed if their 50^th percentile RPKM was greater than 0 or if the 90^th-percentile RPKM value was more than 0.1. Using this criteria they state that 4,329 lncRNAs are detectable in breast cancer—nearly 3 times more than the aforementioned studies using the same data—and that 1,150 are differentially expressed between the 990 breast cancer tumors vs. the 106 normal samples examined [72]. Their results agree with ENCODE data that lncRNAs are expressed or dysregulated in more cell type [60]—or in this case cancer type—specific manner than PCGs. They further found that dysregulated lncRNAs are located in somatic copy number variable regions, their promoters have altered DNA methylation patterns, and they contain cancer-associated SNPs. A cluster analysis of expressed lncRNAs from 817 breast tumors highly correlated with molecular subtypes based on PAM scores—much like the previous studies—and had a high correlation with clinical subtypes [72]. Consistent with the Su et al.[68] report the basal-like subtype formed the strongest cluster apart from all other subtypes. Using an entropy based metric they demonstrated that the expression of lncRNAs are more specific than PCGs for defining subtypes [72]. Because lncRNAs are expressed and dysregulated in a cancer and subtype specific manner and because they are highly structured and potentially more stable in blood and urine, it is possible that lncRNAs will be better biomarkers for diagnosis and prognosis than PCGs. The authors argued that cancer driver lncRNAs can be found by using their comprehensive characterization of lncRNAs with a functional screen. As a proof of concept they designed a siRNA screen of 19 lncRNAs that had cancer specific genomic or epigenomic alterations and that correlated with patient survival. The screen and subsequent assays revealed that reduction of BCAL8 reduces proliferation of breast cancer cell lines, anchorage-independent growth, and tumor growth when injected into nude mice. Using a Guilt-by-association (GBA) analysis [73] they showed that BCAL8 is co-expressed with cell cycle genes at the RNA and protein level [72]. Additional lncRNAs were not examined, but the data and pipeline are a straightforward approach to discovering other lncRNAs that regulate important biological phenotypes of breast cancer cells.

Beyond TCGA

Open data allows for increasing the power of studies through the combination of datasets. TCGA breast cancer data are a resource that have been integrated with other novel and publically-available datasets. Iyer et al. curated 7,256 RNAseq libraries (including 5,847 from TCGA) from tumors, normal tissues, and cell lines and used an ab initio assembly methodology [74–76]—that allows for novel gene discovery—to define the MiTranscriptome compendium including 58,648 expressed lncRNAs, 79% of which were previously unannotated [59]. In breast cancer alone they discovered 1,115 and 134 lineage and cancer specific transcripts respectively. Consistent with other studies, the newly discovered lncRNAs were expressed at lower levels than PCGs and overlapped disease-associated SNPs. BRCAT49 is a previously uncharacterized breast cancer and lineage specific lncRNA that is near a SNP associated to breast cancer in several GWASs [77–81]. This suggests that analyses allowing for novel lncRNA detection have more power to identify functional players in breast cancer biology than analysis techniques that simply align to previously annotated genes. Extending on this, Niknafs et al. identified 63 potentially estrogen-regulated, breast cancer lncRNAs by intersecting lncRNAs upregulated in tumor vs. normal and ER+ vs. ER-tumors [82]. This screen identified DCCAM-AS1 as an estrogen regulated lncRNA that is overexpressed in breast cancer. Silenced expression of DCCAM-AS1 decreased proliferation, invasion, anchorage-independent growth of tested breast cancer cell lines and reduced xenograft growth and metastasis when injected into mice. Overexpression of DCCAM-AS1 promoted growth in the absence of estrogen indicating a possible role in tumor progression and tamoxifen resistance [82]. DDCAM-AS1 was found by another group to be the most abundant lncRNA of 133 that are regulated by ligand-independent ERα activity, further suggesting a role in tamoxifen resistance [83].

Bradford et al. combined TCGA data with RNA-Seq data from 69 breast cancer patients that they generated at Huntsman Cancer Hospital [84]. For consistency, they reanalyzed the TCGA data through the same pipeline they applied for their new samples—StarAlign followed by Cufflinks [74–76] using the GRCh38 genome build with GENCODEv22 gene annotation. They only analyzed a set of 6062 lncRNAs that did not overlap other genes. This is critical as TCGA RNAseq data is not stranded, meaning individual reads can align perfectly to lncRNAs and PCGs when they are orientated in an antisense fashion (Figure 1a). User decisions and algorithm design decide how to treat ambiguous reads. They can be discarded, assigned completely to one gene, or divided in a variety of ways. This choice is often not listed in the methods, but determines the final expression of both antisense genes. If reads are divided, it is possible that RNAseq data will incorrectly state that both antisense genes are expressed.

Figure 1. Difficulties associated with studying lncRNAs.

(A) Unstranded RNAseq reads can be ambiguous when mapping to antisense lncRNAs. (Top) Annotations of lncRNA ZEB1-AS1 (blue) and two ZEB1 PCG isoforms (red; cropped for simplicity). Arrows indicate the direction of transcription. (Bottom) RNAseq reads are accurately assigned when they align to specific exons of the antisense pair, but are ambiguous when they map to the shared exons. (B) Annotations between databases vary. Annotation of SNHG15 from three commonly used databases— GENCODE, RefSeq, and BodyMap—are shown. BodyMap does not include an annotation for SNHG15. (C) LncRNAs are not highly conserved. The conservation of SNHG15 is shown with black bars representing sequence conservation between different species. Note that SNHG15 is only conserved among primates.

Like others, Bradford et al. demonstrated that highly and variably expressed lncRNAs were sufficient to separate breast cancer tumors into their molecular subtypes. Further, they defined a list of six lncRNAs that are significantly overexpressed in basal-like compared to other breast cancer subtypes—CTD-2015G9.2, CTD2527121.15, LINC00393, LINC001198, RP11-10A14.5, RP11-19E11.1[84]. Note the lack of overlap of enriched lncRNAs between studies. This is likely due to the different annotations and methods of analysis used by the independent groups. Interestingly, using a GBA method and integration with cell line expression data in the Cancer Cell Line Encyclopedia [85] they discovered that a high proportion of a specific cell type in the tumor microenvironment caused some tumors to not cluster well into their corresponding subtypes [84]. This suggests that the expression of lncRNAs can potentially be used to study tumor heterogeneity and define cell-type composition. Accordingly, they defined certain lncRNAs as putatively associated with breast cancer stromal and immune cells.

TCGA data has also been used as a validation set. Grembergen et al. [86] repurposed Affymetrix Human Genome U133 Plus 2.0 array data from 823 breast tumors and 172 normal samples across a collection of 7 data sets downloaded from Gene Expression Omnibus (GEO) [87] by BLASTing probes against the LNCipedia database v.2.1 [88]. Unfortunately, repurposing probes potentially introduces the same bias as using unstranded RNAseq for calling expression of antisense lncRNAs. Nevertheless, 3053 annotated lncRNAs had corresponding probes on the array. Of those, 215 were aberrantly expressed in at least 10% of the tumors. The expression of this set of lncRNAs clearly distinguished breast tumors from normal samples and the signature was validated with TCGA expression data. In agreement with all other studies, the expression of lncRNAs separated ER+ (high expression of LINC01297 and low expression of RP11-303E16.2) and ER-tumors and correlated well with molecular subtypes defined by PAM50. With functional assays, they demonstrated that high expression of CYTOR—one of the dysregulated lncRNAs—correlated with poor survival and is critical for proliferation and migration [86].

Two other studies used repurposed microarray data to define lncRNA prognostic signatures. Jiang et al. performed a prospective observational study that demonstrated an integrated mRNA-lncRNA signature had better prognostic value to predict recurrence-free survival and pathologic complete remission after taxane-based neoadjuvant chemotherapy than clinicopathologic parameters in a cohort of 275 TNBC patients [89]. However, this signature was limited by the number of probes covering lncRNA loci (195 lncRNAs were differentially expressed between normal and tumor) on the Affymetrix Human Transcriptome Array 2.0 implying that this signature can be improved if developed with RNAseq data. Functionally, they confirmed that the two lncRNAs in the signature—HIF1A-AS2 and AK124454—promote proliferation, invasion, and paclitaxel resistance [89].

The other was a retrospective study of 164 banked frozen tumors that used repurposed Agilent microarray data (2811 lncRNAs covered) and showed that a lncRNA expression profile could predict metastasis independently of traditional prognostic markers [90]. It is unclear if the lncRNA expression signature is better than a PCG or lncRNA-PCG hybrid classifier, but it is further evidence that lncRNAs are viable biomarkers to define molecular subtypes of breast cancer and to predict recurrence.

Novel lncRNAs defined by regulation by breast cancer driver genes

Other novel breast cancer associated lncRNAs have been discovered through analysis of cell lines treated with estrogen [91] and EGF [92]—a hormone and ligand respectively that drive subsets of breast tumors. One study identified 1,888 lncRNAs in MCF7 cells, 700 of which were not previously annotated in current databases (GENCODE, UCSC, RefSeq, BodyMap, MiTranscriptome). They combined RNAseq data following 3 hours of estrogen treatment with global run-on sequencing (GRO-seq) [93] data at several time points after treatment [91]. This “lncM” set only includes lncRNAs with FPKM > 1 that have evidence of a primary transcript, as determined by GRO-seq. Therefore, this is a conservative estimate of the number of lncRNAs expressed in MCF7 cells. Certainly expanding the time points following estrogen treatment or repeating in other breast cancer cell lines will aid in the discovery of additional breast cancer lncRNAs. On average the lncM lncRNAs were more nuclear and less stable than other lncRNAs or PCGs, which is likely how GRO-seq improved detection. Estrogen regulated the expression of 531 lncM genes indicating they may be important for breast cancer cell biology at least in ER+ tumors. Like all previous studies, the lncM genes were able to cluster normal from cancerous samples and were as effective as PCGs in determining the molecular subtypes. They characterized two lncRNAs—lncRNA152 and lncRNA67—that are regulated by estrogen, have elevated expression in breast tumors, and have expression that is correlated to PCGs associated with cell-cycle regulation. They demonstrated that both of these lncRNAs are necessary and sufficient for proliferation of certain breast cancer cell lines and knockdown of either altered the expression of many cell cycle genes [91]. Interestingly, estrogen signaling downregulates lncRNA152 even though both estrogen and lncRNA152 drive proliferation, suggesting a negative feedback loop that was not explored further.

The EGF study analyzed expression of REFseq annotated lncRNAs with probes on the Agilent SurePrint G3 Human microarray following EGF treatment of the immortalized but non-transformed MCF10a breast epithelial cell line [92]. They reported 346 lncRNAs were significantly (greater than 1.5 FC) altered at a minimum of one time point. Eleven of these lncRNAs correlated with shorter overall and relapse-free patient survival in the METABRIC dataset, suggesting they are involved in breast cancer etiology or are preferentially expressed in HER2+ or basal-like subtypes—as these subtypes have poorer prognosis. Loss of function and gain of function analyses demonstrated that one of the clinically correlated lncRNAs—LIMT (LINC01089)—prevents migration, invasion, and metastasis. LIMT is depleted in basal-like and HER2+ tumors and loss of expression correlates with decreased survival [94]. Finally, another study reported 1,382 lncRNAs were differentially expressed (FC> 2.0; FDR<0.05) in 7 pairs of HER2+ tumors versus adjacent normal tissue [95]. It is unclear what annotation was used, but it would be interesting to determine if these lncRNAs are enriched in HER2+ tumors from TCGA or are regulated by EGF. Likewise, examining lncRNA expression following activation of other breast cancer driver genes or mutations such as PIK3CA or IGF1R will certainly identify more breast cancer associated lncRNAs.

IV. Conclusions

The vast majority of the genome is transcribed even though less than 2% of it encodes for proteins. In fact, there are as many lncRNAs as there are PCGs, but the functions of most are unknown. Over 80% of GWAS SNPs associated with disease are in noncoding regions [96] and early studies revealed that lncRNAs like H19, XIST, HOTAIR, and MALAT1 (Table 1) are critical regulators of breast cancer cell biology. Recent transcriptomic analyses followed by targeted functional validations demonstrate that a large number of novel lncRNAs are dysregulated in breast cancer tissues and play important roles in breast cancer cell biology. Clearly the choice of computational pipeline and annotation database used caused discordant results, but all studies agree that the expression of lncRNAs can be used as biomarkers to define breast cancer subtypes and potentially predict recurrence and survival. lncRNAs are exciting potential biomarkers in breast cancer because 1) they are more cell-type and cancer-type specific than PCGs, 2) the functional unit is easily and directly quantifiable, 3) they are highly structured and stable allowing for potential detection in the blood and urine, 4) they are dysregulated in breast cancer subtypes, and 5) the lncRNA PCAS1 is FDA approved as a biomarker for prostate cancer–thus precedent is set [97].

However, there are many obstacles to studying and utilizing lncRNAs as biomarkers or targets of therapy. 1) Most lncRNAs were recently identified in large transcriptomic studies, but have not been functionally characterized. Large siRNA and CRISPR screens [98] are beginning to demonstrate that many lncRNAs have critical biological roles, but more direct functional studies and mouse models like those of Sauvageau et al. [99] are necessary to prescribe functions to the vast majority of lncRNAs. 2) The annotations between the large databases are inconsistent (Figure 1b) largely because most are annotated based on short RNAseq reads and not on full length transcripts. New sequencing technologies such as single molecule, real-time sequencing [100](Pacific Biosciences) that use long reads have the potential to improve annotations if they become cheap enough to be widely used. 3) Understanding mutations and SNPs in lncRNAs is more difficult than PCGs because we cannot infer function based on reading frames. However, new techniques such as PARIS [101] and SHAPE [102] are defining secondary structures of lncRNAs at a global level. It is possible that once common secondary structures are defined and prescribed functions then we will be able to stratify the significance of a mutation based on disruption of these structures. 4) Because they were recently identified and were not on microarray chips, there are few publically available datasets to mine. Even TCGA data that measured gene expression with RNAseq was not stranded, meaning most antisense lncRNAs cannot be accurately measured because single reads map equally well to the gene on either strand (Figure 1a). 5) Many lncRNAs are unique to primates [103] making functional genetic studies in lower organisms nearly impossible (Figure 1c).

Despite the challenges, new technologies have exponentially increased our understanding of lncRNA expression, dysregulation, and function in breast cancer initiation and progression. With more sequencing under different experimental and developmental conditions new lncRNAs will be discovered and with them excitement will grow for the use of lncRNAs in precision medicine to diagnose and treat breast cancer patients.

Abbreviations

lncRNA

long noncoding RNA

PCG

Protein coding gene

GWAS

Genomewide Association Studies

RNAseq

RNA sequencing

TCGA

The Cancer Genome Atlas

ER

Estrogen Receptor

PR

Progesterone Receptor

TNBC

Triple-negative Breast Cancer

SNP

Single Nucleotide Polymorphism

EMT

epithelial-mesenchymal transition