Uncovering the roles of long noncoding RNAs in neural development and glioma progression

Abstract

In the past decade, thousands of long noncoding RNAs (lncRNAs) have been identified, and emerging data indicate that lncRNAs can have important biological functions and roles in human diseases including cancer. Many lncRNAs appear to be expressed specifically in the brain, and the roles of lncRNAs in neural stem cells (NSCs) and brain development are now beginning to be discovered. Here we review recent advances in understanding the diversity of lncRNA structure and functions in NSCs and brain development. NSCs in the adult mouse ventricular-subventricular zone (V-SVZ) generate new neurons throughout life, and we discuss how key elements of this adult neurogenic system have facilitated the discovery and functional characterization of known and novel lncRNAs. A review of lncRNAs described in other NSC systems reveals a variety of molecular mechanisms, including binding and recruitment of transcription factors, epigenetic modifiers, and RNA-splicing factors. Finally, we review emerging evidence indicating that specific lncRNAs can be key players driving glial malignancies, and discuss next steps towards an in vivo understanding of lncRNA function in development and disease.

Introduction

The human genome transcribes many thousands of lncRNAs – transcripts >200 nucleotides long with no evidence of protein coding potential, and it is now clear that lncRNAs can have critical biological functions and roles in human disease [1–4]. In particular, aberrant lncRNA expression may underlie some of the most devastating human neurological disorders including Alzheimer’s disease [5], schizophrenia [6, 7], developmental delay and autism [8, 9]. However, very few lncRNAs have been characterized in terms of function. This knowledge gap is a critical barrier to progress in this emerging field. New detailed studies of individual lncRNAs are vital to understand how this class of noncoding transcript can contribute to disease and/or serve as therapeutic targets [10].

Before high-throughput sequencing technologies, very few lncRNAs were known, and many fundamental discoveries were made by their individual study. For instance, the lncRNA Xist was discovered in the early 1990’s, and an elegant combination of in vivo and in vitro studies demonstrated its critical role in X-chromosome inactivation [4, 11] and new insights into its mechanism of action [12–14]. In the past decade, many more lncRNAs have been identified. There are over 50,000 human lncRNAs [15–17], and similar catalogs have been generated from various mouse tissues and model organisms [18–22], leading to novel insights into their genomic structure and patterns of expression. While there has been a remarkable surge in their bioinformatic characterization, there is comparatively little data regarding the developmental and molecular function of individual lncRNAs, especially in terms of neural development. For the field of lncRNA biology, the challenge now is to (1) select specific lncRNAs for further study, (2) determine their roles in vivo, and (3) understand their mechanisms of action at the molecular level.

Annotation via the GENCODE consortium suggests that 40% of differentially expressed lncRNAs are specific to the brain [17]. Even within the brain, lncRNA expression patterns are noted to be particularly region-specific [22, 23] and dynamically expressed during neural differentiation in vitro as well as NSC populations in vivo [22, 24, 25]. Here, we describe the use of the V-SVZ to characterize lncRNAs with roles in neurogenesis, including the lncRNA Pnky, which plays a unique role in adult and embryonic NSC lineages. We also review the lncRNAs that have been characterized in other NSC populations and their putative mechanisms of action. Finally, we discuss the potential roles for lncRNAs in the transformation and propagation of central nervous system tumors, and suggest therapeutic strategies informed by careful study of lncRNA biology.

The V-SVZ: a tractable system for molecular-genetic studies of NSC differentiation

In the developing brain, most NSCs are spatially and temporally ephemeral in nature, continually changing their location and developmental potential as the brain grows in term of both size and anatomic complexity. In contrast, NSCs in the adult neurogenic germinal zones are harbored in relatively stable cellular niches and generate a specific subset of neurons for the life of the animal. Furthermore, because adult neurogenesis in the mouse can be studied over longer time periods (e.g., months), it is possible that subtle phenotypes observed in the short-term studies of embryonic neurogenesis might be much more prominent in systems of adult neurogenesis. Thus, the reduced developmental complexity and increased duration of adult neurogenesis provides certain experimental advantages for the study of fundamentally new biological mechanisms.

In the adult mouse brain, several thousand NSCs are located in the V-SVZ of the lateral ventricle [26, 27]. (Fig. 1A, B). Throughout adult life, V-SVZ NSCs (B1 cells) give rise to transit-amplifying cells (C cells), which generate neuroblasts (A cells) that migrate to the olfactory bulb (OB) [28–32] where they differentiate into interneurons (Fig. 1C). Like radial glia – the NSC population of the embryonic brain – B1 cells have many glial cell characteristics [28, 33, 34], including expression of the glial-fibrillary acidic protein (GFAP), glutamate aspartate transporter (GLAST), and brain lipid binding protein (BLBP). In the embryo, radial glia have distinct regional identities, generating different populations of neurons related to the “geographic” location of NSC population in the developing brain. Similarly, B1 cells located in geographically distinct regions of the ventricle wall generate specific interneuron subtypes, and these differences in NSC regional identity appear to be cell-intrinsic [35, 36]. Throughout development, some radial glia generate astrocytes and oligodendrocytes. Indeed, B1 cells are also capable of generating astrocytes and oligodendrocytes [37–39]. Thus, while providing the advantage of generating neurons for long periods of time, B1 cells of the V-SVZ exhibit many characteristics fundamentally important to NSCs in the developing brain.

Figure 1. The V-SVZ lineage.

(A) Coronal view of V-SVZ. (B) Enlarged view of V-SVZ: Type B cells (blue) can contact the ventricle with a thin process extended between ependymal cells (white); type C cells (green) are transit-amplifying cells that give rise to migratory type A cells (red). (C) Sagittal view showing paths of migratory neuroblasts (red, type A cells) to the OB. (D) Schematic of NSC lineages. (E) V-SVZ NSC culture in self-renewal (left) and differentiation (right) conditions. NSCs are GFAP+ (stained in green). Neuroblasts stain for marker Tuj1 (red).

In vivo, B1 cells can exist in a quiescent or activated state [40, 41]. Activated B1 cells express the epidermal growth factor receptor (EGFR) and GFAP. These activated NSCs undergo asymmetric division for self-renewal and the production of C cells (EGFR+,GFAP−), also called transit-amplifying progenitors (TAPS) [42, 43]. C cells express proneural transcription factor Ascl1 and neurogenic Dlx2 and divide symmetrically approximately three times before becoming migratory A cells (Fig. 1D) [27]. A cells travel along the ventricle wall within cellular aggregates called chains that coalesce into the rostral migratory stream (RMS) [44]. DLX2+ migratory neuroblasts are distinguished from TAPs by their expression of doublecortin (DCX) and polysialylated neural cell adhesion molecule (PSA-NCAM) and CD24 [33, 45]. Upon reaching the olfactory bulb, these neuroblasts migrate out radially from the RMS and differentiate into functional interneurons.

V-SVZ NSCs can be efficiently cultured for molecular and biochemical studies, and when grown as monolayers, these cells recapitulate neurogenesis in vitro (Fig. 1E) [46, 47]. In self-renewal conditions, nearly all cells express NSC marker Sox2, and the majority (60–70%) co-stain for additional NSC markers Nestin and GFAP [48]. Upon differentiation, these cultures generate large numbers of neuroblasts (Tuj1+, DCX+, ~40% of cells), oligodendrocytes (Olig2+, O4+, ~15%), and other astrocytes (GFAP+, Nestin-, ~40%) [22, 46, 47]. Importantly, these cultured NSCs transplanted back to the V-SVZ generate neurons for the OB [36], suggesting that these cultured NSCs retain many key in vivo regulatory mechanisms despite their propagation in vitro.

Over the past two decades, by integrating in vivo and in vitro V-SVZ studies, many different laboratories have been able to elucidate key developmental principles regarding the role of signaling molecules, transcription factors, microRNAs, and chromatin modifiers [49]. Such discoveries have often provided unique insights into mechanistic themes relevant to both embryonic and postnatal brain development [35]. Below, we describe how the V-SVZ has been utilized as a model system in which to make new strides into the emerging field of lncRNA biology.

Long Noncoding RNAs in V-SVZ neurogenesis

While several studies of neural lncRNAs had been performed in ESC-derived NPCs or cultured cell lines [24, 50–54], comparatively less is known about the expression and function of lncRNAs in neural lineages in vivo. As noted above, the V-SVZ harbors NSCs with a relatively well-understood daughter cell lineage that generates new neurons for the OB. To generate a reference transcriptome of lncRNAs expressed in V-SVZ neurogenesis, RNA-seq was performed on the V-SVZ and OB from the adult mouse brain [22]. NSCs are also located in the subgranular zone of the dentate gyrus (DG) [55], and DG was also included in the RNA-seq analysis. This sequencing data of the two major adult neurogenic brain regions was combined with data from ESCs and ESC-derived NSCs to generate a catalog of 8992 lncRNAs. Interestingly, 2108 lncRNA transcripts were uniquely recovered from sequencing data of the adult neurogenic brain regions, suggesting that some lncRNAs underlie the long-term self-renewal and neurogenic capacity of the V-SVZ and DG germinal zones.

To confirm the expression and genomic structure of novel lncRNAs RNA Capture-Seq [56] was used to capture novel transcripts and re-sequence them at increased depth. These efforts both confirmed the expression and structure of novel transcripts and revealed that neural lncRNAs can have complex alternative splice variants. For instance, sequences corresponding to an entire coding gene could be identified within the excised “intron” of some lncRNAs. Conversely, some lncRNAs are found within the intron of specific mRNAs. Future studies may provide evolutionary insights into how such complex genomic structures can arise, and whether these genomic arrangements relate to lncRNA function.

After comprehensive annotation of lncRNAs expressed in the tissue or system of interest, an important next step is to identify candidates for functional studies. One method of approaching this issue is to first filter the large set of lncRNAs for those that are differentially expressed among the cells in the lineage under study. While it is certainly possible that lncRNAs expressed ubiquitously have important developmental functions, differential gene expression has generally been a useful first step in the characterization of coding genes. V-SVZ NSCs and their neurogenic daughter cells can be isolated with immunocytochemistry and fluorescent-activated cell sorting (FACS) [45]. To identify which lncRNAs are differentially expressed in the adult V-SVZ neurogenic lineage, a FACS protocol [45] was employed to prospectively isolate EGFR+GFAP+ activated stem cells, EGFR+GFAP-TAPs, and CD24+ young neuroblasts for lncRNA expression analysis. Using the same custom microarrays, the dynamic expression of lncRNAs during neurogenesis in vitro was also determined and correlated with the in vivo lineage. For instance, lncRNA Dlx1as is increased in the TAPs in vivo and exhibit increased expression after 2 days of differentiation in vitro, when the TAP population is rapidly expanding.

The methylation of specific histone lysine residues correlates with transcriptional activity [57] and several analyses of chromatin state maps and transcript expression indicate that histone modifications correlate with lncRNA expression in a manner similar to that of protein-coding genes [21, 22, 24]. Trimethylation of histone 3 lysine 4 (H3K4me3) at promoter regions correlates with active transcription, whereas trimethylation of histone 3 lysine 27 (H3K27me3) corresponds to repressed loci. Loci enriched for both H3K4me3 and H3K27me3 have been termed “bivalent” and are transcriptionally repressed but “poised” for activation [58]. In ESCs, bivalent coding genes are enriched for key developmental regulators. Given that lncRNAs appear to share transcriptional regulatory mechanisms and strategies with mRNAs, those lncRNAs that are bivalent in ESC and/or NSC populations may also be enriched for those that regulate development. Thus, chromatin-state maps of H3K4me3 and H3K27me3 were generated from V-SVZ NSCs, and these data were merged with similar datasets from mouse ESCs, ESC-derived NPCs, and non-neurogenic fibroblasts. Together, with the in vivo expression analysis of FACS-isolated V-SVZ cells, the in vitro analysis of lncRNA expression dynamics during neural differentiation, these chromatin state maps were integrated into an online resource (http://neurosurgery.ucsf.edu/danlimlab/lncRNA/) that allows for browsing and filtering based on chromatin and expression criteria to generate gene signatures for different stages of neurogenesis in the V-SVZ lineage (Figure 2). This resource was used to identify and begin characterization of several lncRNA transcripts, including Six3os, Dlx1as, and Pnky, discussed in sections below.

Figure 2. Workflow for identification and characterization of lncRNAs active in neural development.

First, unbiased RNA-seq is performed and long noncoding RNA transcripts are reconstructed from these reads. To better characterize gene structure and expression levels, RNA Capture-Seq is performed. LncRNAs (shown here in red) are captured and enriched for further sequencing. ChIP-seq is used to characterize chromatin marks across embryonic stem cells, neural stem cells, and non-neural cell types. Finally, expression profiling is performed using FACS-isolated cells of the V-SVZ lineage. Together these data are integrated to choose candidates for further study. Functional assays performed in vitro include shRNA-mediated knockdown in NSC cultures. To determine lncRNA protein-binding partners, RNA pulldown assays are performed in which biotinylated RNA is mixed with cell extract, followed by pulldown and characterization of bound proteins by mass spectrometry. In vivo models for probing lncRNA function include in utero electroporation of embryonic NSCs, and generation of knockout mouse models.

Integration of genome-wide datasets led to the discovery of Pnky, a 825 bp lncRNA divergently transcribed 2.5 kb upstream of the Pou3f2 locus [48]. Pnky is bivalent in ESCs, and this locus becomes monovalent for H3K4me3 in both ESC-NPCs and V-SVZ NSCs, suggesting its importance as a regulator of neural lineage activation. Pnky is only detected in neural tissues, and this transcript is enriched in the nucleus. Analysis of FACS-purified cells indicates that Pnky is most highly expressed in V-SVZ NSCs and becomes downregulated upon differentiation into neuroblasts.

Short-hairpin RNA (shRNA) mediated knockdown of Pnky in postnatal V-NSC cultures increases neurogenesis by 3–4 fold [48]. Time-lapse imaging of individual Pnky-depleted V-SVZ NSCs demonstrated that the potentiation of neurogenesis relates to: (1) an increase in neuronal lineage commitment, (2) an increase in the number of transit-amplifying divisions, and (3) a decrease in death of neuronal progenitors (Figure 3). This phenotype suggests a developmental function for Pnky that is distinct from that of other lncRNAs with known neurodevelopmental function. For instance, as discussed in sections below, knockdown of lncRNAs TUNA, RMST, Six3OS, and Dlx1as all strongly inhibits neurogenesis [22, 50, 54, 59], suggesting that these lncRNAs are required to positively regulate neuronal differentiation. Similarly, genetic deletion of linc-Brn1b results in decreased proliferation of embryonic cortical intermediate progenitors and decreased neurogenesis [60]. lncRNA Evf2 is required for proper expression of Dlx5/6, which encodes key neurogenic transcription factors, and loss of Evf2 results in defective interneuron production [61–63]. Thus, unlike the aforementioned lncRNAs that appear to potentiate neurogenesis, Pnky appears to “restrain” neurogenesis from NSCs, perhaps serving to control their long-term self-renewal and rate of neuronal production.

Figure 3. Knockdown of the lncRNA Pnky enhanced neurogenesis from V-SVZ NSCs.

Left: Normal lineage progression of neuronal production from wild-type V-SVZ NSCs (blue) to transit amplifying cells (TA, green) to neuroblasts (NB, red). Pnky is expressed highest in NSCs and becomes down-regulated during lineage progression. Right: Pnky knockdown promotes neuronal production through two mechanisms: 1) a greater proportion of NSCs commit to the neurogenic lineage, and 2) TA cells undergo more cell divisions, resulting in a greater total number of cell divisions and an increased number of “generations” per initial progenitor.

Although the mechanisms of action for most lncRNAs is not known, an emerging theme is that lncRNAs interact with cellular proteins to form functional ribonucleoprotein complexes. Mass spectrometry analysis of proteins enriched by biotinylated Pnky transcripts led to the identification of Polypyrimidine tract-binding protein 1 (PTBP1) as a protein that associates with Pnky transcripts in the nucleus of V-SVZ NSCs cells. PTBP1 regulates mRNA transcript levels and pre-mRNA splicing during neuronal differentiation [51, 64, 65]. In V-SVZ NSCs, Pnky and PTBP1 regulate an overlapping set of transcripts as well as alternative splicing isoforms, and epistasis experiments with Pnky/PTBP1 double knockdown cells suggest they work in the same genetic pathway to regulate neurogenesis from V-SVZ-NSCs. These studies highlight the interaction between a lncRNA and RNA-binding protein can regulate key aspect of NSC biology. Below, we discuss lncRNAs that have been identified in other NSC populations, including neuroblastoma cell lines, ESC-derived NSCs, and mouse developing cortex (Table 1).

Table 1.

Summary of lncRNAs active in neural development

lncRNA	Tissue/Cell Line	Effect of knockdown/deletion	Protein binding partner	Refs
Cyrano	Zebrafish embryonic brain and notocord	Microcephaly, tail curling, neural tube defects, loss of NeuroD-positive neurons in zebrafish embryos treated with morpholino		[20]
Dali	N2A mouse neuroblastoma cell line	Decreased neurite growth and size during differentiation	POU3F3 DNMT1	[53]
Dlx1as	V-SVZ transit-amplifying cells and neuroblasts Embryonic interneuron progenitors	Decreased neurogenesis from VSVZ-NSCs. Increased DLX1 expression in adult mouse Dlx1as-depleted brains.		[22, 71]
Evf2	Embryonic interneuron progenitors	Decreased number of GABAergic interneurons in early postnatal hippocampus of Evf-2 depleted brains	DLX1 DLX2 BRG1 MECP2	[61–63, 72]
Linc-Brn1b	Cortical progenitors	Reduction of intermediate progenitors in the developing cortex. Reduced cortical layer II-IV thickness and barrel cortex disruption in knockout mice.		[60, 78]
Megamind/TUNA	mESCs and neural progenitors H9 human ESCs Mouse and zebrafish embryonic brain	Reduction of neural commitment of ESCs and neuronal differentiation Hydrocephalus and loss of NeuroD-positive neurons in zebrafish embryos treated with morpholino	PTBP1 HNRNP-K NCL	[20, 54]
Paupar	N2A mouse neuroblastoma cell line	Cell cycle arrest and increased neurite outgrowth upon differentiation.	PAX6	[52]
Pinky	V-SVZ NSCs Cortical progenitors	Increased production of neurons from V-SVZ NSCs and expansion of TA population. Increased production of neuroblasts and depletion of embryonic stem cell population.	PTBP1	[48]
RMST	H9-derived human neural stem cells	Decreased neuronal differentiation	SOX2 HNRNPA2/B1	[50]
Six3os	V-SVZ NSCs Retinal progenitor cells	Decreased generation of neurons and oligodendrocytes from V-SVZ-NSCS. Decreased production of rod bipolar cells and increase in production of Muller glia from retinal progenitors.	EZH2 EYA1 EYA3 EYA4	[22, 59]

LncRNAs associated with homeobox transcription factors

Many high-throughput sequencing efforts and ISH analyses have identified a set of homeodomain associated opposite strand transcripts (HOSTs) [66]. These transcripts can show either reciprocal or coordinated expression with their protein-coding neighbor in vivo. Some homeobox transcription factors such as Six3, Pax6, the Dlx gene family, and the Pou family of transcription factors are critical in neural development. One such lncRNA, Six3os, can modulate SIX3 function in the developing retina, and physically interacts with chromatin-modifier EZH2 as well as the EYA family of transcriptional coactivators [59]. In V-SVZ NSCs, Six3os is specifically enriched in V-SVZ stem cells and is down-regulated upon differentiation; depletion of Six3os results in a decrease in production of both neuronal and oligodendroglial lineages [22].

DLX2 is a homeobox transcription factor that is required for interneuron development in both the OB [67] and forebrain [68], and its proper expression is critical for postnatal neurogenesis in the V-SVZ [46]. Dlx1 and Dlx2 are oriented in an inverted configuration separated by a 8.3 kb intergenic region with several ultraconserved elements that contain known binding sites for pro-neurogenic transcription factors [69]. The lncRNA Dlx1as is also transcribed from this intergenic region. The transcriptional start site overlaps an ultraconserved enhancer region, while its 3′ end partially overlaps Dlx1 on the opposite strand. This transcript has been described in embryonic development [70], where it demonstrates increased expression in more mature cell types relative to the expression of Dlx1/2. In adult neurogenesis, ISH analysis of brain sections and custom microarray analysis of FACS-isolated V-SVZ cell populations demonstrates that Dlx1as is preferentially expressed in the TAPs and migratory neuroblasts of the RMS and throughout the OB of adult mice. In contrast, Dlx1 and Dlx2 are strongly expressed throughout the V-SVZ, RMS, and OB [22, 23]. Dlx1as knockdown in V-SVZ neural progenitors results in decreased neurogenesis without an apparent loss of oligodendroglial differentiation, in contrast to the results observed with Six3os, whose knockdown affected both neuronal and glial lineages. These data demonstrate that lncRNA depletion from a stem cell population can selectively disrupt the neuronal lineage [22].

A Dlx1as-null mouse has been generated by inserting a poly-A cassette in the first intron of Dlx1as [71]. Importantly, this mouse does not efficiently produce full-length Dlx1as, however transcription can still initiate at the TSS. Dlx1as mutant mice are born at expected ratios, survive to adulthood, and are grossly normal. Dlx1as null brains have a slight increase in Dlx1 as measured by ISH, however there is no obvious increase in interneuron generation. The mild phenotype found in the Dlx1as could be caused by several factors: (1) the premature-polyA strategy did not produce a strong enough depletion of the lncRNA, (2) the act of transcription at the Dlx1as promoter is sufficient for its partial function, (3) this straight knockout of Dlx1as could be developmentally compensated by redundant mechanisms. It is, of course, entirely possible that lncRNA Dlx1as is not required for proper neurogenesis or development in vivo, but additional and/or alternative experimental manipulations of this locus may be required to reveal its potential function.

Similar to the Dlx1/2 locus, the Dlx5/6 locus encodes a lncRNA, Evf2, from an ultraconserved intergenic enhancer region. Initial studies in cell lines suggested that Evf2 forms a complex with Dlx2 and enhances activation of the Dlx5/6 intergenic enhancer [63], consistent with models proposed for Paupar and Six3os. An Evf-2 knockout mouse was generated using a premature poly-A signal strategy [62]. Evf-2-null mice have reduced numbers of interneurons in the embryonic hippocampus, however numbers return to normal in the adult. Nevertheless, electrophysiological recordings of the adult hippocampus revealed loss of synaptic inhibition, demonstrating that an early lack of interneurons cause aberrant circuitry in the adult brain.

In contrast to what would be expected based on in vitro results, in vivo deletion of Evf-2 led to an increase of Dlx5 and Dlx6 levels, similar to what was seen at the Dlx1 locus with Dlx1as knockout. Consistent with in vitro findings, Evf-2 null mice had a loss of DLX2 recruitment to the intergenic enhancer, however they also lacked the recruitment of MECP2, methyl-CpG-binding protein and transcriptional repressor of the Dlx5/6 locus. Further studies revealed that Evf-2 exists in a complex with DLX1 and the BRG1 chromatin-remodeling factor. Evf-2 inhibits BRG1 activity, thereby providing a mechanism through which a lncRNA can exert a repressive effect while forming a complex with transcriptional activators [72]. These data demonstrate that the lncRNA Evf-2 can recruit both activators and repressors to the Dlx5/6 locus, and highlights the importance of using genetic knockout strategies to complement in vitro knockdown experiments in cell culture.

Regulation of DNA Methylation

To further characterize the interaction between MECP2, DLX1/2, and Evf-2, the Khotz laboratory conducted genetic epistasis experiments by crossing the Evf-2-null mouse with Mecp2 −/− and Dlx1/2+/− mutant strains [61]. Mecp2-null mice display a 2-fold increase in Dlx5 transcript levels. This increase in Dlx5 expression is abrogated on a Dlx1/2 +/− background, suggesting that DLX1/2 and MECP2 are antagonistic at the Dlx5/6 locus. Analysis of methylation of the DLX5/6 enhancer revealed that Evf-2 inhibits the methylation of the DLX5/6 enhancer in trans. Taken together, these data suggested a model in which Evf-2 can regulate the methylation state of the Dlx5/6 enhancer, which modulates the recruitment of the methyl-CpG repressor MECP2 and DLX1/2 transcriptional activators. Interestingly, a recent study identified hundreds of transcripts that can bind DNA methyltransferase DNMT1 and negatively regulate DNA methylation [73], and the lncRNA Dali has been shown to bind both its neighbor POU3F3 and DNMT1. These studies suggest a similar mechanism may be utilized by Evf-2 at the Dlx5/6 enhancer.

Transcriptional Activation by lncRNAs in Neurogenesis

Genome-wide expression studies and siRNA screens in embryonic stem cells identified lncRNAs required for the maintenance of pluripotency and repression of the neural lineage [24]. Ng, et al. used custom lncRNA microarrays to identify 35 lncRNAs that are upregulated during neuronal differentiation from human embryonic stem cells (ESCs) [74]. Four of these candidates were chosen for shRNA-mediated knockdown, and depletion of any of the four lncRNAs decreases in neurogenesis. One of the identified lncRNAs, RMST, localizes to chromatin at the promoters of key neurogenic regulators, including Sp8 and Dlx2 [50]. Intriguingly, RMST binds directly to SOX2 as well as RNA-binding protein HnRNPA2/B1. RMST co-occupies loci bound by SOX2, and SOX2 occupancy is lost upon RMST knockdown, suggesting that a lncRNA, perhaps through an hnRNP adapter, can modulate the recruitment of transcription factors to downstream targets.

Paupar is a lncRNA transcribed upstream from Pax6, a key homeobox transcription factor expressed in neural progenitor populations [52]. Paupar associates with chromatin, and depletion of Paupar causes cell-cycle arrest and drives the neuronal differentiation of the N2A neuroblastoma cell line. Genome-wide studies indicate that Paupar and Pax6 regulate a common set of transcriptional targets, and genome-wide analysis of Paupar localization on chromatin demonstrates that it binds hundreds of gene promoters, including key neural differentiation genes. Interestingly, Paupar transcript is enriched at regions containing the DNA-binding motif of Pax6, and further studies demonstrate that Paupar directly binds Pax6, and the lncRNA and transcription factor co-occupy several gene promoters. Unlike the role of lncRNAs in targeting epigenetic complexes, knockdown of Paupar does not affect Pax6 localization; rather, Paupar can enhance transcriptional activation or repression by Pax6. These data suggest that PAX6 recruits Paupar to loci genome-wide, where it then acts to modify Pax6 function through as of yet undefined mechanisms.

Similar to RMST and Paupar, the lncRNA Dali binds to a key neurogenic transcription factor, POU3F3 [53]. Depletion of this lncRNA in the N2A neuroblastoma cell line inhibits proper differentiation into neurons. Genome-wide analysis of Dali occupancy on chromatin indicate that Dali binds more than one thousand genes, many involved in cell cycle regulation and neuronal function. Pulldown of biotinylated Dali revealed a direct interaction with DNA methyltransferase DNMT1 as well as POU3F3, and depletion of Dali results in an increase of DNA methylation at target genes. Taken together, these data suggest a model whereby a lncRNA can form a complex with both a neural-specific transcription factor and chromatin-modifying enzyme to affect gene expression and DNA methylation at distant genomic targets.

Conservation across species of lncRNA function in neurogenesis

Sequencing and annotation of lncRNAs in the zebrafish genome led to the identification of two brain-enriched lncRNAs, Megamind and Cyrano [20]. Depletion of Cyrano during development in vivo results in morphological defects in the head and eyes, and neural tube defects. Embyronic depletion of Megamind leads to small heads with hydrocephalus and loss of neuronal populations throughout the brain. Treatment with lower doses allows survival to adulthood, however these fish exhibit locomotor defects [54]. Remarkably, the developmental phenotypes caused by depletion of megamind or Cyrano can be rescued with the mouse or human homologues of these transcripts, suggesting lncRNAs play a conserved role in development of the nervous system despite their relative lack of primary sequence conservation.

Further study of the mammalian homologue of Megamind, called TUNA, revealed that this lncRNA is required for neuronal differentiation from ESCs [54]. TUNA binds RNA-binding proteins PTBP1, hnRNP K, and NCL. These three proteins exist in a complex that requires RNA to form, suggesting that TUNA can serve as a scaffold for complex assembly. TUNA localizes to chromatin at the promoters of several key neurogenic genes, including SOX2. TUNA depletion causes a failure to recruit HNRNP-K to the SOX2 promoter, and subsequent down-regulation of SOX2 and a loss of neurogenesis. TUNA therefore represents another example of neurogenesis being controlled through the cooperation of a lncRNA and RNA-binding proteins.

LncRNAs in cortical development

Neurogenesis in the cortex follows a stereotyped series of cell divisions and transitions similar to those found in the V-SVZ lineage. Radial glial cells are ventricle-contacting stem cells that reside in the ventricular zone (VZ) and can either self renew or divide to give rise to TBR2+ transit-amplifying progenitors that differentiate and migrate through the SVZ and intermediate zone (SVZ/IZ) [75]. These cells divide once or more to give rise to neurons that will then populate the six cortical layers. The layers are born in an “inside out” fashion, with the inner most layers born at the onset of neurogenesis around E11.5, and the outer layers of neurons being born towards the end of the neurogenic period [76]. Aprea and colleagues used a transgenic fluorescent reporter system to separate and purify proliferating progenitors (PPs), differentiating progenitors (DPs), and newborn neurons (N) from the embryonic cortex [25]. These cells were subject to whole-transcriptome sequencing, and lncRNAs from each of these cell types were identified, including 9 ‘switch genes,’ defined as genes either enriched (on-switch) or depleted (off-switch) in DPs compared to other cell types. Among these 9 switch genes were RMST and Gomafu (Miat), which had previously been shown to play key roles in neural development in other experimental systems.

Gomafu overexpression or knockdown via in utero electroporation of developing cortex causes an expansion of TBR2+ progenitors in the VZ, a decrease in the number of differentiating progenitors, and a decrease of neurons reaching the cortical plate [25]. This loss of neurons is chiefly related to an increase in cell death of neurons in the IZ, suggesting that Gomafu regulates progenitor production and neuronal survival in vivo. Mechanistically, Gomafu can affect the splicing of Wnt7b, a known regulator of progenitor proliferation and differentiation in the embryonic cortex. Further studies are required to fully characterize the differential splicing induced by Gomafu manipulation, but differential splicing of Wnt raises the intriguing possibility that a lncRNA can control cell fate decisions through the splicing of the ligands and possibly receptors of key signaling pathways. Interestingly, further studies have implicated Gomafu in aberrant alternative splicing in neurons seen in schizophrenia [6].

Neurog1 is a neurogenic transcription factor expressed in the developing cortex. Like Dlx1as and Evf-2, utNgn1 is a long noncoding RNA transcribed from a conserved enhancer element near Neurog1 [77]. This transcript’s expression is tightly correlated with Ngn1 expression in vivo, and depletion of this lncRNA in acutely dissected neural progenitor cultures results in a failure to upregulate Ngn1 upon differentiation.

The lncRNA Pnky was also studied in the context of cortical development; Pnky transcript is expressed in the embryonic brain of both mouse and human in the VZ [48]. Electroporation of Pnky knockdown constructs at E13.5 and analysis two days later causes a decrease in the proportion of GFP+ cells in the VZ and a corresponding increase of GFP+ cells in the SVZ/IZ. Further analyses demonstrated that Pnky-knockdown decreases SOX2+ stem cells in the VZ and increases the proportion of SATB2+ young neurons in the SVZ/IZ. These data indicate that the lncRNA Pnky can regulate the differentiation of cortical NSCs in vivo, and further demonstrates a lncRNA can play a key role in both embryonic and adult NSC populations.

Loss-of-function studies in mice

In addition to the Dlx1as and Evf-2 mutant mice discussed above, a recent set of studies reported the generation and characterization of 18 lncRNA knockout mice, including 13 with evidence for expression in the brain [60, 78]. In contrast to Dlx1as and Evf-2 mice, these knockouts were generated by complete or near-complete deletion of the targeted lncRNA, and knock-in of a Lac-Z reporter. Of the 20, three lncRNA knockout mice, Fendrr −/−, Mdgt −/−, and Peril −/− demonstrated neonatal or perinatal lethality with variable penetrance. Peril is located 110 kb downstream of key neural transcription factor Sox2 and is expressed in germinal zones of E14.5 and E18.5 brain and spinal cord. RNA-seq of Peril −/− brains vs. controls revealed a down-regulation of pathways involved in cell cycle, energy processing, and protein translation processes, suggesting that Peril may be involved in the regulation of the cell cycle or metabolism of neurogenic progenitors.

Like Pou3f2 (Brn2), the Pou3f3 (Brn1) locus is located near putative lncRNAs. Two lncRNAs are transcribed near the Brn1 locus: linc-Brn1a is bi-directionally transcribed from the Brn1 promoter, while linc-Brn1b is ~10 kb downstream of Brn1. Linc-Brn1b is expressed in germinal zones of the developing brain starting at E13.5 [60]. By E18.5 expression is absent in the VZ and SVZ and is restricted to neurons of the cortical plate, and in adult mice expression is maintained in upper cortical layers of the somatosensory and visual cortex.

Linc-Brn1b-knockout mice exhibit a significant decrease in Brn1 mRNA and protein expression. RNA-seq of null-cortices revealed downregulated genes were enriched in gene ontology terms associated with cellular proliferation and uprgulated genes were enriched for terms related to neuronal differentiation. Consistent with these results, there was a decrease in proliferating intermediate progenitors in the embryonic brain and a selective reduction of upper layer neurons. Interestingly, this was accompanied by an expansion of deep layer neurons, suggesting a mis-specification of cortical progenitor fate.

Long Non Coding RNAs in Gliomagenesis

A deeper understanding of lncRNAs in normal developmental processes can provide insight into the aberrant hijacking of these pathways leading to uncontrolled growth and tumor formation. Glioblastoma multiforme (GBM) is a highly malignant primary brain tumor (WHO Grade IV tumor). The most common of the primary brain tumors, GBM occurs in 7.2 per 100,000 adults per year, and this diagnosis carries a median prognosis of just over one year when treated with surgery, chemotherapy and radiation [79, 80]. Recent analysis [81–83] using microarray gene expression assays has demonstrated lncRNA expression signatures can be used to predict clinical phenotypes and prognosis in glioma. While it is clear that lncRNA expression can be correlated with tumor molecular subtype and clinical behavior, the functional significance of lncRNA in glioma is now just beginning to be explored: Several lncRNAs have been analyzed in preclinical models of GBM, including putative lncRNA tumor suppressors linc-RoR [84] and ADAMTS9-AS2[85], and lncRNA oncogenes CRNDE [86], and H19 [87], however the most well-studied and illustrative example of a lncRNA functioning in tumorigenesis is lncRNA HOTAIR.

HOTAIR, Hox transcript antisense intergenic RNA, is transcribed in the antisense direction of the HOXC gene. HOTAIR is well described for its ability of the 5′ domain to recruit EZH2 to target genes. It further has the ability to function as a molecular scaffold with binding surfaces to assemble histone modification enzymes [88]. HOTAIR overexpression is correlated with multiple cancer lines including breast [89], hepatocellular [90], and colorectal cancer [91], with increased expression correlating with poor prognosis, tumor invasiveness, and metastatic disease. HOTAIR expression is similarly upregulated in GBM samples, with its expression levels serving as an independent prognostic factor in patients with GBM [92]. HOTAIR controls cell cycle progression in glioma via interaction with EZH2, and HOTAIR knockdown or EZH2 inhibition block cell cycle progression in cultured cells [93]. Furthermore, knockdown of HOTAIR is sufficient to block tumor formation in an orthotopic GBM model [94]. Bromodomain and extraterminal (BET) domain proteins represent a class of therapeutic targets in cancer and GBM. Treatment of GBM samples with a BET inhibitor decreases GBM growth and also causes a decrease in HOTAIR expression; furthermore, BET protein family member BRD4 was found to directly bind the HOTAIR promoter [95].

Knockdown of the lncRNA Pnky causes a strong differentiation phenotype and stem cell depletion. Recent study of glioblastoma tumor-propagating cells demonstrated that these “cancer stem cells” rely on a core set of neurodevelopmental transcription factors for their propagation. Included in this set is the gene neighbor of Pnky, BRN2, and chromatin-state maps of the Pnky/Brn2 locus in tumor-propagating cells revealed widespread active chromatin marks at the promoters [96]. Further, the protein-binding partner of Pnky, PTBP1 [97], is upregulated in GBM and functions as a driver of tumor growth. The Pnky/PTBP1 complex is therefore an attractive therapeutic target in GBM that was first discovered through careful study of the lncRNAs of the V-SVZ lineage. Careful characterization of both lncRNAs and their protein-interacting partners in normal developmental lineages and glioma has the potential to broaden therapeutic targets to that may simultaneously target upstream factors, the lncRNA itself, and its protein-interacting partner.

It is clear that lncRNAs can play key roles in both normal neural development and in cancer progression. As we and others [98, 99] have reviewed, studies of lncRNA function in vivo can yield different and sometimes contradictory phenotypes compared to cell culture models. Studies in the V-SVZ and embryonic cortex represent an important first step towards describing the function of lncRNAs in vivo, while the next wave of functional lncRNA discoveries will be made with knockout and conditional knockout mouse models. Here, the challenge will be in the specific design of the lncRNA knockout strategy and careful dissection of RNA function from the potential function of underlying DNA regulatory elements. Finally, transgenic rescue experiments, long considered the ‘gold standard’ for establishing the function of protein-coding genes in vivo, should be performed for any lncRNA expected to exert trans-acting effects [99, 100]. Further understanding of the pathways governed by lncRNAs, and especially their function in vivo, has the potential to unlock new strategies for both regenerative medicine and the treatment of CNS malignancies.