Studies by Derrien T and team (Derrien et al., 2012) aiming at analyzing of lncRNAs from GENCODE V7 catalog revealed that, lncRNAs are produced in a similar way as that of protein-coding genes, whereas they display a striking bias toward two-exon transcripts and they are predominately localized in the chromatin and nucleus, expressed at relatively low levels (Guttman et al., 2009; Quinn and Chang, 2016). In comparison to the protein-coding genes, lncRNAs are under secondary structure conservation, and therefore they are believed to have arisen from different evolutionary pathways (Ponting et al., 2009; Kaessmann, 2010). Yet, origins of lncRNAs are not well understood so far. Due to the fact that lncRNAs harbor low sequence conservations and rapid evolution among mammals, several evolutionary hypotheses could be proposed such as: (1) lncRNAs might be generated by the metamorphosis of protein-coding genes through a gene duplication process; (2) lncRNAs might have evolved from segmental or whole gene duplication of other ncRNA genes; (3) lncRNAs might have originated via de novo generation, such as alternations in genomes including chromosomal rearrangement, generation of splice sites and promoters might transform nonfunctional genomic sequences to functional lncRNAs; and (4) transposable elements (TEs) insertions might be another origin of lncRNAs (Ponting et al., 2009; Kaessmann, 2010; Kapusta et al., 2013; Kazemzadeh et al., 2015). However, it was observed that rarely or only a minority (~15%) of lncRNAs showed significant sequence similarity to other lncRNAs or protein-coding genes on positions other than the shared repetitive elements, suggesting that novel lncRNAs genes are basically originated rather from de novo non-exonic sequences and/or from TEs than duplication (Derrien et al., 2012; Kapusta et al., 2013).

LncRNAs and associated lncRNA transcripts have quite heterogeneousgenomic context, regulation, life cycles, mechanism of action and functional profiles. Broadly, lncRNAs can be classified based on their genomic localization and orientation relative to protein coding genes into several categories: (1) long intergenic noncoding RNAs (LincRNAs), consisting of separate transcript units that are located between but do not overlap with protein-coding genes; (2) intronic transcripts, that are located within intron regions of protein-coding genes (sense or antisense); (3) overlapping lncRNAs that are overlapping with other genes either divergently or convergently transcribed; and (4) bidirectional ncRNAs (BincRNAs) with transcripts that are transcribed from divergent bidirectional promoters (see Figure 1; Guttman et al., 2009; Li and Ramchandran, 2010; Mattick and Rinn, 2015). Despite the diversities of lncRNAs, they share some common features, including: (1) most lncRNAs are transcribed by RNA polymerase II, spliced and modified with a 5′-cap and a poly-A tail, which makes them undistinguishable from protein-coding mRNAs; (2) they are poorly conserved at the sequence level, have a relatively low expression level and display a much more cell-tissue-specific pattern; and (3) they are generally regulated by transcription factors (Xiong et al., 2016). In addition, another subgroup of lncRNAs, the circular RNAs (circRNAs) have recently come into focus with the discovery of their pervasiveness and evolutionary conservation in mammalian and human cells (Jeck and Sharpless, 2014). Most circRNAs are generated during splicing either by spliceosomal machinery or by ribozymes I and II which thus splice out non-coding sequences from exons (exonic circRNAs), introns (intronic circRNAs), or a combination of introns and exons (exon-intron circRNAs; Abdelmohsen et al., 2015). CircRNAs can be differentiated from their linear counterparts by their adoption of a circular form and their lack of 5′ and 3′ ends (Vicens and Westhof, 2014).

LncRNAs have been best described for their participation in regulating gene and genome activity at various levels (see Figure 2). LncRNAs can regulate the expression of nearby genes on the same allele in cis, or in trans to control genes at other genomic locations on different chromosomes, through which they can regulate gene expression at diverse levels, such as transcription, RNA processing and translation (Elling et al., 2016). The majority of lncRNAs are localized in the nucleus, in which they can fulfill their regulatory functions via acting as scaffolds for chromatin modifiers by interacting with chromatin-modifying complexes or as transcriptional co-regulators by binding to transcription factors (Rinn and Chang, 2012; Ulitsky and Bartel, 2013).

The best-known case of lncRNA that regulates transcription mediated through chromatin modification is Xist, a 17 kb lincRNA generated from the inactive X-chromosome (Clemson et al., 1996). It mediates the silencing of the inactive X-chromosome in human female cells through recruiting Polycomb Repressive Complex 2 (PRC2) by the Repeat A motif (RepA) on Xist and thus initiating chromosome-wide silencing via catalyzing Lysine 27 trimethylation on histone H3 (H3K27; Pinter et al., 2012; Jiang et al., 2013; Bergmann and Spector, 2014). Hotair, a 2.2 kb conserved lncRNA transcribed from the human HOXC locus on chromosome 12, is involved in repressing transcription in trans across the HOXD locus. Hotair was shown to physically interact with PRC2 to ensure the PRC2 occupancy and histone H3 lysine-27 trimethylation of HOXD locus (Rinn et al., 2007). The lncRNA Braveheart, prominently expressed in the mouse heart, can interact with Suz12, a subunit of PRC2 (Klattenhoff et al., 2013). The lncRNA Fendrr (Fetal-lethal developmental regulatory RNA), that is also related to cardiac development and heart function, can bind to PRC2 and WDR5, a member of the MLL histone methyl-transferase complex (Grote and Herrmann, 2013; Grote et al., 2013).

There are also cases that many lncRNAs interact with transcription factors. The definitive endoderm-associated lncRNA1 (DEANR1) is crucial for human endoderm differentiation via interaction and upregulation of the endoderm factor FOXA2 (Jiang et al., 2015). The lnc-DC (Lnc dendritic cells) is a lncRNA exclusively expressed in human conventional dendritic cells. It was revealed that lnc-DC can bind to the transcription factor signal transducer and activator of transcription 3 (STAT3) directly in the cytoplasm and induce its phosphorylation on Tyrosine-75 by inhibiting its binding to and dephosphorylation by SHP1, thereby leading to the activation of STAT3 on dendritic cell differentiation (Wang et al., 2014). In addition, the lncRNA breast cancer anti-estrogen resistance 4 (BCAR4), functioning in breast cancer metastasis, was discovered to directly interact with Smad nuclear-interacting protein 1 (SNIP1) and Serine/threonine-protein phosphatase one regulatory subunit 10 (PPP1R10 or PNUTS), thus activating phosphor-GLI2 dependent gene expression (Xing et al., 2014).

Cytoplasmic lncRNAs are also acting as modulators on post-transcriptional regulation of genes through various mechanisms during RNA processing, such as mRNA editing, alternative splicing and others. The antisense intronic lncRNA prostate cancer antigen 3 (PCA3), acting as a dominant-negative oncogene, was demonstrated to interact with and down-regulate an as yet to be determined tumor suppressor gene PRUNE2, by the formation of PRUNE2/PCA3 double-stranded RNAs that allow the adenosine deaminase to edit RNAs via adenosine to inosine editing (A-to-I editing; Salameh et al., 2015). Non-coding RNA activated by DNA damage (NORAD) is an abundant and highly conserved human lncRNA that acts as a multivalent binding platform for RNA binding proteins in the PUMILIO family in order to maintain genomic stability (Lee et al., 2016). Furthermore, the lncRNA-asFGFR2 is an evolutionarily conserved nuclear antisense lncRNA that was generated from within the human FGFR2 locus. It was found to modulate the epithelial- specific alternative splicing of FGFR2 by recruiting PRC2 and histone demethylase KDM2a in PNT2 cells (Gonzalez et al., 2015).

Recent studies also revealed certain lncRNAs acting as “miRNA sponges” that they have the potential to sponge and compete with miRNA target genes for the binding of miRNA response elements (MREs) to relieve miRNA-mediated target mRNA repression (Ebert and Sharp, 2010). The circRNA ITCH (cir-ITCH) was newly discovered to be functionally sponging miR-7, miR-17 and miR-214 and inducing the expression level of ITCH, which induced the ubiquitination and degradation of phosphorylated Dvl2 and thereby the inhibition of the Wnt/β-catenin pathway (Li et al., 2016). The lncRNA urothelial carcinoma-associated 1 (UCA1) was shown to work as an endogenous sponge that can down-regulate miR-216b expression by directly binding to miR-216b (Wang et al., 2015). The lncRNA human ovarian cancer-specific transcript 2 (HOST2) was verified to be a molecular sponge that can modulate the availability of miR let-7b (a potent tumor suppressor) and inhibit miR let-7b functions, thus post-transcriptionally suppressing target gene expressions (Gao et al., 2015).

Based on studies from the GENCODE project in 2012 that 10,000–50,000 lncRNAs in the human genome have been annotated to date, it was remarkably revealed that approximately 40% of lncRNAs (which is about 4000–20,000 lncRNAs) are tissue-specifically expressed in the brain (Derrien et al., 2012). This number is strikingly large in comparison to the number of protein-coding genes in human genome which is approximately 20,000–25,000 in general (Briggs et al., 2015). These brain-specific lncRNAs display the highest signals of evolutionary conservation in comparison with those expressed in other tissues. Studies also found that brain-expressed lncRNAs are enriched in predicted, conserved RNA structures and thus are more likely to possess conserved functions (Ponjavic et al., 2009). In addition, brain-expressed lncRNAs show greater brain region, temporal and spatial specificity than mRNAs. Numerous transcriptome analysis have revealed that various lncRNAs are expressed differentially over time and/or in those brain regions, such as cortex, cerebellum and hippocampus during development and adulthood (Lipovich et al., 2014; Kadakkuzha et al., 2015). LncRNAs are also found to be expressed in a more cell-type-specific manner than protein-coding genes. The transcriptome studies on cortical pyramidal neurons have uncovered 806 of 5195 lncRNAs have differential expression across neuronal types, of which, 55% of lncRNAs are annotated as cell-type signature cluster, 32% of lncRNAs are related to cell-type independent clusters, while around 10% of lncRNAs are found in the mixed cell-type clusters, indicating their role on the specification and maintenance of cell identity (Molyneaux et al., 2015). Another feature of brain-specific lncRNAs is their preferential location adjacent to brain-expressed protein-coding genes that are active in transcriptional modulation or in the nervous system development (Ponjavic et al., 2009). Transcriptome sequencing analysis in the corticogenesis displayed that most lncRNAs overlap with neurogenic genes and share with them a similar expression pattern, indicating that lncRNAs regulate corticogenesis through the tuning of the expression of nearby cell fate determinants (Aprea et al., 2013). Considering the complexity and heterogeneity of the mammalian CNS, the brain would be considered as the largest repertoire of lncRNAs in comparison to other somatic tissues and the tissue and cell-type specificity of these lncRNAs make them greatly contribute to cell fate, lineage specification and maintenance of cell identity during the development of the mammalian brain (Hart and Goff, 2016).

The development of the CNS is a complicated and highly stereotyped process that requires elaborate spatiotemporal regulation of stem/progenitor cell proliferation and differentiation. LncRNAs have been demonstrated to play indispensable roles in CNS development from early neural differentiation to late-stage synaptogenesis (see Figure 3; Briggs et al., 2015).

Accumulated studies have confirmed that dozens of lncRNAs are identified to be functional in establishing pluripotency or driving neural lineage entry in the widely used in vitro model systems—mouse embryonic stem cells (ESCs; Guttman et al., 2011; Ng et al., 2012). These lncRNAs are functional at various stages along the progression from pluripotent cells in the early embryo to the terminal cell types in the mature mammalian brain. The regulatory mechanisms are generally based on modulating these lncRNAs by canonical pluripotency transcription factors, which in turn perform their regulatory effects by directing transcription factors or chromatin remodeling complexes to specific lineage-specifying genes. For examples, the lncRNA Rhabdomyosarcoma 2-associated transcript (Rmst), specifically expressed in the brain and regulated by the transcription repressor REST, was found to modulate neural differentiation in vitro. It was shown that Rmst interacts with SOX2 to co-regulate a large pool of downstream genes (such as DLX1, HEY2 and SP8) implicated in neurogenesis. The absence of Rmst can block the exit from the ESC state and the initiation of neural differentiation (Ng et al., 2013). Tuna, is also a lncRNA that regulates neuronal gene expression in a similar mechanism as Rmst. Tuna interacts with three RNA-binding proteins (NCL, PTBP1 and hnRNP-K) and together target promoter regions of neural genes in differentiating mouse ESCs. The knockdown of Tuna or any of the three RNA-binding proteins can sufficiently suppress neural differentiation (Lin et al., 2014). This is also conserved from relatively related vertebrates, mice and zebrafish, indicating that the neural lineage commitment driven by lncRNAs displays the highly evolutionary conservation (Briggs et al., 2015). Furthermore, the lncRNA Dali, transcribed downstream of the transcription factor Pou3f3, was shown to drive the expression of an essential gene involved in neuronal differentiation in neuroblastoma cells. Dali can regulate transcription of the Pou3f3 locus locally and physically interact with Pou3f3 protein to regulate the expression of neural differentiation genes distally (Chalei et al., 2014). Paupar, a chromatin-associated intergenic lncRNA expressed in the CNS, is divergently transcribed from a locus upstream from the gene encoding the transcription factor Pax6. It was shown that the knockdown of Paupar destroys the normal cell cycle profile of neuroblastoma cells, thus enhancing neuronal differentiation. The function of Paupar is performed via locally interacting with and transcriptionally regulating Pax6, as well as via distally controlling neural gene expression on a large scale (such as SOX2, HES1 and EVF2), which in part requires physical association with Pax6 protein (Vance et al., 2014). Besides that, one of the latest studies has identified an lncRNA, termed LncND (Neurodevelopment) that harbors 16 miRNA response elements for miR-143-3p in primates. It performs its role in neuronal development by sequestering miR-143-3p, and in doing so, modulates Notch signaling pathways via regulating expression of Notch receptors, NOTCH-1 and NOTCH-2. Also, NOTCH-1 and NOTCH-2 show the same expression pattern as LncND in early neurogenesis process when Notch expression is indispensable. While, later in neural differentiation process, reduction of NOTCH expression is followed by LncND, leading to the release of miR-143-3p and the decrease of Notch signaling specifically in LncND-expressing cells (Rani et al., 2016). Interestingly, knocking down LncND in neuroblastoma cells can inhibit NOTCH-1 and NOTCH-2 and make cells differentiate to neurons, which show similar phenotypes as what can be observed by miR-143-3p overexpression. These findings suggested a role of LncND in miR-mediated regulation of Notch signaling to sustain the neural progenitor pool during cerebral cortex expansion in primates (Rani et al., 2016).

There are also several important lncRNAs being identified to control stem cell turnover and the specification of particular lineages in the embryonic mouse brain in vivo. For example, Evf2, the firstly identified nervous system-specific lncRNA, has been shown to have a significant regulatory role in neural development. Depletion in EVF2 can cause imbalance of the excitatory to inhibitory neurons in the postnatal hippocampus and dentate gyrus. This imbalance is caused by the failure of GABAergic interneuron specification (Bond et al., 2009). It was revealed that Evf2 recruits DLX and methyl CpG-binding protein (MECP2) to modulate the transcriptional activity of two transcription factors Dlx5/6 and the glutamate decarboxylase 67 (Gad67, which is required for the conversion of glutamate to GABA), and thereby regulates the gene expression of GABAergic interneurons in the developing mouse forebrain (Feng et al., 2006; Bond et al., 2009). Pnky is an evolutionally conserved and neural-specific lncRNA that modulates neurogenesis of neural stem cells (NSC) in the embryonic and postnatal brain. The Ramos AD group (Ramos et al., 2013) has unveiled that knockdown of Pnky can increase neuronal differentiation and deplete the NSC population in the embryotic mouse cortex. In addition, it was shown that Pnky interacts with PTBP1, a splicing regulator expressed in NSCs that represses the inclusion of neural exons in non-neural cells. Knockdown of PTBP1 also reinforce neurogenesis, indicating that Pnky and PTBP1 function similarly to modulate the alternative splicing of a group of transcripts that are involved with cellular phenotypes (Aprea and Calegari, 2015; Ramos et al., 2015). Together, these studies indicate that lncRNAs regulate cell-fate determination and progenitor cell turnover during neural development both in vitro and in vivo.

Another key process during CNS development is the late-stage synaptogenesis. Accordingly, several lncRNAs have been discovered to make great contributions to this process (Briggs et al., 2015). Malat1 is a well-known lncRNA that have been implicated in the regulation of neurite elaboration. Malat1 is expressed in multiple tissues but is highly abundant in neurons. It was revealed that Malat1 can actively recruit SR-family splicing proteins to transcription sites in order to regulate synaptogenesis-related gene expression. The knockdown of MALAT1 leads to a decreased level of synaptic density; while in contrast, overexpression can reciprocally increase synaptic density (Bernard et al., 2010). Primate-specific BC200 RNA (Brain cytoplasmic RNA, 200 nt) and its rodent counterpart BC1 are evolutionarily conserved lncRNAs that have been firstly identified to be expressed in both developing and adult nervous systems (Muslimov et al., 1997). Studies later reported that BC200/BC1 is functional in the neuronal excitation-repression equilibria via protein-synthesis-dependent implementation (Zhong et al., 2009). In addition, several antisense lncRNAs have been recently presented as core protein regulators, for example Brain-derived neurotrophic factor (BDNF), glial-derived neurotrophic factor (GDNF) and ephrin receptor B2 (EPHB2) that control neurite elaboration. The expression of BDNF is repressed by its antisense lncRNA BDNF-AS. The inhibition of BDNF-AS can release and induce the expression of BDNF by 2- to 7-folds, which is in line with the reduction of EZH2 recruitment and an alternation of the chromatin state at the BDNF locus. This correspondingly induces neuronal outgrowth, differentiation, survival and proliferation both in vitro and in vivo (Modarresi et al., 2012). These observations, along with the current understanding of lncRNA mechanisms of action, imply that lncRNAs have a critical role in regulating neural gene expression and brain development.

LncRNAs have been proposed to play diverse roles in the CNS, such as in neural differentiation, in synaptogenesis and others. As a result, the expression profiles of lncRNAs in the CNS were recently studied. The CNS has prominent cellular diversities owning hundreds of distinct cell types. In particular, neurons and neuroglia cells (mainly including astrocytes and oligodendrocytes) are the major cell types. A recent study has successfully isolated nuclear RNAs from different CNS cell types (neurons, astrocytes and oligodendrocytes) and compared both protein-coding and noncoding nuclear transcriptome profiles in these three cell types (Reddy et al., 2017). For non-coding transcriptome, it was revealed that approximately 300 transcripts at a level of >5 CPM (counts per million) can be observed in one of the three different cell types, and the majority is transcribed at the level of >1 CPM. Several other highly expressed transcripts are also found in all three cell types, such as Xist and its regulators Ftx and Jpx (Tian et al., 2010; Chureau et al., 2011; Reddy et al., 2017). Nevertheless, 169 lincRNAs were shown to have a >10-fold difference in one of the pairwise comparisons, such as Mirg and other adjacent maternally expressed lncRNAs (Meg3 and Rian) that are found to be highly enriched in neuronal nuclei. Mirg was found to be precisely localized within bright subnuclear puncta in neurons; the expression of Meg3 was shown as neuronal selective that it was enhanced in gray matter where neurons are found but depleted from white matter areas where only glia are found; and Rian was found to have similar expression pattern as Meg3 (Zhang et al., 2003; Balik et al., 2013; Reddy et al., 2017). In astrocytes, various transcripts were found but most of them do not possess known functions. Whereas, the lncRNA Rmst, a key co-regulator of neurogenesis with the SOX2 transcription factor, is an exception that it has been shown to be expressed robustly in the nuclei of astrocytes (Ng et al., 2013; Reddy et al., 2017). Meanwhile, it was shown that there are at least 16 lncRNAs with >5 CPM found in oligodendrocytes, such as Neat1 and DLeu2. Notably, although Neat1 was shown to have the highest expression level in oligodendrocytes, it was also present in astrocytes (Klein et al., 2010; Reddy et al., 2017). These data demonstrated different lncRNA expression patterns in three major cell types in the brain, providing a crucial clue for evaluating functions of various lncRNAs in the brain and CNS-associated disorders.