Long non-coding RNA-dependent transcriptional regulation in neuronal development and disease

As previously stated, many lncRNAs have been proposed to function through interactions with chromatin modifiers. In fact, it is estimated that ~30% of all lincRNAs expressed in mouse ES cells interact with one or more of 11 particular chromatin modifiers ( Khalil et al., 2009 ; Guttman et al., 2011 ). This has been extrapolated to suggest that interaction with chromatin regulators is the major mechanism by which lncRNAs regulate transcription. However, many of these lncRNAs display predominantly cytoplasmic expression, suggesting instead that they may have additional cellular functions. Furthermore, there is reason to suspect that the selectivity of lncRNA–chromatin modifier interactions may have been overestimated. Experiments using overexpressed, tagged lncRNAs followed by mass spectrometry do not take into account the low transcript abundance levels seen for most lncRNAs. Chromatin-modifying enzymes are likewise abundantly expressed in virtually all cell types, particularly in comparison to transcription factors. It is thus possible that weak and possibly non-physiological interactions between lncRNAs and chromatin-modifying proteins may be detected using mass spectrometry. This may include weak interactions of highly expressed proteins that have known RNA binding potential, such as PRC2 complex proteins. Furthermore, recent reports suggest that the PRC2 protein complex is quite promiscuous in its RNA binding specificity ( Davidovich et al., 2013 ). A more systematic interrogation of potential lncRNA–protein interactions using techniques that control for the abundance of both lncRNA and protein, such as protein microarrays, will help clarify this issue.

Many groups are taking advantage of RNA-Seq and lncRNA microarray technologies to identify altered transcript expression levels between control and diseased states within various human neurological and psychiatric disorders ( Dharap et al., 2012 , 2013 ; Petazzi et al., 2013 ; Ziats and Rennert, 2013 ). While useful, with few exceptions, these studies have not functionally implicated these lncRNAs in disease progression ( Petazzi et al., 2013 ; Ziats and Rennert, 2013 ). Here we review the limited number of studies that directly link altered lncRNA function to the development and progression of neurological disease. One of the better studied lncRNAs associated with human disease is ANRIL (also known as CDKN2B-AS ). Genome-wide association studies have associated numerous polymorphisms on human chromosome 9p21 that segregate with diseases including cardiovascular disease, Type-2 diabetes, Alzheimer’s disease (AD), primary open angle glaucoma, endometriosis, periodontitis, and several cancers (reviewed in Congrains et al., 2013 ). Polymorphisms map to both the promoter and transcribed region of ANRIL , including many transcription factor-binding sites located throughout the locus. ANRIL has been shown to bind CBX7 and SUZ12 of the PRC1 and PRC2 complexes, respectively, to regulate the histone modification status of the nearby CDKN2A and CDKN2B genes ( Yap et al., 2010 ; Aguilo et al., 2011 ; Kotake et al., 2011 ). As both increased and decreased ANRIL expression levels correlate with disease states ( Congrains et al., 2013 ), the fine control of CDKN2B/CDKN2A transcript abundance seems paramount to normal development. Kcna2AS is an antisense ncRNA to the voltage-gated potassium channel Kcna2. Expression of Kcna2AS was observed in dorsal root ganglia (DRG) and was expressed at higher levels in ganglia exhibiting lower levels of Kcna2 protein expression, or after spinal nerve injury ( Zhao et al., 2013 ). Spinal nerve injury causes an increase of myeloid zinc finger protein 1 (MZF1) binding to the proximal promoter of Kcna2AS , causing an increased expression of Kcna2AS with a concomitant decrease in Kcna2 transcript and protein abundance ( Zhao et al., 2013 ). Additional experiments found that expression of Kcna2AS causes a decrease in voltage-gated potassium currents and an increase in membrane resting potential, suggesting that pain hypersensitivity or neuropathic pain can be caused by altered Kcna2AS levels ( Zhao et al., 2013 ). Recent studies have also implicated altered lncRNA expression as associated with AD progression. AD is characterized by a progressive neurodegeneration that leads to memory and cognitive impairment. A hallmark component of the pathological condition is the buildup of extracellular beta amyloidal plaques. The amyloid precursor protein (APP) is cleaved in the initial and rate-limiting step by β-secretase enzyme (BACE1) to form the amyloid β precursor proteins Aβ 1–40 and Aβ 1–42. In pathological conditions, the Aβ 1–42 proteins oligomerize and contribute to the plaques that participate in AD ( Abramov et al., 2004 ; Ohyagi et al., 2005 ; Snyder et al., 2005 ; Esposito et al., 2006 ; Zhu et al., 2006 ; Lacor et al., 2007 ; Matsuyama et al., 2007 ). As a result, it has been suggested that BACE1 misregulation can contribute to excess Aβ 1–42 protein production and the development of amyloid plaques. Recent work has identified an antisense transcript to BACE1 ( BACE1-AS ) that encodes a conserved ~2 kb lncRNA with a 104 bp overlap with the human BACE1 transcript ( Faghihi et al., 2008 ). Both overexpression and knockdown experiments indicated that BACE1-AS is a positive regulator of BACE1 transcript and protein abundance ( Faghihi et al., 2008 ). Mechanistically, BACE1-AS stabilizes the BACE1 transcript, protecting it from RNA degradation through RNA–RNA pairing of the BACE1-AS and BACE1 homologous regions ( Faghihi et al., 2008 ). Importantly, BACE1-AS and BACE1 transcripts were induced by many cell stressors that are implicated in the initiation of AD, suggesting a direct mechanism by which cell stress can lead to increased Aβ precursor protein production ( Faghihi et al., 2008 ). The importance of BACE1-AS in AD was further supported through examinations of primary tissues from multiple brain regions, where BACE1-AS transcript abundance was elevated twofold in confirmed AD patient brain samples compared to age- and sex-matched controls ( Faghihi et al., 2008 ). Further characterization of BACE1-AS in a transgenic mouse model of AD indicated that BACE1-AS inhibition reduces the insoluble fraction of Aβ 1–40 and Aβ 1–42 precursor proteins ( Modarresi et al., 2011 ), suggesting that increased BACE1-AS expression does directly contribute to AD pathology. Other aspects of AD are also potentially regulated through lncRNA function. Recent work on neurotrophin levels in diseases of the brain have indicated that reduced neurotrophin levels (BNDF and glial derived neurotrophic factor – GDNF) correlate with the onset of neurodegenerative disorders such as Parkinson’s disease, AD, and HD (reviewed in Allen et al., 2013 ). This has led to potential therapeutics aimed at increasing neurotrophin levels ( Weinreb et al., 2007 ; Straten et al., 2011 ; Allen et al., 2013 ). However, as both BDNF and GDNF display complex splicing regulation ( Airavaara et al., 2011 ; Modarresi et al., 2012 ), other mechanisms of therapeutic intervention than exogenous neurotrophin replacement may be better suited to treating the diseases. Interestingly, both BDNF and GDNF have corresponding anti-sense or OS transcripts ( BDNF-AS and GDNF-OS ), however, one of the three GDNF-OS transcripts is likely protein coding ( Airavaara et al., 2011 ). Knockdown of either BDNF-AS or GNDF-OS results in an increase in corresponding protein-coding gene transcript abundance, implying that these lncRNAs negatively regulate neurotrophin expression ( Modarresi et al., 2012 ). Further characterization of BDNF-AS indicates that BDNF-AS recruits EZH2 and the PRC2 complex to the BDNF promoter to repress BDNF transcription through H3K27me3 histone modifications ( Modarresi et al., 2012 ). Combined with studies in which treatment with exogenous BDNF rescued HD phenotypes in mice ( Xie et al., 2010 ), these experiments suggest that inhibition of neurotrophin antisense transcripts may provide a novel target for treatment of neurodegenerative disease. LncRNAs have also been implicated in nervous system disorders through their role in pre-mRNA splicing. The lncRNAs Gomafu and Malat1 are both highly expressed in the nervous system and regulate splicing through interactions with splicing factors ( Sone et al., 2007 ; Tripathi et al., 2010 ; Tsuiji et al., 2011 ; Zong et al., 2011 ). Interestingly, aberrant splicing of the genes DISC1 and ERBB4, among others, is associated with disease pathology in schizophrenia (SZ; Law et al., 2007 ; Nakata et al., 2009 ; Morikawa and Manabe, 2010 ). Additionally, the Gomafu -bound splicing factor QKI is downregulated in SZ brains and is proposed to contribute to disease pathology ( Aberg et al., 2006a , b ; Haroutunian et al., 2006 ; McCullumsmith et al., 2007 ). Recently, Gomafu has been shown to interact with multiple splicing factors, including a strong interaction with QKI ( Barry et al., 2013 ). Gomafu expression is also significantly decreased shortly after neuronal depolarization in the cortical neurons in mice, and in human induced pluripotent stem cell (iPSC)-derived neurons ( Barry et al., 2013 ). Combined with GWAS studies linking Gomafu with eye movement disorders in SZ ( Takahashi et al., 2003 ), this led to the hypothesis that loss of function of Gomafu may directly contribute to SZ disease pathology. Indeed, Gomafu is significantly reduced in superior temporal gyrus of SZ brain samples compared to controls ( Barry et al., 2013 ). Knockdown of Gomafu in iPSC neurons also results in an increase in rare splice variants of DISC1 and ERBB4 ( Barry et al., 2013 ), matching splicing patterns observed in vivo from human SZ brains ( Law et al., 2007 ; Nakata et al., 2009 ). With the increased use of whole exome sequencing and copy number variations (CNV) for genetic analysis of patients with neurological diseases, our understanding of the importance of lncRNAs in neurodevelopment will only be further increased. For example, one patient that displayed a cognitive developmental delay possessed a chromosomal translocation that affected linc00299 ( Talkowski et al., 2012 ). Further examinations of patient databases identified an additional four patients that displayed developmental delay and disruption of the linc00299 locus ( Talkowski et al., 2012 ), suggesting that linc00299 is vital for proper neuronal development. Further characterization of lncRNA function in animal models and in vitro will continue to expand our knowledge on the importance of lncRNAs in both human development and disease.

Natural antisense RNAs transcripts are lncRNAs that are transcribed from the opposite strand (OS) of protein-coding genes, and therefore, share sequence complementarity. The degree of complementarity of NATs with corresponding sense transcripts varies greatly, however, genome-wide analysis suggests that localization of antisense transcription is generally confined to 250 bp upstream of the sense transcript’s transcription start site and 1.5 kb downstream of the sense gene ( Sun et al., 2005 ; Core et al., 2008 ; Seila et al., 2008 ). As previously reviewed, NATs mediate their function through transcriptional and epigenetic regulation, RNA–DNA interactions, and RNA–RNA interactions ( Faghihi and Wahlestedt, 2009 ; Magistri et al., 2012 ). While there are clear examples of antisense transcripts that directly inhibit protein coding gene expression ( Werner et al., 2014 ), the inhibition is probably not mediated by complementary base-pairing of sense–antisense transcripts. Since most lncRNAs are expressed at much lower levels than neighboring protein-coding genes, the stoichiometry between sense–antisense pairs is insufficient to simply block splicing or translation of protein coding genes. One topic of particular current interest is the role of NATs that work in conjunction with epigenetic modifiers. Many imprinted genes are found in genomic clusters and have NATs located within the same locus ( Verona et al., 2003 ; Katayama et al., 2005 ; Wan and Bartolomei, 2008 ; Mohammad et al., 2009 ). The imprinted locus is facilitated through allele specific expression of NATs and corresponding interactions with epigenetic modifiers. For example, the NAT Air interacts with HMT G9a while Kcnq1ot1 interacts with PRC2 components and HMT G9a ( Nagano et al., 2008 ; Pandey et al., 2008 ; Terranova et al., 2008 ). Through complementary base pairing and RNA–protein interactions, the NAT transcript allows sequence-specific recruitment of chromatin modifiers to the locus. For both Air and Kcnq1ot1 , NAT expression from the paternal allele corresponds to paternal allele silencing through chromatin condensation and bidirectional spreading of epigenetic marks ( Nagano et al., 2008 ; Kanduri, 2011 ). Epigenetic control of protein coding genes by NATs is also observed in non-imprinted loci. For example, brain-derived neurotrophic factor (BDNF) is regulated by the NAT BDNF-AS . Loss of BDNF-AS is accompanied by increased BDNF transcript abundance, facilitated through an altered chromatin state ( Modarresi et al., 2012 ).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

While reports suggest that up to 35% of lncRNAs localize to intronic sequences, little is known about the function of these sequences ( St Laurent et al., 2012 ). Surprisingly, intronic ncRNAs are predominantly associated with the sense strand of the unprocessed mRNA, but often show expression patterns that are inversely correlated with the processed mRNA ( Katayama et al., 2005 ; Nakaya et al., 2007 ; Dinger et al., 2008 ; Mercer et al., 2008 ). This suggests a complex regulatory relationship in which intronic ncRNA transcription may be independent of transcription of the protein coding pre-mRNA. In some cases, these intronic ncRNAs are precursor transcripts to miRNAs. Recent work has also suggested that many intron-derived RNAs bind to Ezh2 of the PRC2 complex, thus recruiting chromatin structure modifiers to the locus to silence transcription ( Guil and Esteller, 2012 ; Guil et al., 2012 ).

Other lncRNAs do not overlap with either protein coding genes or promoter or enhancer sequences. These are collectively termed long intergenic ncRNAs (lincRNAs). Analyses of the correlated expression patterns of lincRNA and transcripts of neighboring protein-coding genes imply that lincRNAs participate in similar biological processes to neighboring protein-coding genes ( Luo et al., 2013 ). This has been interpreted that many lincRNAs may function in cis to regulate expression of nearby genes ( Luo et al., 2013 ). However, this finding also raises the possibility that lincRNAs might act in trans to directly or indirectly regulate the activity of co-expressed protein coding genes through RNA–protein interactions. One classical example of lincRNA function comes from studies of HOTAIR . HOTAIR is transcribed from an intergenic region in the HOXC locus and is involved in recruitment of chromatin modifiers to hundreds of genomic loci ( Rinn et al., 2007 ; Tsai et al., 2010 ; Chu et al., 2011 ). Through interactions with the PRC2 and LSD1 complexes, HOTAIR promotes H2K27 methylation and H3K4 demethylation, respectively, resulting in the leading to gene silencing ( Rinn et al., 2007 ; Tsai et al., 2010 ; Chu et al., 2011 ). More specifically, HOTAIR expression silences expression of genes from the HOXD locus, thereby facilitating HOXC locus gene expression specifying positional identity of the HOTAIR -expressing cells ( Rinn et al., 2007 ). Knockout of Hotair in mice causes skeletal defects including homeotic transformation of vertebrae resulting from de-repression of multiple HoxD cluster genes, increased expression of ~30 genes from imprinted loci and loss of vertebral boundary specification during development ( Li et al., 2013a ).

In general, lncRNAs function either in cis , within the same genomic locus, or in trans , affecting gene transcription in a different locus or even on different chromosomes. Many lncRNAs, including the intensely studied Xist and HOTAIR ncRNAs, function through recruitment of the Polycomb repressive complex 2 (PRC2) by binding to PRC2 component histone-lysine N-methyltransferase Ezh2, leading to a local increase in H3K27me3 content and subsequent transcriptional repression ( Zhao et al., 2008 , 2010 ; Tsai et al., 2010 ; Guil and Esteller, 2012 ). However, other lncRNAs, like the BORDERLINE lncRNA, are shown to inhibit repressive histone modifications either solely through their transcription or by binding to and removing the heterochromatin protein 1 (HP1/Swi6) from the locus ( Keller et al., 2013 ). The diverse functions observed for the handful of characterized lncRNAs studied so far underscore the importance of analyzing lncRNA function on an individual basis.

Some lncRNAs are transcribed from the proximal promoters in the opposite direction of protein coding genes, and have been termed “opposite strand” transcripts. Conservative estimates suggest that one-third of brain-enriched transcription factors express corresponding OS transcripts and that many of these may act in cis to regulate protein-coding gene transcription ( Alfano et al., 2005 ; Rapicavoli et al., 2010 ). Many OS transcripts display correlated expression patterns with neighboring protein-coding genes as a result of bi-directional promoters initiating transcription of both the lncRNA and protein-coding gene ( Uesaka et al., 2014 ). Recent reports analyzing the function of Six3OS and Vax2OS , however, indicate that some OS transcripts function in trans , and not by regulating expression of their neighboring protein-coding gene ( Rapicavoli et al., 2011 ; Meola et al., 2012 ). Other promoter-associated lncRNAs overlap proximal promoter sequences but are transcribed from the sense strand relative to the protein-coding gene. The transcription of the lncRNA itself can positively impact transcription in cis of the protein-coding gene, by changing chromatin conformation to permit transcription factor recruitment, leading to initiation of protein-coding gene transcription. Alternatively, promoter-associated lncRNAs can inhibit protein-coding gene transcription through one of two different proposed mechanisms. Chromatin de-condensation that occurs as a result of transcription of a lncRNA within the promoter region of a protein-coding gene may inhibit transcription of nearby genes by altering DNA supercoiling. Conversely, it was recently shown that transcription of the CCND1 promoter-associated lncRNA ( CCND1-pncRNA ) recruits the TLS protein to the promoter of CCND1 during DNA damage. The recruitment of TLS reduces transcription of CCND1 by inhibiting the histone acetyltransferase activity of CBP/p300 at the gene’s promoter ( Wang et al., 2008 ; Kurokawa, 2011 ). This further suggests that some promoter-associated lncRNAs may regulate transcription of neighboring protein-coding genes through recruitment of chromatin-modifying complexes. Past efforts in comparative genetics have identified thousands of sequences that display high sequence constraints across evolution ( Bejerano et al., 2004 ; Woolfe et al., 2005 ; Pennacchio et al., 2006 ). These ultra-conserved regions (UCRs) identified from Fugu rubripes and human ( Woolfe et al., 2005 ; Pennacchio et al., 2006 ) or human, mouse, and rat ( Bejerano et al., 2004 ) are at least 200 bp and display >90% sequence conservation. The UCRs tend to cluster around genes pertinent to the regulation of organism development. Therefore, the preferential localization and high-degree of sequence conservation has led to the hypothesis that these UCRs are vital to the regulation of development. Further studies analyzing these sequences have identified that many function as enhancer sequences ( Woolfe et al., 2005 ; Pennacchio et al., 2006 ). However, in these studies, many UCRs also overlapped known expressed sequence tag (EST) transcripts that were rationalized as genomic contamination or incompletely spliced pre-mRNA ( Bejerano et al., 2004 ; Woolfe et al., 2005 ). Roughly 240 (50%) and 84 UCRs (6%) showed evidence for transcription in the Bejerano et al. (2004) and Woolfe et al. (2005) studies, respectively. Additional work on UCRs has since confirmed that these enhancers/UCRs can indeed be transcribed into non-coding sequence. The discovery that many enhancers or ultra-conserved elements are not only platforms for transcription factor binding but also are transcribed themselves has stimulated studies of the role played by eRNA transcription in the regulation of neighboring genes ( De Santa et al., 2010 ; Kim et al., 2010 ; Licastro et al., 2010 ; Wang et al., 2011 ). Most eRNAs are short sequences, resulting from bi-directional transcription of enhancer sequences. They exhibit H3K4me1-enriched sequences and lack poly-adenylation signals. One proposed mechanism of eRNA function in transcriptional regulation is a ripple effect, a process where growth factor-induced immediate-early gene transcription triggers initiation of transcription at nearby promoters ( Ebisuya et al., 2008 ). One could postulate that eRNAs may function in a similar manner, with transcription factors binding to enhancers, recruiting the transcriptional machinery to the enhancer to induce eRNA transcription and chromatin modifications, leading to activation of neighboring genes. To date, however, no evidence is available to suggest such a mechanism exists for eRNA-dependent regulation of transcription. The expression of eRNAs generally correlates with activation of the neighboring gene(s; Lee, 2012 ; Li et al., 2013b ; Melo et al., 2013 ; Memczak et al., 2013 ). Examples exist, however, in which lncRNA sequence overlaps the ultraconserved enhancer sequence of neighboring genes ( Feng et al., 2006 ). Transcription of lncRNA sequences from ultraconserved sequences, in sharp contrast to eRNAs, may actually inhibit antisense gene transcription of neighboring targets ( Bond et al., 2009 ). Analyses of lncRNAs that overlap ultraconserved element sequences have been shown to possess characteristics more similar to lincRNAs, and therefore, are not typically classified as eRNAs. These signatures include H3K4me3 and modification by 3′-polyadenylation. Although many of the transcribed ultra conserved elements overlap known enhancer sequences; only a minority (~4%) are transcribed bi-directionally and are unlikely to encode short RNAs ( Licastro et al., 2010 ). Together, this provides further indication that these lncRNAs are more similar to lincRNAs than eRNAs. One possible mechanism of function for lncRNAs that overlap ultraconserved enhancer regions comes from recent studies of the lncRNA CCATT1-L . CCATT1-L is expressed from a super enhancer region 515 kb upstream of the MYC locus and positively regulates MYC transcription by facilitating chromatin interactions between the MYC proximal promoter and enhancer elements ( Xiang et al., 2014 ). This suggests that lncRNAs may facilitate transcription factor recruitment to specific DNA sequences, a potential mechanism discussed in further detail below. Circular RNAs (circRNA) define a more unconventional and less well understood class of functional ncRNAs. These unique transcripts were originally identified in plants where they function to encode subviral components ( Sanger et al., 1976 ). In animal species, these transcripts are thought to arise from joining of 5′ and 3′ splice sites within a single exon to form the circular transcript ( Nigro et al., 1991 ; Capel et al., 1993 ; Cocquerelle et al., 1993 ; Chao et al., 1998 ; Burd et al., 2010 ; Hansen et al., 2011 ; Salzman et al., 2012 ). Recent profiling of mouse, human, and Caenorhabditis elegans identified thousands of conserved circRNAs ( Memczak et al., 2013 ). The identified circRNAs are often highly conserved, leading the authors to hypothesize that the circRNA transcripts function as molecular decoys for RNA-binding proteins and miRNAs ( Memczak et al., 2013 ).

Advances in sequencing technologies and the appreciation of functional non-coding elements have resulted in the rapid identification of a plethora of lncRNAs expressed in both vertebrate and invertebrates, alike. Systematic characterization of temporally and spatially restricted expression patterns in the developing nervous system has provided the groundwork for hypotheses regarding lncRNA function. As we understand more about the mechanism by which lncRNAs are regulating transcription, we are beginning to understand the biological significance of what once was labeled as “junk DNA.” Many lncRNAs regulate transcription through regulation of epigenetics and interactions with chromatin-modifying complexes, although the mechanism by which lncRNAs are recruited to specific genomic loci is still unclear. Recently developed technologies have the potential to greatly expand our understanding of the mechanism by which lncRNAs function. The advent of techniques such as ChIRP and CHART allow for systematic characterization of DNA binding sites of lncRNAs throughout the genome ( Chu et al., 2011 ; Simon et al., 2011 ). Additionally, protein arrays, as used to identify Six3os binding partners ( Rapicavoli et al., 2011 ), allow for an unbiased approach to identifying physiologically relevant protein binding partners. These techniques will further our understanding of how lncRNAs function as molecular scaffolds and will enable the functional characterization of lncRNAs working in trans . While not the focus of this review, it is also essential to consider the function of lncRNAs that display cytoplasmic expression, which represent a large fraction of lncRNAs and whose function is poorly understood (reviewed in Batista and Chang, 2013 ). Further characterization of lncRNA–protein interactions through protein arrays will help facilitate these discoveries. As many cytoplasmic lncRNAs associate with ribosomes ( van Heesch et al., 2014 ), it is intriguing to speculate that lncRNAs function as scaffolds to regulate localized protein synthesis and/or degradation, a concept vitally important in the control of synaptic function. As we continue to understand the molecular basis of lncRNA function, it is imperative that studies move from in vitro , homogeneous cell populations and begin to examine the consequence within individual cell types. Neuronal diversification has exhibited a multitude of examples in which transcriptional regulation and cell-fate decisions are very context and cell-type specific. Therefore, it is plausible that individual lncRNAs may display diverse functions that are dependent on their spatial and temporal expression pattern. Inherent to the examination of specific cell types is that epigenetic marks may display vast temporal and/or cell-type specific signatures. In vivo experiments continue to shed light on the importance of lncRNA function throughout neuronal development. As mouse models for genetic loss of lncRNAs such as Evf2 , Dlx1AS , Malat1 , and Neat1 produce modest phenotypes or fail to recapitulate phenotypes observed in knockdown experiments ( Bond et al., 2009 ; Zhang et al., 2012 ; Kraus et al., 2013 ), it is important to consider that lncRNAs may have evolved to function as a fine-tuning mechanism to ensure proper regulation of neuronal cell type proportions in the highly complex mammalian nervous system. Genetic compensation may mask phenotypes resulting from conventional gene knockout approaches, which conditional or acute loss of function studies may readily detect. Furthermore, efforts need to be made to carefully examine genetic models of lncRNA loss-of-function, however, being constantly mindful of the fact that many lncRNAs overlap conserved regulatory elements that may have function independent of the lncRNA itself, complicating interpretation of any observed phenotypes. Further exploration of lncRNA function will only continue to add to our appreciation of the complexity of transcriptional regulation, especially within the context of the seemingly endlessly complex development of the nervous system.

Annotation and high-throughput deep sequencing of the transcriptomes of multiple species have led to the belief that much of the genome is transcribed; however, only a minority of transcribed sequences contain evolutionarily conserved open reading frames ( Okazaki et al., 2002 ; Maeda et al., 2006 ; Kapranov et al., 2007 , 2010 ; Derrien et al., 2012 ; Dunham et al., 2012 ). Many of the transcribed sequences are thus unlikely to encode proteins. Among all human non-coding transcripts, at least 10,000 are estimated to be >200 nucleotides, and are accordingly designated as long non-coding RNAs (lncRNAs; Derrien et al., 2012 ). Based on transcriptome analysis of protein coding genes ( Okazaki et al., 2002 ), transcripts are typically classified as lncRNAs when they do not contain any open reading frame >100 amino acids in length. Although few lncRNAs contain ORFs longer than predicted by pure chance ( Dinger et al., 2008 , 2011 ), they also show relatively low levels of evolutionary conservation overall, suggesting that they may encode short, evolutionarily divergent proteins similar to those observed in Drosophila ( Kondo et al., 2010 ). Recently, researchers detected a number of evolutionarily conserved sequences that do encode small proteins through both ribosome profiling and mass spectrometry ( Bazzini et al., 2014 ). However, analysis of other mass spectrometry experiments reveals that lncRNAs rarely produce detectable protein products ( Banfai et al., 2012 ; Slavoff et al., 2013 ). Furthermore, ribosome profiling experiments have indicated that while lncRNAs can associate with ribosomes, ribosome occupancy of lncRNAs displays features more congruent with untranslated regions (5′ UTRs) and other classical ncRNAs, such as small nucleolar RNAs (snoRNAs) and microRNAs (miRNAs; Guttman et al., 2013 ). Combined with data showing that a large fraction of lncRNA transcripts are retained in the nucleus ( Derrien et al., 2012 ), it suggests that lncRNAs impart functions as RNA transcripts. LncRNAs are distinguished from other ncRNAs subtypes by several different features. Inherent to the name, lncRNAs are classified as such based on a length of >200 nucleotides, distinguishing them from many ncRNAs including miRNAs, snoRNAs, and others. They are also distinct from transfer RNAs (tRNAs) as they are typically transcribed by RNA polymerase II (RNA Pol II), as opposed to RNA Pol III. Moreover, lncRNAs share many features with protein-coding messenger RNAs (mRNAs) – they are capped and polyadenylated. Many lncRNAs also contain multiple exons and are subjected to alternative splicing. However, in comparison to protein-coding transcripts, lncRNAs are roughly one-third as long, contain fewer exons (~2.8 exons in lncRNAs compared to 11 exons for protein coding genes), and are expressed at 10-fold lower levels on average ( Guttman et al., 2010 ; Cabili et al., 2011 ; Pauli et al., 2012 ). In addition, lncRNAs show a higher degree of tissue-specific expression than do protein-coding genes ( Cabili et al., 2011 ). Compared to protein-coding genes, retrotransposon sequences and tandem repeat elements are more frequently included in lncRNA sequences ( Ulitsky et al., 2011 ; Kelley and Rinn, 2012 ). These elements have been proposed to facilitate lncRNA function through either base pairing with other RNAs with similar repeat sequences, or through as yet unidentified mechanisms ( Gong and Maquat, 2011 ; Carrieri et al., 2012 ). The discovery that much of the genome is transcribed bi-directionally has led to a diverse and still not fully standardized categorization of lncRNAs based on genomic localization. Included in the class of lncRNAs are enhancer-related lncRNAs (eRNAs) or transcribed ultra-conserved region lncRNAs ( Figure 1A ), intronic lncRNAs ( Figure 1B ), large/long intergenic or intervening non-coding RNAs (lincRNAs; Figure 1C ), promoter associated lncRNAs ( Figure 1D ), and natural antisense transcripts (NATs; Figure 1E ). LincRNAs have been identified through examination of sequencing reads that map expressed transcripts without clearly defined ORFs to intergenic regions. These lincRNAs usually also possess signatures of active transcription including H3K4me3, polyadenylation signals, and RNA polymerase II occupancy ( Guttman et al., 2009 ). LncRNAs not localized to intergenic regions have been less readily identified and originally described as transcription “noise” due to overlap with protein-coding transcripts or known DNA-regulatory elements such as enhancers. Classification of lncRNAs based on genomic localization . Schematic examples of the classification of lncRNAs based on genomic localization. (A) Enhancer-associated RNAs result from direct, bi-directional transcription of enhancer elements Ultra-conserved enhancer elements are frequently transcribed as part of lncRNA sequences. (B) Intronic lncRNAs localize to the introns of protein-coding genes and are transcribed from the anti-sense (pictured) or sense strand (not shown). (C) LincRNAs are localized to gene deserts, far removed from proximal promoter elements from neighboring protein-coding genes. (D) Promoter associated lncRNAs are transcribed from segments within proximal promoters of the associated protein-coding gene on the anti-sense (opposite strand lncRNAs; shown) or sense (not shown) strand relative to the protein-coding gene. (E) Natural antisense lncRNAs are transcribed for the antisense strand of protein-coding genes and contain complementary sequences to segments of the mature mRNA. Protein-coding exons shown in yellow; lncRNA exons shown in red; overlapping sequence shown in purple. Reports suggest that intronic lncRNAs, which comprise up to 35% of non-coding transcripts, form the largest single class of lncRNAs ( Birney et al., 2007 ; St Laurent et al., 2012 ). Although intronic lncRNAs were originally thought to be unprocessed pre-mRNAs of protein coding genes, current estimates suggest that up to 80% of protein coding loci have transcriptionally active introns that are expressed independently from the protein coding pre-mRNA ( Dermitzakis et al., 2005 ; Louro et al., 2008 ; St Laurent et al., 2012 ). Further confirmation of the presence of intronic lncRNAs comes from reports that find many intronic lncRNAs localized to the cytoplasm, excluding the possibilities that intronic lncRNAs are the result of genomic DNA or unspliced pre-mRNA contamination in deep sequencing studies ( Kampa et al., 2004 ; Kapranov et al., 2007 ; Mercer et al., 2008 ). Intronic lncRNAs are transcribed from either the sense or antisense strand of the protein-coding gene in which they are encoded, further supporting independent transcriptional regulation ( Rinn et al., 2003 ; Bertone et al., 2004 ; Kampa et al., 2004 ). A relatively new class of ncRNAs, eRNAs, result from bidirectional transcription of enhancers. These sequences display H3K4me1 and H3K27ac modifications, and p300/CBP and RNA polymerase II occupancy, and thus show signatures of open or poised chromatin ( Heintzman et al., 2007 , 2009 ; Visel et al., 2009 ; De Santa et al., 2010 ; Kim et al., 2010 ; Orom et al., 2010 ). While many eRNAs are short (<200 nt), there are a considerable number of lncRNAs that display lincRNA-like chromatin signatures and overlap known enhancer sequences. Because of the shared enhancer sequence, these have been further classified as transcribed ultra-conserved region-associated lncRNAs to distinguish them from shorter eRNAs. NATs, on the other hand, were identified in bacteria and eukaryotes in the 1990s ( Wagner and Simons, 1994 ; Vanhee-Brossollet and Vaquero, 1998 ). More recent studies indicate that 50–70% of protein coding genes are also transcribed in the antisense direction, with half of these antisense transcripts being non-coding ( Carninci et al., 2005 ; Katayama et al., 2005 ; Galante et al., 2007 ). Studies have shown that many NATs display localized expression patterns that correspond inversely with their sense transcript counterparts, suggesting possible negative regulation of sense transcripts by NATs ( Vanhee-Brossollet and Vaquero, 1998 ; Alfano et al., 2005 ). In contrast, many lncRNAs without overlapping sequence display expression patterns that correlate with nearby protein-coding transcripts ( Luo et al., 2013 ). Despite the tremendous diversity of lncRNAs, their functional importance has been underappreciated and relatively understudied, in part due to the fact that they often fail to show clear evolutionary conservation ( Ulitsky et al., 2011 ; Basu et al., 2013 ). However, previous comparative genomic analyses have identified thousands of non-coding intergenic and intronic ultra-conserved sequence elements (UCEs) in the human genome ( Bejerano et al., 2004 ; Sandelin et al., 2004 ). Analysis of the genomic localization of UCEs shows that UCEs are preferentially localized to loci encoding DNA-binding proteins ( Sandelin et al., 2004 ). A recent study that incorporated transcriptome data from many different vertebrate species revealed that 4–11% of lncRNAs are conserved across the vertebrate lineage, and many of these map to UCE loci ( Ulitsky et al., 2011 ; Basu et al., 2013 ). Additionally, although the primary sequence of lincRNAs that are localized in close proximity to protein-coding genes often shows little sequence conservation, synteny between vertebrate lincRNAs and protein-coding genes is often conserved during vertebrate evolution ( Ulitsky et al., 2011 ; Qu and Adelson, 2012a , b ). Combined, this suggests that the synteny and evolutionary conservation of these non-coding elements helps facilitate the regulated expression of transcription factors through enhancer activity, functional ncRNA transcripts, or both. Analysis of sequence conservation within transcribed and regulatory regions of individual lncRNAs suggest that the proximal promoters display highest levels of evolutionary conservation ( Carninci, 2007 ; Ponjavic et al., 2007 ; Marques and Ponting, 2009 ; Chodroff et al., 2010 ). Peak conservation is observed ~43 bp upstream of the transcription start site, similar to the level of conservation seen across mouse and human protein coding genes ( Taylor et al., 2006 ; Chodroff et al., 2010 ). Furthermore, exonic sequence of lncRNAs is more highly conserved than intronic sequence, with exon splice sites showing highest evolutionary constraint ( Chodroff et al., 2010 ). Short sequences within the lncRNAs are also frequently conserved. Stringent identifications of miRNAs localized within lncRNA sequence identified 97 lncRNAs that function as potential precursors to miRNA clusters ( He et al., 2008 ). These miRNA sequences display a minimum 98% homology between rat and mouse, a far greater sequence conservation than observed for lncRNAs as a whole ( He et al., 2008 ). Quantification of the number of lncRNAs present across multiple species has elicited a wide range of estimates for the number of vertebrate lncRNAs. Stringent estimates suggest that 1133 lncRNAs are expressed during zebrafish development ( Pauli et al., 2012 ). Consistent with an evolutionary increase in number, size, and divergence of regulatory elements as species become more complex ( Mazumder et al., 2003 ; Frith et al., 2005 ; Taft et al., 2007 ), conservative estimates from recent GENCODE sequencing builds in mouse and humans (July, 2013) indicate the presence of 4074 and 13,870 lncRNAs, respectively ( Derrien et al., 2012 ; Harrow et al., 2012 ). Estimates from mouse suggest that 849 of the 1328 lncRNAs examined by in situ hybridization show specific expression patterns in the adult brain ( Mercer et al., 2008 ). More comprehensive analysis using RNA deep-sequencing technologies will help further elucidate and identify the exact number of lncRNAs expressed during neuronal development. Of the 13,870 identified human lncRNAs, approximately one-third are unique to the primate lineage ( Derrien et al., 2012 ), suggesting that ncRNA-dependent regulation of brain development may have contributed to the evolution of higher cognitive functions ( Barry and Mattick, 2012 ; Barry, 2014 ). Consistent with this idea, 47 of 49 conserved sequences across evolution displayed sequence substitution rates statistically higher between human and chimpanzees than rates compared to other sequences across amniote evolution ( Pollard et al., 2006 ). Of these human accelerated regions (HARs) that are non-coding, a quarter of these mapped to locations adjacent to genes that regulate neural development ( Pollard et al., 2006 ). HAR1F, the most rapidly evolving sequence of all, encodes a lncRNA that is prominently expressed in the developing and adult brain. Although the function of HAR1F is still unknown, this presents a tantalizing link between lncRNAs and the formation of the proportionally larger and more complex human brain ( Pollard et al., 2006 ). The large number of lncRNAs that display neuronal-specific expression suggests an important role of lncRNAs in the neuronal diversification seen in higher vertebrates ( Cao et al., 2006 ; Amaral et al., 2008 ; Chodroff et al., 2010 ; Qureshi et al., 2010 ). Additionally, the spatially and temporally restricted expression patterns of many lncRNAs indicate that their expression is tightly regulated, suggesting that lncRNAs may control the specification and function of individual neuronal subtypes ( Mercer et al., 2008 ). While functional characterization of neuronal-enriched lncRNAs is still limited, broader studies of lncRNA function have implicated lncRNAs as regulators of transcription through both epigenetic regulations of chromatin structure and RNA-transcription factor interactions. Here we focus on reviewing recent advances in the identification and functional analysis of lncRNAs implicated in transcriptional regulation control of neural development.