Long non‐coding RNAs in motor neuron development and disease

Abstract Long non‐coding RNAs (lncRNAs) are RNAs that exceed 200 nucleotides in length and that are not translated into proteins. Thousands of lncRNAs have been identified with functions in processes such as transcription and translation regulation, RNA processing, and RNA and protein sponging. LncRNAs show prominent expression in the nervous system and have been implicated in neural development, function and disease. Recent work has begun to report on the expression and roles of lncRNAs in motor neurons (MNs). The cell bodies of MNs are located in cortex, brainstem or spinal cord and their axons project into the brainstem, spinal cord or towards peripheral muscles, thereby controlling important functions such as movement, breathing and swallowing. Degeneration of MNs is a pathological hallmark of diseases such as amyotrophic lateral sclerosis and spinal muscular atrophy. LncRNAs influence several aspects of MN development and disruptions in these lncRNA‐mediated effects are proposed to contribute to the pathogenic mechanisms underlying MN diseases (MNDs). Accumulating evidence suggests that lncRNAs may comprise valuable therapeutic targets for different MNDs. In this review, we discuss the role of lncRNAs (including circular RNAs [circRNAs]) in the development of MNs, discuss how lncRNAs may contribute to MNDs and provide directions for future research.


| INTRODUC TI ON
Over the past decade, the advent of high-throughput sequencing technologies coupled with advanced bioinformatics analysis has allowed an in-depth look into the non-coding part of the genome with exceptional resolution and scale. This has revolutionized our understanding of mammalian genome architecture, activity and regulation and led to the discovery that only ±2% of the human genome encodes proteins. This is remarkable given that the majority of nucleotides are transcribed, thus giving rise to an exquisitely high number of non-coding RNAs (ncRNAs) (Carninci et al., 2005;Djebali et al., 2012;Dunham et al., 2012). NcRNAs are subdivided on basis of their length into small ncRNAs (<200 nucleotides; e.g. microRNAs (miRNAs), small interfering RNAs (siRNAs) and piwi-interacting RNAs (piRNAs) and long ncRNAs (lncRNAs; >200 nucleotides), that differ in their biogenesis and function. LncRNAs show enormous functional diversity, influencing (post-)transcriptional and translational diversification of specific genes and gene networks (Qureshi & Mehler, 2012;. It is estimated that approximately 2% of the human genome is transcribed into lncRNAs (Managadze et al., 2013;Palazzo & Lee, 2015). LncRNAs are mainly transcribed by RNA polymerase II (Guttman et al., 2009) and based on their location in the genome can be classified into the following five categories: (a) sense lncRNAs, transcripts transcribed from the sense strand of protein-coding genes and that contain one or more exons; (b) antisense lncRNAs, transcripts transcribed from the opposite strand of protein-coding genes and that contain one or more exons; (c) bidirectional lncRNAs, transcripts transcribed from the opposite strand that share the same promoter as another gene and that are usually located <1,000 base pairs (bp) away from each other in close genomic proximity; (d) intronic lncRNAs, transcripts transcribed from intronic regions without showing overlap with exons; and (e) intergenic lncRNAs, independent transcripts located in between two genes (Derrien et al., 2012;Ma et al., 2013). Another interesting class of lncRNAs are circular RNAs (circRNAs). CircRNAs are produced by back-splicing events that result in a closed loop structure, where a covalent bond is formed by linking 5' (splice donor) and 3' (splice acceptor) RNA splice sites. CircRNAs can be transcribed from both exonic and intronic regions, but the majority of circRNAs are exonic (Chen, 2016;Jeck et al., 2013;Memczak et al., 2013).
Mechanistically, lncRNAs can act at the transcriptional and post-transcriptional level to affect transcription, RNA processing and translation (Elling et al., 2016). Mediation of chromatin interactions, involvement in the formation of nuclear structures (e.g. paraspeckles), formation of decoys for transcription factors or as scaffolds for DNA-binding proteins and RNA-binding proteins (RBPs), functioning as miRNA sponges and control of mRNA decay are core functions of lncRNAs ( Figure 1; Marchese et al., 2017;Quinn & Chang, 2016). Even though the majority of lncRNAs is present in the nucleus, some lncRNAs are found in the cytoplasm (Kung et al., 2013;Noh et al., 2018;Rinn & Chang, 2012;Ulitsky & Bartel, 2013). In the cytoplasm, lncRNAs are involved in processes such as RNA editing, RNA splicing and protein machinery assembly for translation initiation (Noh et al., 2018;Rashid et al., 2016). The majority of lncRNAs are of low abundance. In general, expression of lncRNAs is approximately 10-fold lower as compared to protein-coding transcripts (Cabili et al., 2011;Ravasi et al., 2006). This difference can in part be explained by prominent differences in lncRNA expression between cell types and tissues, and the strong spatiotemporal regulation of lncRNA expression as compared to mRNA expression (Cabili et al., 2011;Field

K E Y W O R D S
long non-coding RNA, circular RNA, motor neuron development, spinal muscular atrophy, amyotrophic lateral sclerosis F I G U R E 1 Schematic illustration highlighting the roles of long non-coding RNAs (lncRNAs) in motor neuron (MN) development and disease. (a) Non-coding RNA (lncRNA and circRNA) expression is enhanced during MN differentiation suggesting roles in the regulation of MN development from progenitor stages to mature motor neurons (MNs). An interesting link has been established between lncRNA expression and homeobox (Hox) gene expression. LncRNAs influence Hox gene expression but can also be derived from the Hox gene cluster. A role for lncRNAs in the specification of neuron subtypes has been proposed. (b) LncRNAs and circular RNAs have been implicated in a wide range of cellular and molecular functions in developing MNs, all of which (in)directly impact on gene expression regulation. LncRNAs have been reported to play a role in transcription regulation, epigenetic modulation, paraspeckle formation, RNA processing, translational control, miRNA sponging and synaptic enrichment of RNAs and proteins. CircRNAs have been reported to play a role in transcription regulation, RNA processing, RBP and miRNA sponging, RBP and miRNA transport and synaptic enrichment of RNAs and proteins. (c) Several lncRNAs have been linked to MN disease, three examples of which are shown here. Nuclear-enriched abundant transcript 1 (NEAT1)_2 is a lncRNA that regulates paraspeckle assembly and that clusters with several RNA-binding proteins (RBPs) and inhibits their function. Changes in NEAT1_2 and paraspeckle assembly in general may contribute to the pathogenesis of amyotrophic lateral sclerosis (ALS). C9ORF72-AS is an antisense lncRNA that forms RNA foci and toxic dipeptide repeat (DPR) proteins in different cellular compartments because of the presence of hundreds to thousands hexanucleotide repeats in the 5' region of the C9ORF72 locus in C9-ALS patients. In RNA foci C9ORF72-AS aberrantly interacts with RBPs which may contribute to cellular toxicity. ATAXIN2-AS is an antisense lncRNA that forms CUG repeatcontaining RNA aggregates recruiting RBPs leading to neurotoxicity. In ALS, patients may carry an expanded CAG repeat region in the 5' coding part of the gene. Overall, lncRNA dysregulation may lead to altered RNA metabolism as a result of interference with RBPs and in case of C9-ALS the accumulation of toxic DPR proteins. These defects contribute to MN degeneration and muscle atrophy as observed in ALS patients et al., 2019;Yan et al., 2013;Yunusov et al., 2016). Ubiquitously expressed lncRNAs are generally expressed at high levels, while cell type-or tissue-specific lncRNAs, such as those in MNs, are often expressed at lower levels (Jiang, Li,, et al., 2016). In addition to relatively low levels of expression, many lncRNAs show low levels of evolutionary and sequence conservation (Basu et al., 2013;Ulitsky et al., 2011). It has been suggested that evolutionary conserved lncRNAs may have important general functions, while species-specific lncRNAs could play relevant roles in biological processes that are crucial for a specific species (e.g. expansion of the brain in primates). Most of the lncRNAs discussed in this review are conserved among mammals (Table 1).
A remarkably large number of annotated lncRNAs, that is, approximately 40% (4,000 -20,000 lncRNA genes), is expressed specifically in the brain (Derrien et al., 2012). In situ hybridization data from the Allen brain atlas show the expression of hundreds of lncRNAs in specific regions of the mouse brain, in specific types of neural cells and even in specific subcellular compartments. Several of these expression patterns are similar to those described for coding mRNAs, hinting at important neuronal functions for lncRNAs . Indeed, lncRNAs have been linked to processes such as neuron development, neurite growth, synaptic transmission, memory consolidation and ageing (Derrien et al., 2012;Mehler & Mattick, 2007;Mercer et al., 2008;Pereira Fernandes et al., 2018;Shi et al., 2017) (Figure 1a,b, see also

| CAT7
Another screen designed to detect lncRNAs co-precipitating with CAT7 shares high sequence similarity to a non-syntenic Danio rerio (zebrafish) analog, zcat7l. Depletion of zcat7l by morpholino interference resulted in microcephaly at 48 hr post-fertilization (hpf) and mortality starting on day 5. These phenotypes were successfully reverted by co-expression of the human CAT7 lncRNA isoform, indicating possible conserved functions between these non-syntenic lncRNAs. The phenotypes found following zcat7l depletion were also observed following knockdown of the zebrafish polycomb complex proteins bmi1a/b, and zcat7l and bmi1a/b genetically interacted in combined knockdown experiments. Overall, these data indicate a functional link between CAT7/zcat7l lncRNAs and PRC1 in human and zebrafish. More generally, these findings show that lncRNAs can regulate and promote transcription activation at specific loci (Ray et al., 2016).

| Gm12688 and Gm14204
In another study, single-cell topological RNA sequencing analysis was used to characterize the mechanisms underlying in vitro differentiation of mouse embryonic stem cell (mESC)-derived MN cultures (Rizvi et al., 2017). Specific temporal regulation of lncR-NAs was found during pluripotency, at the neural progenitor stage and during the transition to neuronal maturation. Different stages of MN differentiation are characterized by waves of Hox gene activation ( Figure 1a). Intriguingly, these waves were accompanied by an up-regulation of lncRNAs that originate from the opposite strand of these genes. This co-expression is in line with previous reports identifying lncRNAs as regulators of Hox gene clusters Ponjavic et al., 2009;Rinn et al., 2007;. In addition, several lncRNAs were detected in other neuron types in the cultures, which contain MNs and concomitant cell types (i.e. those found in the spinal cord). For example, Gm12688, an intergenic lncRNA located near Foxd3 and transcribed from the opposite strand, was found in V1 interneurons. Gm14204, an intergenic lncRNA located near Solute carrier family 32 member 1 (Slc32a1) and transcribed from the opposite strand, was restricted to V1 and V2b GABAergic interneurons (Rizvi et al., 2017). The discovery of these intergenic lncRNAs suggests that lncRNAs may play a role in neuronal diversification in the spinal cord.

| Lhx1os, Lnc-MN1, LncMN-2 and LncMN-3
Stage-specific expression of lncRNAs was also detected in an analysis of the long non-coding transcriptome of mESC-and human-

| Malat1, Meg3, Rmst, Xist and Miat
In addition to studies reporting general changes in lncRNA expression, other work has focused on the spatial enrichment of these noncoding RNAs in specific subcellular compartments. Initially, lncRNAs had predominantly been detected in the nucleus and were therefore studied for potential roles in the regulation of gene expression (Sun et al., 2018;Ulitsky & Bartel, 2013). The presence of lncRNAs in other subcellular compartments hinted at the exciting possibility of additional and different molecular functions. Briese et al. (2016) investigated the abundance of several lncRNAs in the axonal or somatodendritic compartment, including that of metastasis-associated lung adenocarcinoma transcript 1 (Malat1), Meg3, rhabdomyosarcoma 2-associated transcript (Rmst), X-inactive-specific transcript (Xist) and Myocardial infarction-associated transcript (Miat). While some lncRNAs showed similar levels in different compartments, Meg3 and Rmst were enriched in somatodendritic and axonal fractions, respectively (Briese et al., 2016). The molecular mechanisms dictating this highly specific distribution are unknown. However, a number of lncRNAs can interact with proteins and it is plausible that protein:RNA complexes play a role in compartment-specific lncRNA distribution (Briese et al., 2016;Quan et al., 2017).

| Lncrps25
While many of these studies reveal interesting lncRNA expression patterns that hint at roles during MN differentiation, recent work has also begun to explore the functional role of lncRNAs in MNs in vivo. For example, the lncRNA lncrps25 was found to be strongly expressed in the embryonic zebrafish CNS and co-expressed with mnx1, while lncrps25 knockdown induced body abnormalities and striking changes in locomotor behaviour, that is, decreased swimming distance and speed (Gao et al., 2020). These phenotypes were caused by defects in primary MNs that showed reduced axon length and branching. Lncrps25 depletion led to a reduced expression of oligodendrocyte transcription factor 2 (olig2) in brain and spinal cord. Olig2 co-expression could partially rescue the motor phenotypes caused by lncrps25 depletion. Olig2 is part of the Olig gene family that encode for basic helix-loop-helix transcription factors that are required for MN and oligodendrocyte development (Lu et al., 2002;Park et al., 2002). Based on the effect of lncrps25 depletion, and lncrps25 and olig2 expression patterns, it was proposed that lncrps25:olig2 interactions occur at the neural plate stage to eventually affect MN development (Gao et al., 2020). However, the precise mechanistic details of this lncRNA-mRNA pathway remain to be defined. A recent study focusing on defining lncRNA networks in oligodendrocytes identified the oligodendrocyte-specific lncRNA LncOL1. LncOL1 is involved in oligodendrocyte differentiation and CNS (re)myelination (He et al., 2017). Interestingly, a peak in genomic occupancy by Olig2 was detected around the transcription start site of LncOL1, suggesting a role for Olig2 in the transcriptional regulation of LncOL1. Therefore, Olig2 appears to be a shared factor in the regulation of MN and oligodendrocyte differentiation through interactions with different lncRNAs (lncrps25 and LncOL1). This is intriguing since the specification of MN and oligodendrocytes from ventral progenitor cells is a sequential process mediated by olig2 (Ravanelli & Appel, 2015). Further, oligodendrocytes are known to play a role in MNDs such as ALS. Oligodendrocyte dysfunction contributes to MN excitability and death, and consequently to subsequent oligodendrocyte-related pathology (Ferraiuolo et al., 2016). It will be interesting to determine how Olig2:lncRNA networks interact during MN and oligodendrocyte development, and how their dysfunction may lead to disease.

| CircRNAs in developing motor neurons
CircRNAs are highly enriched in the brain, dynamically regulated and conserved across species (Rybak-Wolf et al., 2015). The mechanisms that cause brain-enriched circRNA expression remain to be determined. One explanation is that many circRNAs derive from genes that are enriched in brain tissue, but the existence of neuron-specific biogenesis mechanisms has also been proposed (van Rossum et al., 2016). Another interesting observation is that whereas linear lncRNAs often display cell type-specific expression (e.g. in MNs), cell type-specific circRNAs are rare. This difference may arise from the fact that whereas many lncRNAs originate from intergenic stretches of the genome that lack protein-coding genes, circRNAs are generally produced from protein-coding transcripts (Memczak et al., 2013;Ransohoff et al., 2018;Salzman et al., 2012). MN-specific circRNAs therefore mainly arise from MN-specific genes, while the intergenic regions that produce other lncRNAs may be susceptible to different kinds of (post)transcriptional regulation. CircRNAs are differentially expressed across brain regions, subcellular compartments and regulated at pre-and postnatal stages (Chen & Schuman, 2016;Memczak et al., 2013;Rybak-Wolf et al., 2015;Venø et al., 2015;You et al., 2015). Many of these RNA molecules derive from genes that encode for synaptic proteins or proteins with essential roles in neural development. Further, the expression of several circRNAs is influenced by neuronal activity and they are shown to accumulate during ageing (Venø et al., 2015;Westholm et al., 2014). Together, these observations hint at roles for circRNAs in a wide range of neuronal processes in brain development, function and disease (van Rossum et al., 2016).
The functional roles and mechanism-of-action of most circRNAs remain poorly understood. A detailed discussion of the function of circRNAs is beyond the scope of this review, but several other recent reviews cover our current knowledge of the molecular effects and functional roles of circRNAs (Ebbesen et al., 2016;Meng et al., 2019;van Rossum et al., 2016;Salzman, 2016;Sekar & Liang, 2019).
Translation into small peptides, miRNA sponging and sequestration of RBPs are some of the currently accepted mechanisms by which circRNAs can act ).

| FUS-linked circRNAs
CircRNAs have also been detected in MNs. Hundreds of circR-NAs were found in mESC-derived MNs and, similar to that has been demonstrated for other lncRNAs, their expression increased during MN differentiation (Errichelli et al., 2017). The majority of identified circRNAs was conserved in hiPSC-derived MNs. In addition to expression profiling, this study also provided insight into the mechanisms underlying circRNA biogenesis in MNs. The ALS-associated protein fused in sarcoma (FUS) was shown to promote circRNA formation in MNs without affecting their linear counterparts. FUS is an RBP with roles in RNA splicing and may contribute to circRNA formation by regulating the back-splicing events that lead to RNA circularization. In line with this idea, crosslinking immunoprecipitation (CLIP) assays revealed a specific enrichment of FUS at circularizing exon-intron regions of selected circRNAs (c-1, c-2, c-13, c-16, c-48, c-80, c-82, c-84, c-88) (Errichelli et al., 2017;Verheijen & Pasterkamp, 2017).

| SMN circRNAs
Although the biogenesis and functions of circRNAs are often evolutionarily conserved, in some instances the abundance, the presence of specific isoforms or functions of circRNAs are species-specific (Rybak-Wolf et al., 2015). Here, it is worth mentioning the human survival motor neuron (SMN) locus, which is able to originate a surprisingly large number of different exonic circRNAs (Ottesen et al., 2019). SMN protein is essential for MN development and survival and its loss causes SMA (Chaytow et al., 2018;Fallini et al., 2012;Lefebvre et al., 1995). Most of the reported functions of SMN are linked to RNA processing. However, it is known that SMN protein accumulates in the axons of MNs during development hinting at roles in, for example, local gene expression regulation (Gabanella et al., 2005;Giavazzi et al., 2006;Groen et al., 2013;Zhang et al., 2006). In addition to SMN protein, the SMN locus also generates a vast array of circRNAs. Using in-

| LN CRNA S IN MOTOR NEURON DISE A SE
LncRNAs are strongly regulated by exposure to drugs, environmental factors (e.g. toxins) or as a consequence of pathological situations.
In many instances, regulation of lncRNA expression confers neuroprotection or induces anti-apoptotic responses (Table 3). Given this regulation of lncRNAs, their dynamic expression patterns in MNs and their emerging roles in MN development and function, it is not surprising that lncRNA dysregulation has been implicated in MNDs.
Understanding the mechanisms-of-action and functions of lncRNAs may assist the development of new therapies for MNDs. Here, we discuss and summarize the expression and functional role of lncR-NAs in different MN-related diseases (see also Table 4), extending on previous reviews (Chen & Chen, 2020;.

| Amyotropic lateral sclerosis (ALS)
ALS is a fatal disease characterized by the progressive loss of MNs in brain and spinal cord. About 90%-95% of cases lack a clearly iden- In vivo (mice)  (Continues)

| NEAT1
NEAT1 is a well-characterized lncRNA that organizes nuclear structures called 'paraspeckles'. Paraspeckles contain proteins involved in transcription and RNA processing (Clemson et al., 2009;Fox & Lamond, 2010). So far, two isoforms of NEAT1, NEAT1_1 and NEAT1_2, have been described that are spliced by alternative processing of the 3'-untranslated region Naganuma et al., 2012;Yamazaki et al., 2018). NEAT1 is classified as one of the architectural RNAs (arcRNA), that function by providing essential scaffolds for the formation of nuclear bodies, with NEAT1_2 playing a predominant role in paraspeckle formation (Chujo et al., 2016;Naganuma et al., 2012). Paraspeckle formation is a tightly regulated process that occurs parallel to RNApolII transcription of NEAT1_2 and binding of several paraspeckle proteins (PSPs) to NEAT1 (Mao et al., 2011;Yamazaki et al., 2018). So far, more than 60 PSPs have been identified, most of which are RBPs (Yamazaki & Hirose, 2015).
RNAi studies identified seven essential PSPs that are able to form a paraspeckle with ~ 50 NEAT1_2 molecules (Chujo et al., 2017;Naganuma et al., 2012). Alu repeats, implicating NEAT1 function in mRNA export (Chen & Carmichael, 2009 (Sunwoo et al., 2009;Zhang et al., 2013). Further, proteasome inhibi-  (Shelkovnikova et al., 2014). This disrupts neuroprotective mechanisms and may contribute to the aggressive disease phenotype (early onset and fast progression) observed in FUS-ALS cases . Although these studies suggest that ALS-related RBP mutations affect NEAT1 expression and may thereby influence paraspeckle formation, the precise role of NEAT1 in ALS pathogenesis is incompletely understood. A recent study shows that the activation and up-regulation of endogenous NEAT1 levels using the CRISPR-Cas9 system in NSC-34 cells induces neurotoxicity.
NEAT1_2, but not NEAT1_1, was proposed to mediate this neurotoxic effect (Suzuki et al., 2019). These data support the idea that altered NEAT1 expression in ALS leads to defects in paraspeckle formation causing cell death and neurodegeneration (Figure 1c). However, a neuroprotective role for NEAT1 is proposed in Parkinson's disease (PD) (Simchovitz et al., 2019). Increased NEAT1 levels were found in the substantia nigra of PD patients and in drug-induced oxidative stress in vitro PD models. This increase in expression was also associated with formation of NEAT1 paraspeckles both in vitro and in vivo. Further, neuroprotective agents, like fenofibrate and simvastatin, were found to induce NEAT1 expression, whereas NEAT1 depletion increased oxidative stress-induced cell death (Simchovitz et al., 2019).
Several of the genes associated with ALS give rise to sense or antisense RNAs which are not translated into protein but act as ln-cRNAs. These include naturally occurring (anti)sense RNAs but also RNAs that gain lncRNA functions as a result of genetic changes (e.g. repeat expansions). A few prominent examples are discussed below.

| C9ORF72
As indicated in previous sections, the expansion of a repeat region of six-nucleotide motifs (GGGGCC) n (G 4 C 2 ) in the 5' region of the In addition to different sense isoforms of C9ORF72, one or more antisense transcripts that arise from the same promoter have been detected. The mutant hexanucleotide repeat regions of C9ORF72 are transcribed in a bidirectional fashion, producing sense (C9ORF72-S), with G 4 C 2 repeats, and antisense (C9ORF72-AS), with GGCCCC (G 2 C 4 ) repeats, RNAs (Donnelly et al., 2013;Mizielinska et al., 2013;Renton et al., 2011). Both expanded repeat-containing transcripts (sense and antisense) can be translated into poly-dipeptides (dipeptide repeat [DPR] proteins) which accumulate in MNs of C9-ALS patients (Figure 1c; Freibaum & Taylor, 2017;Gendron et al., 2017;Saberi et al., 2018;Wen et al., 2014;Zu et al., 2013). Several studies have focused on understanding the function of C9ORF72-S RNA and its associated proteins (reviewed in Moens et al., (2017)). Overexpression of G 4 C 2 repeats in cell culture, C. elegans, Danio rerio and Drosophila melanogaster led to deleterious phenotypes, which provided evidence that C9ORF72-associated toxic gain-of-function contributed to the disease (Burguete et al., 2015;Freibaum et al., 2015;Kramer et al., 2016;Mizielinska et al., 2014;Wen et al., 2014;Xu et al., 2013;Zu et al., 2011). Similarly, adeno-associated virus-mediated over-expression of 66 G 4 C 2 repeats in mice led to an accumulation of RNA foci and DPR proteins together with neuronal cell loss, TDP-43 pathology and cognitive and motor dysfunction (Chew et al., 2015). The functional relevance of C9ORF72-AS is less well understood. Interestingly, C9ORF72-AS RNAs harbour a few miRNA-binding sites in the first exon (Douglas, 2018). Similar to

C9ORF72-S, C9ORF72-AS forms RNA foci in MNs and sporadically in interneurons (in frontal cortex and spinal cord) of ALS and FTD
patients (DeJesus-Hernandez et al., 2011;Gendron et al., 2013;Mizielinska et al., 2013;Zu et al., 2013). However, a higher frequency of C9ORF72-AS as compared to C9ORF72-S RNA foci and DPR proteins is observed in MNs of C9-ALS patients (Cooper-Knock et al., 2015). C9ORF72-AS RNA foci are found in the peri-nucleolar regions, suggesting that nucleolar defects or stress may contribute to C9-ALS pathogenesis (Kwon et al., 2014;Tao et al., 2015). It is important to note, however, that RNA foci formed by either C9ORF72-S C9ORF72-S expansions are reported to form G-quadruplexes, which regulate transcription and gene expression (Fratta et al., 2012;Haeusler et al., 2014;Reddy et al., 2013;Wang, Goodrich, et al., 2019). C9ORF72-AS repeats do not form G-quadruplexes but can form i-Motifs and hairpin structures, that can affect genome stability and transcription (Assi et al., 2018;Kovanda et al., 2015).
Further, the crystal structure of C9ORF72-AS repeat RNA showed that C9ORF72-AS RNA forms a double helix structure with tandem C:C mismatches, that could attract proteins recognizing these structures (Dodd et al., 2016). Therefore, repeat expansions in C9ORF72 may not only affect ALS pathogenesis through the formation of RNA foci or production of DPR proteins, but also indirectly by regulating the secondary structure of DNA and RNA, and thereby gene expression.
To determine how ALS-associated C9ORF72 repeat expansions cause ALS, different in vivo models have been generated. Expression of C9ORF72 repeat expansions that form DPR proteins in Drosophila is toxic leading to adult-onset neurodegeneration with reduced lifespan (Mizielinska et al., 2014). Expression of repeat expansion RNA that is not translated into DPR proteins did not induce neurotoxicity, suggesting that the toxic effect of C9ORF72 repeat expansions may be mediated by DPR proteins (Mizielinska et al., 2014). To better understand the role of C9ORF72-S and C9ORF72-AS repeat RNA expansions, another study created RNA-only Drosophila models, to avoid the effects of DPR proteins. Flies were generated to express 1) small (~100) RNA repeats in sense or antisense directions either as a processed polyadenylated transcript (named polyA) or as intronic versions fused with green fluorescent protein exons, or 2) large RNA-only repeats (>1,000) in the sense direction . Both, sense and antisense polyA repeat RNA was found to accumulate in the cytoplasm of adult Drosophila neurons, whereas intronic RNAs with small and large repeats accumulated in the nucleus forming RNA foci, mimicking pathological conditions. Even though the RNA foci sequestered endogenous Drosophila RBPs, no neurodegenerative phenotypes or toxicity were observed following either sense or antisense RNA expression (small and large repeats). This suggests that mutant RNA is not sufficient to induce disease-relevant toxicity . However, expression of repeat expansions in a Danio rerio (zebrafish) model induced neurodegenerative phenotypes (Swinnen et al., 2018). Injection of ~70 C9ORF72-S (G 4 C 2 ) or C9ORF72-AS (G 2 C 4 ) repeats in zebrafish embryos induced motor axonopathy. A dose-dependent effect was observed on axon outgrowth and branching. Further, it was shown that toxicity induced by repeat RNA was independent of DPR proteins, suggesting that the presence of repeat RNA in the absence of DPR proteins is sufficient to induce MN toxicity in zebrafish (Swinnen et al., 2018). These differences in sensitivity for RNA-induced toxicity may be explained by species differences or differences in experimental approaches. However, even within one species (mouse) different effects of the expression of repeat RNAs were found. Even though in most models RNA foci and DPR protein inclusions formed from both sense and antisense C9ORF72 RNAs were observed, some models developed motor or cognitive defects, whereas others were unaffected (Jiang, Zhu, et al., 2016;Liu et al., 2016;O'Rourke et al., 2015;Peters et al., 2015). Further work is needed to carefully assess the role of RNA toxicity in C9-ALS. Expression of longer repeat sequences (in the pathological range) may be required to model Targeting of C9ORF72-S by antisense oligonucleotides (ASOs) and of the G-quadruplex structure formed by C9ORF72-S by specific binding molecules have been explored as therapeutic strategies for C9-ALS (Donnelly et al., 2013;Jiang, Zhu, et al., 2016;Lagier-Tourenne et al., 2013;Simone et al., 2018;Zamiri et al., 2014;Zhang et al., 2015). ASOs were effective in reducing RNA foci and reversed gene expression changes in C9-ALS hiPSC-derived MNs (Sareen et al., 2013). Even specific targeting of the repeat expansion by ASOs of ASOs could lead to adverse side effects, because of their off-target effects depending on the sequence and modification chemistry of the motifs used to stabilize the ASO-molecule for protection against cellular degradation . Off-target effects of ASOs are caused by binding to other mRNAs leading to an undesirable biological effect and activation of the immune system.
Non-specific target binding is avoided by careful selection of candidate ASOs. Human clinical ASO candidates are routinely screened in silico for potential mismatches and tested empirically for their effect on mRNAs (Kamola et al., 2015;Monia et al., 1992). Activation of  et al., 2004). However, adaptations in the ASO design have largely mitigated this effect (Bennett & Swayze, 2010;Scoles et al., 2019;Sewell et al., 2002). Despite extensive research in ASO development, unrecognized off-target issues remain . So, it is important that each ASO molecule is evaluated independently and considered for a complete toxicological assessment.
First, formation of double-stranded RNA structures by hybridization of overlapping complementary RNA sequences in sense and antisense transcripts can activate the RNA interference pathway (Polikepahad & Corry, 2013;Werner et al., 2014). Second, antisense transcripts may cause transcriptional interference via the displacement of sense transcription factors in the sense promoter region (Wight & Werner, 2013). Third, antisense transcripts may epigenetically regulate sense transcripts through recruitment of chromatin remodelling factors and modulation of their activity (Kanduri, 2008;Malecová & Morris, 2010;Yu et al., 2008). Given these possible roles of antisense transcripts, sense oligonucleotides (SOs) designed against antisense transcripts may be a valuable therapeutic strategy.
However, further studies are needed to better define the role of C9ORF72-AS in C9-ALS.

| ATAXIN
The ALS risk gene ataxin-2 (ATXN2) also gives rise to antisense transcripts, ATXN2-AS, with increased expression in ALS patient tissue (Li, Sun, et al., 2016). ATXN2 protein is ubiquitously expressed in neuronal and non-neuronal tissues and is localized to the cytoplasm (Huynh et al., 1999). ATXN2 associates with polyribosomes in the endoplasmic reticulum and regulates mRNA translation (Satterfield & Pallanck, 2006). Further, it influences mRNA maturation, energy metabolism and endocytosis (Huynh et al., 1999;Magaña et al., 2013). In ALS, mutant ATXN2 interacts with FUS and TDP-43 in an RNA-dependent interaction manner which contributes to disease pathogenesis (Elden et al., 2010;Van den Heuvel et al., 2014;Yokoshi et al., 2014). The interaction of ATXN2 and TDP-43 was first identified in a screen for toxic modifiers of TDP-43. This work showed that an increase in the length of a polyQ repeat in ATXN2 (from 22 normal to 27-33 glutamines in ALS) is significantly associated with enhanced risk for developing ALS (Elden et al., 2010).
ATXN2 polyQ repeat expansions of more than 33 repeats are associated with another neurodegenerative disease spinocerebellar ataxia type 2 (SCA2; Li, Sun, et al., 2016;Van den Heuvel et al., 2014). SCA2 is an autosomal-dominant disorder caused by the expansion of CAG repeats in the N-terminal coding region of ATXN2. Clinical features of SCA2 are characterized mainly by limb and gait ataxia, dysarthria and abnormal eye movements (Cancel et al., 1997;Geschwind et al., 1997). The symptoms of SCA2 are found to be clearly of cerebellar origin and involve cerebellar circuits. However, a typical oculardependent feature is also observed in SCA2, where slow or absent saccades are displayed arising from the degeneration of oculomotor neurons. Other frequently observed symptoms include dystonia, myoclonus as well as neuropathy, muscle spasticity and frontal-executive dysfunction (Geschwind et al., 1997;Scoles & Pulst, 2018).  (Li, Sun, et al., 2016). These mutant CAG/CUG transcripts aberrantly interact with RBPs involved in, for example, ribosomal RNA (rRNA) processing and splicing (Figure 1c; Swinnen et al., 2020). Despite these promising results, whether and if so how ATXN2-AS contributes to ALS pathogenesis remains to be established.

| Other lncRNAs in ALS
Whole transcriptome RNA sequencing (RNA-seq) analysis in peripheral blood mononuclear cells (PBMCs) and spinal cord from sALS and fALS patients identified several differentially expressed lncR-NAs . Several of these lncRNAs were antisense RNAs that were deregulated in sALS, for example, ZEB1-AS and ZBTB11-AS, and with sense counterparts that have reported functions in transcription regulation (Keightley et al., 2017;Li, Xie, et al., 2016). Interestingly, a few of these antisense lncRNAs can potentially target genes associated with neurodegenerative diseases, that is, UBXN7-AS, ATG10-AS and ADORA2A-AS (Guo & Qi, 2017;Lee et al., 2015;Villar-Menéndez et al., 2014). The detection of lncRNAs in PBMCs and blood raises the possibility that lncRNAs may be used as biomarkers for assessing clinical disease course or characteristics.

| Spinal muscular atrophy (SMA)
SMA is a neurodegenerative disease characterized by profound muscle weakness attributed to progressive spinal MN degeneration, observed within weeks to months after birth or even at birth.
SMA is classified as a rare disease that has a worldwide incidence of one in 6,000 -10,000 newborns. However, it is the second most common autosomal recessive disease after cystic fibrosis and is the most common monogenic defect leading to infant mortality (Lunn & Wang, 2016). SMA is caused by reduced levels of SMN protein because of recessive mutations of the SMN1 gene and retention of a variable copy number of the highly homologous SMN2 gene (Lefebvre et al., 1995;Lorson et al., 1999). The critical sequence difference between SMN2 and SMN1 is one nucleotide (C → T) in exon 7 at a splice enhancer site. This change results in alternative splicing of SMN2 pre-mRNA Monani et al., 1999). This leads to a preferential exclusion of exon 7 from most transcripts (SMNΔ7), leading to a truncated unstable SMN protein, contributing to reduced levels of overall functional SMN protein. Individuals affected by SMA show loss of SMN protein and carry at least one copy of SMN2. The number of SMN2 copies inversely correlates with disease severity (Lefebvre et al., 1995). In addition to SMN2 copy number, other modifiers are known to modulate SMA clinical severity (Crawford et al., 2012 Ydewalle et al., 2017). These studies provide an important proof-of-concept for targeting lncRNAs as a means of inducing SMN expression and modulating disease progression.

| CircRNAs in motor neuron disease
Several studies have investigated circRNAs in the context of human disease . In recent years, transcriptome analyses have revealed that dysregulation of circRNAs is associated with complex neurological diseases such as Alzheimer's disease, PD, multiple system atrophy and ALS (Chen, Mills, et al., 2016;D'Ambra et al., 2019;Errichelli et al., 2017;Hanan et al., 2020;Huang et al., 2018;Wang et al., 2018). As outlined in previous sections, FUS is required for circRNA biogenesis (Errichelli et al., 2017). HiPSCderived MNs harbouring ALS-linked FUS mutations (FUS-P525L) also showed deregulation of specific circRNAs suggesting that these mutations affect circRNA biogenesis. CircRNAs may not only play a role in ALS pathogenesis but could also be exploited as therapeutic agents. A genome wide loss-of-function screen for modifiers of TDP-43 toxicity identified (intronic) lariats as decoys for TDP-43.
Reduction of debranching enzyme 1 activity led to an increase in intronic lariats in this screen. The newly formed lariats acted as decoys and sequestered toxic cytoplasmic TDP-43 aggregates, possibly preventing them from interacting with other RNAs and RBPs (Armakola et al., 2012). This study suggests the possibility of using circular lncRNAs (such as circRNAs) as decoys for toxic proteins in ALS. CircRNAs were also investigated as potential biomarkers in ALS using microarray analysis (Dolinar et al., 2019). Expression profiling of circRNAs in ALS was performed by comparing leukocyte samples from 12 patient and eight age-and sex-matched healthy controls.
Interestingly, several of the dysregulated circRNAs were located in gene loci that play key roles in ALS pathologies, for example, hsa_ circ_0001173 in vesicle-associated membrane protein-associated protein B/C (Nishimura et al., 2004). Three of the circRNA candidates (hsa_circ_0023919, hsa_circ_0063411 and hsa_circ_0088036) showed similar differential expression patterns in a replication experiment as in the original microarray experiment. Where, hsa_circ_0023919 was down-regulated, hsa_circ_0063411 and hsa_circ_0088036 were up-regulated in ALS patients, with good sensitivity and specificity for detection. This suggests that these candidate circRNAs could be used as diagnostic markers for sALS (Dolinar et al., 2019). While microarrays only allow detection of selected RNAs, RNAseq approaches can facilitate a more unbiased detection of circRNAs in the future, including of low abundant and novel transcripts, and in identification of their biomarker potential.
Overall, these studies hint at roles for circRNAs in the pathogenesis of ALS, and as diagnostic and therapeutic targets.   (Hoye et al., 2017;Tung et al., 2019). For example, reduced expression of the miR-17-92 cluster of miRNAs in limb innervating lateral motor column-motor neurons (LMC-MNs) is associated with increased vulnerability of LMC-MNs to degenerate (Kanning et al., 2010;Tung et al., 2019). LncRNAs may subserve similar functions, especially given their ability to influence miRNA function.

| CON CLUS ION
Sex-specific expression of lncRNAs also has been reported and it is interesting from a disease perspective as MNDs such as ALS and X-linked SMA occur more frequently in males as compared to females (Dressman et al., 2007;Wijesekera & Leigh, 2009).
Currently, not much is known about potential sex-specific expression of the lncRNAs but establishing such expression patterns would be the first step to probe sex-specific lncRNA functions in healthy and diseased MNs.
Overall, accumulating experimental evidence hints at important roles for lncRNAs in MN development and disease. Although in many instances further work is needed to dissect how lncRNAs function in developing MNs and why lncRNA dysregulation contributes to MNDs, it is clear that such knowledge could aid the development of diagnostic and therapeutic approaches for treating MN pathologies like ALS and SMA.

ACK N OWLED G EM ENTS
The authors thank Jan Veldink for critically reading the manuscript.

CO N FLI C T O F I NTE R E S T
The authors declare that there is no conflict of interest regarding the publication of this article.