mRNA isoform balance in neuronal development and disease

Balanced mRNA isoform diversity and abundance are spatially and temporally regulated throughout cellular differentiation. The proportion of expressed isoforms contributes to cell type specification and determines key properties of the differentiated cells. Neurons are unique cell types with intricate developmental programs, characteristic cellular morphologies, and electrophysiological potential. Neuron‐specific gene expression programs establish these distinctive cellular characteristics and drive diversity among neuronal subtypes. Genes with neuron‐specific alternative processing are enriched in key neuronal functions, including synaptic proteins, adhesion molecules, and scaffold proteins. Despite the similarity of neuronal gene expression programs, each neuronal subclass can be distinguished by unique alternative mRNA processing events. Alternative processing of developmentally important transcripts alters coding and regulatory information, including interaction domains, transcript stability, subcellular localization, and targeting by RNA binding proteins. Fine‐tuning of mRNA processing is essential for neuronal activity and maintenance. Thus, the focus of neuronal RNA biology research is to dissect the transcriptomic mechanisms that underlie neuronal homeostasis, and consequently, predispose neuronal subtypes to disease.


| Neuronal cell type specification and diversity
Traditionally, neurons were classified by morphology, spatial position, and electrophysiological behavior (Migliore & Shepherd, 2005;Molyneaux et al., 2007;Zeng & Sanes, 2017). Morphological properties include the size of the soma, patterning of dendrites and axons, as well as dendritic spine density. These features, combined with spatial position, were used to classify cortical neurons into two groups: neurons with local connections and neurons that extend axons to distant targets. Electrophysiological characterization utilizing patch clamp recordings could further segregate cortical neurons by their resting potential, firing rate, and spike patterns (Migliore & Shepherd, 2005;Zeng & Sanes, 2017). However, these classification methods do not account for the molecular diversity of mammalian neurons. Recently, over 100 transcriptionally-distinct neuronal subtypes were identified in the brain using advanced single-cell RNA sequencing techniques (Saunders et al., 2018). Molecular neuronal diversity arises from unique expression of transcription factors and alternative processing of mRNA, producing gene expression modules enriched in neuron-specific functions, including axon guidance, synaptic transmission, and transmembrane potential (Furlanis et al., 2019;M. B. Johnson et al., 2009;Song et al., 2017;Tasic et al., 2016;Zeisel et al., 2018). For example, Furlanis and colleagues identified genes with differential splicing regulation across neocortical cell classes were enriched in gene ontology terms in five categories: adhesion complexes, voltage-gated ion channels (i.e., calcium and potassium channels), presynaptic release machinery, postsynaptic neurotransmitter receptor complexes, and scaffolding proteins (Furlanis et al., 2019). Moreover, neuronal genes are disproportionately long and alternative mRNA processing produces an abundant diversity of isoforms (Gabel et al., 2015;King et al., 2013;Lerch et al., 2012;X. Liu et al., 2011;Miura et al., 2013;Schreiner et al., 2014;Takeuchi et al., 2018;Ushkaryov & Südhof, 1993). Thus, in many cases, molecular diversity alone can be used to distinguish neuronal subtypes (Oldham et al., 2008;Saunders et al., 2018;Tasic et al., 2016;Zeisel et al., 2015).
While spontaneous activity prior to synaptogenesis contributes to synapse formation, the expression of neurotransmitters by mature neurons shapes the output of neural circuitry (de Graaf-Peters & Hadders-Algra, 2006;Hua & Smith, 2004). Neurotransmitters are released by the presynaptic neuron into the chemical microenvironment and are recognized by ligand-gated ion channel receptors on the postsynaptic neuron (B. Lu et al., 2009;L. Nguyen et al., 2001). Three main classes of neurotransmitters are found in the human nervous system-excitatory, inhibitory, and modulatory neurotransmitters. Excitatory neurotransmitters, such as glutamate, are released into the synapse to increase neuronal activity while inhibitory neurotransmitters, such as γ-aminobutyric acid (GABA), are used to temper neuronal activity. Modulatory neurotransmitters influence other neurotransmitters, working together to enhance the excitatory or inhibitory function. The generation of neurotransmitter receptors is controlled by distinct activity-regulated gene expression programs including cell type-specific transcription factors and downstream effectors, such as growth factors (Hua & Smith, 2004;B. Lu et al., 2009;L. Nguyen et al., 2001). For example, AMPA-type glutamate receptors (AMPAR) are composed of four subunits, which contain mutually exclusive exons, "flip" and "flop" (Sommer et al., 1990). Alternative splicing of AMPAR subunits in the hippocampus is regulated by neuronal activity and calcium influx and can alter the kinetic properties of the receptors resulting in variable responses to input signals (Lambolez et al., 1996;Penn et al., 2012;Tanaka et al., 2000).

| Neuronal migration
The adult human central nervous system (CNS) contains approximately 86 billion neurons and over 160 trillion synapses (Herculano-Houzel, 2009;Silbereis et al., 2016;Tang et al., 2001). It is estimated that the number of synapses per neuron in the CNS is between 7200 and 80,000 (DeFelipe et al., 2002;Huttenlocher, 1979;Pakkenberg & Gundersen, 1997;Silbereis et al., 2016;Tang et al., 2001). How these neurons are spatially organized in the nervous system throughout development and how synapses are established is an area of intense investigation. Neuroepithelial cells that line the central cavity of the neural tube form the ventricular zone (VZ) and subventricular zone (SVZ) to serve as pools of progenitors for all neurons of the CNS. In the brain, these neuroepithelial cells are called radial glia, which undergoes asymmetric division to maintain the progenitor pool and produce immature migrating neurons. Radial glia extends fibers radially outward from the VZ and SVZ through the intermediate zone (IZ) to form the cortical plate and serve as scaffolds for neuronal migration. Newly born neurons migrate along the radial glia forming the cortical layers in an inside-out order ( Figure 1a and Box 1; de Graaf-Peters & Hadders-Algra, 2006;Silbereis et al., 2016). RNA binding proteins, such as NOVA and RBFOX family proteins, are implicated in regulating the mRNA processing changes that contribute to the establishment of polarity and migration of immature neurons (Conboy, 2016;Leggere et al., 2016;Yano et al., 2010). For example, knockout of Nova1/2 delays the migration of immature neurons in the spinal cord via disruption of Netrin-Dcc signaling (Leggere et al., 2016). Furthermore, migration of Purkinje neurons in the cerebellum of Nova2 knockout mice was inhibited via disruption of Reelin-Dab1 signaling (Yano et al., 2010).

| Synaptogenesis
Neurons are functionally mature following dendritic and axonal outgrowth, and the formation of synapses. Cortical neurons vary by layer in key features, such as dendritic spine density and arborization, and axon length (Silbereis et al., 2016). Spine projections that develop from the cell body form the basal dendrites while apical dendrites are found at the distal end of the axon. Axons arise from extensions of the growth cone, a complex cytoskeletal structure, guided by chemoattractants or chemorepellents in the cerebrospinal fluid (de Graaf-Peters & Hadders-Algra, 2006;Guérout et al., 2014). Synaptogenesis occurs between the apical dendrites of one neuron and the basal dendrites of another ( Figure 1b). The specificity of these synaptic connections is based on cell surface markers (e.g., neurexins, neuroligins, and protocadherins), locally secreted molecules (e.g., semaphorins and WNT glycoproteins), and neuronal activity ( Figure 1c; Heckman & Doe, 2021;Y. Pan & Monje, 2020;Silbereis et al., 2016). Trans-synaptic cell adhesion proteins, such as neurexins/neuroligins and protocadherins, promote the development of functional synapses via compatible cell-cell interactions (Heckman & Doe, 2021; J. L. Lefebvre et al., 2012;B. Lu et al., 2009;Oku et al., 2020;Schreiner & Weiner Joshua, 2010). These synaptic proteins are important for self/nonself discrimination and interact in specific combinations (e.g., heterophilic versus isoform-specific homophilic) to increase the diversity of synaptic contacts while maintaining the specificity of the neuronal circuit. Furthermore, locally secreted molecules, including semaphorins and WNT glycoproteins, act as signals to guide subcellular synapse specificity (Fiore & Püschel, 2003;C.-W. He et al., 2018;Heckman & Doe, 2021;Park & Shen, 2012). Shatz, 1996;B. Lu et al., 2009;Y. Pan & Monje, 2020). The primary function of the nervous system is to convey and process information from the environment while maintaining a homeostatic level of activation (P. R. Lee & Fields, 2020;Thalhammer et al., 2020). Activity-regulated gene expression programs are induced by membrane depolarization and encode downstream effectors and regulators of synaptogenesis. The continuous remodeling of synaptic biochemistry and morphology in response to stimuli is known as synaptic plasticity, the cellular basis for learning and memory (P. R. Lee & Fields, 2020;Richter & Klann, 2009). Rapid and robust changes in gene expression and mRNA translation contribute to the strengthening or weakening of synaptic connections (Donlin-Asp et al., 2021;Kelleher III et al., 2004;Mirisis & Carew, 2019;Rajgor et al., 2021). Donlin-Asp and colleagues identified an increased association of synaptic Camk2a and Psd95 transcripts with dendritic spines during two forms of plasticity-long-term potentiation and long-term depression-in hippocampal neurons, but this localization was uncoupled from local translation. These findings suggest that mRNA dynamics and protein synthesis are unique contributors to synaptic plasticity ( Donlin-Asp et al., 2021). In addition to synaptic plasticity, neural circuits are also maintained via attenuation of apoptosis, which is initiated at neuron birth. Neuron-specific splicing of BAK, a pro-apoptotic BCL-2 family gene, via PTBP1 downregulation leads to retention of a poison microexon resulting in nonproductive translation and impaired apoptotic capacity (Lin et al., 2020;Y.-F. Sun et al., 2001). Changes in gene expression related to aging have also provided insight into changes in memory and cognition (Colantuoni et al., 2011;Loerch et al., 2008;T. Lu et al., 2004). During fetal and postnatal development, genes associated with synapse development (e.g., GABRA1/2, CHRNG, and KCNMA1) and energy metabolism (e.g., GSS, GFPT1, and AACS) rise to meet the demands of the developing brain and processing of novel stimuli, however, these genes reverse their expression during aging (Colantuoni et al., 2011;T. Lu et al., 2004;Somel et al., 2010). While neurons and synapses are not distinctly lost during aging, the genetic programs that control synaptic plasticity, such as synaptic vesicular transport (e.g., Sortilin, DNCH1, and CLTB) and ion channel homeostasis (e.g., CACNB2, GAD1, and KCNJ9), are greatly reduced (Colantuoni et

BOX 1 Key mRNA processing events in neuronal development
Neuronal genes undergo alternative processing to guide specific changes in development, including establishing polarity, migration, and synaptogenesis. The majority of these events arise from alternative splicing regulated by neuronal RBP families, such as NOVA1/2, PTBP1/2, and RBFOX1/2/3 (Su et al., 2018). The following alternative splicing events have been identified as key processing changes related to specific neuronal functions.

| mRNA ISOFORM DIVERSITY WITHIN AND BETWEEN NEURONAL SUBTYPES
Over 75% of human genes are subject to co-or posttranscriptional mRNA processing by AS and/or APA ( . Accurate regulation of gene expression is essential in neurons. As cell types with unique morphological and physiological phenotypes, neurons require a precise balance of mRNA isoforms to maintain cellular identity and electrophysiological function. The neuronal heterogeneity of the central nervous system associated with location, morphology, and function implies multiple transcriptomic regulatory layers (i.e., gene expression, alternative splicing, and alternative polyadenylation) are responsible for highly specific neural patterning and cell type specification throughout development (Furlanis et al., 2019;Guvenek & Tian, 2018;Ha et al., 2021).

| Alternative splicing
Alternative splicing encompasses regulated changes in the coding information of a transcript, including skipped exons, retained introns, alternative 5 0 or 3 0 splice sites, and alternative first or last exons (Figure 2; E. T. Wang et al., 2008). Alternatively, regulated exons are the most abundant type of neuron-specific AS events (Ellis et al., 2012). The inclusion/exclusion of coding sequence can lead to altered protein domain structure, which in turn, can affect proteinprotein interactions (Figure 3; Buljan et al., 2012). During neurodevelopment, highly conserved, switch-like AS at microexons (3-15 nucleotides in length) modulates the interaction domains of proteins involved in neurogenesis, including signaling pathways and cytoskeleton organization (M. Irimia et al., 2014). Inclusion of a six nucleotide microexon in the transcriptional coregulator Apbb1, which adds two charged residues to the PTB1 domain required for its interaction with the chromatin-modifying enzyme Kat5, resulted in enhanced interaction by extending the binding interface. Genes with neuron-specific splicing are often enriched in neuronal functions, such as neurotransmitter receptors, ion channels, adhesion, and scaffold proteins (Ule et al., 2005). Despite sharing similar gene expression programs, neuronal subclasses can be distinguished by AS of synaptic proteins, such as Nrxn3, Nlgn2, Gria1/2, and Cacna1d/g (Furlanis et al., 2019). Cell type-specific splicing programs function to fundamentally shape synapse assembly and intrinsic neuronal properties, which can be organized into five categories: (1) adhesion complexes, (2) voltage-gated ion channels, (3) presynaptic release machinery, (4) postsynaptic neurotransmitter receptor complexes, and (5) , 2015). Furthermore, PCDHs serve as receptors of Reelin, an extracellular matrix protein involved in cell adhesion, and combine via trans homophilic interactions to establish neuronal networks in the developing cortex (D'Arcangelo et al., 1995;Senzaki et al., 1999). Over 50 PCDH genes have been identified and organized into three tandem gene clusters-α, β, and γ (H. Sugino et al., 2000;Wu & Maniatis, 1999;Wu et al., 2001). Within each cluster, PCDH genes contain a large, unique 5 0 exon, encoding the variable extracellular domains, with smaller 3 0 exons, encoding the identical intracellular domain (Wu & Maniatis, 1999), and cell type-specific expression of PCDH isoforms is achieved through alternative transcription initiation and splicing of 5 0 exons (Esumi et al., 2005;Hirano et al., 2012;Kaneko et al., 2006;Tasic et al., 2002;X. Wang et al., 2002). Interestingly, PCDH isoforms are differentially localized within neurons (Junghans et al., 2008;Kallenbach et al., 2003;Phillips et al., 2003) and play distinct roles in dendritic and axonal development and connectivity (W. V. Chen et al., 2017;Katori et al., 2017;Mountoufaris et al., 2017;Suo et al., 2012). For example, Junghans and colleagues demonstrated differential localization of Pcdhb16 and Pcdhb22 isoforms in retinal neurons in which Pcdhb16 was localized to the postsynapse and Pcdhb22 was less synapse-restricted (Junghans et al., 2008). Additionally, Kallenbach and colleagues showed subcellular localization of Pcdh-γ was dynamically regulated during spinal neuron development; early in differentiation, Pcdh-γ is broadly localized to both dendritic and axonal growth cones, but the expression is progressively restricted to the somatodendritic compartment as the neurons mature (Kallenbach et al., 2003). Neurexins (NRXNs) are a family of presynaptic cell-adhesion molecules-NRXN1, 2, and 3-crucial for defining synaptic connectivity through differential interactions with neuroligins at the postsynaptic membrane. NRXN genes undergo alternative transcription initiation to produce three distinct isoform subclasses-α, β, and γ-which are extensively alternatively spliced to generate over 1300 unique isoforms that are neuronal subtype-specific, and neurexin splicing profiles have been shown to correlate with marker genes of neurodevelopmental origin. (Fuccillo et al., 2015;Lukacsovich et al., 2019;Schreiner et al., 2014;Treutlein et al., 2014;Ushkaryov & Südhof, 1993). For example, a major splice isoform switch at alternatively spliced segment 4 in all three NRXN isoform subclasses distinguishes excitatory and inhibitory neuronal populations in the hippocampus (T.-M. Nguyen et al., 2016). Regulation of NRXN splicing is controlled by neuronal RBPs, including the general neuronal splicing regulator PTBP2 (Resnick et al., 2008) and neuronal subtypespecific STAR family proteins (Ehrmann et al., 2016;Iijima et al., 2011Iijima et al., , 2014Traunmüller et al., 2014Traunmüller et al., , 2016.

| Alternative polyadenylation
Alternative polyadenylation refers to regulated changes in the 3 0 UTR of an mRNA molecule, most often observed as tandem 3 0 UTRs, in which multiple polyadenylation signals are available within a single transcript for cleavage and polyadenylation (Figure 2; E. T. Wang et al., 2008). However, some AS events can coincide with APA events, namely, alternative last exons or splicing that retains a functional intronic or coding sequence polyadenylation signal. Furthermore, U1 snRNP, a spliceosome component, has been shown to dose-dependently block polyadenylation at proximal or upstream polyA sites in a process called telescripting, which serves to maintain the transcriptional integrity of differentiated cell types (Berg et al., 2012;Kaida et al., 2010). General lengthening of mRNA 3 0 UTRs is associated with differentiation and development, with neurons expressing the longest 3 0 UTRs (Agarwal et al., 2021; Guvenek & Tian, 2018;Hilgers et al., 2011;Miura et al., 2013). Over 70% of neuron-enriched genes express multiple 3 0 UTR isoforms (Tushev et al., 2018), each containing an array of regulatory information, including miRNA binding sites, sites of RBP recognition, and molecular "barcodes" which determine subcellular localization (Figure 3). In addition to its association with development and cell type specification, APA is also induced by neuronal activity (Berg et al., 2012;Flavell et al., 2008;Fontes et al., 2017;Lau et al., 2010;Tushev et al., 2018). Long-term potentiation of neurons is coupled with 3 0 UTR shortening and activation of intronic polyadenylation sites, contributing to neuronal plasticity (Tushev et al., 2018). Shortening of 3 0 UTRs in response to neuronal activation often results in truncations of the coding sequence, leading to alternative protein function, or removal of length-dependent inhibitory mechanisms, allowing for rapid and robust mRNA translation (Flavell et al., 2008;Lau et al., 2010). 3.2.1 | miRNA-mediated regulation and translation efficiency miRNAs bind to 3 0 UTRs and posttranscriptionally regulate gene expression via transcript degradation or inefficient translation (Filipowicz et al., 2008;Hoffman et al., 2016;Majoros & Ohler, 2007). Two-thirds of genes targeted by miRNAs have alternative 3 0 UTRs and 40% of these binding sites are located in alternative 3 0 UTR segments (Majoros & Ohler, 2007). While miRNA expression is often ubiquitous, the mRNA targets of miRNA-mediated repression are often alternatively 3 0 end processed to modulate the presence of the miRNA binding site within the 3 0 UTR, suggesting regulation of gene expression by miRNAs is cell type-specific Lewis et al., 2005;Lianoglou et al., 2013;Nam et al., 2014;J. Wang et al., 2021). For example, the binding site for miRNA miR-33-3p is present in alternative 3 0 UTRs of Camk2a expressed in mammalian hippocampal neurons of the CA1 region (Tushev et al., 2018). Furthermore, neurons express long 3 0 UTRs, leading to ample opportunity for RBP and miRNA binding to have a strong effect on mRNA translation efficiency (Blair et al., 2017;Miura et al., 2013). For example, MECP2, a global regulator of transcription required for proper brain development, is posttranscriptionally regulated during neurodevelopment (Li, Dong, et al., 2016;Rodrigues et al., 2016). In differentiated cortical neurons, the long 3 0 UTR of MECP2 (MECP2-L) shows enhanced stability due to the absence of pluripotent cell-specific miRNAs (Rodrigues et al., 2016). Furthermore, MECP2-L is translationally silenced in stem cells by the binding of TIA1, and translationally activated in neurons by the binding of ELAVL3/HuC.

| Subcellular localization and isoform-specific function
Targeted mRNA subcellular localization is utilized by neurons as a means for distributing transcripts to specific cellular compartments for rapid local translation (Kleiman et al., 1990;Paradies & Steward, 1997). The molecular "barcodes" that direct a transcript to a specific cellular compartment are typically located within the 3 0 UTR, and therefore, can be affected by alternative 3 0 end processing Berkovits & Mayr, 2015;Ciolli Mattioli et al., 2019;Minis et al., 2014;Taliaferro et al., 2016;Tushev et al., 2018). In neurons, transcripts with long 3 0 UTRs often localize to the neurites, including BDNF and CAMK2A, however, localized transcripts, in general, have longer 3 0 UTRs than nonlocalized transcripts (Taliaferro et al., 2016;Tushev et al., 2018). Furthermore, localized neuronal transcripts show enrichment for temporal or activity-regulated cell type-specific functions, such as synaptic maintenance and cytoskeletal organization (Cajigas et al., 2012;Gumy et al., 2011). For example, brain-derived neurotrophic factor (BDNF) transcripts contain alternative 3 0 UTRs that dictate subcellular localization and isoform-specific function Lau et al., 2010). The short 3 0 UTR of BDNF is localized to the soma and is responsible for maintaining basal protein levels, while the long 3 0 UTR is localized to the neurites for activity-dependent rapid initiation of translation. Subcellular localization is also often associated with location-specific function. In particular, CDC42, important for establishing neuronal polarity via cytoskeleton reorganization, is differentially localized to the soma and neurites to participate in local axonogenesis and dendritic spine development, and this isoform-specific function is mediated by alternative 3 0 UTRs (C. Chen et al., 2012;Ciolli Mattioli et al., 2019;Yap et al., 2016).

| RNA binding proteins
RNA binding proteins are integral to the regulation of alternative mRNA processing. Sequence motifs present within nascent RNA are recognized by various RBP family proteins, such as PTBP, FOX, NOVA, and MNBL, which are crucial for tissue-and temporally regulated alternative mRNA processing, especially within neurons (Weyn-Vanhentenryck et al., 2018). Moreover, RBPs with splicing functions often regulate their own expression, and that of other RBPs, through the usage of poison exons (Boutz et al., 2007;Dredge et al., 2005;Jangi et al., 2014;Spellman et al., 2007;, however, for some splicing modulators, autoregulation is achieved through the generation of a dominant negative isoform (Damianov & Black, 2010). Neuron-specific RBPs operate in a position-dependent manner (Licatalosi et al., 2008;Llorian et al., 2010;Xue et al., 2009) and the targets of these RBPs are often linked to key neuronal functions, including synapse assembly and axon guidance (Flavell et al., 2008;Jelen et al., 2007;Traunmüller et al., 2014;Ule et al., 2005;S. Zheng, 2020).

| NOVA family RNA binding proteins
The NOVA family of RBPs (NOVA1/2) were the first tissue-specific splicing factors found to be restricted to neurons (Buckanovich et al., 1993;Yang et al., 1998). The association of NOVA proteins to RNA is sequence-specific, preferentially binding to YCAY-rich elements near splice sites, and leads to a location-dependent effect on splicing (i.e., binding of NOVA within an alternative exon leads to repression of exon inclusion, whereas binding of NOVA within the intron downstream of the regulated exon results in enhanced inclusion; Dredge et al., 2005;Jensen et al., 2000;Licatalosi et al., 2008;Ule et al., 2006). NOVA proteins have also been implicated in the regulation of APA in the brain, promoting the usage of long 3 0 UTRs for dozens of genes, including Cugbp2 and Slc8a1 (Hwang et al., 2017;Licatalosi et al., 2008). The targets of NOVA-mediated splicing and polyadenylation are enriched in core neuronal functions, including synaptic adhesion, ion channels, and cytoskeletal proteins (Jelen et al., 2007;Ule et al., 2005;S. Zheng, 2020). For example, NOVA2 binding inhibits the inclusion of DAB1 exons 7b/c, associated with dysregulated Reelin signaling, which is important for proper neuronal migration in the mammalian brain (Yano et al., 2010).

| ELAVL/Hu family RNA binding proteins
The neuron-specific members of the ELAVL/Hu (ELAVL2-4/HuB-D) family (hereby referred to as nELAVL) play important roles in neuronal development. Individual nELAVL proteins are expressed at distinct time points during neurogenesis to suppress progenitor cell proliferation and guide neural progenitors to a neuronal cell fate (Akamatsu et al., 1999;Okano & Darnell, 1997). Specifically, Elavl2 expression is high in the early stages of neurodevelopment within the ventricular zone, extending through the intermediate zone, and decreases in neurons of the cortical plate (Okano & Darnell, 1997). In contrast, Elavl3 is absent in immature neurons of the ventricular and intermediate zones but is highly expressed in neurons of the cortical plate. Elavl4 expression is mainly restricted to the immature neurons of the intermediate zone, and decreases as neurons reach the cortical plate. Binding of nELAVL proteins to RNA is sequence-specific, preferentially associating with GU-rich sequences at exon-intron junctions, and the effect on splicing is location-dependent (i.e., binding of nELAVL to the intron downstream of an alternative exon promotes exon inclusion, while binding within the alternative exon represses exon inclusion; Ince-Dunn et al., 2012). Binding of nELAVL proteins to GU-rich sequences within the 3 0 UTR is also involved in the maintenance of mRNA steady-state levels by enhancing mRNA stability (Ince-Dunn et al., 2012;Mukherjee et al., 2011;Pascale et al., 2005). Furthermore, although specific binding of nELAVL proteins to the ubiquitously-expressed ELAVL1/HuR to promote expression of a long 3 0 UTR and a less stable transcript has been demonstrated, recent bioinformatic analyses have suggested that nELAVL proteins may promote global usage of distal polyA sites during neuronal differentiation by binding to U-rich sequences (S. Lee et al., 2021;Mansfield & Keene, 2012). The targets of nELAVL-mediated splicing are enriched in protein complexes and cytoskeletal dynamics at the synapse and axon, while the transcripts with regulated steady-state levels are enriched for amino acid and sugar biosynthetic pathways ( Ogawa et al., 2018). For example, ELAVL3/HuC is required for the splicing of AnkyrinG (ANK3) at exon 34, of which inclusion of this exon is reduced with neuronal maturity and the shorter isoform contributes to cytoskeletal rearrangements required for establishing the axon initial segment during neuronal polarization (Ogawa et al., 2018).
such as TDP-43, EWSR1, and CLP1, have been associated with splicing and 3 0 end processing defects within a subset of genes in AD neurons, and correlate with disease pathology and cognitive impairment, respectively (Barbash et al., 2017;E. C. B. Johnson et al., 2018;Marques-Coelho et al., 2021;Raj et al., 2018).

| Parkinson's disease
Parkinson's disease (PD) is a progressive neurodegenerative disease characterized by asymmetric resting tremor, bradykinesia, and nonmotor features, including anosmia, autonomic dysfunction, and cognitive decline (Simon et al., 2020). The pathological symptoms of PD include loss of dopaminergic neurons in the substantia nigra and cytoplasmic inclusions containing alpha-synuclein aggregates called Lewy bodies. Additionally, neurons in the substantia nigra exhibit impaired calcium homeostasis, lysosomal and mitochondrial dysfunction, and metabolic stress (Duda et al., 2016). The majority of PD cases are sporadic, suggesting lifestyle and environmental factors affect PD risk (Goldman et al., 2019). Genes causative for familial PD, such as LRRK2, have been associated with increased mRNA translation via repression of miRNA activity and positive regulation of ribosomal protein transcripts. Recently, specific mRNA processing changes in PD-associated genes, LRPPRC and PINK1, have linked mRNA processing defects to PD pathology (Gaweda-Walerych et al., 2016;Martin, 2016;Simon et al., 2020). For example, alternative 3 0 end processing of SCNA, encoding the presynaptic neuronal protein α-synuclein, has been associated with increased protein expression and mislocalization, away from synaptic terminals (

| Cancers of the nervous system
Nervous system tumors are a heterogenous group of diseases characterized by abnormal cell growth and metabolic reprogramming affecting regions of the brain and spinal cord (Bielli et al., 2020;Y. Zheng et al., 2021). Aberrant mRNA processing is an established hallmark of many cancers (Frankiw et al., 2019;Singh et al., 2009). Systematic profiling of the glioblastoma multiforme transcriptome identified over 45,000 alternative splicing events in 10,000 genes with the goal of identifying biomarker events to improve prognosis and guide clinical treatment (X. Chen, Zhao, et al., 2019;Zhang et al., 2021). Another study found aberrant splicing of tumor suppressor annexin A7 (ANXA7) in glioblastoma due to loss of miRNA, miR-124, responsible for suppressing the expression of PTBP1 in neural tissues (Ferrarese et al., 2014). In addition to alterations in splicing, misprocessing of mRNA 3 0 ends can also drive pathogenicity or serve as a prognostic biomarker (Masamha et al., 2014;Mayr & Bartel, 2009;Ogorodnikov et al., 2018). In glioblastoma, CFIm25 expression is reduced, leading to global 3 0 UTR lengthening and enhanced cell proliferation (Masamha et al., 2014). Similarly, PCF11 expression in neuroblastoma was shown to correlate with a cancer stem cell phenotype by driving highly proliferative, embryonic programs (Ogorodnikov et al., 2018). Thus, although mRNA misprocessing is typically secondary to pathogenic mutations in nervous system cancers, some changes appear unique to specific cancer subtypes, and can be used as prognostic biomarkers and therapeutic targets.

| Huntington disease
Huntington's disease (HD, OMIM #143100) is an adult-onset neurodegenerative disorder characterized by chorea, dystonia, incoordination, dementia, and behavioral difficulties, which can be preceded by up to a decade of mild psychotic and behavioral symptoms (Table 1). HD is caused by a CAG-repeat expansion in the Huntingtin gene (Maslon et al., 2019)-unaffected individuals typically have approximately 9-35 repeats, whereas affected individuals typically have ≥40 repeats-and the number of repeats correlates with the age of onset of symptoms (Andrew et al., 1993;Aronin et al., 1995;Brinkman et al., 1997;Duyao et al., 1993;Nance & Myers, 2001;Snell et al., 1993;Trottier et al., 1994). HTT mRNA is alternatively processed via alternative inclusion of exons 28 and 29 (Hughes et al., 2014;Mort et al., 2015;Ruzo et al., 2015) and variable 3 0 UTR length (B. Lin et al., 1993;Romo et al., 2017). The functional consequences of HTT alternative splicing are unclear (skipping of exon 29 is in-frame, while skipping of exon 28 is out-of-frame, resulting in a premature termination codon within exon 29), but variation in 3 0 UTR length is associated with cell typespecific expression (Hughes et al., 2014;Lin et al., 1993;Romo et al., 2017), and the longest HTT isoform is most abundant in the brain. Protein fragments produced by proteolysis of mutant HTT localize to pathogenic intranuclear and cytoplasmic inclusions in neurons (DiFiglia et al., 1997;Lunkes et al., 2002). Similar to other repeat expansion disorders, mutant HTT sequesters RBPs, notably the MBNL1 splicing factor, within these nuclear inclusions (Mykowska et al., 2011), leading to a broad pattern of splicing defects (Becanovic et al., 2010;Lin et al., 2016;Mykowska et al., 2011) and changes in 3 0 UTR length (Romo et al., 2017) of the RBP targets.

| Pontocerebellar hypoplasia
Pontocerebellar hypoplasia (PCH) is a heterogeneous group of pediatric neurodegenerative disorders frequently caused by mutations in proteins involved in RNA and protein metabolism (van Dijk et al., 2018). Over 20 subtypes of PCH have been described to date. Each subtype exhibits the classical features of cerebellar and brainstem atrophy, combined with unique phenotypes, such as cortical malformations and dysmorphic facial features (van Dijk et al., 2018). Mutations causing PCH are enriched in genes required for RNA metabolism and processing-two subtypes of PCH with mutations in mRNA processing proteins are described below. PCH subtypes 1B (OMIM # 614678), 1C (OMIM # 616081), 1D (OMIM # 618065), and 1F (OMIM # 619304) are caused by mutations in subunits of the RNA exosome, EXOSC3, EXOSC8, EXOSC9, and EXOSC1, respectively. PCH1 patients exhibit progressive muscle weakening, microcephaly, hearing and vision impairment, impaired myelination, and motor neuron degeneration often leading to respiratory failure (Boczonadi et al., 2014;Burns et al., 2018;Donkervoort et al., 2017;Rudnik-Schöneborn et al., 2013;van Dijk et al., 2018;Wan et al., 2012). The RNA exosome is composed of a nine-subunit core and an exoribonuclease (Q. Liu et al., 2006;Mitchell et al., 1997) and is involved in the turnover of inherently unstable mRNAs. Unstable mRNA molecules contain AU-elements (ARE) within the 3 0 UTR, which are recognized by ARE-binding proteins, such as AUF1, TTP, and KSRP, and trafficked to the exosome via interactions with the DEVH box family of RNA helicases, such as SKI2W or MTR4 (C.-Y. Chen et al., 2001;Kilchert et al., 2016). Mutations in exosome subunits impair complex assembly by altering stoichiometric ratios or inhibiting protein-protein interactions with other subunits (Morton et al., 2018). Furthermore, mutation of exosome complex subunits caused increased levels of p53 transcript and protein stabilization, resulting in G2/M cell cycle arrest and increased cell death (Müller et al., 2020), likely contributing to the severe clinical phenotypes observed in PCH1 patients (Boczonadi et al., 2014;Burns et al., 2018;Donkervoort et al., 2017;Rudnik-Schöneborn et al., 2013;van Dijk et al., 2018;Wan et al., 2012).
PCH subtype 10 (PCH10, OMIM # 615803) is caused by homozygous p.R140H mutation in CLP1 (Karaca et al., 2014;Schaffer et al., 2014;Wafik et al., 2018). PCH10 patients exhibit severe intellectual disability, progressive microcephaly, limb spasticity, refractory seizures, and progressive spinal motor neuron disease. CLP1 is a subunit of the CFII m subcomplex of the CPA complex (de Vries et al., 2000), and promotes the expression of RBPs involved in mRNA 3 0 end formation, specifically PCF11, CFIm68, CPSF100, and CstF64 (LaForce et al., 2022). The p.R140H variant disrupts the CFII m subcomplex and also impairs the association of the CPSF and CstF subcomplexes with RNA polymerase II (LaForce et al., 2022;Schaffer et al., 2014). Recently, CLP1 was implicated in the regulation of mRNA isoform diversity by repressing proximal polyadenylation sites, and this function is augmented by the PCH10-associated p. R140H variant (LaForce et al., 2022;Monaghan et al., 2021). In PCH10 patient-derived motor neurons, mRNA 3 0 UTR lengthening was correlated with long gene overexpression, including genes enriched in ion channel and synaptic functions, which likely promotes motor neuron differentiation and alters neuronal activity, potentially contributing to the neuropathy observed in PCH10 patients (Karaca et al., 2014;LaForce et al., 2022).
Selective neuronal vulnerability attempts to describe the differential vulnerability of neurons or specific neuronal subpopulations to disease (Fu et al., 2018;Morrison et al., 1998;Saxena & Caroni, 2011;Wieloch, 1985). At present, the contributions of gene expression to selective neuronal vulnerability are three-fold: (1) metabolic requirements, (2) activity and communication, and (3) transcriptional processing and protein availability. (1) Neuronal activity is metabolically expensive. The metabolic cost scales linearly with the number of neurons and with body size (Fonseca-Azevedo & Herculano-Houzel, 2012). Many neurons transmit information to and from brain regions or distal organs, and the speed of this transmission is metabolically demanding (Laughlin et al., 1998). Furthermore, neurons rely on ion channels to drive electrical activity and maintenance of firing, while ion gradients govern neuronal biophysics (Hasenstaub et al., 2010). Thus, misprocessing of genes that alter the metabolic load on a neuronal subpopulation likely contribute to the loss of that population. (2) Neurons communicate via neurotransmitters. This feature is largely unique to neurons and thus represents an important axis of regulation. Neurotransmitters, such as the excitatory molecule glutamate and the inhibitory molecule γ-aminobutyric acid (GABA), rely on protein transporters and receptors to transmit signals between cells at the synapse. In many cases, transporters and receptors are ligand-specific and loss-of-function due to aberrant co-transcriptional processing can lead to the collapse of a signaling pathway (Furlanis et al., 2019;C.-L. G. Lin et al., 1998;Schreiner et al., 2014;Ule et al., 2005). For example, abnormal splicing of glutamate transporter EAAT2 has been identified in ALS samples, and translation products are either rapidly degraded or have a dominant negative effect on the wildtype protein, resulting in loss of protein and activity. This loss of EAAT2 function leads to increased extracellular glutamate and excitotoxic neuronal degeneration (Lin et al., 1998). (3) The unique morphological and physiological properties of neurons are achieved and maintained by cell type-specific expression of transcription factors and RBPs. The altered expression of many transcriptional regulators can affect neuronal specification, maturation, and metabolism. To achieve proper splicing of long, neuronal genes, such as genes important for synapse formation, the rate of transcriptional elongation is critical for kinetic coupling with co-transcriptional processes (Maslon et al., 2019). Furthermore, if RBPs required for efficient gene expression are sequestered or unavailable, protein complex stoichiometry can be affected and result in aberrant mRNA processing (Braunschweig et al., 2014;Chou et al., 2018;Jenal et al., 2012;Liu-Yesucevitz et al., 2010). Together, these features of neuronal gene expression emphasize the essential processes that, when disrupted, contribute to selective neuronal vulnerability to disease.
The consequences of altered gene expression in neurons are well-documented. While molecular mechanisms have been dissected for several diseases (e.g., TDP-43-associated ALS, PCH1, and PCH10), many of the molecular mechanisms that underlie disease pathology remain unclear. The contributions of multiple regulatory layers (i.e., gene expression, splicing, and 3 0 end processing) are increasingly important for defining, not only disease mechanisms but also proper neuronal development and maintenance of neuronal plasticity (Ha et al., 2021). Therefore, understanding neuronal gene regulatory networks and how the balance of gene isoforms is maintained in health and altered in disease will be invaluable for expanding our understanding of the human nervous system. AUTHOR CONTRIBUTIONS Geneva R. LaForce: Conceptualization (equal); funding acquisition (supporting); visualization (lead); writingoriginal draft (lead); writingreview and editing (equal). Polyxeni Philippidou: Funding acquisition (lead); visualization (supporting); writingoriginal draft (supporting); writingreview and editing (equal). Ashleigh Schaffer: Conceptualization (lead); funding acquisition (lead); writingoriginal draft (equal); writingreview and editing (equal).

RELATED WIREs ARTICLE
Post-transcriptional regulation in corticogenesis: how RNA-binding proteins help build the brain