mRNP assembly, axonal transport, and local translation in neurodegenerative diseases

Abstract

The development, maturation, and maintenance of the mammalian nervous system rely on complex spatiotemporal patterns of gene expression. In neurons, this is achieved by the expression of differentially localized isoforms and specific sets of mRNA-binding proteins (mRBPs) that regulate RNA processing, mRNA trafficking, and local protein synthesis at remote sites within dendrites and axons. There is growing evidence that axons contain a specialized transcriptome and are endowed with the machinery that allows them to rapidly alter their local proteome via local translation and protein degradation. This enables axons to quickly respond to changes in their environment during development, and to facilitate axon regeneration and maintenance in adult organisms. Aside from providing autonomy to neuronal processes, local translation allows axons to send retrograde injury signals to the cell soma. In this review we discuss evidence that disturbances in mRNP transport granule assembly, axonal localization, and local translation contribute to pathology in various neurodegenerative diseases, including spinal muscular atrophy (SMA), amyotrophic lateral sclerosis (ALS), frontotemporal dementia (FTD), and Alzheimer's disease (AD).

INTRODUCTION

The biological function of RNA localization to the cell periphery is to generate an asymmetry within the cell, which provides a mechanism for regulating gene expression with precise patterns of spatiotemporal control. This evolutionary conserved mechanism is a driving force behind many biological processes, including mating type switching in yeast, body axis establishment and cell fate determination in Drosophila embryos and Xenopus oocytes, cell motility and migration in fibroblasts, and axonal pathfinding and synaptic plasticity in polarized neurons (Martin and Ephrussi, 2009). While most studies have focused on the localization of a small number of well characterized transcripts, recent studies suggest that mRNA localization may be more the rule, than an exception. Based on RNAseq studies in hippocampal neuropil samples, it has been estimated that ca. 2,550 mRNAs are present in axons and dendrites of hippocampal neurons, suggesting that many of the proteins that form the synapse may originate from local translation that is independent of the cell soma (Cajigas et al., 2012). Recent studies using RNAseq on compartmentalized cultures of sensory neurons and motor neurons in vitro (Briese et al., 2016; Minis et al., 2014), and on retinal ganglion cell axons in vivo (Shigeoka et al., 2016), have identified a highly complex axonal transcriptome that includes transcripts of multiple classes, similar to the somatodendritic compartment. Overcoming limitations of previous studies using micro-arrays, these studies have expanded considerably the known axonal transcriptome (reviewed by (Kar et al., 2017)).

Early studies in the 1960s using metabolic labeling in isolated synaptosomes (Autilio et al., 1968) and axons (Koenig, 1967) led to the hypothesis that mRNA localization and its local translation may provide a mechanism for autonomous temporal and spatial control of the local proteome in nerve terminals (Holt and Schuman, 2013). Due to technical limitations, it has been widely assumed for a long time that transcripts and the machinery necessary for their translation is only present in dendrites, but not in axons. Although studies found polysomes at the base of spines (Steward and Levy, 1982), ribosomes were rarely observed in adult axons, possibly due to their localization in F-actin-rich structures at the periphery of the axoplasm termed periaxoplasmic ribosomal plaques (PARPs) that were described in several related studies (Calliari et al., 2014; Koenig et al., 2000; Sotelo-Silveira et al., 2008; Sotelo-Silveira et al., 2004). Similar axonal clusters of ribosomes are also present in C. elegans neurons, suggesting that clustering of ribosomes may be a common feature throughout evolution (Noma et al., 2017). Furthermore, super-resolution microscopy has revealed eukaryotic ribosomes in mature interneuron axon terminals in mouse hippocampal slices (Younts et al., 2016). Additional studies with tagged ribosomal components will be required to confirm the presence of these structures in other model systems in vivo. The advent of new technologies, including compartmentalized chambers and animal models that allow for cell-type specific tagging and isolation of polysomes, have revealed the presence of an increasingly complex repertoire of several thousand of axonally localized transcripts (reviewed by (Kar et al., 2017)). There is growing evidence that axonal translation is present in developing axons and in axons responding to nerve injury (Briese et al., 2016; Gumy et al., 2011; Saal et al., 2014; Shigeoka et al., 2016; Willis et al., 2007; Zivraj et al., 2010). An approach based on the RiboTag method (Sanz et al., 2009; Shigeoka et al., 2018) has made it possible to compare axonal ribosome-bound mRNAs, or ‘translatomes’, from axons of retinal ganglion cells in both developing and adult mice (Shigeoka et al., 2016). The identification of a complex translatome in adult axons, encoding proteins with a role in axon survival, neurotransmission, and neurodegenerative disease, provides compelling evidence for an important role of axonal translation in synapse function and maintenance in vivo.

Beyond mechanisms of mRNA localization, there are many additional factors and mechanisms to regulate axonal homeostasis. The RNAi pathway (Hengst et al., 2006; Murashov et al., 2007) and dynamic N6-methyladenosine (m6A) modification (Yu et al., 2017) are functional in axons and regulate local translation. Local protein synthesis of Robo3.2 is regulated by axonal nonsense-mediated mRNA decay, influencing axonal pathfinding (Colak et al., 2013). The protein synthesis machinery present in axons includes components of the endoplasmic reticulum (ER) and Golgi apparatus required for the synthesis of most secreted and transmembrane proteins (Gonzalez et al., 2016; Luarte et al., 2017; Merianda and Twiss, 2013; Merianda et al., 2009), as reviewed by (Cornejo et al., 2017). Axons also harbor lysosomes required for autophagosome turnover (Farías et al., 2017). Local protein synthesis is coupled with local protein degradation as a major feature that is needed to maintain growth cone responses (Deglincerti et al., 2015; Verma et al., 2005).

Because of their extraordinary length and energetic demands, highly polarized neurons are particularly vulnerable structures and are at continuous risk of damage. In the case of lower motor neurons in the spinal cord, axons can measure up to 1 m in length in adult humans, and considerably more in larger extant and extinct organisms (Smith, 2009; Wedel, 2012). The axonal volume can exceed that of the cell soma by 1000-fold or more (Goldstein, 2001). Other reasons to explain the exquisite vulnerability of neurons compared to other cell types are that postmitotic and long-lived cells are likely more vulnerable to the accumulation of cell damage, and that their interconnectedness facilitates spread of the pathology along neurites (Ramaswami et al., 2013).

Local translation in mammalian axons serves primarily two functions: (1) it provides cells with the means to respond to environmental cues and express proteins with a role in axon elongation and pathfinding during development, and synaptic function and maintenance in adults (Andreassi et al., 2010; Aschrafi et al., 2017; Cosker et al., 2013; Hengst et al., 2009; Kar et al., 2014; Welshhans and Bassell, 2011; Wu et al., 2005). (2) it enables cells to respond to injury and communicate with the cell soma, by expressing transcription factors and nucleocytoplasmic transport factors that are assembled into retrograde signaling complexes that regulate transcription in the nucleus (Baleriola et al., 2014; Ben-Yaakov et al., 2012; Cox et al., 2008; Ji and Jaffrey, 2014; Kar et al., 2013; Riccio et al., 1997; Yan et al., 2009). Beyond the scope of this review, local translation has been shown to play an important role in axon regeneration, as reviewed by (Twiss et al., 2016). Impaired axonal trafficking of RNA-granules has also been implicated in the pathogenesis of chemotherapy-induced peripheral neuropathy (Pease-Raissi et al., 2017). The profound effect that changes to axonal translation may have on the pathology of various neurodegenerative diseases has only recently been recognized, as reviewed by (Baleriola and Hengst, 2015; Costa and Willis, 2017; Kar et al., 2017; Wang et al., 2016). Potentially disease-relevant examples for (1) are discussed below in the sections on spinal muscular atrophy (SMA) (β-actin, Gap43, Nrn1) (Fig. 1) and amyotrophic lateral sclerosis (ALS) (Nef-L, MAP1B) (Fig. 2), and an example for (2) is discussed in the section on Alzheimer’s disease (AD) (Atf4) (Fig. 3).

Figure 1. mRNP assembly, axonal transport, and local translation defects in SMA.

SMA is caused by deletions or mutations in the SMN1 gene locus, leading to reduced SMN protein levels. SMN deficiency causes reduced assembly of mRNAs (e.g. β-actin, Gap-43, Nrn1) and mRBPs (e.g. HuD and IMP1) into mature mRNA transport granules, and decreased axonal transport of mRNPs. This impairs mRNA localization and translation at axon terminals, causing synaptic defects at NMJs.

Figure 2. mRNP assembly, axonal transport, and local translation defects in ALS/FTD.

An important hallmark of ALS/FTD pathology is the loss of TDP-43 or FUS mRBPs from the nucleus, and their accumulation in cytoplasmic aggregates. This defect is associated with, and perhaps triggered by reduced nuclear protein import through nuclear pore complexes (NPCs), which in turn further contribute to nucleocytoplasmic transport defects. The most common form of fALS is caused by GGGGCC hexanucleotide repeat expansions in the first intron of the C9orf72 gene locus, leading to the accumulation of nuclear RNA foci and DPR aggregates formed by RAN translation. These TDP-43 or FUS aggregates, as well as nuclear RNA foci and DPR aggregates, can trap mRNAs and mRBPs, impairing the assembly and transport of mRNA transport granules in axons. Affected transcripts include those encoding Neurofilament-L and MAP1B. This deficiency causes reduced local translation at axon terminals, leading to instability of microtubules (MTs) and synaptic defects.

Figure 3. Transmission of a neurodegenerative signal from axon to the cell body in AD.

Aggregates of oligomeric Aβ_1-42 trigger, via a currently unknown signaling mechanism, an increased anterograde transport of specific mRNAs from the cell soma into axon terminals. One of these mRNA encodes the transcription factor ATF4 that is locally translated. A locally assembled signaling complex containing ATF4 is retrogradely transported to the cell soma, where it enters the nucleus and induces the expression of the transcription factor CHOP, causing cell death.

1.1. Assembly and transport of neuronal mRNPs

mRNAs do not exist as isolated molecules within eukaryotic cells. From the moment that an mRNA molecule is transcribed from DNA in the nucleus, they are decorated by a set of mRNA-binding proteins (mRBPs) that determine its fate and activity, by regulating pre-mRNA splicing, capping, and poly-adenylation, followed by mRNA nuclear export, packaging, trafficking, quality control, translation, and decay (Singh et al., 2015). These processes are spatially and temporally closely linked and do not occur in a strictly separated sequential manner, e.g. transcription elongation is coupled to pre-mRNA splicing and mRNP assembly and nuclear export (Moore and Proudfoot, 2009). These interconnected steps in the life time of mRNAs explain why mRBPs typically participate in multiple steps of mRNA processing, and are often found to regulate splicing and trafficking, as well as turnover and translation of their target transcripts.

The compartmentalization of mRNA is determined by a complex network of interacting cis- and trans-regulatory elements that modulate mRNA trafficking. Cis-acting elements within the nucleotide sequence, often referred to as “zipcodes” (Kislauskis and Singer, 1992), are most commonly located in the 3' untranslated region (UTR) of the mRNA. This allows for encoding spatial information about the gene product without changing its amino-acid sequence. The variety and specification of zipcode regions are largely products of alternative splicing and polyadenylation. An analysis of local transcriptomes within neural projections and cell soma of primary neurons and neuronal cell lines found that alternative last exons often confer isoform-specific localization (Taliaferro et al., 2016). Interestingly, a shift toward gene-distal last exon isoforms during neuronal differentiation led to a coordinated induction of mRNA isoforms that preferentially localize to neurites. Alternative 3′-UTR isoforms were also found enriched in axonal ribosome-associated transcripts in mouse neurons in vivo (Shigeoka et al., 2016). Axonal transcripts found to be mislocalized in neurodegenerative diseases are listed in Table 1.

Table 1. Mislocalized axonal transcripts in neurodegenerative diseases.

mRNA	Protein function	Disease	Axonal transport/translation defects	Reference
Gap43	Axonal growth	SMA	Reduced in motor neuron axons upon SMN depletion	(Fallini et al., 2016)
β-actin	Cytoskeletal protein	SMA	Reduced in motor and sensory neuron axons upon SMN depletion, impaired local axonal translation upon SMN depletion	(Jablonka et al., 2006; Rathod et al., 2012; Rossoll et al., 2003)
Nrn1	Neuron projection extension, synapse formation and stability	SMA	Reduced in soma and neurites of cortical neurons upon SMN depletion	(Akten et al., 2011)
Anxa2a	Ca2+-binding protein associated with plasma membrane and cytoskeleton	SMA	Reduced in NSC-34 neurites upon SMN depletion	(Rage et al., 2013)
Cox4i2	Cytochrome C oxidase subunit	SMA	Reduced in NSC-34 neurites upon SMN depletion	(Rage et al., 2013)
195 mRNAs	Axon growth and synaptic activity	SMA	Mostly down-regulation in axons of primary motor neurons upon SMN depletion	(Saal et al., 2014)
Nef-L	Neuronal cytoskeletal protein	ALS-TDP	Impaired transport of Nef-L-containing granules in motor neuron axons	(Alami et al., 2014)
futsch/MAP1B	Microtubule assembly	ALS-TDP, ALS-C9orf72	Reduced transport and translation of futsch mRNA in motor neuron axons	(Coyne et al., 2014)
Ddr2	Cell growth and migration	ALS-TDP, ALS-FUS	Mislocalization of Ddr2 mRNA	(Yasuda et al., 2017)
kinesin-1	Cargo transport	ALS-FUS	Sequestered in FUS aggregates	(Yasuda et al., 2017)
565 mRNAs	Diverse axonal functions and metabolic processes	ALS-TDP, ALS-SOD1	Altered axonal levels of mRNAs and miRNAs	(Rotem et al., 2017)
Atf4	Transcription factor	AD	Aβ1-42 exposure increases axonal localization and local translation followed by retrograde transport	(Baleriola et al., 2014)

In addition to these RNA processing events and information encoded in cis-elements, transcript stability and localization is tightly regulated by the association of zipcode regions with trans-acting mRBPs. Classical mRBPs contain RNA-binding domains (RBDs) such as RNA recognition motifs (RRMs), heterogeneous nuclear RNP K-homology domains (KHs), or zinc fingers (Znf). However, an ‘interactome capture’ approach has led to the identification of additional classes of protein that directly bind mRNA but lack these RBDs, expanding the repertoire of known mRBPs (Castello et al., 2012; Castello et al., 2016). The majority of these newly identified RNA interacting proteins show intrinsically disordered regions that lack stable secondary or tertiary structure and are enriched in short amino acid motifs, such as RGG and SR repeats (Hentze et al., 2018). mRBPs recognize secondary structures formed within zipcode regions, and assemble into transport-competent RNPs. A single mRBP can bind several structurally or functionally related mRNAs and coordinate sequential steps in mRNA processing. Once assembled, RNPs interact with molecular chaperones, adaptor proteins, and motor proteins to be actively transported along microtubules into axons and dendrites (Buchan, 2014; Xing and Bassell, 2013). Formation of RNP complexes protects mRNA from premature degradation and represses translation and ensures that transcripts arrive at their appropriate destination. Axonal RNP localization and cue-dependent release of translational repression contribute to the precise spatial and temporal gene regulation that supports the development, regeneration, and plasticity of neuronal circuits (Sasaki et al., 2010; Taylor et al., 2013; Tcherkezian et al., 2010; Welshhans and Bassell, 2011; Wong et al., 2017). mRBPs mislocalized in neurodegenerative diseases are listed in Table 2.

Table 2. Mislocalized mRBPs in neurodegenerative diseases.

mRBP	Disease	Axonal transport/translation defects	Reference
hnRNP R	SMA	Reduced in motor neuron axons upon SMN depletion	(Rossoll et al., 2003)
HuD/ELAVL4	SMA	Reduced in motor neuron axons upon SMN depletion	(Fallini et al., 2011)
IMP1/ZBP1	SMA	Reduced in motor neuron axons upon SMN depletion	(Fallini et al., 2014)
TDP-43	ALS-FUS	Sequestered in FUS aggregates	(Kamelgarn et al., 2016)
	ALS-TDP	Altered TDP-43 granule size, dynamics and mobility	(Gopal et al., 2017; Liu-Yesucevitz et al., 2014)
FUS	ALS-FUS	Altered FUS granule dynamics	(Murakami et al., 2015; Patel et al., 2015)
Staufen1	ALS-TDP, ALS-SOD1	Reduced synaptic levels	(Gershoni-Emek et al., 2016)
SMN	ALS-FUS	Sequestered in FUS aggregates	(Murakami et al., 2015)
	ALS-FUS	Sequestered in FUS aggregates	(Groen et al., 2013; Murakami et al., 2015)
APC	ALS-FUS	Sequestered in FUS aggregates	(Yasuda et al., 2013)
FMRP	ALS-FUS	Sequestered in FUS aggregates	(Yasuda et al., 2013)
	ALS-C9orf72	Sequestered in G4C2-containing granules, increased synaptic levels	(Burguete et al., 2015)
ELAVL2	ALS-TDP, ALS-SOD1	Increased translation	(Rotem et al., 2017)

A growing body of evidence suggests RNP granules form through the process of liquid-liquid phase separation (Brangwynne et al., 2009; Han et al., 2012; Kato et al., 2012; Mitrea and Kriwacki, 2016). As related to intracellular RNP assembly, this phenomenon refers to the ability of specific proteins to spontaneously separate into a demixed liquid phase, forming a membrane-less compartment within the cell. Unlike other membrane-less organelles, RNP granules are highly transient, assembling and dissembling in response to changes in the environment. The dynamic nature of RNP granule formation is largely owed to the presence of intrinsically disordered regions (IDRs) and low-complexity domains (LCDs) within the protein sequence (Molliex et al., 2015). These domains establish weak and multivalent interactions between RBPs, promoting the assembly of oligomeric structures. While this propensity toward fibrillization plays a fundamental role in the assembly of stress granules, persistent cellular stress or fibril-promoting mutations in the LCD can result in excess pathological fibrillization as seen in ALS, FTD, and AD (Lin et al., 2015; Molliex et al., 2015; Patel et al., 2015). While most studies have focused on larger RNP granules, such as cytoplasmic stress granules and P-bodies, neuronal transport granules share some of their mRBP components, and TDP-43-containing axonal mRNA transport granules display liquid-like properties (Gopal et al., 2017). Heat-shock proteins have been implicated in P-body and stress granule assembly, but the only evidence for the involvement of chaperones in the assembly of mRNA transport granules comes from the identification of SMN as a molecular chaperone for the assembly of axonally-transported mRNPs (Donlin-Asp et al., 2017).

1.2. Mechanisms of mRNA trafficking and local translation in axons

Once an RNP granule is formed, it must be translocated to the axon terminal. Axons employ two main classes of axonal transport based on the overall speed of movement, namely fast axonal transport (up to 400 mm/day, or 1 μm/s) and slow axonal transport (<8 mm/day or 0.1 μm/s) (Maday et al., 2014). Both classes of transport utilize the same microtubule-associated molecular motors from the kinesin and dynein protein families (Hirokawa and Takemura, 2005). Most cargos are bound to both plus-end-directed kinesin motors and minus end-directed cytoplasmic dynein motors, and their overall net directionality towards the periphery or cell body is achieved through back-and-forth movements (Rezaul et al., 2016). Unique domains in RBPs determine their affinity for kinesin family members, providing an additional determinant of transcript axonal localization (Chevalier-Larsen and Holzbaur, 2006). This process of cargo-loading is often facilitated by additional molecular chaperones and adaptor proteins. Transport of the motor-cargo complex depends on the hydrolysis of ATP to generate the force necessary for movement.

Localized RNAs are often kept in a translationally-repressed state until they reach their destination, whereupon they are released and may be translated to give a localized pool of protein product. At the axon terminal, post-translational modifications of RBPs are often necessary to dissociate the RNP complex and relieve translational repression of the localized transcript. For example, IMP1/ZBP1 (insulin-like growth factor 2 mRNA binding protein 1 / zipcode-binding protein 1) is a component of RNP transport granules and regulates the axonal localization of β-actin mRNA in neurons (Donnelly et al., 2011; Salerno et al., 2008; Welshhans and Bassell, 2011). The nonreceptor tyrosine kinase Src phosphorylates a key tyrosine residue on IMP1, interfering with its ability to bind RNA and destabilizing the RNP complex (Huttelmaier et al., 2005; Sasaki et al., 2010).

Examples for extrinsic signals that can drive localized protein synthesis in axonal growth cones include netrin-1 and brain-derived neurotrophic factor (BDNF), which regulate IMP1/ZBP1 phosphorylation (Eom et al., 2003; Kalous et al., 2014; Sasaki et al., 2010; Welshhans and Bassell, 2011; Yao et al., 2006). Stimulation of local protein synthesis in neurons via netrin-1 (Piper et al., 2015), NGF (Gracias et al., 2014), BDNF and glutamate (Hsu et al., 2015) is achieved by activating the mammalian target of rapamycin (mTOR) signaling pathway. As discussed below, deficiency in the assembly and axonal delivery of IMP1-containing mRNPs may contribute to pathology in spinal muscular atrophy (SMA), and restoring local translation in axons by stimulating the mTOR pathway may represent a target for future therapy (Kye et al., 2014; Ning et al., 2010).

The recent advent of methodologies with increased sensitivity and precision has allowed researchers to greatly expand our knowledge of mRNA abundance and diversity within axons. Microarray and RNAseq analyses of the axonal transcriptome in multiple model organisms have provided a more elaborate and comprehensive picture of the axonal transcriptome during embryonic development and in response to injury or pathological conditions, as reviewed by (Kar et al., 2017). The cell-type specific isolation of ribosome-bound mRNAs in mouse retinal ganglion cell axons allowed for the identification of over 2,000 mRNAs in developing and mature axons in vivo (Shigeoka et al., 2016). Analysis of axonal “translatome” profiles offers invaluable insight into axonal transport and local translation that occurs in vivo, and provides direct evidence for the occurrence of developmental stage-specific mRNA translation in axons of the developing and mature CNS. As discussed below, disease-associated disruptions to the axonal localization or translation of mRNA may contribute to a wide variety of pathological phenotypes observed in neurodegenerative diseases.

2. Spinal muscular atrophy

Spinal muscular atrophy (SMA) is the second most common fatal autosomal recessive disorder after cystic fibrosis. SMA is characterized by early synaptic defects and dying back axonopathy that is followed by a progressive degeneration and loss of spinal motor neurons, and skeletal muscle atrophy (Kolb and Kissel, 2011). Disease classification is based on age of symptom onset and highest motor milestone achieved (Russman, 2007). SMA is the leading genetic cause of infant mortality, and occurs in approximately 1:10,000 live births (Prior et al., 2010). In the most common and severe form of SMA (Type I SMA/ Werdnig Hoffmann disease), patients are diagnosed in the first 6 months of life and never gain the ability to sit unassisted. Death from respiratory failure typically occurs within the first 2–4 years of life. Irrespective of clinical severity, all SMA patients possess homozygous deletions or mutations in the survival motor neuron (SMN1) gene (Lefebvre et al., 1995). Humans have a nearly identical copy of this gene (SMN2), which carries a single nucleotide substitution in a splice site, leading to preferential alternative splicing and exclusion of exon 7 in the SMN transcript (Lorson et al., 1999; Lorson and Androphy, 2000; Monani et al., 1999). This C-terminally truncated SMNΔ7 protein is highly unstable and rapidly degraded. Since SMN2 produces only 10–20% full-length transcript, the variable number of SMN2 copies modulates the levels of full-length SMN protein, and consequently the severity of the disease. Thus, SMA is directly caused by a pathological reduction of SMN protein levels.

The pathological hallmark of SMA is the loss of alpha-motor neurons in the spinal cord. However, several phenotypic consequences of low SMN occur prior to cell death. The earliest defects in SMA patients and animal models are found at central synapses in the spinal cord and neuromuscular junctions (NMJ). In mice, NMJs begin to form at embryonic day 12 and undergo a critical stabilization and maturation process until E17 (Boido and Vercelli, 2016). Despite establishing NMJs at a normal frequency, SMA patients and animal models display a lack of synapse maturation. Arrested maturation and a gradual retraction of motor neuron axons are early, detrimental events in SMA pathogenesis that strongly suggest that SMA is a synaptopathy that affects primarily motor neurons, but also other parts of the motor circuitry.

Why motor neurons are uniquely vulnerable to SMN deficiency remains unclear. As its best characterized molecular function, SMN facilitates snRNP assembly and the formation of spliceosomal complexes (Liu et al., 1997). SMN-dependent splicing defects have been described in a number of SMA mouse models, suggesting a direct link between SMN deficiency and impairments in spliceosome assembly and function (Baumer et al., 2009; Custer et al., 2016; Doktor et al., 2017; Garcia et al., 2016; Lotti et al., 2012; Praveen et al., 2012; See et al., 2014; Zhang et al., 2008; Zhang et al., 2013). However, since these defects in spliceosome assembly are found ubiquitously throughout various tissues, it is not clear how such deficits lead to motor neuron degeneration and other SMA phenotypes (Doktor et al., 2017; Shababi et al., 2014). It has been suggested that additional roles of SMN in axonal mRNA processing may cause a specific motor neuron susceptibility and contribute to the pathophysiology of SMA, as reviewed by (Briese et al., 2005; Burghes and Beattie, 2009; Donlin-Asp et al., 2016; Fallini et al., 2012b; Jablonka et al., 2004; Rossoll and Bassell, 2009). This hypothesis is based on the finding of mRNA mislocalization in SMN-deficient axons in vitro, as outlined in more detail below. A recent study has found that SMN interacts with the neuron-specific RBP HuD/ELAVL4 in zebrafish motor neurons in vivo, and reduction of both SMN and HuD lead to reduced axonal localization Gap43 mRNA encoding growth-associated protein 43, a protein with a role in neurite outgrowth, regeneration, and plasticity (Frey et al., 2000). Overexpression of HuD rescued Gap43 mRNA levels as well as morphological motor axon and movement defects, suggesting that reduction in axonal RNPs contributes to the SMA phenotype in vivo, at least in the zebrafish model (Hao et al., 2017). The specific vulnerability of motor neurons may come from unique requirements of RNP localization and local translation, perhaps related to their long axons and specialized synapses (NMJs), or from cell-type specific differences in axonally localized transcripts, such as the reported enrichment of Acta and Actg in axons of cultured primary motor neurons (Moradi et al., 2017).

SMN is a ubiquitously expressed 38kDa protein that oligomerizes to form a multimeric protein complex with Gemins 2-8 and Unrip (SMN Complex). SMN and other components of the SMN complex (gemin6, gemin7, gemin2, and gemin3) have been observed in the axons and dendrites of neurons, independent of splicing-related proteins (Sm proteins) (Sharma et al., 2005; Zhang et al., 2006). Multiple studies have demonstrated a role of SMN in the development, function, and maintenance of motor neuron axons and NMJs (Farrar et al., 2017). The extensive axons and specialized nature of the NMJs may render spinal motor neurons particularly susceptible to defects in RNP assembly and trafficking. In addition to SMN-dependent alterations in splicing, inhibition of RNP assembly and axonal localization may contribute to downstream pathogenic consequences. In the following section we will discuss evidence for the hypothesis that in addition to snRNP assembly, SMN also mediates assembly of mRNPs and their targeting for axonal transport.

2.1. SMN acts as a molecular chaperone for the assembly of mRNPs

Accumulating evidence shows that besides facilitating snRNP assembly, SMN plays also a role in the assembly of other RNPs with a role in histone mRNA processing, mRNA decay, and mRNA localization (Li et al., 2014). The first evidence for its involvement in mRNA localization was the observation of reduced β-actin mRNA levels in axons of motor neurons from a severe SMA mouse model (Rossoll et al., 2003). In support of a role in mRNP assembly and trafficking, numerous studies demonstrated that SMN localizes also to axons and growth cones in vitro (Dombert et al., 2014; Hao le et al., 2015; Jablonka et al., 2001; Pagliardini et al., 2000; Rossoll et al., 2003; Sharma et al., 2005; Zhang et al., 2006; Zhang et al., 2003). In addition, SMN was also shown to localize in Zebrafish motor axons during the period of robust axonal development, as well as in NMJs from E18 mouse embryos (Dombert et al., 2014; Hao le et al., 2015). Moreover, in axons SMN is localized to granules that are actively transported along cytoskeletal elements (Fallini et al., 2010; Zhang et al., 2003).

Another line of evidence comes from studies on SMN-interacting mRBPs, and their involvement in SMA-related phenotypes. SMN was found to associate with several mRBPs, including hnRNP-U, hnRNP-R, hnRNP-Q, KSRP/ZBP2/FBP2, IMP1/ZBP1, HuD/ELAVL4, FMRP, TIAR, SBP2 and FUS/TLS (Akten et al., 2011; Fallini et al., 2011; Fallini et al., 2014; Hua and Zhou, 2004; Hubers et al., 2011; Liu and Dreyfuss, 1996; Mourelatos et al., 2001; Piazzon et al., 2008; Rossoll et al., 2002; Tadesse et al., 2008; Wurth et al., 2014; Yamazaki et al., 2012), and SMN depletion results in reduced axonal localization of mRBPs and their associated mRNAs in cultured motor and sensory neurons (Fallini et al., 2016; Jablonka et al., 2006; Rossoll et al., 2003). One of those mRBPs, HuD/ELAVL4, is a neuron-specific member of the Hu protein family of mRBPs that is known to stabilize its mRNA targets upon binding (Bronicki and Jasmin, 2013). SMN associates with HuD in the context of an mRNP complex and is required for its localization into RNP granules and its transport in axons of cultured motor neurons (Fallini et al., 2011; Hubers et al., 2011). Knockout of HuD in Zebrafish results in axonal defects in motor neurons, similar to the effect of SMN knockdown (Hao et al., 2017; McWhorter et al., 2003), whereas HuD expression in motor neurons rescues axonal defects caused by SMN deficiency (Hao et al., 2017). SMN also interacts and co-localizes in actively transported axonal granules with the RBP IMP1/ZBP1 in primary motor neuron cultures (Fallini et al., 2014). HuD and IMP1 interact in an mRNA-dependent manner, and in primary dorsal root ganglion (DRG) neurons both are required for axonal localization of Gap43 mRNA, which encodes a protein with an important role in the regulation of presynaptic terminal function and axonal growth and plasticity (Yoo et al., 2013). In motor neurons from a severe SMA mouse model, Gap43 mRNA is mislocalized from axons and growth cones, and overexpression of either HuD or IMP1 rescues this defect (Fallini et al., 2016).

SMN was also shown to associate with Nrn1 mRNA which is one of the target mRNAs for HuD (Akten et al., 2011). In primary motor neuron cultures Nrn1 co-localizes with SMN in motor neuron axons, and SMN knockdown leads to decrease of Nrn1 in both soma and neurites of cultured cortical neurons (Akten et al., 2011). 3’UTR of Nrn1 is able to provide axonal localization and translation for its transcript in cortical neurons (Akten et al., 2011). Interestingly, expression of Nrn1 in zebrafish could rescue the axonal defects resulted from Smn knockdown (Akten et al., 2011). The Nrn1 gene encodes a GPI-anchored protein that is involved in neuronal and synaptic development, regeneration and survival (Zhou and Zhou, 2014). As one of its functions, NRN1 was shown to promote axonal branching and neuromuscular synaptogenesis in motor neurons in Xenopus (Javaherian and Cline, 2005). Thus, SMN depletion can potentially lead to deficiency in Nrn1 transport and translation in motor axons that would possibly result in defects in axonal growth and maturation of synapses formed by such axons. Intriguingly, Nrn1 mRNA is localized to axons in both hippocampal and DRG neurons (Taylor et al., 2009; Willis et al., 2007), but the mechanisms of its axonal transport differ in these cells since in hippocampal neurons axonal localization of Nrn1 depends on its 3’UTR, while in DRG neurons it is driven by motifs present in its 5’UTR (Merianda et al., 2013). This difference is likely to be caused by competition for binding of limiting amounts of HuD between 3’UTRs of Nrn1 and Gap43 in DRG neurons, which results in Nrn1 displacement from HuD by the more abundant Gap43 transcripts despite the two-fold higher affinity of HuD for Nrn1 (Gomes et al., 2017). This competition does not occur in embryonic hippocampal neurons that contain ~4-fold more HuD protein than adult DRG neurons (Gomes et al., 2017). These data also suggest the existence of additional mechanisms of Nrn1 axonal transport that are HuD-independent (Gomes et al., 2017). Whether Nrn1 axonal transport is impaired in SMA patients, and how SMN contributes to HuD-dependent and HuD-independent Nrn1 transport in motor neurons, are yet to be revealed.

Another SMN-interacting mRBP is hnRNP R that is required for axonal translocation of β-actin mRNA (Glinka et al., 2010; Rossoll et al., 2002). Knockdown of hnRNP R in motor neurons results in axonal defects and defective clustering of voltage-gated Ca²⁺ channels in growth cones, resembling the phenotypes observed in motor neurons from severe SMA mouse models (Glinka et al., 2010; Jablonka et al., 2007; Rossoll et al., 2003). Local axonal translation of β-actin is impaired in motor neurons from severe SMA mouse model (Rathod et al., 2012), while it is known that it is required for axonal development (Leung et al., 2006; Wong et al., 2017; Yao et al., 2006). β-actin mRNA is reduced in motor neuron axons from SMA severe mouse model, suggesting that impairment of β-actin local translation results from mRNA axonal transport defects (Rossoll et al., 2003). Interestingly, SMN knockdown in differentiated NSC-34 cells results in reduction of neuritic localization of Anxa2 mRNA (Rage et al., 2013). Anxa2 encodes protein Ca²⁺-binding protein Annexin A2 that was shown to regulate actin remodeling (Hayes et al., 2006). Thus, it can be anticipated that lack of β-actin and Anxa2 axonal transport can result in perturbation of local actin concentration and dynamics in growth cone.

Although currently little is known about the exact molecular mechanisms leading to mRNA transport defects in SMA pathology, several findings indicate that similar to the known snRNP assembly defects, mRNP assembly is also deficient in SMA. IMP1-containing mRNP granules isolated from cultured fibroblasts from SMA patients are reduced in size comparing to healthy controls, and motor neurons from an SMA mouse model and SMA patient fibroblasts show diminished binding of the mRBP IMP1/ZBP1 to β-actin zipcode region in reporter assays. Moreover, mRNP granules from patient fibroblasts show decreased association with both microtubules and microfilaments (Donlin-Asp et al., 2017). Taken together, these data support a role of SMN as a molecular chaperone for mRNP assembly that is required for the assembly of mRNP transport granules and their association with the cytoskeleton (Donlin-Asp et al., 2017). Metabolic labeling in compartmentalized cultures of SMN-deficient primary neurons shows a reduction of global mRNA translation in axonal growth cones (Fallini et al., 2016). This defect may result from insufficiency in mRNP assembly and axonal transport, although a direct role of SMN in translational regulation has also been postulated (Bernabo et al., 2017; Sanchez et al., 2013). Another mechanism of how SMN-deficiency can affect local translation involves miRNAs and the mTOR pathway. In particular, upregulation of miRNA-183 in response to SMN deficiency caused subsequent downregulation of mTOR via direct binding to its 3'UTR (Kye et al., 2014).

According to current models, mRNA is transported along the cytoskeleton in a form of membrane-less mRNP granules that are driven by associated molecular motors (Martin and Ephrussi, 2009). Emerging evidence suggests that mRNA can also be co-transported in association with membranous vesicles (Jansen et al., 2014). SMN interacts with the α-subunit of vesicle coat protein COPI (α-COP), and co-localizes with this protein in axonal growth cones of murine primary motor neurons (Peter et al., 2011). Moreover, these proteins are co-transported in differentiated PC12 cells (Peter et al., 2011). Expression of α-COP in SMN mutant Zebrafish was able to rescue axonal defects observed in this model (Li et al., 2015). Since COPI is present in axons and was shown to associate with a set of mRNAs, including β-actin mRNA (Bi et al., 2007; Peter et al., 2011; Todd et al., 2013), it is plausible that the interaction between SMN and COPI mediates a vesicle-associated mode of axonal mRNA transport.

Currently, a direct link between observed insufficiency in axonal mRNA transport in SMA models and molecular mechanisms of the disease development has not been established. However, axonal defects that have been observed in multiple SMA animal models (Carrel et al., 2006; McWhorter et al., 2003; Rossoll et al., 2003; Ymlahi-Ouazzani et al., 2010) may result from inefficient axonal transport of SMN target mRNAs. Of note, local axonal translation of both β-actin and Gap43 mRNAs, but not their expression in the cell soma, are required for axonal growth in primary DRG neurons, supporting branching and elongation of the axon, respectively (Donnelly et al., 2013a). Both elongation and branching defects were observed in zebrafish upon SMN depletion (McWhorter et al., 2003), which could be explained by a direct effect of β-actin and Gap43 axonal transport failure. Importantly, transgenic expression of HuD in motoneurons of SMN mutants restored Gap43 mRNA levels and rescued the observed motor axon and locomotor defects (Hao et al., 2017). SMN depletion has been shown to lead to a widespread reduction of mRNA content in axons (Fallini et al., 2011; Saal et al., 2014). While β-actin, Nrn1, and Gap43 mRNAs encode proteins with important functions in regulating axon outgrowth and presynaptic terminal function and maintenance and may play a role in SMA pathophysiology, it is likely that additional mRNAs are mislocalized in response to SMN-deficiency. It will be important to gain a better understanding of the full repertoire of affected transcripts in SMA models in vivo, the effect of these changes on the local proteome, and how these defects are linked to axonal and synaptic pathology in SMA.

3. Amyotrophic lateral sclerosis and frontotemporal dementia

The considerable clinical, pathological, and genetic overlap between amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) has led to their classification as part of the ALS/FTD disease spectrum (Mackenzie and Feldman, 2005; Ng et al., 2015). ALS is a neurodegenerative disease characterized by the dysfunction and loss of both upper and lower motor neurons. Patients mostly suffer from progressive disability, which ultimately leads to complete paralysis and death 2–5 years after onset. Contrary to the conventional view of ALS as a disease affecting only the motor system, cognitive impairment has been reported in 40–50% of ALS cases, with 12–15% of patients also showing signs of dementia (Montuschi et al., 2015; Phukan et al., 2012; Ringholz et al., 2005). FTD is characterized by progressive neurodegeneration in the frontal and temporal lobes, leading to changes in behavior and personality, frontal executive deficits, and language dysfunction. While 90% of ALS cases are sporadic, the remaining 10% are familial (fALS) and caused by mutations in several genes. A stronger genetic contribution is present in FTD, where ca. one third of cases are familial (Ling et al., 2013). The growing number of ALS/FTD disease-related proteins containing RNA-recognition motifs (RRMs) and glycine-rich LCDs, including TDP-43, FUS/TLS, hnRNPA1, hnRNPA2B1, EWSR1, TAF15, and TIA-1, provides a strong link to dysfunction in RNA metabolism and impaired dynamics of membrane-less RNP granules (Kapeli et al., 2017; Mackenzie et al., 2017; Taylor et al., 2016).

Evidence from ALS patients and animal models suggests that similar to SMA, distal axonal degeneration begins very early in this disease, long before the onset of symptoms and the loss of motor neurons (Fischer et al., 2004; Fischer and Glass, 2007). Defective axonal transport is considered one of the earliest most common pathological mechanisms in ALS, as studies on post-mortem subjects have shown an abnormal accumulation of hyperphosphorylated neurofilaments, molecular motor kinesin and other cargoes (mitochondria, vesicles, lysosomes) in the perinuclear region of motor neurons (De Vos and Hafezparast, 2017). Widespread axonal disorganization and disruption of RNA metabolism, nucleocytoplasmic transport and axonal transport in ALS (Taylor et al., 2016) are likely to affect RNA processing in axons. Splicing defects observed in ALS models may also affect the localization of alternative spliced transcripts in axons (Coyne et al., 2017). The relevance of these findings for ALS pathogenesis in human patients is currently not known.

3.1. TDP-43

The TDP-43 protein, encoded by the TAR DNA-binding protein (TARDBP) gene, is a central determinant in ALS pathogenesis. TDP-43 belongs to the family of hnRNPs and is composed of two RRMs, a nuclear localization signal (NLS), a nuclear export signal (NES) and a C-terminal glycine-rich LCD, which regulates TDP-43 interaction with other proteins (including other RBPs) and harbors most of the ALS-linked mutations (Kapeli et al., 2017). Mutations in the TARDBP gene are linked to 5% of fALS (Ingre et al., 2015). Importantly, truncated TDP-43 (TDP-43 C-terminal fragment, TDP-CTF) accumulates in abnormal phosphorylated and ubiquitinated aggregates in the cytoplasm of motor neurons and other neuronal populations in ca. 97% of ALS and 45% of FTD patients (sporadic and familial cases combined) (Kapeli et al., 2017). This most common pathological subtype of FTD is now commonly referred to as frontotemporal lobar degeneration with TDP-43 pathology (FTLD-TDP). Pathological TDP-43 aggregates are considered a hallmark of ALS/FTD, and are also frequently present in other neurodegenerative diseases (i.e. secondary TDP-43 proteinopathies) (Wilson et al., 2011), which include AD, dementia with Lewy bodies (Amador-Ortiz et al., 2007; Arai et al., 2009; Higashi et al., 2007), and Huntington’s disease (Schwab et al., 2008). The cytoplasmic accumulation of TDP-43 is thought to be caused by reduced nuclear import and increased aggregation of TDP-43, and it may induce toxicity through both loss and gain of function mechanisms (Ederle and Dormann, 2017).

TDP-43 regulates different steps of mRNA metabolism from pre-mRNA splicing to mRNA translation and decay, but also regulates non-coding RNAs (Ratti and Buratti, 2016). TDP-43 has been shown to associate with a large number of nuclear and cytoplasmic RBPs with diverse roles in RNA processing in HEK-293T cells (Freibaum et al., 2010). Although its localization is mainly nuclear, TDP-43 also shuttles to the cytoplasm where it regulates stress granule assembly and dynamics, as well as mRNA transport and translation in different neuronal compartments, including dendrites and distant axon terminals. TDP-43-containing RNP granules have been detected along the length of motor and cortical neuron axons (Alami et al., 2014; Fallini et al., 2012a; Ishiguro et al., 2016) and hippocampal neuron dendrites (Wang et al., 2008), but also in NMJs (Alami et al., 2014; Narayanan et al., 2013). This distal localization of TDP-43 can be induced by physiological signals or neuronal injuries. For instance, TDP-43 shuttles to dendrites in rat hippocampal neurons following KCl-induced depolarization (Liu-Yesucevitz et al., 2014) and to axons in mouse motor neurons upon stimulation by BDNF (Fallini et al., 2012a). In Xenopus, BDNF activates local protein translation at the synapse in order to regulate growth cone guidance in isolated retinal growth cones (Campbell and Holt, 2001) and potentiation of neurotransmitter release in nerve-muscle cultures (Zhang and Poo, 2002). Despite the lack of such data in mammalian models, this suggests that TDP-43 may participate in regulating BDNF-induced axonal protein translation. Moreover, TDP-43 binds to mRNAs involved in synaptic activity (Narayanan et al., 2013; Polymenidou et al., 2011; Sephton et al., 2011; Xiao et al., 2011) and its depletion from mouse or fly brains impairs the expression of several synaptic proteins (Polymenidou et al., 2011; Romano et al., 2014).

TDP-43-positive mRNA transport granules in primary cultured neurons display liquid droplet behavior, with distinct biophysical properties and maturational states depending on their localization in the axon (Gopal et al., 2017). While granules in the proximal axon exhibit an irregular shape, and are more stationary with fewer dynamic fission/fusion events, those localized in the mid axon are more circular and dynamic, and can travel long distances. In addition, TDP-43 granules in the mid axon are maintained by weak hydrophobic interactions, which may facilitate mRNA release at the target site, whereas those located in the proximal axon are maintained through more stable interactions and show less internal molecular mobility (Gopal et al., 2017). Granules containing mutant TDP-43 are bigger, more viscous and less mobile than wild-type TDP-43 granules in rat hippocampal and cortical neurons (Gopal et al., 2017; Liu-Yesucevitz et al., 2014). This could be due to the fact that mutant TDP-43 is more abundant in the axons of primary cultured motor neurons than wild-type TDP-43 (Fallini et al., 2012a), hence disrupting the tightly regulated mRNA transport process.

An elegant study showed in mouse cortical neurons and in Drosophila motor neurons that TDP-43 assembles with translationally-repressed mRNAs into cytoplasmic granules which are trafficked bidirectionally along the axons in a microtubule-dependent manner (Alami et al., 2014). On the other hand, A315T and M337V mutations in TDP-43 cause a decrease in the anterograde movement of TDP-43 granules in fly motor neuron axons versus an increase in their retrograde movement, leading to an accumulation of mutant TDP-43 granules in the proximal axon compartment and their depletion from axon terminals (Alami et al., 2014). Among the mRNAs present in these granules is Neurofilament-L (Nef-L) (Alami et al., 2014) which has been shown to bind TDP-43 in vitro (Strong et al., 2007; Volkening et al., 2009) and in mouse brain tissues (Polymenidou et al., 2011). Two distinct fractions of Nef-L-positive mRNP granules were identified: a fraction positive for TDP-43 which moves preferentially in the anterograde direction, and another fraction lacking TDP-43, which travels retrogradely. This indicates that Nef-L mRNA is transported by different types of granules in axons for different purposes, a process that remains to be clarified. As expected, Nef-L granules exhibited a bidirectional transport defect in induced pluripotent stem cell-derived motor neurons carrying different TDP-43 mutations (Alami et al., 2014). It should also be noted that TDP-43-positive Nef-L granules are also positive for Staufen1 (Volkening et al., 2009), an RBP with a well-characterized role in neuronal mRNA transport and localization in axons and dendrites (Heraud-Farlow and Kiebler, 2014).

Another mRNA component of TDP-43 granules is futsch, the Drosophila homologue of MAP1B. TDP-43 associates with MAP1B mRNA in fly and mouse brains (Godena et al., 2011; Sephton et al., 2011). Interestingly, in axon growth cones and dendrites of cultured mouse hippocampal neurons, MAP1B transcripts were detected in RNP granules containing FMRP (fragile X mental retardation protein) (Antar et al., 2005; Antar et al., 2006), an RBP which also colocalizes with TDP-43 granules (Fallini et al., 2012a; Wang et al., 2008). This indicates that both RBPs might be incorporated in the same type of transport granules that regulates the fate of MAP1B mRNA in axons and dendrites. Futsch levels are decreased at the NMJ when Drosophila TDP-43 (TBPH) is depleted, leading to synaptic defects (Godena et al., 2011). In fly motor neurons overexpressing wild-type or mutant TDP-43^G298S, futsch transcript and protein levels were reduced at the NMJ and increased in the cell body. MAP1B also accumulates in the motor neuron cell bodies of ALS subjects with TDP-43 pathology. Furthermore, TDP-43 overexpression blocked futsch protein synthesis within polysomes (Coyne et al., 2014). Altogether, these findings suggest a model where TDP-43 forms a complex with futsch mRNA and regulates its axonal transport and translation in motor neurons, whereas this process is dysregulated in the context of TDP-43 proteinopathy. Since MAP1B regulates the organization and stability of microtubules, which are needed to ensure proper delivery of mRNAs to synapses, disrupting MAP1B expression could ultimately perturb the localization and local translation of other synaptic mRNAs (Roos et al., 2000). Restoring futsch levels in TDP-43-expressing motor neurons is neuroprotective, since it recovers microtubule stability at the NMJ and ameliorates the abnormal phenotype of flies (Coyne et al., 2014).

TDP-43-positive transport granules contain other factors involved in mRNA transport, including molecular motors and RBPs such as FMRP, IMP1, Staufen1, HuD/ELAVL4, but also SMN, which are colocalized with these TDP-43 positive granules (Fallini et al., 2012a; Wang et al., 2008). HuD co-aggregates with TDP-CTF inclusions in primary cultured motor neurons (Fallini et al., 2012a) and TDP-43 has been shown in vitro to form an RNP complex with FMRP and Staufen1 (Yu et al., 2012). Overexpression of mutant TDP-43^A315T caused a decreased synaptic localization of Staufen1 in axons of primary spinal cord neurons in vitro. Reduced Staufen1 levels at the NMJ were also found in the mutant SOD1^G93A mouse model of ALS, although the functional link between SOD1 function and this axonal RNP mislocalization is currently unclear (Gershoni-Emek et al., 2016). A recent study reported that the expression and localization of several axonal-enriched RNAs (both mRNAs and miRNAs) in mouse motor neurons are impaired in mutant TDP-43^A315T and SOD1^G93A models. These transcripts are mainly involved in mitochondrial functions, synapse assembly, and axon extension (Rotem et al., 2017). Axonal and synaptic miRNAs have been shown to regulate local mRNA translation events (Sasaki et al., 2014; Schratt et al., 2006), therefore they can be used as potential drugs or drug targets in pathological conditions such as ALS. Strikingly, one of the altered miRNAs in both ALS models happens to regulate the expression of the mRBP HuB/ELAVL2. HuB binds to mRNAs with specific axon targeting motifs and is particularly enriched at the NMJ. Since HuB protein levels are significantly increased in mouse brains and cultured motor neurons expressing mutant TDP-43^A315T or SOD1^G93A, this suggests that its role in axonal mRNA transport and translation may be disturbed in ALS (Rotem et al., 2017).

Taken together, the findings discussed above indicate that TDP-43 can bind to synaptic mRNAs and deliver them to distal neuronal sites to regulate their local translation. Disrupting this process may eventually lead to neuronal loss of activity and health, and to the development of neuromuscular diseases such as ALS. Overexpression of wild-type or mutant TDP-43 (M337V or Q331V) causes reduced axon outgrowth and neurotoxicity in mouse and chicken embryonic motor neurons (Fallini et al., 2012a; Tripathi et al., 2014). The NMJ organization and activity are also highly affected in flies in which TDP-43 was either downregulated or overexpressed (Godena et al., 2011; Khalil et al., 2017; Romano et al., 2014). These alterations could result from TDP-43 losing its capacity to transport mRNAs or RBPs required for synapse maintenance. Overexpressing the C-terminal but not the N-terminal fragment of TDP-43 causes axonal degeneration in mouse motor neurons, which suggests that perturbing TDP-43 interaction with RBPs mediates its toxic effect on axon health in TDP-43 proteinopathy (Fallini et al., 2012a). Reduced levels of futsch/MAP1B and synaptic defects have been reported in a mutant TDP-43 Drosophila model, and TDP-43 RNA-binding ability is required to prevent futsch downregulation and therefore to restore microtubule organization and NMJ growth (Godena et al., 2011). Additional work is required to confirm if and how such axonal and NMJ defects are specifically due to TDP-43 losing its ability to regulate RNA/RBP transport and translation.

3.2. FUS

Similar to TDP-43, FUS/TLS is a predominantly nuclear hnRNP that also exerts cytoplasmic functions, including the regulation of mRNA transport and local translation. FUS, EWSR1 and TAF15 are members of the FET protein family that is characterized by an N-terminal QGSY-rich LCD that mediates protein-protein interactions, an RRM, a zing-finger domain, multiple arginine-glycine-glycine (RGG) domains and a PY-NLS at the extreme C-terminus (Schwartz et al., 2015). In ALS, a toxic gain of function and a loss of function of FUS have both been proposed as possible pathogenic mechanisms. Numerous mutations in FUS, as well as rare mutations in EWSR1 and TAF15, have been linked to fALS (Kapeli et al., 2017). These FUS mutations disrupt its role in mRNA metabolism and cause it to accumulate in the cytoplasm and form ubiquitinated inclusions (Neumann et al., 2009), suggesting that a defect specifically in this protein can cause ALS. The majority of disease-causing FUS mutations are clustered at the C-terminal NLS and disrupt its function (Dormann et al., 2010). In FTD-FUS, FUS-positive cytoplasmic inclusions contain also the other FET protein family members EWSR1 and TAF15, and no disease-causing FUS mutations have been identified. Taken together, this suggests a more general breakdown of nuclear import pathways for PY-NLS proteins in FTD as compared to the FUS-specific pathology observed in ALS (Ederle and Dormann, 2017). FUS can also co-aggregate with polyQ proteins and associate with neuronal intranuclear inclusions in Huntington’s disease human brain (Doi et al., 2008).

Aside from its role in regulating transcription, pre-mRNA splicing, and mRNA stability and translation, FUS can form RNP granules that facilitate mRNA delivery to dendrites in rat and mouse hippocampal neurons (Belly et al., 2005; Fujii et al., 2005; Fujii and Takumi, 2005; Schoen et al., 2015). Indeed, FUS accumulates actin-related mRNAs at dendritic spines upon mGluR5 activation in order to maintain spine morphology and density (Fujii and Takumi, 2005; Yoshimura et al., 2006). Molecular motors kinesin and myosin are both required for the translocation of FUS granules to dendritic spines in hippocampal neurons (Kanai et al., 2004; Yoshimura et al., 2006). Moreover, several studies demonstrated that this process depends on FUS interaction with other proteins involved in RNA transport, such as TDP-43, SMN (Groen et al., 2013), IMP1 (Kamelgarn et al., 2016), Sam68 (Belly et al., 2005) and adenomatous polyposis coli (APC). APC ensures proper microtubule organization and axonal growth in neurons (Yasuda and Mili, 2016) by regulating the transport and protein synthesis of specific mRNAs such as β2B-tubulin, and anchoring them at the plus ends of detyrosinated microtubules (Mili et al., 2008). APC is thought to promote microtubule polymerization by spatially directing the local translation of its components near the dynamic microtubule plus ends, suggesting a self-organizing principle for these polarized cellular structures (Preitner et al., 2014). FUS associates with APC and promotes the translation, but not the transport, of APC-bound transcripts at cell protrusions in mouse fibroblasts (Yasuda et al., 2013). FUS has also been shown to colocalize with RBPs Staufen and FMRP within RNA granules isolated from mouse brains (Kanai et al., 2004).

FUS granules exhibit liquid droplet-like features similar to TDP-43 (Murakami et al., 2015). ALS mutations modify the biophysical properties of these granules in C. elegans models, causing their transition from soluble dynamic liquid droplets into insoluble stable fibrillary hydrogels (Murakami et al., 2015), whereas in vitro this phenomenon occurs only within “aged” FUS granules (Patel et al., 2015). These abnormal assemblies can sequester RNP granule components, such as Staufen1 and SMN, and cause impairment of new protein synthesis by cytoplasmic RNP granules in axon terminals, motor dysfunction and shortened lifespan in C. elegans. SMN or Staufen1 downregulation can reduce the formation of mutant FUS^P525L aggregates and mitigate neurotoxicity in this model (Murakami et al., 2015). In neuronal cell lines, SMN is sequestered into cytosolic FUS aggregates, where its association with FUS is increased by ALS-linked mutations, and inhibits axonal localization of SMN (Groen et al., 2013). SMN overexpression in mouse cortical neurons expressing mutant FUS^R521C rescued FUS-induced axonal defects, restoring axon length, branching and growth cone size. These findings suggest that FUS mutations my cause axonal defects via aberrant sequestration of SMN, leading to reduced localization in axons, and SMN-dependent defects in mRNP assembly, localization, and local translation (Groen et al., 2013). It is intriguing that FUS toxicity could be rescued by either up- or downregulation of SMN, although both P525L and R521C mutations increase FUS binding to SMN, therefore altering SMN functions (Sun et al., 2015). Despite these findings, reports on defective SMN-mediated RNA transport and translation caused by such alterations are still lacking. FUS aggregates can trap other RBPs such as APC, FMRP (Yasuda et al., 2013) and IMP1 (Kamelgarn et al., 2016), as well as the motor protein kinesin-1 (Yasuda et al., 2017). Both kinesin-1 mRNA and protein are sequestered into mutant FUS inclusions, causing the loss of detyrosinated microtubules and the mislocalization of APC-dependent mRNA within mouse fibroblast protrusions and hippocampal neuron axons (Yasuda et al., 2017). Interestingly, these effects are inclusion-dependent since they were absent when mutant FUS was diffusely distributed, and they were also detected in the presence of mutant TDP-43^A315T aggregates. However, while overexpression of either kinesin-1 or Hsp104 (which dissolves FUS inclusions) rescued loss of detyrosinated glutamate microtubules and RNA misdistribution, this was not observed in fibroblasts containing TDP-43^A315T inclusions, and kinesin-1 was not sequestered into these aggregates (Yasuda et al., 2017). This indicates that both mutant FUS and TDP-43 aggregates alter RNA localization and translation, but through distinct mechanisms.

3.3. C9orf72 repeat expansion

An expansion of a GGGGCC (G₄C₂) hexanucleotide repeat in the first intron of the C9orf72 gene has been identified as the most common genetic cause of familial ALS and FTD (DeJesus-Hernandez et al., 2011; Renton et al., 2011). Three different disease mechanisms have been proposed, 1) a loss of function of C9orf72 caused by the loss of one splicing isoform, 2) toxic gain of function via the formation of nuclear RNA foci, and 3) proteotoxicity via unconventional translation of the repeat RNA (Edbauer and Haass, 2016; Gitler and Tsuiji, 2016; Wen et al., 2017). While the function of the C9orf72 protein is still not well understood, it has been shown to participate in cellular trafficking, autophagy, and actin dynamics (Farg et al., 2014; Sivadasan et al., 2016; Sullivan et al., 2016; Zhang et al., 2012). The G₄C₂ intronic expansion reduces C9orf72 expression mainly in the frontal cortex and cerebellum of patients (Belzil et al., 2013; Donnelly et al., 2013b; Waite et al., 2014), although recent studies have shown that C9orf72 null mice do not exhibit motor neuron pathology but pathology related to the immune system (Atanasio et al., 2016; Burberry et al., 2016; Koppers et al., 2015; O'Rourke et al., 2016). Expanded G₄C₂ transcripts can form distinct G-quadruplex structures and aggregate in intranuclear RNA foci that may trap RBPs, causing defects in RNA processing, nucleolar stress, and disruption of nucleocytoplasmic transport (DeJesus-Hernandez et al., 2011; Donnelly et al., 2013b; Haeusler et al., 2014; Haeusler et al., 2016; Zhang et al., 2015). Based on an earlier discovery in the CAG repeat disease spinocerebellar ataxia 8 (SCA8) (Zu et al., 2011), several groups have found that the G₄C₂ repeat can undergo repeat associated non-AUG (RAN) translation, in which initiation is thought to occur internally in any of the six reading frames within expanded repeats and form aggregates of dipeptide repeat (DPR) proteins (Ash et al., 2013; Mori et al., 2013a; Mori et al., 2013b; Zu et al., 2013). There is accumulating evidence from several labs that at least poly-GA, poly-GR, and poly-PR DPR proteins significantly contribute to the disease progression of C9orf72-mediated ALS/FTD (Freibaum and Taylor, 2017).

The involvement of C9orf72 repeat expansion pathology in RNA axonal transport and translation has not been widely explored. Expanded G₄C₂ transcripts might indirectly interfere with the role of TDP-43 and FUS granules in this process either by trapping them in RNA foci (Ishiguro et al., 2016; Lee et al., 2013; Rossi et al., 2015; Sareen et al., 2013; Xu et al., 2013), or by sequestering factors regulating TDP-43 and FUS import to the nucleus (Zhang et al., 2015), therefore facilitating their cytoplasmic aggregation. Interestingly, expanded G₄C₂ RNAs have been shown to disrupt mRNP transport granule function. The expression of RBPs forming transport granules is highly dysregulated in brains of patients with C9orf72 expansion (Donnelly et al., 2013b). Expanded G₄C₂ transcripts cause significant branching defects in rat spinal cord and fly epidermal sensory neurons, and this effect was not due to RNA foci or RAN translation. Interestingly, G₄C₂ repeat RNAs were found to incorporate into granules that move bidirectionally along the length of neurites in patient-derived iPS neurons and rat spinal cord neurons (Burguete et al., 2015). Furthermore, expanded G₄C₂ RNA foci in rat spinal cord neurons were shown to colocalize with FMRP, and its protein level was increased in iPS neurons. Consistent with these findings, downregulation of FMRP or its target CPEB3 (Cytoplasmic Polyadenylation Element Binding Protein 3) rescued the axonal branching defects in fly neurons expressing expanded G₄C₂, whereas their upregulation significantly worsened this phenotype (Burguete et al., 2015). Expanded G₄C₂ repeats have also been shown to induce significant cell death in Neuro2A cells and eye degeneration in flies, which is reversed by the overexpression of Pur-α (Xu et al., 2013), an RBP that is present in RNA transport granules in dendrites (Kanai et al., 2004). Another connection between C9orf72 pathology and mRNP dynamics comes from an interactome study of DPR proteins, where poly-GR and poly-PR proteins were found to interact with RBPs and proteins with LCDs that often mediate the assembly of membrane-less organelles, including TDP-43 and FUS, and impair RNP assembly, dynamics, and function (Lee et al., 2016). However, in the CNS of C9-ALS cases, a high degree of co-localization of poly-GR with phosphorylated TDP-43 was found in dendrites within the motor cortex, but not in axons (Saberi et al., 2017).

Further studies are required to determine if C9orf72 expansions and other forms of ALS alter mRNA transport/translation and if so, via similar or distinct mechanisms. There is accumulating evidence that expanded G₄C₂ repeats cause nucleocytoplasmic transport defects (Kim and Taylor, 2017), which may also contribute to axonal RNP localization defects. It may affect both the localization of mRNAs to axons, but also retrograde signaling from the axon to the cell body. Components of the nuclear import machinery, including importin β1 (Hanz et al., 2003) and RanBP1 (Yudin et al., 2008), have been found to be locally translated in axons, and associate with dynein to form a complex with a role in long-distance retrograde injury signaling (Rishal and Fainzilber, 2014).

4. Alzheimer’s disease

Alzheimer’s disease (AD) is the most common neurodegenerative disease characterized by the progressive loss of cognitive and functional abilities. The two defining hallmark pathological features of AD are neurofibrillary tangles (NFTs) consisting of abnormally phosphorylated tau, and senile plaques composed of the β-amyloid peptide (Aβ) (Bloom, 2014). According to the ‘amyloid hypothesis’, the accumulation of extracellular Aβ_1-42, a cleavage product of amyloid precursor protein (APP), is considered causative for most neurodegenerative alterations in AD (Hardy and Selkoe, 2002). In support of this hypothesis, most cases of autosomal dominant familial AD can be attributed to mutations in genes encoding APP and presenilins 1 and 2, which affect expression of Aβ_1-42 (Ryan and Rossor, 2010). While the contribution of Aβ to AD pathology remains controversial, it is clearly a key feature of the disease.

Axonal transport plays a key role in the pathophysiology of AD. Similar to motor neuron diseases, axonal defects represent an early event in AD pathogenesis, and axonal swelling precede other known pathological changes (Stokin et al., 2005). Axonal transport impairments are found in transgenic mice overexpressing AD-related APP and apolipoprotein E4 (ApoE4) (Christensen et al., 2014; Tesseur et al., 2000). In addition, AD progression is characterized by the spread of pathology to neurons downstream in the synaptic circuit. In mouse models overexpressing either APP or tau in the entorhinal cortex, a primary anatomical site for the development of AD pathology, trans-synaptic propagation of pathology to neurons downstream in the synaptic circuit has been found (de Calignon et al., 2012; Harris et al., 2010).

Although the connection of AD to RNA processing defects is not as compelling as in the case of SMA and ALS/FTD, U1-70K and other U1 small nuclear ribonucleoprotein (U1 snRNP) spliceosome components were found enriched in insoluble aggregates in AD patient brains, implicating abnormal RNP assembly and RNA splicing in AD pathogenesis (Diner et al., 2014; Hales et al., 2014). Several studies have found that TDP-43 pathology is present in 30%–70% of AD case series, and associated with memory loss and hippocampal atrophy (Josephs et al., 2014; Josephs et al., 2016; Lippa et al., 2009; Ryan and Rossor, 2010; Uryu et al., 2008). It has been reported that TDP-43 pathology may associate with tau expression (Amador-Ortiz et al., 2007) and lower the threshold for AD related dementia (Wilson et al., 2013). Currently, the role of TDP-43 in AD pathogenesis is still unclear. However, the overlapping presence of TDP-43 and tau pathology appears important for our understanding of AD disease development.

4.1 Axonal synthesis of a neurodegenerative signal after Aβ1-42-treatment

As a central hallmark of AD pathology, Aβ-derived oligomers disrupt synaptic plasticity and induce abnormalities in spine composition and morphology (Lacor et al., 2007; Lambert et al., 1998). Application of soluble Aβ impairs BDNF/TrkB-mediated retrograde signaling (Poon et al., 2013). Local application of Aβ_{1 42} to isolated axons in tripartite microfluidic chambers was found to trigger the selective recruitment of a specific cohort of mRNAs into axons, and their local translation (Baleriola et al., 2014). Transcriptomic analysis of Aβ_1-42-treated axons identified mRNAs of many AD-related genes, including transcripts encoding APP, ApoE, Clu, which regulate Aβ_1-42 production and metabolism, and FERMT2, which has been implicated in tau pathology (Shulman et al., 2014). The axonal upregulation of these transcripts after Aβ_1-42 treatment suggests that these proteins might function downstream of amyloid pathology (Baleriola et al., 2014). Among the axonally enriched transcriptome after Aβ_1-42 treatment, the transcript coding for activating transcription factor 4 (ATF4) was also identified. Locally synthesized ATF4 was retrogradely transported to the neuronal soma where it induced the expression of the CCAAT-enhancer-binding protein homologous protein (CHOP), triggering cell death. Furthermore, axonal ATF4 protein and transcripts were also found increased in the subiculum and entorhinal cortex of AD patients. Since these regions of the brain are especially vulnerable in AD (Khan et al., 2014), these findings closely mirror results from hippocampal neurons and the adult mouse brain. Taken together, this study indicates that axonal mRNA transport and translation in response to a neurodegenerative stimulus may play an important role in spreading the pathological changes of AD. This finding provides an example of changes to local translation as a mechanism for transmitting a neurodegenerative signal from the periphery of neurons to the soma, instead of serving as a factor in cellular homeostasis (Baleriola and Hengst, 2015).

5. Concluding remarks

While this review is focused on current evidence from studies on SMA, ALS/FTD, and AD disease models, it is likely that other diseases will show similar axonal defects in RNA processing. Accumulating evidence suggests that defects in the regulation of ribostasis and proteostasis, which is physically and functionally linked to the formation of insoluble inclusions by aggregation-prone RBPs, are key features of neurodegenerative diseases (Ramaswami et al., 2013; Shukla and Parker, 2016). Work on tauopathies and α-synucleinopathies have uncovered surprising connections to mRNA processing and stress granule dynamics (Chung et al., 2017; Vanderweyde et al., 2016). A common disease mechanism observed across a wide spectrum of neurodegenerative diseases is a defect in axonal transport (Brady and Morfini, 2017; Chevalier-Larsen and Holzbaur, 2006). Impaired nucleocytoplasmic trafficking has emerged as another common mechanism contributing to aging, ALS pathology, as well as other related neurodegenerative diseases, such as Huntington's and AD (Kim and Taylor, 2017). It appears likely that all these processes will impact axonal mRNA processing, and therefore synaptic function and maintenance. In all these cases it remains an open question if the identified defects contribute to pathogenesis in human disease. Rescue experiments in animal models that closely mimic the disease process will be needed to establish these axonal defects as valid targets for therapy.

At least for SMA and ALS, it appears that axonal mRNA localization and local translation defects are caused by underlying defects in RNP assembly. Several studies have found connections between disease mechanisms in SMN and ALS/FTD, such as the association of SMN with the ALS disease protein FUS, suggesting that common pathways may be affected across different motor neuron diseases (Yamazaki et al., 2012). However, their observed molecular pathology differs in very fundamental ways. SMA is characterized by a deficiency in formation of different RNP classes, including stress granules, in the absence of pathological RNA-protein aggregations (Donlin-Asp et al., 2016), whereas an exaggerated aggregation and persistence of RNP granules that leads to insoluble aggregates, is characteristic for ALS/FTD and other neurodegenerative disorders (King et al., 2012). Thus one could classify SMA as an RNP hypo-assembly disease, whereas ALS/FTD belongs to hyper-assembly disorders (Donlin-Asp et al., 2016; Shukla and Parker, 2016). This also suggests that a precise balance between association and disassembly of RNPs needs to be maintained for neuronal health.

Despite dramatic progress in our understanding of the mechanisms that regulate axonal mRNA transport and local translation at axon terminals, many open questions about the basic biology of mRNP transport and local translation in axons, as well as its impact on neurodegenerative diseases remain. While more sensitive methods have led to the discovery of diverse and complex axonal transcriptomes, as well as ribosomes and other components of the translation machinery in axons, the significance of local translation for the maintenance and function of mature axons is less well understood. Of note, while a subset of proteins required for the expression of membrane proteins is present, this compartment lacks a fully formed ER and Golgi apparatus (Cornejo et al., 2017). This raises the possibility that locally translated proteins differ from those synthesized in the cell soma, as suggested by the discovery that proteins at the dendritic plasma membrane have a specific pattern of glycosylation (Hanus et al., 2016).

Most studies in the field of axonal RNP transport and local translation are based on candidate approaches and in vitro studies in isolated cell culture models of disease. An important methodological advance for the study of mRNA localization and local translation in vertebrate axons was the development of compartmentalized chambers that enable a physical separation of axons and somatodendritic components. However, while this allows for the investigation of developing neurons and regenerating neurons after injury, it is less clear how these findings relate to mature neurons in vivo. While these methods are indispensable for certain mechanistic studies, they do not provide the context of complex CNS tissues with pre- and postsynaptic components and glia, and cannot reveal the full extent of pathological changes in disease. The relative importance of certain regulatory mechanisms may differ between cell types and organisms. Whereas BDNF and Netrin-1 mediated local translation of β-actin mRNA is required for growth cone guidance in Xenopus neurons in vitro (Leung, 2006; Yao, 2006), trajectories of β-actin depleted RGC axons in the optic tract did not show major guidance defects or abnormal outgrowth rates, indicating that axon pathfinding is not sensitive to the level of β-actin translation under these specific experimental conditions (Wong et al., 2017). However, this study also shows that inhibition of axonal β-actin mRNA translation disrupts axon branching, similar to what has been found in cultured rat neurons in vitro, where depletion of β-actin mRNA from DRGs decreased the number of axon branches, and in chick spinal cord in vivo, where axonally targeted β-actin mRNA increased branching of sensory axons growing (Donnelly et al., 2013a). This demonstrates that in vitro findings need to be compared to in vivo experiments and across different experimental paradigms. The generation of improved animal models and the advent of new genetic tagging technologies makes it now possible to uncover the full spectrum of axonal RNA processing defects via unbiased profiling of axonal transcriptomes, translatomes, and proteomes in vivo (Glock et al., 2017).