Synaptic control of local translation: the plot thickens with new characters

Introduction

Neurons typically have thousands of synapses, whose composition and strength are dynamically controlled by their history of impulse transmission. Synaptic stimulation activates multiple signaling pathways that ultimately lead to post-translational modifications of key proteins and to the regulated production of vastly diverse molecules. All these processes modulate the strength of synaptic transmission and allow for memory formation. Long-lasting memories require durable changes in synaptic strength, either reinforcement or debilitation, which are known as long-term potentiation and long-term depression, respectively. Both forms of synaptic plasticity largely depend on the coordinated translation of messengers localized at the post-synapse that encode multiple receptor subunits, scaffold proteins, adhesion molecules, signaling molecules, and others [1–11]. Previous reports demonstrate through multiple experimental approaches that neuron activity modulates protein synthesis in dendrites and synapses. Balanced translation is key to synaptic homeostasis, and the dysregulation of local protein synthesis may provoke malfunctions or pathogenic conditions [2, 10, 12–21]. Both transcript-specific and global effects were identified through the incorporation of precursor amino-acids or puromycin-derivatives and the use of fluorescent proteins, among other strategies. A clear outcome of the stimulation of brain-derived neurotrophic factor (BDNF) or metabotropic receptors is an increase in overall protein synthesis, which is accompanied by an enhancement in the translation of a number of messengers and the repression of others [13, 22–25]. In contrast, N-methyl-d-aspartate (NMDA) receptor stimulation decreases the local rate of protein synthesis and, at the same time, activates the translation of a number of messengers [2, 6, 10, 15, 26–29]. Response diversity involves multiple molecular mechanisms in which non-coding RNAs and a growing number of RNA binding proteins (RBPs) play a pivotal role [8, 14, 22, 29–34]. An expanding concept is that mRNA repression implies the formation of specific cytoplasmic bodies collectively known as mRNA-silencing foci, which are non-membranous, microscopically visible assemblies of repressed mRNAs and associated proteins [35, 36]. Neuron-specific mRNA-silencing foci related transport as well as the ubiquitous processing bodies (PBs) are present in dendrites and synapses, and their significance for local translation is just beginning to be understood [35, 37–41]. Among other novel regulatory molecules, Pumilio and Smaug reversibly aggregate and form dendritic foci that contribute to the modulation of translation upon synaptic stimulation [14, 42–44].

Translational regulation at the synapse is an important aspect of gene expression control and it is followed by protein modification and degradation, which occur locally and under the control of neuronal activity [9, 11, 45, 46]. Calcium/calmodulin-dependent protein kinase II alpha (CaMKII alpha) is a key protein required for synaptic plasticity under the control of multiple regulatory pathways. Smaug 1 foci, cytoplasmic mRNA polyadenylation and miRNA-mediated silencing as well as post-translational mechanisms locally regulate CaMKII upon synaptic stimulation [14, 47, 48]; reviewed in [49]. The activity-regulated cytoskeleton-associated protein (ARC/Arg3.1) is another example of multi-step regulation. ARC/Arg3.1 transcription is activated by synaptic stimulation. ARC/Arg3.1 mRNA is transported along dendrites in association with granules that contain specific RBPs that trigger ARC/Arg3.1 mRNA decay upon ectopic translation outside stimulated synapses [1, 3, 19, 50, 51]; reviewed in [20].

In this work, we provide an overview of the current knowledge on the mechanisms involved in translational regulation in dendrites and synapses, with a focus on specific RBPs and on the dynamics of relevant mRNA-silencing foci. Most of the studies examined here were performed in mammals; the hippocampus is a frequent experimental model due to its known involvement in memory consolidation. Substantial information has also been gathered from invertebrate organisms and is reported here as well.

The synaptic transcriptome: a nuclear screenplay by local storytellers

The presence of ribosomes at dendrites and post-synaptic surroundings has been known for a long time [52], and our knowledge of the identity of the localized transcripts has been progressively growing. Initially limited to a few messengers serendipitously discovered to have a polarized distribution, the list of localized mRNAs has expanded to thousands using microarray analysis [7]. In a comprehensive deep-sequencing analysis, E. Schuman’s lab identified more than 2,000 transcripts localized in hippocampal dendrites. Notably, 86 % of these transcripts overlap with previously identified localized mRNAs, strongly suggesting that the list is now quite complete [53]. Among other cellular functions, localized mRNAs encode glutamate and gamma-aminobutyric acid (GABA) receptors, voltage-gated ion channels, post-synaptic density (PSD) components and scaffolds; cytoskeleton components and regulators; kinases, phosphatases, and signaling molecules of the calcium and phosphatidylinositol pathways; RBPs that regulate translation, transport, and localization; and molecules involved in protein synthesis, processing, and ubiquitination. Although axons are mostly devoid of mRNAs and polysomes, the translation of specific transcripts is known to occur at specific domains with important consequences for axonal growth, survival, and regeneration; these findings are comprehensively reviewed in [54, 55].

The transcriptome at the synapse comprises nearly 50 % of the mRNA species present in a mature hippocampal neuron, implying that a large number of mRNA molecules are actively transported along relatively long distances—up to hundreds of micrometers—from the cell nucleus [53]. RNA transport has been intensely investigated. Recent articles systematically review how cis-acting elements and specific RBPs dictate the packaging of messengers in large RNA transport granules. These granules move by the action of molecular motors along microtubules and microfilaments [56–61]. Neuron activity influences RNA motility, thus ensuring the efficient delivery of selected transcripts to stimulated synapses and the response at the synapse may involve granule remodeling and mRNA release [37, 38, 52, 56, 61–64]. A number of key RBP components of RNA transport granules are nuclear; the recent discovery of a novel pathway for nuclear mRNA export has renewed the field. Speese and colleagues discovered that several transcripts encoding post-synaptic components of the fly neuromuscular junction leave the nucleus packaged in granules [65]. These granules are relatively large and cannot pass through the nuclear pore; rather, they exit the nucleus through the budding of the nuclear envelope in a mechanism that parallels the export of viral particles (Fig. 1). In a first step, the inner nuclear membrane ensheaths the RNA granule in a vesicle and delivers it to the perinuclear space. A subsequent fusion with the outer nuclear membrane releases the RNA granule into the cytoplasm [65]. The composition of these granules is only partially known. The presence of a C-terminal fragment of the wingless-type MMTV integration site family member 1 (Wnt1) receptor that carries a PDZ-binding motif may help granule targeting to the post-synapse, where PDZ-containing proteins concentrate [65]. An intriguing speculation is that neuron activity enhances the exit of nuclear RNA granules. This might be particularly relevant for ARC/Arg3.1 mRNA, whose transcription increases dramatically upon synaptic stimulation; mRNA floods the cytoplasm and dendrites in relatively short time, potentially involving a massive export pathway such as nuclear envelope budding [3, 20].

Different mRNAs show different localization patterns along proximal and distal dendrite segments, and specific motifs directing their transport have been identified. These RNA signals are preferentially located in the 3′UTR, can be repetitive and redundant, are highly diverse, and include both sequence motifs and structural elements. These properties provide specificity for RBP recognition and diversity of localization patterns [66, 67]; reviewed in [59, 61, 68, 69]. The use of alternative UTRs may affect mRNA subcellular localization; the transcript encoding BDNF, an important neurotrophin linked to several disorders that regulates synapse maturation, is a clear example [70–72]. BDNF is synthesized from several alternatively processed transcripts, and BDNF messengers with long 3′UTRs are transported and translated in the dendrites. The localized production of BDNF is important for dendrite branching and spine growth, and the high levels of neurotrophin synthesized from the BDNF mRNA variants located at the soma cannot meet this requirement [70, 71]. RNA transport is also influenced by RNA editing. The mRNA encoding the α-amino-3-hydroxy-5-methyl-4-isoxazolepropionate receptor (AMPAR) is alternatively spliced and edited, thus generating a number of mRNA molecules that localize differentially; for instance, unedited AMPAR transcripts are more efficiently targeted to dendrites [73]. Furthermore, the splice variants termed FLIP are present in dendrites and dendritic spines, whereas the FLOP AMPAR mRNA is largely restricted to the soma.

Finally, cytoplasmic splicing is an emerging regulatory mechanism, and recent work suggests that it may contribute to shaping the local transcriptome (Fig. 1). Intronic sequences are retained in a number of messengers referred to as cytoplasmic intron sequence-retaining transcripts (CIRTs). The mRNAs encoding CaMKII beta, fragile X mental retardation protein (FMRP) and other relevant synaptic molecules are CIRTs. Remarkably, intronic sequences retained in CIRTs have repetitive elements that direct the mRNA to dendrites [74]. It is speculated that CIRTs are spliced locally, and this seems to be the case for the calcium-activated big potassium channel (BKCa) [46]. Cytoplasmic splicing of BKCa mRNA at dendrites allows for the space-restricted production of BKCa protein, thus modulating channel currents in response to local stimulation [46]. The molecular mechanism for cytoplasmic splicing at dendrites is poorly understood but likely involves spliceosome components and/or IRE1-dependent cleavage [75, 76]. Taken together, these findings show that regulated transcription, transport, and local mRNA processing dynamically affect the transcriptome at dendrites and synapses.

Regulation of translation by synaptic activity: multiple subplots make the story

Cis-acting signals present in localized mRNAs direct their transport and trigger multiple mechanisms for translational control that involve non-coding RNAs and a growing number of RBPs. In addition, the stimulation of neurotransmitter or neurotrophin receptors modulates the translational apparatus associated with the post-synapse through several parallel mechanisms that differentially affect different transcripts; the mammalian/mechanistic target of rapamycin (mTOR) pathway has a pivotal role. In this section, we summarize recent advances in these pathways and discuss how translation specificity is achieved through the coordinated regulation of multiple molecular networks.

RBPs: new plays for old actors

In this section, we summarize the current knowledge on the markedly small number of RBPs with well-known effects on mRNA translation at dendrites and synapses. We briefly discuss the relevance of FMRP and cytoplasmic polyadenylation element (CPE) binding protein (CPEB), two molecules extensively discussed in the literature, and we emphasize novel pathways involving Smaug, Nanos, and Pumilio (Fig. 2).

FMRP and CPEB are among the most studied RBPs, and compelling evidence for their important role in synaptic translation arises from multiple independent studies [5, 21, 34, 48, 64, 97–109]. Fragile X syndrome (FXS) and several forms of autism are associated with mutations in the genes encoding FMRP or FMRP-associated proteins [97, 108]. FMRP is produced at the post-synapse upon group 1 mGluR (mGluR1/5) activation and acts as a translational brake. This balance is extremely important, and a loss of FMRP function provokes serious neurological defects in human diseases and animal models [21, 25, 97, 109, 110]. FMRP mRNA, numerous transcripts encoding synaptic components, general translation factors including mTOR and molecules linked to autism are regulated at the translational level by FMRP or related proteins. This protein family binds secondary structures such as “kissing complexes” and G-quadruplexes. Additional sequence motifs have been identified, and it is expected that additional targets will be discovered [21, 34, 97, 99, 108, 111]. FMRP forms dendritic RNA granules, and a large body of evidence indicates that FMRP associates with polysomes. FMRP affects translation initiation and/or elongation with the help of specific miRNAs in a number of cases [83, 99]. FMRP-mediated mRNA repression is regulated by methylation and phosphorylation and may involve ubiquitination followed by FMRP decay [83, 99]. FMRP is a versatile protein, and in addition to RNA transport, translation, and stability, FMRP modulates the nuclear editing of transcripts encoding important synaptic functions, which may affect their trafficking (Fig. 2) [73]. The dysregulation of RNA editing is associated with neuropsychiatric and neurodevelopmental conditions, further positioning FMRP at the center of neuron health.

Members of the CPEB protein family are expressed in different cell types, where they control mRNA translation by regulating the length of the polyA tail and additional mechanisms that directly affect translation [21, 34, 48, 100, 102, 104, 106, 112]. Like its Xenopus oocyte counterpart, mammalian CPEB forms a bifunctional activator-repressor complex in the dendrites and synapses that contains both the poly(A) polymerase GLD2 and the deadenylase poly(A)-specific ribonuclease (PARN) [98]. Thus, CPEB provides a switch that alternatively activates or inactivates translation (Fig. 2). NMDAR stimulation induces the ubiquitination and phosphorylation of CPEB and provokes the exit of PARN and the activation of GLD2 [48, 98, 101]. More than 100 mRNA transcripts appear to be affected by this dual activity complex, including CaMKII alpha and the NMDAR subunit NR2A, which are also regulated by miRNA and FMRP; the AMPA receptor subunits GluA1 and GluA2; the RBP Hu protein D (HuD); and other signaling molecules that contribute to synaptic plasticity and learning [98, 101]. CPEB forms dendritic granules, and later in this review, we discuss how the regulated oligomerization of CPEB by specific protein domains is emerging as an evolutionarily conserved feature relevant to memory formation and persistence [102–104, 106, 112].

Among other proteins recently linked to translational regulation in dendrites, the RBP eIF4AIII forms dendritic RNA granules and contributes to the spatial restriction of ARC/Arg3.1 translation by a novel mechanism [50]. eIF4AIII is a component of the exon-junction complex (EJC), which is a multiprotein label of exon–exon junctions that is removed during the pioneer translation round. The EJC associated with an exon–exon junction located at the 3′UTR of ARC/Arg3.1 mRNA directs the transcript to the non-sense-mediated decay (NMD) pathway, which is a housekeeping mechanism that dictates the degradation of messengers with premature stop codons located upstream of exon–exon junctions. The NMD pathway requires the interaction of the EJC with translation factors, and it is proposed that ARC/Arg3.1 mRNA is degraded when translated in improper locations, whereas ARC/Arg3.1 mRNA is selectively protected from NMD at activated synapses [20, 50]. NMD affects the fate of additional transcripts encoding several synaptic proteins, and this may involve a trans-acting factor termed Nova. Nova is present in the somatodendritic compartment, where it mediates the transport of specific messengers. It is also present in the nucleus, where it represses splicing modes that ultimately trigger NMD. The Nova-NMD pathway is regulated by neuron activity. Interestingly, a number of synaptic components regulated by Nova at the mRNA level are linked to familial epilepsy and related animal disease models [45]. Another RBP connected to neuronal diseases with an emerging role in dendritic translation is TAR DNA binding protein of 43 kDa (TDP-43). TDP-43 is typically localized to the nucleus in most cell types but forms dendritic RNA granules directly or indirectly linked to translation regulation upon synapse stimulation (reviewed in [113]).

In comparison with the large number of localized transcripts, the battery of synaptic proteins that interact with RNA is only partially known. Remarkably, two independent screens aimed to identify RBPs in non-neuronal cell types yielded 300 novel molecules among a total of 800 RBPs [114, 115]. These studies concluded that a high proportion of these proteins are not predictable as RNA binders by current computational methods. In addition, a significant number of identified proteins were shown to be causative in neurological diseases when mutated. These striking observations suggest that novel RBPs with potentially important roles in mRNA translation and stability are expected to be identified among the numerous but poorly described proteins present at the synapse [116].

A choir of RNA granules sings a varied repertoire: SyAS-foci and self-aggregation motifs

Synaptic activity affects RNA granule motility and integrity, reflecting their role in mRNA transport and repression. The significance of neuronal granules in RNA movement has been reviewed comprehensively [3, 56–61, 68]. In this section, we focus on the relevance of granules in translational regulation and how specific protein domains present on RNA repressors govern their aggregation.

It is an increasingly supported concept that mRNA repression is linked to the formation of mRNA-silencing foci, which harbor the transcripts in association with their repressor RBPs. PBs and stress granules (SGs) were the first mRNA-silencing foci identified. PBs are constitutive, whereas SGs are induced upon acute stress insults [35, 36, 41, 146–148]. SGs contain abortive translation initiation complexes that include a number of translation initiation factors and small ribosomal subunits, and PBs concentrate proteins involved in mRNA decapping and silencing and exclude ribosomes and most translation factors. Polysomes are in close contact with PBs, suggesting that translation occurs in their surroundings [149]. Moreover, PBs dynamically exchange mRNAs with the cytosol and may dissolve, thus releasing the transcripts to allow their translation [35, 36, 41]. Similarly, in vertebrate oocytes, the dissolution of large RNA granules containing dormant mRNAs precedes their translational activation [150]. In the Drosophila oocyte, strongly repressed transcripts that reside in the PB inner core move to the periphery to begin translation [151]. Similarly, when monitoring protein synthesis at the single-molecule level, Carson and coworkers found that mRNA translation occurs in the vicinity of dendritic RNA granules [16].

The relevance of PBs to mRNA repression in dendrites and synapses is emerging. Vertebrate and invertebrate neurons display several types of small dendritic foci. These foci contain subsets of classical PB components, including the decapping protein 1a (DCP1a) and the co-activators Rck/p54, Hedls, and Pat1b (Fig. 3d–f) [22, 35, 37, 39, 47]. Large PBs are also present and are likely formed by the fusion of small foci through homotypic and heterotypic interactions between PB components [22, 35–37, 39, 41, 47, 146, 152–154]. An additional type of foci, which are different from canonical PBs, contain the PB scaffold molecules LSm1 and LSm4, the nuclear cap-binding protein CBP80 and small ribosomal subunits, and exclude 4E and large ribosome subunits. This composition suggests the presence of transcripts stalled at the pioneer translation initiation round and broadens the spectrum of dendritic RNA granules [38].

Independent work from several laboratories has demonstrated that neuron activity regulates the integrity of dendritic PBs. NMDAR activation enhances the exchange of DCP1a between the foci and the cytosol, and prolonged NMDAR stimulation ultimately provokes their dissolution. This activity links NMDAR-dependent translation to DCP1a foci dynamics [37, 39] (Fig. 4). Both insect and vertebrate PBs are connected to miRNA-mediated silencing. Fly dendritic PBs contain Me31B/Rck/p54 and trailer hitch (Tral), another protein linked to the miRNA pathway. Similarly, a number of mammalian miRNAs colocalize with dendritic foci containing Rck/p54, GW182, Ago2, and DCP1a [37, 47, 152]. BDNF stimulation activates the miRNA-mediated silencing of a number of transcripts and enhances the formation of dendritic DCP1a foci, further linking their dynamic assembly to the miRNA pathway [22].

Additional RNA granules containing various mRNA repressors are present in dendrites and synapses. These granules respond to specific stimuli with changes in motility or integrity. Mammalian Smaug1/Samd4A forms specific mRNA-silencing foci associated with synapses, termed S-foci, as they are distinct from PBs and other neuronal RNA granules hitherto described (Fig. 3a–f). As expected if repressed mRNAs are reversibly stored in the S-foci, they disintegrate when mRNAs are captured into polysomes and assemble when the pool of non-translated transcripts increases. Synaptic activity enhances the cycling of mRNAs between the S-foci and polysomes; NMDAR stimulation triggers a rapid and reversible dissolution of the S-foci to release transcripts, thus transiently allowing their translation [14, 63, 140] (Fig. 4). In addition to PBs and S-foci, a few other dendritic RNA granules, including FMRP and Caprin1/RNG105/GPIAP1 granules, also behave as mRNA-silencing foci. These foci respond to synaptic activity, and are collectively termed synaptic-activity regulated silencing foci (SyAS-foci) [5, 34, 63, 107, 155] (Fig. 4).

SyAS-foci formation is mediated by protein self-aggregation, which is a common feature in RBPs linked to translational regulation in dendrites as well as in SG and PB components. A growing concept is that the formation of SGs and abnormal protein aggregates may influence the normal dynamics of RNA granules, thus affecting synaptic plasticity [35, 36, 41, 146, 148]. Multimerization often depends on disordered regions that tend to assemble into ordered aggregates, and a number of self-oligomerization domains display low amino acid diversity, which may include imperfect repeats of 3–4 amino acids, frequently with a high proportion of glutamine (Q) and/or asparagine (N), conferring a prion-like behavior [35, 36, 41, 146, 147, 156, 157] (Table 2). Aplysia CPEB is the first example of the physiological prion-like self-aggregation of an RBP involved in translational regulation. A QN domain specific to the neuronal isoform governs Aplysia CPEB oligomerization upon neuron activation, thereby stimulating the binding to target transcripts [102, 106, 112]. The multimerization of CPEB molecules by QN domains is evolutionarily conserved, and the fly orthologue Oo18 RNA Binding 2 (Orb2) also forms amyloid-like fibers. Orb2 oligomerization temporally correlates with memory formation, and neurons with Orb2 oligomers are thought to be the storage sites of long-lasting memories [103, 104]. Recent work with transgenic flies highlights the complexity of Orb2 regulation by QN-mediated oligomerization. The Orb2 gene generates two isoforms, a less abundant variant termed Orb2A and the more abundant Orb2B, which differentially contribute to memory formation. Persistent learning strictly requires the Orb2A QN domain and the Orb2B RNA binding domain (RBD), whereas the Orb2B QN domain and the Orb2A RBD are dispensable. Genetic and biochemical approaches indicate that the Orb2A QN domain dictates the formation of an oligomer that is highly resistant to physicochemical treatments including boiling, high salt, detergents, and denaturing agents, thus providing a molecular mechanism for enduring effects at the synapse [104]. CPEB oligomerization also occurs in vertebrates. Mammalian CPEB forms dendritic granules through the N-terminal polyQ region and this domain functionally substitutes for the Drosophila Orb2A QN domain in fly-training models [104]. NMDAR stimulation triggers the phosphorylation and monoubiquitination of CPEB3, resulting in the polyadenylation and translation of GluA1 and GluA2 mRNAs, which has important consequences for synaptic plasticity and memory formation [48, 98, 101]. It is an open possibility that the integrity or composition of CPEB granules is affected by these post-translational modifications.

Following a related mechanism, Pumilio molecules aggregate in several species via a conserved QN-rich region (Table 2). Mammalian Pumilio 2 forms dendritic granules and assembles into particles when transfected into non-neuronal cells, and the QN-domain mediates Pumilio 2 recruitment to SGs [42]. Similarly, Drosophila Pumilio forms microscopically visible aggregates when expressed in yeast cells, and the QN domain regulates the synaptic function of fly Pumilio [43]. Whether Pumilio aggregation responds to synaptic activity remains to be investigated. Recent work in vertebrate oocytes demonstrated the relevance of the reversible Pumilio 1 aggregation via the QN domain. Oocyte maturation triggers the translational activation of dormant mRNAs stored in Pumilio 1 granules. This translational activation requires granule disassembly and is impaired when the granules are stabilized by overexpression of the QN domain [150].

FMRP granules reversibly dissolve upon AMPAR or mGluR stimulation [5, 14, 107], and different domains determine FMRP aggregation in vertebrates and invertebrates. Like CPEB and Pumilio, Drosophila FMRP provides an example of a functionally relevant QN domain, which is present at the C-terminus of the most abundant FMRP isoform. This QN domain mediates protein–protein interactions and is required for short-term memory in flies [110]. The mammalian molecule lacks QN regions and oligomerizes through a specific domain termed NDF (N-terminal domain of FMRP) with yet unknown relevance for translational regulation [105] (Table 2). The mRNA repressor Caprin1/RNG105/GPIAP1 forms dendritic RNA granules that respond to BDNF by dissolving and enabling for the interaction of released mRNAs with the translational apparatus. Caprin1/RNG105/GPIAP1 aggregates through specific domains, including an RGG box and a coiled-coil region that are also required for translational repression [155]. TDP-43 is another important RBP forming dendritic granules that remodel upon neuron depolarization. TDP-43 contains a conserved C-terminal region rich in glycine and uncharged polar amino-acids, mostly asparagine, glutamine, and tyrosine, which facilitates protein aggregation in normal and pathological conditions (Table 2) [113, 148]. Fused in sarcoma/translocated in liposarcoma (FUS/TLS) also forms dendritic RNA granules and joins the growing list of RBPs linked to neurodegeneration [62, 158, 159]. In the FUS/TLS N-terminus, there are 27 imperfect repeats of G/S-Y-G/S surrounded by a high percentage of Q residues. This region defines a protein domain that allows oligomerization in vitro and the recruitment of FUS/TLS to SGs in several cell types [148, 156, 159] (Table 2). The dynamics of FUS/TLS granules upon neuron activity has not been investigated.

Heterogeneous nuclear ribonucleoprotein A2 (hnRNPA2) is involved in mRNA transport and translation in dendrites and forms dendritic granules. A conserved short prion-like region rich in GYN mediates hnRNPA2 aggregation in vitro, and point mutations that exacerbate aggregation lead to pathogenic inclusions [148, 160]. In addition, a protein termed tumor overexpressed gene (TOG) contains multiple hnRNPA2 binding sites that facilitate hnRNPA2 granule formation [161]. Finally, the RBDs of a number of neuronal RBPs, including Staufen, HuD, and the Drosophila orthologue Elav, contribute to oligomerization. Speculatively, interaction with the target mRNA may modulate aggregation [162–164] (Table 2). The molecular determinants of S-foci formation are unknown. The SAM domain is not required and it is expected that novel motifs enabling regulatory diversity will be discovered [14]. In an innovative approach in vitro, McKnight and coworkers found that low-complexity regions present in RBPs govern the spontaneous formation of RNA granules. FUS/TLS, FMR1, TDP-43, and several other RBPs reversibly aggregate into fibers with hydrogel properties that are able to capture native RNPs containing localized transcripts [156, 165]. A preference for homotypic protein interactions and selectivity in heterotypic interactions are observed, thus providing a molecular mechanism for granule diversity in neurons.

In addition to the self-aggregation of translational repressors, RNA–RNA interactions and additional factors such as EJC components and the scaffold protein TOG contribute to mRNA packaging and granule assembly [161, 166, 167]. Whether the aggregation of mRNA-silencing foci directly contributes to mRNA repression is debated. Current evidence favors the notion that SG and PB aggregation follows translation inhibition. The assembly of the various types of mRNA silencing foci that populate dendrites and synapses open important questions and a working model is that translation repression triggers mRNP aggregation, which may facilitate a long-lasting response. RNA granules then facilitate the transport from the cell soma and mRNA storage in the synaptic surroundings. When required, regulated granule dissolution precedes translation activation [63, 150]. It is expected that aggregation is regulated by post-translational modifications triggered by synaptic stimulation, and increasing phosphorylation of the FUS/TLS aggregation domain impairs fiber formation in vitro [165]. An emerging concept is that RNA granules behave as liquid droplets, and their dissolution can be conceptualized as a phase transition event [147, 168]. Thus, in addition to eventual post-translational modifications that directly govern protein–protein interactions, physicochemical factors that influence SGs, as for example calcium levels, pH, or molecular crowding [147], may affect SyAS-foci dynamics and integrity with consequences for mRNA translation.

Resetting the stage for the next scene: how to achieve translational specificity

The overall translational capacity at the post-synapse is modulated by multiple networks, which in concert with the control of RNA granule motility and integrity, help non-coding regulatory RNAs and RBPs to fine-tune translation. Relevantly, the dendritic translational apparatus subtly differs from that at the cell body, providing a basis for specific regulation. A number of ribosomal proteins are encoded by multiple genes, and the mRNAs for specific variants concentrate up to tenfold in the dendrites of Aplysia neurons [169]. Ribosome heterogeneity is common in several cell types of vertebrates and invertebrates and is a source of translation specificity [170]. Additional mechanisms for spatial control include the interaction of the translational apparatus with membrane receptors. The cytoplasmic tail of the netrin receptor DCC associates with eIF4E and ribosome subunits, which locally blocks their activity eventually releasing them upon ligand binding. This mechanism limits the activation of the translational apparatus and allows for synapse-specific responses [18]. Furthermore, initiation and elongation factors as well as ribosomal proteins are locally regulated by post-translational modifications and additional mechanisms dictated by multiple signaling pathways. All these processes have different effects on different transcripts, which commonly display different initiation and elongation rates and form polysomes of varied sizes [16].

The mTOR kinase is a major regulator activated upon multiple synaptic stimuli and differentially affects the translation, repression or stability of several transcripts [12, 29, 81, 171–174]; reviewed in [175, 176]. mTOR modulates translation initiation and elongation through various parallel mechanisms that involve the eIF4E binding protein (4EBP) and S6 kinase (S6K) (Fig. 5). mTOR phosphorylates and inactivates 4EBP, thus releasing eIF4E and facilitating the translation of transcripts with 5′ terminal oligopyrimidine tracts (TOP mRNAs) or related motifs [171, 175] (Fig. 5). Several translation factors are encoded by TOP mRNAs, and multiple studies have demonstrated that the translation of eukaryotic elongation factor 1A (eEF1A) mRNA is stimulated in dendrites upon activation of the mTOR pathway by synaptic activity [172, 173]. Simultaneously, mTOR activates S6K, which in turn differentially affects the activity of three important translation factors: eukaryotic initiation factor 4B (eIF4B), eukaryotic elongation factor 2 (eEF2) kinase (eEF2 K) and the ribosomal protein S6 (reviewed in [175]) (Fig. 5). The phosphorylation of eIF4B and eEF2K results in stimulated initiation and elongation with different effects on different transcripts [12, 81, 171] (Fig. 5). Although the effect of S6 phosphorylation is poorly understood, it is clear that BDNF stimulation triggers S6 phosphorylation and increases dendritic protein synthesis [12, 175]. It is also known that the mTOR/S6K pathway preferentially stimulates the translation of newly processed transcripts through the direct recruitment of activated S6K to the EJC by an S6K target protein termed S6K1 Aly/REF-like target (SKAR) (reviewed in [175]). ARC/Arg3.1 mRNA and other transcripts carrying EJCs in the 3′UTR and messengers before the pioneer translation round associated with CBP80 are expected to be translationally activated upon local mTOR activation [38, 50] (Fig. 5). Finally, mTOR positively affects the stability of specific transcripts, namely, CaMKII alpha, Homer, and GAP43 mRNAs, by yet unknown mechanisms [29]. In addition to its significance for mRNA translation and turnover, mTOR is a central regulator of major cellular networks connecting autophagy, lipid biogenesis, cytoskeleton remodeling, energy, and growth balance. All these networks undoubtedly contribute to the serious neurological defects associated with the dysregulation of the mTOR pathway in human diseases and animal models [171, 175].

Synaptic activity governs the translational activation or repression of specific mRNAs through the coordination of multiple pathways that affect translation factors, regulatory molecules and SyAS integrity. Pioneer studies using biochemical manipulations of synaptoneurosomes and more recent analyses by in situ visualization of protein synthesis using the FUNCAT (fluorescent non-canonical amino acid tagging) strategy have demonstrated that NMDAR stimulation rapidly reduces the local production of proteins [13–15, 27, 177]. This massive translational silencing involves the activation of eEF2K upon calcium entry, followed by eEF2 inactivation and impaired elongation [26]. This mechanism allows poorly initiated transcripts to compete for initiation factors more successfully; as a result, the translation of CaMKII alpha and ARC/Arg3.1 mRNAs increases [2, 19, 20, 27]. The translation of CaMKII alpha mRNA upon NMDAR stimulation is additionally facilitated by the dissolution of the S-foci, which is triggered by calcium entry and PI3K and likely involves mTOR signaling [14]. Calcium entry upon NMDAR stimulation also activates calpain, a protease that degrades the PABP inhibitor PABP interacting protein 2A (PAIP2). Together with the elongation of the polyA tail of specific transcripts, PAIP2 degradation enhances PABP recruitment, thereby facilitating translation [178]. NMDA-specific translation is further assisted by the inhibition of the miRNA-mediated repression of specific transcripts. NMDAR stimulation dictates the degradation of MOV10, a protein linked to the miRNA pathway, which facilitates the translation of CaMKII alpha mRNA and other miRNA targets [81]. At the same time, the mRNA encoding the voltage-gated potassium channel Kv1.1 is translationally repressed upon NMDAR stimulation through a novel network that involves miR-129, the RBP HuD and the mTOR pathway. In this mechanism, Kv1.1 mRNA repression by miR-129 is antagonized by HuD, which is titrated by a number of HuD target transcripts upon mTOR activation [17, 22, 29] (Fig. 5).

In contrast to the global inhibitory effect of NMDAR stimulation, the activation of group I mGluR globally induces protein synthesis at the post-synapse. Two important pathways, the mitogen-activated protein kinase (MEK)/extracellular signal-regulated kinase (ERK) and PI3K/Akt/mTOR, converge to stimulate Cap-dependent initiation upon group I mGluR stimulation. In addition, eEF2K is activated via calcium-calmodulin, thereby slowing elongation. This facilitates ARC/Arg3.1 translation and promotes a relative increase in monosome-associated translation [16, 20, 24, 25, 97, 173]. The translational activation by group I mGluRs is limited below deleterious levels by FMRP. This pathway involves FMRP autoregulation and reversible changes in FMRP granule motility and integrity. The regulation of additional networks contributes to DHPG-specific translation [19, 20, 83, 107, 158].

BDNF strongly stimulates local protein synthesis through the activation of the mTOR pathway, and translation specificity upon BDNFR stimulation is assisted by a dual effect on miRNA-mediated silencing [12, 22]. Compelling evidence was recently presented in which BDNF induces the maturation and the decay of distinct pools of miRNAs. MicroRNA regulation by BDNFR stimulation involves the simultaneous activation of DICER and a protein factor termed Lin28a, which mediates pre-miRNA decay. BDNF activates the MAPK/ERK pathway, which conveys the signal to the HIV-1 TAR RBP (TRBP), a subunit of the DICER complex that enhances its activity [22]. As a consequence, pre-miRNA processing is globally enhanced, a large proportion of the 195 miRNAs present in hippocampal dendrites increase their levels and active miRNA repression correlates with PB assembly [22]. Simultaneously, BDNF post-transcriptionally induces the expression of Lin28a, which facilitates the 3′ polyuridylation of the Let-7 family of pre-miRNAs. This modification triggers the rapid turnover of Let-7 pre-miRNAs by the 3′–5′ exoribonuclease DIS3l2 [179]. As a result, BDNF stimulation dramatically changes the miRNA repertoire and a number of localized transcripts show enhanced translation, whereas many others are repressed. Among other targets, the messengers for CaMKII alpha, Homer 2, and GluA1 are repressed by miRNAs of the Let-7 family and translationally stimulated by BDNF in a Lin28a-dependent manner [22]. It is a common theme that the signaling downstream of neuron stimulation affects a combination of translation factors, regulatory ncRNAs and RBPs as well as RNA granule dynamics, all of which contribute to translational specificity at the post-synapse.

Concluding remarks

Regulated translation at the synapse contributes to the vast computational capacity of the neuronal network by locally affecting the strength of synaptic transmission. A comprehensive analysis of the local transcriptome now seems possible by the almost unlimited power of deep-sequencing techniques and is only restricted by the feasibility of sampling the appropriate tissues or subcellular fractions [53]. The study of non-coding regulatory RNAs has the same possibilities and limitations. The analysis of RBPs and RBP binding elements will be facilitated by high-throughput strategies recently developed to investigate the mRNA interactomes of non-neuronal cells, in which hundreds of novel RBPs were found [111, 114, 115]. Once all these components are mapped, mechanistic insights will require substantial effort to be placed on functional studies.

The dysregulation of localized mRNAs is linked to neurodegeneration. Several RBPs and regulatory RNAs, as well as several pathways that affect their expression, are causative of a number of pathologies and disorders [32, 53, 68, 79, 108, 159, 174]. FMRP, Smaug1/Samd4A, Pumilio, CPEB, Staufen, and Caprin1/RNG105 affect spine morphology and excitability in cultured neurons and model organisms and provoke serious defects paralleling those observed in several human conditions. Several of these important RBPs form dendritic mRNA-silencing foci, collectively termed SyAS-foci, and future studies are pending on how synaptic activity controls their aggregation by modifying specific protein domains or the physicochemical milieu. Multiple mechanisms involving mRNA-silencing foci, RBPs, non-coding RNAs, and the modulation of the translational apparatus concertedly regulate the translational activation or repression of thousands of localized messengers with high spatial and temporal precision. Future work will contribute to understanding the relevance of these multiple pathways in neuron activity and homeostasis.

Electronic supplementary material

Below is the link to the electronic supplementary material.