Keywords

The requirement of de novo transcription for long-term synaptic plasticity thought to underlie memory consolidation has been known for over 30 years (reviewed in Davis and Squire 1984 and Goelet et al. 1986). Yet, the coordinated role and function of these newly synthesized transcripts still remains largely unclear. With more recent advancements in high throughput methodologies to assess activity-dependent gene expression, such as microarray and next generation sequencing (NGS), the number of distinct RNA transcripts induced by activity is growing, and consists of transcription factors, neurotransmitters, receptors, scaffolding and structural proteins, as well as many others. In addition to the transcripts that are retained and translated in the cell soma, there is a subset of RNAs that are trafficked into dendrites, often hundreds of microns away from the cell body. These dendritic RNAs are thought to play a role in synapse maintenance and modification in response to internal or external stimuli, such as growth factor or neurotransmitter receptor activation. Dendritic RNAs are thought to be in a translationally repressed state, which poises their translation to be locally regulated by compartment-specific cues such as those found at individual synapses. A recent study in rats suggested that there are upwards of 2,500 RNAs present in CA1 hippocampal projections (Cajigas et al. 2012). This number is tenfold greater than the number of dendritic RNAs previously found through microarray studies (Zhong et al. 2006; Poon et al. 2006) and underscores the important role(s) these RNAs might have in terms of spatially and temporally controlling synapse-specific plasticity. However, the question remains as to how a transcript packaged and shipped from the nucleus is able to find the synapse that initially signaled for its transcription.

One attractive model is synaptic tagging and capture (Frey and Morris 1997), where a synapse is “tagged” to indicate that it had recently undergone some form of plasticity in order to be able to “capture” the correct gene products required for long-term stabilization. This model maintains input specificity if new RNA or protein synthesis is required for stabilization. Synaptic tagging has been shown to occur for both long-term potentiation (LTP, Frey and Morris 1997; Barco et al. 2002; Dudek and Fields 2002; Fonseca et al. 2004; Young and Nguyen 2005) and long-term depression (LTD, Kauderer and Kandel 2000; Sajikumar and Frey 2004). However, the search for the identity of such a tag is still an area of heavy investigation, where no tag quite fits all conceptual requirements. Here we discuss whether the signal sent from the synapse to the nucleus plays a critical role in the synaptic tag and capture hypothesis, or whether the nucleus to synapse signal is sufficient to stabilize synapses that have been tagged. We discuss potential types of tags and how local protein synthesis might play a role in distinguishing inactive versus active synapses.

1 Synapse-to-Nucleus Signaling: What Counts to the Nucleus?

We have previously proposed that synaptic activity-induced signaling from the synapse to the nucleus does not happen on a fast enough time scale to account for some of the transcriptional events that happen within minutes, namely immediate early gene (IEG) transcription (Adams and Dudek 2005). For example, the well-characterized IEG Arc, also known as Arg3.1, can be detected in hippocampal pyramidal cells within 2 min after neuronal activity (Guzowski et al. 1999). Instead, we proposed the idea that calcium influx evoked from synaptic Excitatory Postsynaptic Potentials (EPSPs), or more likely EPSPs together with action potentials, could account for the early transcriptional events occurring in the nucleus, as these can induce calcium-dependent signals instantaneously and within close proximity to the nucleus (Saha and Dudek 2008; Dudek and Fields 2001; Zhao et al. 2005). That stated, since the time we first considered these time, distance, and volume constraints (Adams and Dudek 2005), several examples of synapse-to-nucleus signals (SNSs) that display activity-dependent synaptonuclear shuttling have been described, many of which translocate to the nucleus specifically in response to LTP-inducing stimulation. These include TORC1 (CRTC1), CREB2, JACOB, and ERK1/2 (Zhou et al. 2006; Ch’ng et al. 2012; Lai et al. 2008; Behnisch et al. 2011; Karpova et al. 2013; Davis et al. 2000; Patterson et al. 2001). In the case of ERK1/2, Yasuda and colleagues recently reported that stimulating as few as 3–7 spines on the dendrites of neurons in culture with uncaged glutamate is sufficient to lead to nuclear ERK activation as measured by a nuclear fluorescent reporter (Zhai et al. 2013). Nuclear ERK activation was inhibited by the NMDAR antagonist APV but not by blockade of Voltage Gated Calcium Channels with CdCl2 (and experiments were performed in tetrodotoxin to block action potentials). Therefore the authors reasoned that uncaging-induced nuclear ERK activation was not caused by direct membrane depolarization in close proximity to the nucleus. In the same study, the experiments above were performed within 200 μm from the cell soma and when synapses activated at greater distances, the authors noted that nuclear ERK activation took ~40 min to reach the same level of activation, for example (Zhai et al. 2013). Thus, even in this study, evidence that SNS nuclear import occurs on a sufficiently short time scale required for many IEG transcription (<2 min) is lacking. Furthermore, evidence that comes from studies investigating the mechanism of ERK dendritic trafficking supports the idea that rather than being transported actively (such as with a molecular motor), ERK1/2 is propagated by passive diffusion and is imported to the nucleus via facilitated diffusion, which would make it an unsuitable synapse-to-nucleus signal unless activated in very close proximity to the cell soma (Wiegert et al. 2007).

How would genes be transcribed rapidly (with 2 min) in response to neuronal activity? We found that promoter regions of the fastest of the IEGs such as Arc (rapid IEGs) come “pre-charged” with RNA polymerase II (Pol II), in effect poising the genes for a rapid response (Saha et al. 2011). In these cases the Pol II is proposed to have already initiated transcription, but is paused, apparently awaiting a signal, a process mediated by the Negative Elongation Factor (NELF) complex (Adelman and Lis 2012). There is little question that signaling from the synapse to the nucleus is likely to have profound effects on the later transcriptional output of the nucleus, such as for the slower “delayed IEGs” (Saha et al. 2011), or the so-called second wave of transcription. Also possible is that some of the activity-dependent synaptonuclear proteins are acting as sites of integration for the nucleus, as has been proposed for αCaMKII, whose active form has been shown in vitro to increase as a function of number of inputs and frequency, whereas activity of calcineurin increases with number of inputs (Fujii et al. 2013). Despite the compelling evidence for activity-dependent SNSs and the information they may carry on the type and amount of stimulus, it still begs the question as to how the nucleus integrates incoming signals to produce a coordinated change in gene expression that can selectively modify tagged synapses to impact synaptic function.

We propose, that as long as the synaptic tag is created, the signal that makes it to the nucleus need not contain locale-specific information. In our model, the nucleus acts as a calculator of incoming signals from activated synapses, either in the form of an electrical signal, through calcium, or as part of a transported signal. To our knowledge, evidence in support of the idea that incoming signals can specify information regarding the location of the tag, or a “follow the trail of breadcrumbs back to the synapse” model has yet to be reported. Our proposed model is not only independent of what synaptic signals come in but it also allows for the nucleus to integrate multiple electrical signals imposed upon the neuron from the network. For instance, the nucleus can integrate signals from excitatory and inhibitory neurons, on both the soma and distal dendrites, and if the threshold for repeatedly firing action potentials is reached, the nucleus can respond accordingly (Saha and Dudek 2013). Who gets the product? That is up to the tag!

2 Activity-Dependent Transport of mRNAs to the Synapse: What’s in a Tag?

2.1 Dendritic mRNA Transport and Its Role in Tagging

Despite considerable investigation, surprisingly very little is known about how dendritically targeted mRNAs get docked at synaptic sites. Previous studies have shown that dendritic mRNAs display bidirectional transport in dendrites with rates consistent with microtubule-based transport (Köhrmann et al. 1999; Dynes and Steward 2007; Dictenberg et al. 2008; Tübing et al. 2010). Bidirectional dendritic transport suggests that mRNAs, present as ribonucleoprotein particles (RNPs) consisting of one or a few mRNA transcripts and their RNA binding proteins, might shuttle from synapse to synapse or within dendritic compartments. Another critical finding was that many of the RNA binding proteins associated with dendritically targeted RNPs are translational repressors, suggesting that mRNAs are translationally silent as they are being transported (Krichevsky and Kosik 2001; Napoli et al. 2008; Fritzsche et al. 2013). This led to the appealing notion that synaptic activity could locally remodel dendritic RNPs to allow for translation to occur in a synapse-specific manner. In the example of Arc mRNA, there is evidence for Arc RNPs being translationally repressed when associated with the RNA-binding protein fragile X mental retardation protein (FMRP) and cytoplasmic FMRP-interacting protein 1 (CYFIP1). However after brain-derived neurotrophic factor (BDNF) treatment, Arc RNPs are remodeled by Rac1-dependent phosphorylation of CYFIP1 that recruits proteins critical for cytoskeletal arrangement, such as the WASp-family verprolin homologous protein (WAVE) regulatory complex (De Rubeis et al. 2013). These data suggest that dendritic RNPs are dynamic and highly regulated both spatially and temporally depending on the extracellular cues.

Because it is likely that the synaptic tag(s) differs depending on the cell type and form of plasticity, we reason that seeding the materialization of variable tags is a common underlying process. For example, local actin remodeling and/or local protein synthesis are both processes that have been shown to be required for many forms of synaptic plasticity and might be a mechanism by which tags can be built upon within a common framework depending on the microenvironment it encounters (Martin and Kosik 2002). Similarly, the sushi belt model (Doyle and Kiebler 2011) proposes that RNPs do not statically anchor to synaptic sites but rather patrol a dendritic compartment until recruited into a synapse that recently underwent activity-dependent tagging. This model complements ours in that how the signal reaches the nucleus is relevant for neither tag generation nor the synapse’s ability to recruit plasticity-related proteins (PRPs). Rather, the output of the nucleus, whether transported RNPs or PRPs, gets recruited to the tag independently of the signal to the nucleus in order to stabilize plasticity of synaptic transmission.

2.2 The Functional Role of ARC in Inverse Synaptic Tagging

Until recently, the synaptic function of the IEG ARC has been at odds with studies implicating ARC in both the strengthening and weakening of synaptic contacts. Arc transcription is known to be strongly induced with neuronal activity that produces both LTP and long-term memory and Arc mRNA is transported to and localized near activated synapses, presumably to be locally translated (Lyford et al. 1995; Link et al. 1995; Guzowski et al. 2000; Steward et al. 1998). Furthermore, Arc mRNA is degraded in an activity- and translation-dependent manner in dendrites, consistent with a tight temporal and spatial burst of translation near activated synapses (Farris et al. 2014). ARC protein may play a role in stabilizing F-actin during consolidation, as depletion of ARC via antisense oligonucleotides 2 h after LTP resulted in a rapid decay of LTP and loss of F-actin at synaptic sites (Messaoudi et al. 2007). ARC protein has also been shown to interact with members of the endocytic machinery, dynamin, and endophillin, leading to the internalization of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPARs) and therefore a weakening of synapses during both LTD and homeostatic forms of synaptic plasticity (Chowdhury et al. 2006; Shepherd et al. 2006; Rial Verde et al. 2006). How one molecule could participate in so many forms of plasticity remains unclear. Recently, however, Okuno and colleagues have discovered that ARC protein, induced from strong synaptic activation, formed a tight interaction with the inactive (calmodulin-unbound) β isoform of CaMKII at unpotentiated or weak synapses (Okuno et al. 2012). Synaptic levels of ARC were correlated with the removal of surface AMPARs, an effect that was abolished with a lack of CaMKIIβ (Okuno et al. 2012). Some evidence that such a widespread weakening of synapses, developing 1 h after action potential firing has been presented (Bukalo et al. 2013), even though the same stimulus “rescues” early-LTP at tagged synapses from decay (Dudek and Fields 2002). These data together suggest that Arc induced by activity may contribute to late-phase long-term plasticity by preventing the enhancement, or even by inducing depression of unstimulated synapses, to maintain balance of synaptic weights.

What is fascinating about this idea of inverse tagging, though, is that it unites plasticity-induced gene expression, dendritic mRNA localization, and local translation with tagging and synapse-specific remodeling. CaMKII is an enzyme consisting of two subunits (α and β) in the brain. The regulatory subunit (CaMKIIβ) is thought to be translated in the soma and then transported to synapses, whereas CaMKIIα is locally translated only at active synapses. This situation is well suited for CaMKIIα to act as an initial tag at activated synapses to capture CaMKIIβ to form the active CaMKII enzyme which can then initiate downstream pathways such as actin remodeling, which can then serve as a tag for further PRPs. However, for inactive synapses without CaMKIIα, CaMKIIβ presence serves as the tag for locally translated ARC to be captured and prevent the undesired enhancement of weak synapses (Nonaka et al. 2014). This example of inverse synaptic tagging illuminates novel and interesting avenues for future tagging research as well as providing a mechanism for how ARC can participate in multiple forms of synaptic plasticity. Remaining unclear, though, is the function of ARC protein enriched in the dendritic shaft, which in the presence of calcium and calmodulin, prefers to interact with CaMKIIα.

Interestingly, inverse tagging may be consistent with the concept of “cross-tagging,” a positive associative interaction of LTP and LTD in which late-LTP (or late-LTD) at one synaptic input is able to promote stabilization of the opposite form of plasticity (LTD or LTP, respectively) at another independent synaptic input where only the early phase was induced (Sajikumar and Frey 2004; reviewed in Frey and Frey 2008). The functional crosstalk between various, and seemingly opposing, forms of plasticity at separate synapses imposes additional requirements for the tagging and capture of PRPs. Namely, the tags must be specific for each type of plasticity and the genes required for both LTP and LTD must be induced either by the late-inducing stimulus or at least available from prior transcriptional history. Consistent with cross-tagging, inverse tagging also requires that PRPs are “captured” by both active and inactive synapses that were “tagged” by different mechanisms (see Fig. 9.1). Both mechanisms expand the synaptic tagging hypothesis to include a cell-wide capture process that integrates plasticity at different afferents and considerably strengthens the argument that PRPs or signals from the nucleus do not rely on locale-specific signals from their inputs. The inverse tagging model does not, however, specifically address whether or how ARC, or other IEGs might play a role in stabilizing synaptic plasticity at modified synapses or whether any input specificity to LTD can be maintained. We look forward to future studies on this front.

Fig. 9.1
figure 1

ARC: the proverbial “tag along” in synaptic and inverse tagging. Having ARC be at the right place at the right time is critical for multiple forms of synaptic plasticity. Therefore it is not surprising that ARC is heavily regulated at every level from birth (transcription) to death (protein degradation). Here we depict the steps necessary for ARC protein to make it to a potentiated or tagged synapse. (1) A strong afferent stimulus is needed to generate (2) action potentials, which are sufficient to (3) induce Arc. Arc mRNA is then packaged into a ribonucleoprotein (RNP) particle and (4) transported out of the nucleus and trafficked via microtubule-based transport into dendrites. Arc’s translation is suppressed until its RNP is remodeled by extracellular cues, such as signals from a potentiated synapse, whereby it gets (5) rapidly translated by polyribosomes present in dendrites. How ARC protein finds a potentiated or tagged synapse is not known, although it may be directly due to the docking of its mRNA near active synapses. ARC co-immunoprecipitates with F-actin and CaMKIIα and has been suggested to play a role in stabilizing F-actin during consolidation. CaMKIIα mRNA is present constitutively in dendrites, but has also been shown in vivo to localize near stimulated synapses. It is likely, that the local translation of CaMKIIα occurs soon after stimulation since it does not need to be transported from the nucleus in response to activity. Alternatively, (illustrated here), ARC protein has been shown in vitro to behave like an “inverse tag” by way of its association with the inactive form of CaMKIIβ at unpotentiated synapses, possibly ensuring the synapse remains weak by removal of AMPARs. Similarly, ARC has been shown to be required for synaptic depression by removing AMPARs from the synapse, although unlike with inactive synapses, it is not known what recruits ARC to depressed synapses and whether this occurs after a strong stimulus

Full size image

3 Conclusions

Recent data describing mechanisms underlying nuclear to synapse signaling can shed light on our understanding of synaptic tagging and capture and how mRNA localization and local protein synthesis might fit into the picture. The extensive bidirectional dialogue between the nucleus and the synapse underscore the complexity and amount of regulation concerning these processes as they relate to plasticity and learning. However, what is even less clear is how these processes fit into the specific networks in the context of complex behaviors. New models that can integrate how mRNA targeting and local protein synthesis contribute to synaptic tagging and how multiple inputs can be calculated to form a coordinated output of gene expression to modify synaptic transmission will be critical for understanding the mechanisms of learning and memory.