Keywords

1 Relevance of Synaptic Tagging in Late Plasticity

The synaptic tagging hypothesis was proposed in 1997 by Frey and Morris as a mechanism underlying late associative plasticity which depends on new protein synthesis (Frey and Morris 1997). Although activation of synaptic tagging can be detected by associative late-phase long-term potentiation (L-LTP) in two-pathway experiments (for details see other articles in this book such as one by Redondo and Morris), it was difficult to define synaptic tagging in a cell biological sense. Martin and Kosik (2002) listed three conditions for a synaptic tag: it should be (1) spatially restricted, (2) time-limited and reversible, and (3) able to interact with cell-wide molecular events that occur after strong stimulation to produce long-term synapse-specific strengthening. Now that associative late-phase long-term depression was reported to involve a synaptic tagging mechanism (Sajikumar and Frey 2004), and taking into account of the fact that the essential difference of late plasticity from early phase is dependent on protein synthesis, the third condition can be substituted with the following; a synaptic tag should be able to interact with cell-wide molecular events that occur after particular stimulation to produce a protein synthesis-dependent, synapse-specific, and persistent modification of synapse strength.

Nevertheless, wide variety of cell biological activities can be nominated as candidates of synaptic tag. Thus, to figure out the molecular entity and cell biological mechanism of synaptic tagging, we had to start from considering its neurobiological relevance. In this short review, we would first like to discuss the relevance of synaptic tagging as an input-specificity mechanism of late plasticity and relationship to the expression mechanism of late plasticity.

1.1 Synaptic Tagging as an Input-Specificity Mechanism of Late Plasticity

Studies on the early phase of long-term potentiation (E-LTP) have revealed major molecular mechanisms for the input-specificity (Collingridge 2003) and the expression (Nicoll 2003). Of the two, the input-specificity mechanism is an upstream component that launches the expression mechanism in restricted synapses.

E-LTP can be evoked by various methods such as high frequency stimulation (Bliss and Lømo 1973), theta burst stimulation (Larson et al. 1986), pairing of presynaptic activation, and postsynaptic depolarization (Magee and Johnston 1997), and synchronized firing of both pre- and postsynaptic cells (spike timing-dependent plasticity) (Markram et al. 1997). In these protocols, both pre- and postsynaptic cells are coincidently excited, as Hebb (1949) assumed in his famous monograph. Depolarization input-specifically releases the NMDA receptor channel from blockade by magnesium, and another immediately following glutamate activates the same receptor to elicit calcium influx through it (Novak et al. 1984). In the hippocampal CA1 pyramidal-Schaffer-collateral synapses, or the dentate gyrus granule cell-perforant path synapses, calcium influx through NMDA receptors is known to be necessary and sufficient to elicit E-LTP (Malenka et al. 1988) and believed to be the coincidence detector for the input-specificity mechanism (Collingridge 2003).

On the other hand, given the dependency of late plasticity on induction of a new set of gene expression (Krug et al. 1984; Frey et al. 1988), the late expression mechanism must involve synaptic functions of new PRPs. Competition among associative L-LTP of multiple synapses was observed when protein synthesis was limited (Fonseca et al. 2004), supporting the idea that late expression depends on synaptic functions of PRPs. The late input-specificity mechanism should trigger synaptic functions of PRPs. Synaptic tagging was proposed as a late input-specificity mechanism that enables PRP functions in the restricted synapses.

The synaptic tagging hypothesis assumes that a synaptic tag is activated by synaptic inputs that evoke early-phase plasticity. Furthermore, an important feature of PRP was revealed by the experiments comparing with a counter-hypothesis, the mail hypothesis, in which each PRP is born to be delivered specifically to the synapse that had received the plasticity-evoking inputs. Two-pathway experiments excluded the mail hypothesis (Frey and Morris 1997), and supported the view of synaptic tagging hypothesis in which PRPs are transported in an unspecific manner along dendrites thereby available in all synapses as suggested in Fonseca et al. (2004). Thus, synaptic tagging hypothesis proposes not only an input-specificity mechanism of late plasticity, but also a thorough story for PRPs, in which PRPs synthesized in soma prevail in all dendrites and function only in the “tagged” synapse to express late plasticity.

1.2 Two Possibilities of Synaptic Tagging Action

The idea that the input-specificity mechanism triggers the expression mechanism suggests that two processes involved in late expression can be regulated by synaptic tagging, namely, synaptic localization of PRP and synaptic adaptation. A simpler possibility of the cell biological activity for synaptic tagging is input-specific regulation of PRP localization in synaptic area, because PRP transported unspecifically along dendrites should be captured by the tagged synapse before functioning there. Our research showed an example of this mechanism (Okada et al. 2009).

On the other hand, synaptic adaptation involves input-specific modulation of the synapse. The modulation is induced by expression mechanisms of early plasticity and results in PRP acceptance in the synapses. Increase in surface expression of synaptic GluA receptors (electrophysiological plasticity) and enlargement of spine head (morphological plasticity) are considered as major expression mechanisms of the early plasticity, whereas up to present, consensus on the expression mechanism of late plasticity has not been reached. It is likely that late expression mechanisms modulate the functional structure of the synapse built by the early expression and make it persistent. For example, synaptic F-actin increased in an L-LTP-dependent manner (Fukazawa et al. 2003). This increase may be preceded by a transient decrease in synaptic F-actin which was NMDA receptor-dependent (Ouyang et al. 2005), suggesting remodeling of F-actin in spines was promoted by early plasticity. Thus, early plasticity likely alters molecular composition, size, and activity of the postsynaptic protein complex. If the altered state of the complex is a prerequisite for integration of new PRPs, only synapses expressing early plasticity can express late plasticity, and we refer to this alteration of molecular states in the synapse as synaptic adaptation. Although a special case of synaptic adaptation was experimentally excluded (Frey and Morris 1998a), we guess that the general idea of synaptic adaptation mechanisms may be consistent with synaptic functions of some PRPs. Furthermore, synaptic adaptation and synaptic localization are not mutually exclusive as principles of synaptic tagging.

2 Strategy to Reach the Cell Biological Activity of Synaptic Tagging

In this chapter, we would like to describe our strategy to define cell biological activity for synaptic tagging.

2.1 Advantage and Limitations of Two-Pathway Protocol

The two-pathway experiment, which measures associative L-LTP, has been the exclusive method to detect activation of synaptic tagging (Frey and Morris 1997). By the use of inhibitors, two-pathway experiments revealed many molecules involved in associative L-LTP and L-LTD, such as PDE4 (Navakkode et al. 2004), PKMζ (Sajikumar et al. 2005), PKA, ERK1/2, CaMKII and PKMζ (Sajikumar et al. 2007), AKAP (Huang et al. 2006), PKA (Young et al. 2006), neuropsin (Ishikawa et al. 2008), and F-actin (Ramachandran and Frey 2009). However, involvement of these molecules in synaptic tagging is difficult to be concluded from these experiments.

Accumulating knowledge on mechanisms underlying late plasticity suggests that concerted actions of multiple internal processes are required for late expression. Internal processes include preceding early plasticity, induction of new PRP gene expression, PRP production and transport, synaptic tagging and capture, and PRP functioning in the targeted synapses (Reymann and Frey 2007). Furthermore, one molecule can be involved in many of the multiple internal processes. For example, NMDA receptors are known to contribute to multiple processes involved in late plasticity such as early plasticity (Collingridge 2003), induction of gene expression (Cole et al. 1989), and PRP transport into spine (Okada et al. 2009). Under such a complicated situation, an inhibitor specific to a molecule that is involved in only synaptic tagging process should be used to conclude involvement of a molecule in synaptic tagging.

To circumvent such an agnostic difficulty, it is obvious that synaptic tagging activity itself should be directly measured. But, what activity should be measured? We decided to first hypothesize a possible activity that conforms to synaptic tagging hypothesis, and then accumulate experimental evidence supporting it.

2.2 Controlled Transport Across Dendrite–Spine Boundary

As discussed in previous section, the late expression is achieved by function and localization of PRPs in synapses and synaptic tagging enables PRPs to do so. We paid much attention to PRP transport, which is a prerequisite process for synaptic functions of PRPs. Since PRP transport from soma to dendrites is unspecific (Frey and Morris 1997), we focused PRP transport from dendrite to synapse. This transport may have two steps, namely, entry from dendrite to synaptic area and integration to preexisting synaptic protein complex. Both of them can be activity-dependent (as discussed in Sect. 6.1.2), thereby promising candidate for synaptic tagging.

Proteins and vesicles are transported in dendrite along cytoskeletons, a major part of which is microtubule–kinesin system (Brady 1995), while dendritic transmembrane proteins are carried by myosin motors (Lewis et al. 2009). Microtubules are not found in spines where F-actin–myosin system is predominant (Lebeux and Willemot 1975). Cargoes should switch from microtubule to F-actin to enter spines, and actually such switching was observed in neurons (Shakiryanova et al. 2006; Correia et al. 2008) and non-neuronal cells (Kuroda and Fukuda 2004), while direct interaction between kinesin and myosin was also observed (Huang et al. 1999). Protein trafficking in and out of a spine is limited, and sometimes activity-dependent (Bloodgood and Sabatini 2005). Spine localization of several proteins such as CaMKII (Shen and Meyer 1999), AMPA receptors (Ehlers et al. 2007; Matsuo et al. 2008), profilin (Ackermann and Matus 2003), Arc (Moga et al. 2004), and neurabin/Lfc (Ryan et al. 2005) was reported to be regulated by synaptic activity.

These observations suggest that transport of some synaptic proteins from dendrite to spine depends on synaptic activity, which suggests an example of synaptic tagging in accordance with the synaptic localization theory discussed in Sect. 6.1.2). This idea can be directly tested by tracking the movement of a PRP from soma to spines. We decided to ask whether the PRP prevails in all dendrites before it enters spines activity dependently.

2.3 Critical Assumptions of Our Hypothesis

We hypothesized that a synaptic tagging mechanism activity dependently captures PRP in dendrite to translocate it into spines. To test our hypothesis, we referred to known characteristics of synaptic tagging (Frey and Morris 1997, 1998a, b), which can be summarized as follows: (1) PRPs prevail in all dendrites, namely, they are transported unspecifically in dendrites. (2) E-LTP evoking stimulus, such as NMDA receptor activation, activates synaptic tagging input-specifically. (3) Activation of synaptic tagging is independent of protein synthesis. (4) Once activated, synaptic tag is active for a few hours. These are essentially identical to the synaptic tagging conditions described in Kelleher et al. (2004).

2.4 Use of Vesl-1S for the Tracer PRP

Next, we selected a PRP suitable for the tracer of input-specific transport to spines. The tracer protein should be newly synthesized in soma after intense neuronal activity evoking late plasticity, localized in the synapse, and needed for late plasticity. We selected Vesl-1S (Homer-1a) protein as the tracer among our list of genes induced during late plasticity (Matsuo et al. 2000). Vesl-1S protein is one of the proteins that fulfill all of these criteria for the tracer PRP. Furthermore, its role in the late plasticity was known as following. Vesl proteins are one of the postsynaptic scaffolding proteins which bind to mGlu1/5 receptors (Brakeman et al. 1997; Kato et al. 1998), IP3 receptors (Tu et al. 1998), Shank (Tu et al. 1999), TrpC channel (Yuan et al. 2003) and so on. Long forms of Vesl/Homer proteins are tetrameric (Hayashi et al. 2006), make a postsynaptic network with Shank (Hayashi et al. 2009), and regulate constitutive activity of mGluRs (Ango et al. 2001). On the other hand, short form of Vesl/Homer protein such as Vesl-1S/Homer-1a is an immediate-early gene product inducible in L-LTP in the dentate gyrus (Kato et al. 1997; Brakeman et al. 1997). Knock in mice lacking Vesl-1S shows abnormal long-term memory (Inoue et al. 2009). Its mRNA was detected in soma, but so far not in dendrites (Kato et al. 1997; Matsumoto et al. 2007). Vesl-1S/Homer-1a works as a dominant negative form of long form Vesl/Homer proteins (Xiao et al. 1998). Glutamate stimulation caused biphasic changes in long-form Homer-1c clusters in neurons; first reduced then later increased (Inoue et al. 2007). This transient decrease was coincided with Homer-1a accumulation and biphasic changes were totally dependent on Homer-1a expression. These observations strongly suggested that Vesl-1S/Homer-1a dissociated the preexisting Homer clusters, which was required for later enhancement of the cluster. Thus, Vesl-1S/Homer-1a is not directly involved in persistency of late plasticity, but it plays a key role as an initiator of PSD rearrangement prerequisite for late plasticity.

3 Results

3.1 Activity-Dependent Regulation of Spine Translocation of Vesl-1S/Homer-1a Protein as a Synaptic Tagging

We tested our hypothesis by measuring fluorescence of EGFP-fused Vesl-1S/Homer-1a (VE) proteins in dendritic spines of primarily cultured neurons of rat hippocampus (Okada et al. 2009). We tested whether synaptic activity regulates Vesl-1S localization in the activated spines, and whether this activity-dependent spine translocation fulfills above-mentioned four known characteristics of synaptic tagging.

Cells were seeded on a coverslip at a high density (105 cells/1.8 cm2) and transfected with the Vesl-promoter-driven VE plasmid on DIV9-12. Fluorescence was observed after growth of spines on DIV18-25. The Vesl promoter has moderate induction power restricted in neurons, thus, a limited portion of neurons expressed moderate levels of VE protein. VE fluorescence was observed homogeneously in entire neurons including soma, axon, and dendrites. Mushroom-shaped spines of pyramidal cell-like neurons and granule cell-like neurons were tested, and difference among them was not observed. Spine head size was not usually changed by the applied stimuli, and spines with significant change in size and shape were excluded from analysis.

During growing-up incubation, elaborate synaptic connection was autonomously built among neurons. These neurons fire spontaneously, evoking glutamatergic transmission in many spines. Since this transmission is spontaneous, synaptic NMDA receptors can be selectively activated by changing the extracellular medium into magnesium-free artificial cerebrospinal fluid (Mg-free ACSF). Brief application of Mg-free ACSF increased spine fluorescence gradually and persistently, which was not affected by inhibitors of proteasome and protein synthesis, suggesting the increase is resulted by translocation of VE proteins from dendrites to spines (#3 of the synaptic tagging characteristics described in Sect. 6.2.3).

Translocation of soma-derived Vesl-1S/Homer-1a into spines was detected by the use of photoactivatable GFP (PAGFP) instead of EGFP. PAGFP-Vesl1S (PAV) was photoconverted only in soma by scanning illumination of small laser spots. The fluorescing PAV prevailed in most dendritic branches, but do not enter into spines (#1 of the synaptic tagging characteristics described in Sect. 6.2.3). PAV entered spines which received local activation of NMDA receptor by means of microperfusion (#2, see Fig. 6.1a). Furthermore, spine translocation activity lasted for 3 h (#4, see the next section for details). Similar movement was not found when EGFP was used instead of VE or PAV, indicating that Vesl-1S/Homer-1a, a PRP, is the target of the transport. We concluded that NMDA receptor-dependent control of VE transport to spine conforms to the synaptic tagging hypothesis (Fig. 6.2). Thus, synaptic tagging hypothesis was demonstrated with an exemplifying PRP, Vesl-1S/Homer-1a.

Fig. 6.1
figure 1

The cartoon illustrates neurons with spines (small circles). Fluorescence protein translocation is shown by darker colors in the spines. (a) Experiments with VPA, indicating input-specific spine translocation of soma-derived VPA. (b) 8-Bromo cyclic GMP (BrcGMP) application for 20-min activated VE translocation. (c) BrcGMP failed VE translocation in the presence of TTX. VE translocation activity regained by TTX washout 0.5–3 h after BrcGMP, but not 4 h, suggesting VE translocation was reversibly and persistently activated

Full size image
Fig. 6.2
figure 2

“Synaptic tagging and capture” by activity-dependent PRP translocation into spine. Strong inputs to one synapse evoke input-specific late-LTP through the following processes. First, synaptic inputs (1) activate somatic PRP synthesis (2). New PRPs are transported along dendrites unspecifically (3). On the other hand, some downstream signals of the strong inputs activate synaptic tagging (4). Activated synaptic tag facilitates PRP translocation from dendrite to spine. We speculate that PRPs are translocated from the dendritic transport system to the intraspine transport. PRPs are thus to be incorporated into PSD or other synaptic machinery and contribute to expression of input-specific late-LTP. PRP translocation takes place also in the synapses received weak inputs because weak inputs that can evoke early LTP are assumed to activate synaptic tagging. Synapses without receiving inputs are not tagged, therefore, PRP translocation does not occur there, resulting in failure in LTP

Full size image

3.2 Molecular Mechanisms Underlying VE Protein Transport to Spine

Further studies revealed that spine translocation of VE protein involved two distinct reactions, protein kinase G (PKG) and a TTX-sensitive component. We searched downstream signals of NMDA receptor activation and found involvement of extracellular calcium, indicating that calcium influx through NMDA receptors activates both early phase expression and the synaptic tagging. Intracellular increases in calcium ions usually trigger multiple reactions and we found that calcium-dependent nitric oxide (NO) production was involved in spine translocation of VE proteins. NO can spread across membranes, however, a water-soluble NO scavenger, N-(dithiocarboxy)-sarcosine complexed with iron, in the extracellular medium (a condition under which extracellular NO is scavenged) did not affect NMDA receptor-dependent spine translocation of VE proteins, suggesting that the generator and target molecules of NO were in the same intracellular space. NMDA receptor activation was mimicked by a membrane permeable cyclic GMP analogue, 8-bromo cyclic GMP (BrcGMP), suggesting that the major contribution of NO in the VE transport was PKG activation through cyclic GMP production. Involvement of NO, a diffusible activator of soluble guanylyl cyclase, indicates an involvement of a moderately remote interaction between the postsynaptic site near the NMDA receptor and the spine–dendrite border, which is consistent with the idea that synaptic activity regulates interaction with material transport in the dendrite. The nature of this interaction is the subject of the future study.

Frey and Morris (1998a) reported that the synaptic tagging has a lifetime of 1–2 h from observations of time-dependent decay of the associativity between weak and strong stimuli for L-LTP. We observed that BrcGMP could not activate VE translocation in the presence of tetrodotoxin (TTX), while washout of TTX within 3 h after BrcGMP restored VE translocation (Fig. 6.1b, c). These results suggested that spine translocation of VE proteins was promoted by some unknown factor released in an activity-dependent manner. These results also suggested that PKG or its downstream was persistently active for 3 h, which is longer than the time window observed in slice preparation (Frey and Morris 1998a). A shorter time window (30 min) for the association between activities of the perforant path and the basolateral amygdala in behaving animals was reported (Frey et al. 2001). Behavioral tagging experiments showed ~1–2 h of the time window (Ballarini et al. 2009). The life time of associativity and VE translocation may be different, and depends on the experimental configuration, such as temperature and complexity of the processes. Higher stringency (shorter lifetime) is expected for more complex systems.

We also tested involvement of mGluR1, mGluR5, PKA, PKCα/β2, and trkB, in VE translocation using specific inhibitors such as CPCCOEt, MPEP, PKAI, GF109203X, and K252a, respectively (Okada et al. 2009). All of these drugs had no effect on VE transport into spines, while some of them were implicated in associative late plasticity, for example, PKA (Sajikumar et al. 2007) and trkB (Lu et al. 2011). They may be involved in the internal processes of late plasticity other than PRP translocation. In addition, AKAP (Huang et al. 2006), CaMKII (Okuno et al. 2012), CaMKIV, MEK1/2 (Sajikumar et al. 2007), PKMζ (Ling et al. 2002; Sajikumar et al. 2005), and PDE4B4 (Navakkode et al. 2004) are reported to be involved in associative late plasticity, but we did not confirm their involvement in VE translocation.

4 Perspectives

4.1 Multiple Mechanisms for Synaptic Tagging

It is noted that late plasticity requires induction of hundreds of new PRPs (Nedivi et al. 1993; Matsuo et al. 2000). These PRPs occur in synapses after varied lags, depending on their promoters or transport systems. It is therefore likely that the synaptic tag should be persistently active to cover lags of all PRPs required for late plasticity. It is not clear whether an exclusive tag is activated during the entire lifetime of synaptic tagging, or multiple tags for synaptic functions of individual PRPs are activated during portions of the lifetime. Since synaptic tagging is closely linked to the expression mechanism, we speculate each PRP may be introduced in the late expression by distinct synaptic tagging mechanisms suitable for each PRP. The idea of the multiple tagging emphasizes that the late plasticity is established as the final consequence of concerted actions of PRPs having distinct localization and roles (Fig. 6.3). PRP may be classified into several groups in the light of tagging mechanism. Proteins interacting each other may be transported in the same vesicles, such as Homer/Vesl and class I metabotropic glutamate receptors (Ango et al. 2000). Cell biological activities other than spine translocalization can be synaptic tagging, such as synaptic adaptation.

Fig. 6.3
figure 3

A hypothesis of multiple PRP-specific tagging for late expression. This hypothesis assumes that each PRP contributes to late expression in distinct manners, because they are produced at discrete timing and location, and require dissimilar localization and conditions for individual synaptic functions. L-LTP inducing inputs activate somatic synthesis of PRPs and the early phase of LTP (E-LTP). E-LTP in turn activates three processes, synaptic adaptation (expressed by brown spine), local synthesis, and synaptic tag activation (indicated by yellow spiny symbols). Local synthesis supplies new PRP at a shortest lag behind synaptic activation, and these PRPs (depicted by orange diamonds) may contribute to synaptic strengthening as well as adaptation. Synaptic tagging depicted here is activity-dependent PRP translocation and cooperates with dendritic transport of PRPs. Since hundreds of PRPs are newly synthesized with ranged lags (depicted as to two kinds of PRPs indicated by magenta circles and green squares, respectively), coincidence of the tag activation and PRP synthesis-transport should be persistent for hours, while the coincidence time window of E-LTP is as narrow as 10 ms. Cell biological activities other than the translocation regulation can also be synaptic tag but not included in this figure. All PRPs are gathered in the tag-activated synapses and contribute in late expression in concert (indicated by the square integrating synaptic functions of classes of PRPs)

Full size image

4.2 Molecular Events Involved in Late Expression

In this section, we discuss molecules implicated as major contributors in the late expression mechanism. These proteins may also characterize distinct groups in synaptic tagging diversity.

4.2 AMPA Receptor Insertion

Increase in surface expression of GluA receptors is considered as a major expression mechanism for E-LTP (Bliss and Collingridge 2013). GluA receptor surface expression is dynamically regulated by membrane-vesicle fusion in a manner depending on the subunit composition (Shi et al. 2001). Extrasynaptic receptors are also highly mobile and GluA receptors conjugated with Stargazin proteins are fixed in the synaptic region by binding to PSD-95 proteins (Opazo et al. 2012). This slot hypothesis for receptor surface expression is a good model for stabilization of membrane-sorted receptors and one of the possible underlying mechanisms of the early expression. Scaffolding proteins such as Homer and Shank also contribute to the receptor stabilization by forming sub-membrane networks that regulate receptor localization, recycling, and postsynaptic reconstruction (Xiao et al. 1998; Lu et al. 2007; Hayashi et al. 2009). While the early expression is reversible and lasts for a few hours in vitro, the late expression lasts far longer. One of the possible late expression mechanisms may be increments in the slot number using new component proteins supplied as new PRPs. This situation is associated with an increased number of total molecules in a spine, leading to PSD increment (Desmond and Levy 1986) and spine head enlargement as observed in matured spines (Matsuzaki et al. 2004). Thus, GluA receptor trafficking is supported and regulated by structural components of postsynaptic density. Activity-dependent regulation of these proteins is one of the possible major expression mechanisms, thereby the targets of synaptic tagging.

4.2 F-actin Network

F-actin content in spines is essential for morphological plasticity. Spine head size is enlarged transiently in E-LTP (Matsuzaki et al. 2004) and persistently in L-LTP in a BDNF- and protein-synthesis-dependent manner in the hippocampus (Tanaka et al. 2008), but not in the cerebellum (Sdrulla and Linden 2007). Accordingly, E-LTP is associated with a dynamic increase in spine F-actin (Okamoto et al. 2004), while F-actin is accumulated in synapse layer of dentate granule cells in L-LTP (Fukazawa et al. 2003).

F-actin is the major cytoskeleton and the transporting rail in the spines; therefore, its increase is necessary in the era of spine enlargement to transport the increasing numbers of molecules for the support of the increased functions. This consideration suggests that F-actin network should be stabilized in the late plasticity. Synaptopodin is an actin-binding protein, expressed a few hours after L-LTP induction as a late PRP (Yamazaki et al. 2001) and enhanced surface expression of GluA receptors through Ca2+ release from the spine apparatus (Vlachos et al. 2009). Overexpression of Synaptopodin stabilized spine head enlargement after NMDA receptor stimulation of cultured hippocampal neurons (Okubo-Suzuki et al. 2008). Synaptopodin-deficient mice lacked spine apparatus, and did not show E- and L-LTP (Deller et al. 2003). Thus, Synaptopodin can be involved in the late expression mechanism through F-actin stabilization.

4.2 Degradation of Preexisting PSD

Postsynaptic protein complexes including receptor scaffolding and F-actin networks are reconstructed during L-LTP by posttranslational modification of preexisting PSD proteins. Phosphorylation regulates protein–protein interaction, while polyubiquitination triggers protein degradation and was involved in LTP maintenance (Fonseca et al. 2006) and memory reconsolidation (Lee et al. 2008). Molecules that destroy or destabilize the integration of the postsynaptic protein network should be immediately degraded after disintegration of preexisting structure. For example, Vesl-1S/Homer-1a possesses a PEST sequence for proteasomal degradation and rapidly degraded in spines (Ageta et al. 2001). Other mechanisms of protein degradation such as Caspase3 (Li et al. 2010), and autophagy (Shehata et al. 2012) were reported to have influence on plasticity and memory.

4.2 Extracellular Component

Extracellular molecules such as Cadherins (Tang et al. 1998) and β-catenin (Murase et al. 2002) were implicated in LTP. These molecules are known to be involved in intracellular signaling as well as a cell adhesion, and affect synaptic size, transmission efficacy, and transcription regulation.

EphB receptor is a cell adhesion molecule interacting with the Ephrin ligand, and involved in spine formation and synaptic plasticity by facilitating glutamate receptor clustering (Henkemeyer et al. 2003). Although regulatory mechanisms of these extracellular molecules are not well known, they may fall in the important category of synaptic tagging or expression mechanism of late plasticity. For example, Neuropsin is an extracellular protease implicated in late associative LTP (Ishikawa et al. 2008) and reported to cleave EphB2 in the amygdala which triggers modulation of anxiety (Attwood et al. 2011).

4.2 Brain-Derived Neurotrophic Factor (BDNF)

BDNF and its receptor trkB are also implicated in synaptic and behavioral tagging. BDNF is essential for protein-synthesis-dependent persistent structural plasticity of dendritic spines (Tanaka et al. 2008). TrkB is transiently activated by E-LTP evoking theta burst in confined synapses in a manner independent of protein synthesis, which was necessary for associative L-LTP and behavioral tagging (Lu et al. 2011). Although TrkB activation seems to fulfill conditions of synaptic tagging, input-specific TrkB phosphorylation implies spatially restricted action of BDNF. The release mechanism of BDNF is not elucidated (Bramham and Messaoudi 2005).

4.2 Distributed Plasticity

Although input-specific expression was considered as an essential feature of E-LTP, distributed expression has been consistently observed, which appeared within the same dendritic branch (Engert and Bonhoeffer 1997). Input-specific E-LTP persistently activated Rho-GTPase within surrounding area of 5 μm distance, suggesting distribution of plastic changes among synapses in vicinity (Murakoshi et al. 2011). Computational and electrophysiological studies suggested that NMDA spike propagation within a dendritic segment is the key for distributed LTP (Polsky et al. 2004). These observations suggest that molecules or activities spreading or propagating within dendrites are involved in tagging or expression mechanisms underlying distributed plasticity. In accordance with these observations, cluster plasticity hypothesis was suggested (Govindarajan et al. 2006) with experimental demonstrations using hippocampal slices (Govindarajan et al. 2011).

4.3 Local Synthesis

PRP supply does not necessarily require the synaptic tagging, when the PRP is translated by local synthesis, another input-specificity mechanism for late plasticity. Subcellular fractionation and electron microscopic studies demonstrated that dendrites and postsynaptic area contained mRNAs for some PRPs which are induced and functioning in late plasticity (Eberwine et al., 2001). Dendritic transport of mRNA involves RNA granules which contain various mRNAs (Anderson and Kedersha 2006), and stability of mRNA is regulated in part by miRNAs (Shouten et al. 2013). Local synthesis was excluded from associative late plasticity in the original report of synaptic tagging (Frey and Morris 1997), while it is involved in persistent plasticity in other conditions. For example, mRNA for GluA1 receptor was found in dendrites, and its translation was activated by synaptic activity, enabling input-specific supply of new receptors even in spines of dendrites without physical connection to the soma (Ju et al. 2004). However, local synthesis and somatic synthesis are not mutually exclusive; rather they may work in concert. For instance, PRP such as α-calcium-calmodulin-dependent protein kinase II may be supplied by both mechanisms (Miller et al. 2002). It is noted that local synthesis achieves PRP delivery to the activated synapses more rapidly than somatic synthesis does. PRPs supplied by these mechanisms may have distinct roles in early and late expression mechanisms. Availability of mRNA for particular PRPs in relation with the translational machinery may function as the synaptic tagging as reported in Aplysia neurons (Wang et al. 2009).

It is noted that protein synthesis dependence of late plasticity has been often shown by the use of protein synthesis inhibitors such as anisomycin, a blocker of peptidyltransferase reaction in eukaryotic mRNA–ribosomal complex. Anisomycin is also known to activate MAP kinases (Takenaka et al. 1998), and enhance noradrenaline release in the amygdala (Sadowski et al. 2011), suggesting its effects on cellular function other than protein synthesis.

4.4 Synaptic Tagging as a Cellular Mechanism of Memory Association

According to Hebb’s theory, the neuronal entity of a brain function is a cell assembly built up through plastic changes in synaptic connection (Hebb 1949). According to sparse coding theory, only a few synapses among 103–105 undergo plastic changes in an activity-dependent manner (Olshausen and Field 2004). Consistently, small portions of neurons were activated during learning episodes, and recall of the late memory reactivated fewer neurons (Reijmers et al. 2007). Optogenetical manipulation showed that reactivation of a set of neurons that had been activated during learning caused recall of the experience, supporting Hebb’s theory (Liu et al. 2012).

We usually do not remember every component of an experience. Especially in the late memory, fewer components can be recalled later. This well-recognized feature of our memory, selection of recallable subjects in an experience, may be closely related to the function of synaptic tagging in the late memory. Original synaptic tagging is proposed for cellular level observation and considered as an input-specificity mechanism for late plasticity, which selects synapses undergoing plastic changes. Under the above-mentioned assumption by Hebb, this cellular role of synaptic tagging can be translated into systems level. Because individual axons transfer separate sets of information composing an experience, selection of emphasized synapses is equivalent to that of memory subjects.

Another feature of our memory is association. Above consideration on the role of synaptic tagging in the subject selection suggests that two distinct memories can be associated when the cell assembly representing each memory contains synapses that are activated commonly by the two memories. This idea for neuronal basis of memory association was also suggested by the behavioral tagging revealed by the “behavioral two-pathway” protocol (Ballarini et al. 2009), in which weak conditioning of a task (causing short memory) was reinforced (becoming persistent memory) by strong conditioning of another task.

Episode memory first represented in the hippocampal network is thought to be sent to the cerebral cortices representing individual modalities of the episode, to form remote memory. During memory transfer from hippocampus to multiple cortices, there must be some rule to recall many of the memory components and reproduce the episode memory as a whole. This rule is not known, but a possibility was suggested as memory tag (Lesburgueres et al. 2011). In this way, synaptic, behavioral and memory tagging contribute to the consistency of restored and recalled memory, by generating proper sets of synaptic connections and neuron networks. Deficits of synaptic tagging may thus disrupt such life-spanning association among experiences. For example, post-traumatic stress disorder may occur through association of a traumatic component and any details of the event.