During learning and memory, it has been suggested that the coordinated electrical activity of hippocampal neurons translates information about the external environment into internal neuronal representations, which then are stored initially within the hippocampus and subsequently into other areas of the brain. A widely held hypothesis posits that synaptic plasticity is a key feature that critically modulates the triggering and the maintenance of such representations, some of which are thought to persist over time as traces or tags. However, the molecular and cell biological basis for these traces and tags has remained elusive. Here, we review recent findings that help clarify some of the molecular and cellular mechanisms critical for these events, by untangling a two-way signalling crosstalk route between the synapses and the neuronal soma. In particular, a detailed interrogation of the soma-to-synapse delivery of immediate early gene product Arc/Arg3.1, whose induction is triggered by heightened synaptic activity in many brain areas, teases apart an unsuspected ‘inverse’ synaptic tagging mechanism that likely contributes to maintaining the contrast of synaptic weight between strengthened and weak synapses within an active ensemble.
1. Input-specificity of late phase plasticity: a fascinating long-term biochemical challenge
Clinical case studies and animal lesion experiments have indicated the critical importance of the hippocampus in associative learning and formation of episodic memory . During learning and memory, it has been suggested that the coordinated electrical activity of hippocampal neurons translates information about the external environment into internal neuronal representations . A widely held hypothesis posits that plasticity at the synaptic level is a key feature that critically modulates such representations of the external environment within a circuit . Thus, ever since long-term potentiation (LTP) [3–5] and long-term depression (LTD) [6–8] were discovered, one of the outstanding questions has been to pin down the cellular or the subcellular location of these plastic changes, and to understand how these events govern the formation of ‘engrams’ (or ensembles of active neurons contributing to the representation of a memory event) within various brain areas [9–11].
While engrams were originally thought to be formed at the cellular level, the discovery of the synaptic origin and the input-specificity of LTP and LTD induction called for a cell biological re-examination of the spatio-temporal dynamics of the initial plastic signal processing. How rapidly induced and sustained is the original plasticity signal? Is this localized to the plastic synapse or does the signal diffuse out? What then is the molecular substrate for the persistence of spine-level input-specific changes following plasticity? How does this relate to the original plasticity signal? What is the relationship between a cellular-level engram and synaptic plasticity?
2. Necessity for a crosstalk between synapses and the neuronal soma: a long-distance, two-way communication at the heart of the persistence of local synaptic changes following plasticity and during memory formation
A large number of past works are in keeping with the idea that once synaptic plasticity is induced, the consolidation and the maintenance of such a new plasticity may involve additional molecular processes that necessitate active mRNA and new protein synthesis at the nucleus and the soma of the cell [9–13]. Indeed, plasticity-inducing synaptic activity was also shown to control the expression of many genes encoding synaptic proteins, ion channels, kinases or immediate early genes (IEGs) [3,13–15]. Some of these have been shown to be important for memory formation [16–18]. Behavioural experiments also suggest the critical role of activity-dependent transcription, new protein synthesis and further translational control in synaptic plasticity and memory [19–24]. Thus, it is likely that the proper regulation of the bidirectional signalling between the synapses and the nucleus is essential for the generation and persistence of memory.
Are the changes induced by synaptic plasticity then somehow rendered persistent through a local, synapse-autonomous mechanism? Or alternatively, are robust and input-specific changes in synaptic efficacy accompanied by input-non-selective new transcription and new protein translation, which then together transform an otherwise transient plasticity into a long-lasting and stable one (figure 1, presence of a ‘many-to-one’-type route between many plastic synapses and one neuronal soma responsible for triggering transcription)? How can a cellwide, synapse-unrestricted, mechanism, such as activity-dependent gene expression or protein synthesis, possibly contribute to preserving input-specificity of the persistence of plastic changes (figure 1, presence of a ‘one-to-many’-type route to account for one soma-driven responses contributing to the input-specific maintenance of plasticity at many synapses)?
A hypothesis called ‘synaptic tagging and capture’ has been proposed to provide a tangible framework to understand these ideas [25–28], based on the intriguing observation that a strong plasticity event, through new gene expression and protein synthesis, could render persistent a temporally close weak plasticity event, which otherwise would have remained transient [25,28]. To account for this, it was speculated that a synaptic ‘tag’ which is long-lasting, but not permanent, is created at or near the synapses where synaptic plasticity is also induced. When a strong stimulus (which could be a plasticity-inducing stimulus on its own) is triggered within a limited time window of the original plasticity (such as within 60 min before or after the original plasticity stimulus), this triggers a strong transcriptional and translational response, which is sufficient to result in the subsequent delivery and targeting of new synthesized plasticity-related proteins (PRPs) from the soma towards the synapses where the original plasticity was induced. In this ‘synaptic tagging and capture’ hypothesis, the state of the synaptic tag, via a functional interaction (or ‘capture’) of the PRP, determines the ultimate persistence of the plastic changes [25–28]. However, the molecular basis for such a ‘synaptic tagging and capture’ hypothesis has remained largely elusive to date (figure 1 ‘synaptic capture’ question), although several candidate molecules have already been postulated as either a synaptic tag or a putative PRP .
Better understanding of these issues will certainly necessitate intensive investigation of the cell biological signalling crosstalk between many active synapses and the neuronal soma. Untangling such a long-distance, two-way communication route is of critical significance, as this may lie at the heart of the persistence of local synaptic changes following plasticity and during memory formation [25,26,29]. Additionally, however, we should keep in mind, and not underestimate, the formal (and non-mutually exclusive) possibility that a tiny amount of well-positioned molecular alterations in the plastic synapse may still strongly influence perpetuation of an ongoing local change in synaptic efficacy [30–32].
3. Addressing the ‘many-to-one’ question: defining a CaMKK–CaMKIV–CREB–SARE–Arc pathway critical for the late phase-plasticity and long-term memory signalling
Over the past years, many groups have attempted to systematically investigate the molecular basis for the signalling from synapses to the nucleus that accompanies plastic changes at the synapses. Among many gene expression pathways present in neurons, the activation pathway of the transcription factor CREB is arguably one of the most studied activity-dependent synapses-to-nucleus signalling mechanisms [9,11–15,33–35].
Our earlier study thus uncovered an activity-dependent protein kinase cascade CaMKK–CaMKIV that critically controls the amplitude and time course of CREB phosphorylation downstream of synaptic activity (figure 2a) [36–38]. Consistently, α/δ-CREB-null mice , CaMKIV-null mice  and knockout mice for either CaMKK-α [40,41] or CaMKK–β  all showed specific defects in long-term memory. Alternative routes of CREB activation include a cAMP–PKA pathway, and a Ras–MEK–Erk pathway [12,13,34], and the molecular dissection of these differential activation routes within the brain at the circuit level is still eagerly awaited.
A large amount of effort has also been spent on identifying an exhaustive list of all the putative target genes for CREB [43–45]. It remains to be clarified, however, how many of these will be critical for the late phase of long-term plasticity and long-term memory. One likely target molecule is brain-derived neurotrophic factor (BDNF), a secreted peptide growth factor, and the role of CREB in activity-dependent stimulation of several alternative promoters for the gene encoding the neurotrophic factor BDNF has indeed been thoroughly investigated . BDNF's role in synaptic tagging and capture has also been shown [47–49].
Another possible candidate is the IEG Arc/Arg3.1 [50,51], a gene whose expression is widely used as a biological marker to map spatial representation of active neurons in fixed brain samples [52,53], and whose deletion has led to loss of long-term memory in mice . By carefully examining the promoter and the distal regulatory elements critical for activity-dependent neuronal expression, we were indeed able to identify a CRE half-site that functioned as a genuine CREB-regulated locus, within the synaptic activity-responsive element (SARE), a distal enhancer region located at about 7-kb upstream of the Arc's transcription start site [54,55]. Strikingly, the SARE of the Arc gene consisted of a unique cluster of binding sites for CREB, MEF2 and SRF/TCF, each of which cooperatively contributed to converting synaptic inputs into a transcriptional output (figure 2b). Multiplexing SARE and fusing this to the minimal promoter of the Arc gene has enabled us to create a synthetic promoter, which we named enhanced SARE (E-SARE). This artificial promoter was about 30 times more potent than the c-fos promoter and is expected to serve as a useful means to map and record from activity-regulated neurons and circuits in various areas of the brain in vivo . Future studies will reveal whether neurons in which synaptic activity-induced CREB activation and Arc expression are enhanced truly represent part of a functional ensemble of active neurons within a memory circuit, as suggested by the IEG mapping analyses [52,53].
4. Synaptic tagging and inverse synaptic tagging: a putative two-bit tagging code for securing the persistence of an input-specific memory engram?
Despite the excitement about IEG mapping results which were consistent with the idea that Arc may be one of the memory trace proteins critical for memory formation, the molecular function of Arc has remained, however, enigmatic. Indeed, several studies showed that this putative memory trace-coding protein Arc, despite being strongly upregulated by synaptic activity that induced persistent forms of plasticity and learning [15,52,53], also critically contributed to weakening synapses by promoting AMPA receptor endocytosis during various forms of synaptic plasticity [18,57–59].
To address this incongruence, we directly imaged plasticity-induced Arc trafficking from the soma to the dendrites and back to the synapses . Contrary to expectations that Arc may be recruited into the potentiated synapses through an orthodox synaptic tagging and capture mechanism, we instead found a preferred targeting of Arc to inactive synapses (figure 3). This unexpected result was mediated via Arc's high affinity interaction with an inactive, CaM-unbound form of Ca2+/calmodulin-dependent protein kinase (CaMK)IIβ . Consistently, the degree of synaptic Arc accumulation was more sustained during a period of inactivity following strong induction, and in fact correlated with removal of surface GluA1 from individual synapses. A lack of CaMKIIβ either in vitro or in vivo resulted in loss of Arc upregulation in the silenced synapses . These findings provide compelling molecular evidence for an ‘inverse’ synaptic tagging mechanism that enables Arc to specifically target the unpotentiated synapses that contains more inactive CaMKIIβ. Arc targeting to inactive synapses will promote the clearance of surface AMPA receptors at the inactive synapses, and thereby help maintain the contrast of synaptic weight between strengthened and weak synapses (figure 3) [60,61]. At the circuit and systems levels, this may subserve memory consolidation by preventing undesired synaptic enhancement at weak synapses, while sparing potentiated synapses.
For inactive CaMKIIβ to be able to fulfil the role of an inverse synaptic tag vis à vis of induced Arc protein, any synaptic activity that is below the threshold of LTP induction should in principle do little to perturb CaMKII inactivity. Directly addressing this question has become possible very recently, through novel techniques that enable direct measurement of enzymatic activities in situ. Using fluorescence resonance energy transfer (FRET) imaging and fluorescence lifetime microscopy measurements, direct measurements of the enzymatic activity of CaMKIIα with single spine resolution demonstrated [62,63] and confirmed some earlier suggestions  that the activation kinetics of CaMKIIα, a critical molecular switch involved in the induction of LTP [4,5,65,66], was very fast (within seconds) and input-specific (figure 4 top panels). Inconsistent with some predictions [30, see also 67], however, the deactivation of CaMKIIα after cessation of LTP induction was also fast, within a minute (figure 4) [62,63]. The same conditions of stimuli that triggered sustained morphological plasticity in a high-frequency- and input-dependent manner also triggered activation of CaMKIIα, but the former far outlasted the latter . Thus, CaMKIIα genuinely functioned as a synaptic sensor for high-frequency input (figure 4 top panels), and was an enzyme that decodes both input frequency and numbers , but its activity was short-lived and did not appear to encode the plasticity induction per se. As an independent control for an alternative synaptic Ca2+ effector, calcineurin was shown to be activated at a much lower frequency and input numbers than CaMKIIα (figure 4 lower panels) . While direct measurements of CaMKIIβ are eagerly awaited as a next step, this recent evidence lends support to the idea that CaMKII activation may indeed gate the role of Arc in inverse synaptic tagging. A role for CaMKII has also been suggested in the original synaptic tagging [27,28].
Evidently, much work lies ahead to definitively establish the mechanisms and physiological significance of synaptic tagging  and inverse synaptic tagging . Perhaps, simultaneous onset of both tagging mechanisms, in conjunction with the induction of LTP per se, may permit synapses to gain the ability to use a robust, two-bit tagging code for securing the persistence of an input-specific memory engram (figure 3). Further investigation on both synaptic tagging and inverse synaptic tagging will undoubtedly shed more light on the fundamental role of new gene expression and of the guided targeting of new protein products to synapses as a molecular basis for memory allocation within an activated neuronal network .
This work was supported in part by grants-in-aid from the Ministry of Education, Culture, Sports, Science and Technology of Japan and the Japan Society for Promotion of Science (M.N., H.O. and H.B.), awards from the Takeda Foundation (H.B.) and the Tokyo Society of Medical Sciences (H.F. and H.O.), a SICPMe JST-CONACyT collaborative grant (H.O. and H.B.) and a CREST investigatorship (H.B.).
We are grateful to Stuart Sharry for critical comments. We apologize to the many authors whom we could not cite due to space limitations.
One contribution of 35 to a Discussion Meeting Issue ‘Synaptic plasticity in health and disease’.
- © 2013 The Author(s) Published by the Royal Society. All rights reserved.