Learning Induced Synaptogenesis and Structural Synaptic Remodeling

Yuri Geinisman, Robert W. Berry and Olga T. Ganeshina Abstract

This chapter analyzes the results of quantitative electron microscopic studies of the vertebrate brain aimed at the elucidation of changes in synaptic unltrastructure that may underlie learning and memory. It has been reported that behavioral learning promotes new synapse formation, including both a net synaptogenesis, which causes a net gain in synapse number, and a specific synaptogenesis. The latter either accompanies a learning-related adult neurogenesis or leads to the formation of multiple-synapse boutons. Other data strongly suggest that behavioral learning also elicits a structural remodeling of existing synapses. This process is revealed morphologically as an increase in the number of perforated axospinous synapses and an enlargement of postsynaptic densities. Further research will show if the observed diversity of structural synaptic alterations reflects differences among various forms of learning and memory or among various consecutive processes underlying the formation, consolidation and long-term storage of each memory trace.

Introduction

Over a century ago, Ramón y Cajal99 and Tanzi114 postulated that changes in the number and structure of synaptic connections might underlie the establishment of long-term memories following behavioral learning. Since that time, cellular mechanisms of learning and memory have been thought to include the formation of new synapses and/or a structural remodeling of existing synapses. Numerous attempts have been made to verify this assumption by defining those modifications of synaptic ultrastructure in the vertebrate brain that result from the learning of new behaviors (for reviews see refs. 4, 6, 34 and 48). The review of the literature presented here examines a growing body of evidence for learning-induced synaptogenesis and synapse remodeling. The data reported so far are also related to the phenomena of adult neurogenesis, spine motility and transformation of postsynaptically silent synapses into functional ones. Finally, the role of these phenomena in shaping the patterns oflearning-induced alterations in synaptic ultrastructure is discussed.

Patterns of Synaptogenesis Elicited by Behavioral Learning

Net Synaptogenesis Resulting from Learning

The search for the net synaptogenesis that may underlie the learning of new behaviors involves determining whether the process of learning is accompanied by a net gain in synapse number. Initial attempts to resolve this issue were based on the use of light microscopic preparations stained according to the Golgi method for counting dendritic spines, which usually synapse with axon terminals. However, since the Golgi method stochastically stains only a small fraction of neurons, subsequent work over the past two decades has focused on the examination of electron microscopic preparations that reveal all the synapses present in a given tissue specimen. Samples of entire synaptic populations, including all morphological synaptic types, were taken from pertinent regions of the vertebrate brain and used for obtaining estimates of synapse number. Animals that had learned a given behavioral task were compared on this measure with controls that had no such learning experience. These electron microscopic studies, however, produced inconsistent results (Table 1).

Several studies found that behavioral learning promotes a net gain in synapse number (Table 1). Additionally, quantitative electron microscopic analyses of dendritic spines showed that their total number was increased in the molecular layer of the rat dentate gyrus as a consequence of either passive avoidance conditioning91 or water maze training.92 Similar results were obtained with the aid of confocal microscopy: an increase in the number of spines per unit length of basal, but not apical, dendrites of CA1 pyramidal cells was detected following spatial learning of rats in a complex environment.82,83 Taken together, these data are consistent with the notion that the process of learning is accompanied by a net synaptogenesis.4,6,48 At variance with such a notion is the observation of other studies that the learning of new behaviors does not cause a net change in synapse number (Table 1).

A number of factors might have contributed to the discrepancy in the results. One of these is that appropriate methods for synapse quantification were not available until recently, and most of the ultrastructural studies referred to in Table 1 used methodologically inadequate procedures. These included the identification of synapses in single (rather than in serial) ultrathin sections, sampling in selective (rather than in systematic, uniformly random) fashion, counting with two-dimensional (rather than with three-dimensional) probes, and estimating the numerical density of synapses (rather than their total number). Each of these procedures involves biases.40,41 The direction and magnitude of the biases were not evaluated, which makes it difficult to interpret the data of the earlier electron microscopic studies. To circumvent these problems, a stereological method was designed for providing unbiased estimates of total synapse number in defined brain regions.41

This method was then used to explore the problem of whether the total number of synapses in the stratum radiatum of hippocampal subfield CA1 is altered by trace eyeblink conditioning.40 The latter is a hippocampus-dependent form of associative learning67,84,107 that is accompanied by increases in the synaptic responsiveness40,96 and postsynaptic excitability21,85 of CA1 pyramidal neurons. For conditioning, rabbits were given daily 80-trial sessions to a criterion of 80% conditioned responses in a session. During each trial, the conditioned stimulus (tone) and the unconditioned stimulus (corneal airpuff) were presented with an intervening trace interval of 500 msec. Brain tissue was taken for morphological analyses 24 hours after the last session. The results showed that the total number of synapses in the CA1 stratum radiatum was not changed in conditioned rabbits as compared to pseudoconditioned controls (Fig. 1A). No trend towards a conditioning-induced increase in total synapse number was observed (Fig. 1A), and the group means (± SEM) for pseudoconditioned and conditioned animals (23,267 ± 915 and 23,250 ± 869 synapses x 106, respectively) differed by only 0.07%. These data provide convincing evidence for the stability of total synapse number in the CA1 stratum radiatum 24 hours after acquisition of the trace eyeblink conditioned response.

If the discrepancy in the results of previous studies is due to their methodological limitations, it is reasonable to expect that the use of modern stereological methodology for synapse quantification following learning would yield consistent data. This, however, was not the case. Conflicting results were also obtained with the aid of the unbiased disector technique. Rusakov et al101 found that spatial learning in the Morris water maze had no effect on the numerical density of synapses per unit volume of the rat CA1 stratum radiatum and dentate gyrus molecular layer. This finding is in agreement with our observation (Fig. 1A). In contrast, the work from the Greenough laboratory demonstrated that the learning of complex motor

Table 1. Results of synapse quantification based on analyses of samples taken from the entire synaptic population of a given brain area following behavioral learning

Behavioral

Species and Brain

Parameter

Paradigm Used

Area Examined

Analyzed

Result

Reference

Brightness discrimination

Rat hippocampus

Na

Increase

Wenzel et al, 1980

conditioning in a Y-maze

(subfield CA1)

Acquisition of male-like

Canary robustus

N

Increase

DeVoogt et al,1985

singing behavior by

archistralis nucleus

females treated with

testosterone

Passive avoidance

Chick hyperstriatum

Nv

Increase

Stewart et al,1987;

conditioning

ventrale and lobus parolfactorius

Hunter & Stewart, 1989; Doubell & Stewart, 1993

Complex motor skill

Rat cerebellar and

Nn

Increase

Black et al, 1990;

acquisition

motor cortices

Kleim et al, 1996; 1997

Visual imprinting

Chick hyperstriatum ventrale

Na, Nv

No change

Bradley et al, 1981; Horn et al, 1985

Visual discrimination

Rabbit visual cortex

Nv

No change

Vrensen & Nunes

conditioning

Cardozo, 1981

One-way active

Rat dentate gyrus

Na

No change

Van Reempts et al,

avoidance conditioning

1992

Spatial learning in the

Rat hippocampus

nv

No change

Rusakov et al, 1997

Morris water maze

(subfield CA1) and dentate gyrus

Trace eyeblink

Rabbit hippocampus

n

No change

Geinisman et al,

conditioning

(subfield CAD

2000

Designations: Na - synaptic numerical density per unit tissue area; Nv - synaptic numerical density per unit tissue volume; Nn - number of synapses per postsynaptic neuron; N - total synapse number.

skills increased the number of synapses per neuron in the rat motor and cerebellar cortex.9'69'70 It appears, therefore, that the methodological limitations alone cannot account for the diversity of the data reported so far.

Another factor that may be of special importance in this respect is the transient nature of a learning-induced net gain in synapse number. In certain brain regions, such a change persists only for a limited period of time following behavioral training. This was clearly demonstrated in experiments involving passive avoidance conditioning of day-old chicks (for a review see ref. 112). Acquisition of the conditioned avoidance response was followed by an increase in the overall numerical density of axospinous synapses that was observed in the intermediate and medial hyperstriatum ventrale (IMHV) only at 1 hour, but not at 24 hours, after training.22 On the other hand, the same change in the lobus parolfactorius (LPO) was detected 24 and 48 hours, but not 1 or 6 hours, post-training.56 The different temporal sequence of alterations in the synaptic numerical density observed in the IMHV and LPO may be explained by differences in the involvement of these brain regions in task acquisition and consolidation. Lesion studies have indicated that the IMHV may be the site of initial registration of the memory trace, which is subsequently transferred to the LPO for long-term storage.43 Similar observations were made more recently in experiments examining the effect of passive avoidance conditioning91 or spatial learning92 on the numerical density of dendritic spines in

Pseudoconditionod ^ Conditioned

Pseudoconditionod ^ Conditioned

All synapses Axospinous synapses

Figure 1. Total number of all synaptic contacts (A) and of axospinous synapses (B) in the CA1 stratum radiatum of pseudoconditioned and conditioned rabbits examined 24 hours after cessation of training (data are from ref. 34). Bars show group means ± SEM.

All synapses Axospinous synapses

Figure 1. Total number of all synaptic contacts (A) and of axospinous synapses (B) in the CA1 stratum radiatum of pseudoconditioned and conditioned rabbits examined 24 hours after cessation of training (data are from ref. 34). Bars show group means ± SEM.

the molecular layer of the hippocampal dentate gyrus of adult rats. Following the cessation of training on either task, spine number was shown to increase at 6 hours and then return to control levels by 72 hours. The latter temporal pattern of ultrastructural changes is consistent with the concept that the hippocampal formation plays only a transitory role in the consolidation of memory (e.g., ref. 109). In the cerebellar cortex of adult rats, on the other hand, an increase in the number of synapses per Purkinje cell persists for at least 4 weeks after motor skill learning.70 This suggests that the cerebellar cortex may be involved in a long-term retention of memories for motor skills.

The data described above underscore the necessity of examining various phases of the acquisition/consolidation process in order to reach a definitive conclusion of whether a net increase in synapse number is characteristic of a given form of behavioral learning. In most studies available in the literature, including the one from this laboratory,40 synapses were quantified only at a single time point relative to behavioral acquisition. If quantitative ultrastructural analyses were performed at various time points along the acquisition/consolidation curve, the presence of additional synaptic contacts might have been detected as well. It is also conceivable, however, that some forms of learning may not involve a net gain in synapse number and, hence, a net synaptogenesis. Work from the laboratories of Rakic and Goldman-Rakic has provided evidence in favor of this supposition (for a review see ref. 11). Their estimates of synaptic numerical density in five major areas of the primate cerebral cortex reveal no ultrastructural sign of net synaptogenesis over the entire period of adulthood in spite of presumably continuous accumulation of long-term memories.

Even if the acquisition and retention of a given behavior does not result in a detectable net synaptogenesis, it is possible that new synapses may be nevertheless formed as a consequence of behavioral learning. In such cases, learning-induced synaptogenesis might be confined to re-arranging only a specific subset of synaptic connections in order to establish a memory trace. Examples of such specific synaptogenesis will be considered next.

Specific Synaptogenesis Related to Learning-Induced Adult Neurogenesis

It has been established that the production of new neurons in the vertebrate brain continues throughout adulthood, and there are observations suggesting that it may occur not only spontaneously but also as a consequence of behavioral learning (for reviews see refs. 47, 50, 75 and 102). These observations are especially relevant to the present discussion because to become functional, adult-born neurons have to form appropriate synaptic connections. This implies that a specific synaptogenesis involving newly generated neurons should accompany the process of adult neurogenesis.

Evidence for the existence of neurogenesis in the adult brain comes from studies employing the incorporation of tritiated thymidine, as well as from immunocytochemical studies using markers for the detection of proliferating cell progenies and their neuronal phenotype. Although these methodological approaches have serious limitations that make the interpretation of the results difficult (for reviews see refs. 44 and 98), there is a general consensus that new neurons are spontaneously added to several regions of the adult mammalian brain, most notably to the dentate gyrus and olfactory bulb. Adult neurogenesis has been most extensively studied in the dentate gyrus of the hippocampal formation, and the latter is known to mediate certain forms of learning and memory. Therefore, the following discussion is focused on data from the dentate gyrus.

Principal neurons of the dentate gyrus (granule cells) are generated in all adult mammalian species studied thus far, including rodents (e.g., see ref. 1, 5, 8 and 62), nonhuman primates46,73 and humans.18,28 New granule cells arise from their precursors residing in the subgranular zone of the dentate gyrus and migrate into the granule cell layer where they assume the morphological features characteristic of neighboring neurons.14,61,62 The majority of the new granule cells die within two weeks of their birth.14 However, adult-generated granule cells do display synapses,61,62 receive functional synaptic inputs similar to those found in mature granule cells118 and rapidly extend their axons through the mossy fiber tract to their natural target area, hip-pocampal subfield CA3.52,81,110 This suggests that neurons newly generated in the adult dentate gyrus may become, at least temporarily, integral components of neural circuits in the hippocampal formation and participate in its functions related to learning and memory.45,50,63

Several experimental approaches have been used to test the validity of this suggestion. A number of studies have addressed the question of whether environmental complexity or aging alter adult hippocampal neurogenesis. Exposure of adult birds and rodents to enriched environments, which presumably offer more opportunities for learning than standard laboratory environments, increases the number of new hippocampal neurons. This does not affect the proliferation of progenitor cells, but rather promotes the survival of their progeny.7,64,65, 9 The enhanced survival of new granule cells in mice kept in an enriched environment is associated with improved spatial learning in the Morris water maze.64,65,89 Conversely, the process of normal aging, which produces deficits in hippocampus-dependent forms oflearning and memory (for a review see ref. 38), is accompanied by a dramatic reduction in the number of granule cells that are born in the dentate gyrus of aged mice and rats.13,74,104

Another approach was employed in experiments designed to determine if adult neurogenesis is modulated by specific learning experiences. An indication that behavioral learning may upregulate adult neurogenesis was initially provided by studies showing that the seasonal modification of song in birds coincides with the increased addition of new neurons to a forebrain vocal center involved in song learning.2,68 A subsequent study of birds demonstrated that spatial learning associated with the first few exposures to the experience of storing and retrieving food augments the rate of neuronal recruitment into the hippocampus and hyperstriatum ventrale.93 More recently, it has been found that learning the trace eyeblink conditioned response or the location of an invisible escape platform in the Morris water maze promotes the survival of granule cells born in the dentate gyrus of adult rats 1-2 weeks prior to training.45 Additionally, the spatial learning ability of rats in the Morris water maze containing an invisible platform correlates with the extent of survival of adult-generated granule cells.3 Unlike these hippocampus-dependent learning experiences, hippocampus-independent experiences, such as learning the delayed eyeblink conditioned response or the location of a visible and cued platform in the Morris water maze, have no effect on the number of new granule cells.45 It should be noted here that functional integration of adult-born neurons into existing or newly developing circuitry occurs on a time-scale of several days. Thus an increased production of nerve cells resulting from an accelerated rate of adult neurogenesis would not have immediate functional consequences while a prolonged survival of functionally competent new neurons would.45

The results summarized above have been extended by the finding that the targeted destruction of the majority of newborn granule cells in adult rats by a toxin for proliferating cells (methylazoxymethanol acetate) impairs hippocampus-dependent trace eyeblink conditioning but does not affect delayed eyeblink conditioning, the latter ofwhich is hippocampus-independent.106 These observations suggest that adult-generated granule cells may play a role in the formation of hippocampus-dependent memories (see ref. 117). The hippocampal formation is considered to be especially important for the acquisition of associations between temporally or spatially discontiguous events121 and for a transient storage of recently acquired memories.109 It is possible that new granule cells, in spite of their limited lifespan, may support such major functions of the hippocampal formation. Since pyramidal cells can be newly generated in the adult hippocampus proper,100 they may also mediate these hippocampal functions.

Newborn neurons in the adult brain may be much more modifiable and readily involved in synaptogenesis than older neurons. In order to survive and appropriately function, the adult-generated neurons that subserve learning and memory must make synaptic connections. Although such specific synaptogenesis associated with learning-induced adult neurogenesis may take place, it is difficult to document it. Markers for newly established synaptic contacts are not yet available for ultrastructural studies, which makes it necessary to demonstrate that learning-induced neurogenesis in a given brain region is accompanied by an increase in total synapse number. This alteration is not likely to be detected because synapses involving adult-born nerve cells constitute only a minute fraction of all synaptic contacts in a brain region. Quantification of the entire synaptic population is not necessary, however, for the detection of the other known form of specific synaptogenesis. This involves the formation of multiple synapses by single axonal boutons, which can be readily identified and selectively quantified as is described below.

Specific Synaptogenesis Related to Learning-Induced Formation of Multiple-Synapse Boutons

The term "multiple-synapse boutons" (MSBs) refers to those presynaptic axon terminals (boutons) that establish separate synaptic appositions with two or more postsynaptic elements instead of only one synaptic apposition with a single postsynaptic element. Earlier studies have reported that acquisition of complex motor skills increases the incidence of MSBs29,59 as well as the overall number of synapses9,69,70 in rat cerebellar and motor cortices. These data indicate that motor skill learning elicits both a specific synaptogenesis selectively producing additional MSBs and a net synaptogenesis. In light of the above observations, we explored the possibility that hippocampus-dependent associative learning, which does not alter total synapse number in the CA1 stratum radiatum40 and hence does not involve a net synaptogenesis, nevertheless induces a specific synaptogenesis leading to the formation of MSBs, at least at the studied time point after training.

Electron micrographs obtained in our earlier study40 were reanalyzed as described in detail elsewhere.35 Trace eyeblink conditioned and pseudoconditioned rabbits were compared, the hippocampi being taken for morphological analyses 24 hours after the last training session. Unbiased stereological methods were used for obtaining estimates of the total number of MSBs in the CA1 stratum radiatum. Inspection of electron micrographs revealed that a typical MSB in this layer is a single presynaptic bouton that forms separate synapses with two spine heads (Fig. 2). Such MSBs can be unequivocally identified only in serial sections because in single

Figure 2. Electron micrographs of consecutive ultrathin sections (a-f) through the rabbit CA1 stratum radiatum demonstrating a typical multiple-synapse bouton (MSB). The bouton makes two synapses (black and white arrows), each one involving a separate dendritic spine (S1 or S2). Scale bar — 0.5 |tm.

Figure 2. Electron micrographs of consecutive ultrathin sections (a-f) through the rabbit CA1 stratum radiatum demonstrating a typical multiple-synapse bouton (MSB). The bouton makes two synapses (black and white arrows), each one involving a separate dendritic spine (S1 or S2). Scale bar — 0.5 |tm.

synapses constitute only about 2% of the entire synaptic population of the rabbit CA1 stratum radiatum.40 Therefore, only those MSBs that formed synapses exclusively with spines were quantified. The results showed that the mean total number of MSBs in the CA1 stratum radiatum was significantly increased in the group of conditioned rabbits (1,236 ± 43 x 106) as compared with the pseudoconditioned group (1,047 ± 28 x 106). Conditioned rabbits also had significantly more MSBs relative to untrained controls (1,077 ± 52 x 106), while the two control groups did not differ significantly from each other with respect to total MSB number.

Although MSBs in the rabbit CA1 stratum radiatum usually form synapses with two spines, some MSBs synapse with three or four spines. Therefore, we addressed the question of whether trace eyeblink conditioning alters the number of axospinous synapses per MSB and found that the mean numbers of axospinous synapses per MSB were the same (2.05) for the pseudoconditioned and conditioned groups. Another characteristic of MSBs is that they form both perforated synapses,16,94 which exhibit a discontinuous profile of the postsynaptic density in at least one serial section, and nonperforated synapses that show a continuous postsynaptic density profile in all consecutive sections (Fig. 3). The perforated subtype has been implicated in synaptic plasticity associated with behavioral learning and hippocampal LTP (for reviews see refs. 34, 58 and 86). Our estimates of perforated synapse number per MSB showed, however, that the groups of pseudoconditioned and conditioned animals did not differ significantly on this measure.

The major finding of the cited study regarding an increase in total MSB number following trace eyeblink conditioning is in accord with those of earlier reports that motor skill learning induces the addition of MSBs.29,59 Taken together, these results suggest that various forms of learning may promote MSB formation. Interestingly, this kind of morphological alteration is not unique to behavioral learning and to related phenomena such as hippocampal LTP115 or an exposure to enriched environments.60 Rather, the incidence of MSBs has been reported to increase as a consequence of various experimental manipulations that induce plasticity (for a review see ref. 35), indicating that the formation of MSBs may represent a general form of structural synaptic plasticity.

Individual MSBs in the CA1 stratum radiatum can synapse with spines arising from the same or different dendrites.108,123 The LTP-induced increase in the proportion of activated boutons synapsing with two or more spines is essentially due to the formation of those MSBs that synapse with spines originating from the same dendrite.115 However, we have been unable to reliably trace many multiple spines to their dendritic origins and to obtain representative samples for quantitative analyses. It is not known, therefore, whether the MSBs that are newly formed as a result of trace eyeblink conditioning make synapses with spines arising from the same dendrite. If this is the case, the strength of the conditioned synaptic input to target CA1 neurons may be amplified. If, however, the multiple postsynaptic spines synapsing with additional MSBs emanate from dendrites of different neighboring neurons, this may contribute to a synchronous activation of the latter and hence to the assembly of functional multineuronal units tuned to the synaptic input activated by conditioning stimulation. In either case, the effect of conditioning stimulation would be facilitated.

The results of our recent study and those reported in the literature suggest three models of structural plasticity that may underlie MSB formation after trace eyeblink conditioning (Fig. 4). For the purpose of simplicity, the following assumptions were incorporated into each model: 1) existing single-synapse boutons are transformed into MSBs by the conditioning and 2) the spines that are postsynaptic to newly formed MSBs originate from the same dendrite. The models also take into account our observation that trace eyeblink conditioning does not alter the total number of axospinous synapses in the CA1 stratum radiatum at the studied time point (24 hours) after acquisition of the conditioned response (Fig. 1B). This parameter would have been increased by about 2% if new axospinous synaptic contacts were added in the process of MSB formation. However, no trend towards such an increase was detected (Fig. 1B), suggesting that MSB formation in the conditioned animals may not be due to the recruitment of new axospinous synapses.

Figure 3. Electron micrographs of consecutive ultrathin sections (a-d) through the rabbit CA1 stratum radiatum demonstrating a multiple-synapse bouton (MSB) forming a perforated synapse (black arrows) with a large spine (S1) and a nonperforated synapse (white arrows) with a small spine (S2). The PSD of the large spine exhibits perforations (arrowheads) in some serial sections (a, b) whereas the PSD of the small spine shows no perforation in all sections. Scale bar — 0.5 |im.

Figure 3. Electron micrographs of consecutive ultrathin sections (a-d) through the rabbit CA1 stratum radiatum demonstrating a multiple-synapse bouton (MSB) forming a perforated synapse (black arrows) with a large spine (S1) and a nonperforated synapse (white arrows) with a small spine (S2). The PSD of the large spine exhibits perforations (arrowheads) in some serial sections (a, b) whereas the PSD of the small spine shows no perforation in all sections. Scale bar — 0.5 |im.

Figure 4. Models of conditioning-induced MSB formation that presumably involves spine motility and does not increase synapse number. A) Relocation of existing spines (broken lines) from non-activated boutons for specific synaptogenesis with boutons activated by conditioning stimulation. B) Emergence of new spines (asterisk) and their outgrowth for specific synaptogenesis with activated boutons, coupled with the resorption of spines (broken lines) postsynaptic to non-activated boutons. C) Splitting of spines with completely partitioned segmented PSDs (arrowhead) that produces double or multiple spines establishing synaptic contacts with single activated boutons. This is accompanied by the retraction of spines (broken lines) from non-activated boutons into parent dendrites.

Figure 4. Models of conditioning-induced MSB formation that presumably involves spine motility and does not increase synapse number. A) Relocation of existing spines (broken lines) from non-activated boutons for specific synaptogenesis with boutons activated by conditioning stimulation. B) Emergence of new spines (asterisk) and their outgrowth for specific synaptogenesis with activated boutons, coupled with the resorption of spines (broken lines) postsynaptic to non-activated boutons. C) Splitting of spines with completely partitioned segmented PSDs (arrowhead) that produces double or multiple spines establishing synaptic contacts with single activated boutons. This is accompanied by the retraction of spines (broken lines) from non-activated boutons into parent dendrites.

Additionally, these models incorporate the recently discovered phenomenon of spine motility (for reviews see refs. 51, 80, 103 and 122). This phenomenon was established by labeling CA1 pyramidal neurons in cultured hippocampal slices with vital fluorescent markers and time-lapse two-photon imaging of their spines. Such experiments demonstrated that spines are highly dynamic structures constantly undergoing formation and resorption under normal conditions and that the process of new spine formation is markedly augmented by local high-frequency stimulation of dendrites, a manipulation which elicits LTP.2 The ability of spines to rapidly elongate or retract is especially prominent during early postnatal development, but is retained to a certain degree after the maturation of CA1 pyramidal neurons in slices obtained from developing animals and maintained in culture.17,2 ,31,72 Moreover, spines exhibiting larger and probably more mature synaptic contacts are no less motile than those forming smaller synapses.111 A concordant movement of activated boutons and their postsynaptic spines23 may also contribute to the spatial alignment of pre-and postsynaptic elements during MSB formation.

Accordingly, the first model (Fig. 4A) posits that, following trace eyeblink conditioning, some postsynaptic spines contacting non-activated boutons leave their presynaptic partners, relocate to boutons activated by conditioning stimulation and synapse with them. The second model (Fig. 4B) postulates that conditioning stimulation may induce the emergence of new spines and their outgrowth for a specific synaptogenesis with activated single-synapse boutons, probably in response to a signal emitted by the boutons. This model encompasses the retraction of some postsynaptic spines from non-activated boutons into parent dendrites, a process that keeps the total number of axospinous synapses constant. Finally, the third model (Fig. 4C) proposes that a specific synaptogenesis producing MSBs involves the splitting of large spines, which exhibit a PSD consisting of multiple segments separated from each other by complete spine partitions. The process of spine retraction maintaining the constancy of axospinous synapse number is also incorporated into the third model. This model was suggested by the findings of Toni et al116 who examined the effect of hippocampal LTP on synaptic ultrastructure and observed a temporal coincidence between the disassembly of synapses involving such spines and the addition of MSBs that synapse with double spines arising from the same dendrite. However, the spatial arrangement of spines originating from the same dendrite and receiving synapses from the same MSB indicates that MSBs of this kind are unlikely to be formed by spine splitting.30

Further studies are needed to establish which of the three models is valid. In any event, our data described above demonstrate that trace eyeblink conditioning elicits specific synaptogenesis resulting in the formation of MSBs. Although the latter change does not require a net synaptogenesis to take place, it may facilitate the effect of conditioning stimulation.

Pattern of Structural Synaptic Remodeling Elicited by Behavioral Learning

Increase in the Number of Perforated Axospinous Synapses following Learning: A Possible Morphological Correlate of the Conversion of Synapses into More Efficacious Subtypes

An increase in the proportion or number of perforated axospinous synapses is perhaps the most notable and consistent change among activity-dependent alterations in synaptic ultrastructure (for a review see refs. 34, 58 and 86). There are reports in the literature that behavioral learning is also associated with such structural synaptic modification. The first indication that this may be the case came from the study of Greenough et al49 showing that rats reared in a complex environment have more perforated synapses in the visual cortex than their counterparts kept in isolated conditions. The same effect was later found to be characteristic of visual discrimination conditioning.120 Additionally, a significant correlation between the spatial learning ability of rats tested in an 8-arm maze and the number of perforated synapses per postsynaptic neuron was observed in the molecular layer of the dentate gyrus.36 The numerical density of perforated synapses involving concave spines in the same synaptic layer was also estimated to increase following active avoidance conditioning.119 It has been established that large perforated PSDs contain more AMPA receptors than do small nonperforated PSDs. 9 This finding implies that perforated synapses are more efficacious than nonperforated ones, and an augmentation of synaptic efficacy is believed to be essential for learning.54'71'114

A recent study from this laboratory demonstrated, however, that the total number of perforated axospinous synapses and their various morphological subtypes remained stable in the rabbit CA1 stratum radiatum 24 hours after trace eyeblink conditioning.40 A plausible explanation for this negative finding is provided by data strongly suggesting that the addition of perforated synapses is an early and transitory event associated with the induction of the NMDA receptor-dependent form of hippoca is widely regarded as a synaptic model of memory.10 For example, the proportion of perforated synapses in the rat CA1 stratum radiatum increases at 30 min but returns to the control level at 60 min after potentiating stimulation of Schaffer collaterals in cultured hippocampal slices.12,115 It is possible, therefore, that an increase in the number of perforated synapses was not observed by us 24 hours after trace eyeblink conditioning because it occurred at an earlier time point.

In any event, the formation of perforated axospinous synapses is a rapid process that is completed within 15 min or contrast, the assembly of new excitatory hippocampal synapses takes 1-2 hours following an initial contact of pre- and postsynaptic elements.32 Comparison of the time frames of the two processes indicates that a rapid formation of perforated synaptic contacts results from a structural remodeling of existing synapses and not from synaptogenesis. In discussing such a remodeling, which may also occur as a result of behavioral learning, it is necessary to note that perforated axospinous synapses are morphologically heterogeneous and may be subdivided into several distinct subtypes (reviewed in ref. 33). This suggests that synaptic plasticity might be mediated by the conversion of some synaptic subtypes into others.15,25,88

Of special importance in this regard are the data demonstrating that an LTP-related increase in the number of perforated synapses is essentially due to the addition of their particular subtype distinguished by the presence of multiple, completely partitioned transmission zones.37,116 In synapses belonging to this subtype (Fig. 5E), complete spine partitions provide barriers between two to four discrete transmission zones, each one being formed presynaptically by a separate axon terminal protrusion and delineated postsynaptically by a separate PSD segment.33 It has been postulated that these synaptic contacts evolve from existing synapses.33,34 The process is proposed to commence with an enlargement of small nonperforated synapses (Fig. 5A) and their conversion into atypically large ones (Fig. 5B). This is followed by the consecutive formation of perforated synapses that have initially a focal spine partition with a fenestrated PSD (Fig. 5C), then a sectional partition with a horseshoe-shaped PSD (Fig. 5D) and finally a complete partition(s) with a segmented PSD (Fig. 5E).

The structural features of the latter synaptic subtype suggest that it may be especially efficacious. Multiple transmission zones may function as independent units, provided that there is a barrier in the synaptic cleft preventing the diffusion of neurotransmitter and that each PSD segment is associated with an activated or newly inserted receptor cluster.26,33 Under these conditions, an amplification of impulse transmission would be expected to take place. Mathematical modeling has also shown that the formation of multiple, completely partitioned transmission zones may facilitate synaptic transmission by altering calcium diffusion within the presynaptic bouton and enhancing thereby the probability of release.42

As was mentioned before, the number of axospinous synapses with multiple, completely partitioned transmission zones increases soon after LTP induction and then returns to the control level.39,116 The reversal of the initial morphological change may reflect the transformation of such synapses into other synaptic subtypes. Three possibilities, which are not mutually exclusive,

Figure 5. Diagram illustrating a hypothetical synapse restructuring that may underlie activation-dependent alterations in synaptic efficacy as explained in the text. The schematic shows the following synaptic subtypes: (A) typical (small) and (B, L) atypical (large) nonperforated axospinous synapses; perforated axospinous synapses that have (C) a focal spine partition and fenestrated PSD, (D) a sectional partition and horseshoe-shaped PSD, or (E) a complete partition(s) and segmented PSD; (F) an axospinous perforated synapse involving the postsynaptic spine that is partially retracted into a parent dendrite; (G) an asymmetrical axodendritic synapse; (H) two nonperforated axospinous synapses formed by a multiple-synapse bouton; perforated axospinous synapses that lack spine partitions and exhibit (I) a segmented, (J) horseshoe-shaped, or (K) fenestrated PSD. The sequence of synapse restructuring from A through B, C, D to E is proposed to be a rapid process that supports an initial maximal level of synaptic enhancement. The conversion of synapses from E through F to G and/or from E to H may underlie an enduring retention of synaptic enhancement at a relatively low level. Additionally, the consecutive transformation of synapses from E through I, J, K, L to A may account for the return of synaptic responses to the control level.

Figure 5. Diagram illustrating a hypothetical synapse restructuring that may underlie activation-dependent alterations in synaptic efficacy as explained in the text. The schematic shows the following synaptic subtypes: (A) typical (small) and (B, L) atypical (large) nonperforated axospinous synapses; perforated axospinous synapses that have (C) a focal spine partition and fenestrated PSD, (D) a sectional partition and horseshoe-shaped PSD, or (E) a complete partition(s) and segmented PSD; (F) an axospinous perforated synapse involving the postsynaptic spine that is partially retracted into a parent dendrite; (G) an asymmetrical axodendritic synapse; (H) two nonperforated axospinous synapses formed by a multiple-synapse bouton; perforated axospinous synapses that lack spine partitions and exhibit (I) a segmented, (J) horseshoe-shaped, or (K) fenestrated PSD. The sequence of synapse restructuring from A through B, C, D to E is proposed to be a rapid process that supports an initial maximal level of synaptic enhancement. The conversion of synapses from E through F to G and/or from E to H may underlie an enduring retention of synaptic enhancement at a relatively low level. Additionally, the consecutive transformation of synapses from E through I, J, K, L to A may account for the return of synaptic responses to the control level.

have to be considered in this respect. One of these is that some additional axospinous synapses with a segmented PSD and complete spine partitions (Fig. 5E) may be converted into asymmetrical axodendritic synapses (Fig. 5G). Such conversion is suggested by the observations that synapses of the latter kind are selectively increased in number during the maintenance phase of LTP and that there is a synaptic subtype (Fig. 5F), which appears to be transitional between axospinous and axodendritic junctions.39 Another possibility alluded to in the preceding section was originally envisioned by Carlin and Siekevitz15 who postulated that large segmented synapses (Fig. 5E) might split into small nonperforated ones (Fig. 5H). A temporal coincidence between the disassembly of synapses with multiple transmission zones and an increase in the proportion of MSBs after LTP induction116 gives credence to this notion (but see ref. 30). Finally, it is also feasible that the transformation of additional synapses with multiple transmission zones into typical nonperforated synaptic contacts may be accomplished through the consecutive formation of perforated axospinous synapses that lack spine partitions and exhibit a segmented (Fig. 5I), horseshoe-shaped (Fig. 5J) and fenestrated (Fig. 5K) PSD.

These different patterns of the proposed remodeling of perforated synapses with multiple transmission zones (Fig. 5) may account for both the decay of synaptic responses (the sequence from E through I, J, K, L to A) and the sustained retention of a low level of synaptic enhancement (the sequences from E through F to G or from E to H) during LTP maintenance. Behavioral learning is accompanied, as is hippocampal LTP, by increases in the number of perforated axospinous synapses and of MSBs. It is tempting to speculate, therefore, that some forms of perforated synapse restructuring as outlined above may be characteristic of both phenomena.

Enlargement of Postsynaptic Densities following Learning: A Possible Morphological Correlate of the Conversion of Postsynaptically Silent Synapses into Functional Synapses

Recent work using time-lapse confocal imaging of hippocampal spines expressing a prominent PSD protein (PSD95) tagged with green fluorescent protein revealed that PSDs may rapidly (<15 min) expand or shrink.79 It is not surprising, therefore, that previous electron microscopic studies indicated that the length of PSD profiles increases following acquisition of new behaviors.20,55,112,119,120 In our experiments with trace eyeblink conditioning, we measured the length of PSD profiles on electron micrographs of consecutive sections through each synapse sampled from the CA1 stratum radiatum of the rabbit hippocampus to obtain estimates of the PSD area in conditioned and control animals.40 These measurements showed that nonperforated axospinous synapses had a significantly larger PSD area in conditioned animals (30.3 ± 0.8 nm2 x 103) than in pseudoconditioned (27.5 ± 0.9 nm2 x 103) or unstimulated (26.1 ± 1.0 nm2 x 103) controls.

The PSD contains signal transduction proteins, such as postsynaptic receptors and ion channels.66,125 The recent discovery of "silent" hippocampal synapses leads to the suggestion that the conditioning-induced enlargement of nonperforated PSDs might reflect an addition of AMPA receptors. Electrophysiological experiments have revealed that a high proportion of synaptic contacts in the rat CA1 stratum radiatum exhibit functional NMDA receptors, but not functional AMPA receptors.57,76 This makes such synapses postsynaptically silent, in that they do not generate a synaptic response to a release of a neurotransmitter, because NMDA receptor channels are blocked by extracellular magnesium at normal resting membrane potentials. Correspondingly, immunocytochemical studies have provided evidence for the existence of hippocampal synapses that exhibit only NMDA, but not AMPA, receptor immunoreactivity.19,5 ,95 A lack of AMPA receptors, and not their inactive state, accounts for this phenomenon.90,113 Notably, silent synapses acquire AMPA-type responses after LTP induction in the rat CA1 stratum radiatum, 7,76 indicating that they may be transformed into functional synaptic contacts due to an insertion ofAMPA receptors into their PSDs. Based on these findings, it seems reasonable to hypothesize that the same synaptic modification may be induced by trace eyeblink conditioning (Fig. 6). This hypothesis appears to be inconsistent with the data showing that gene-targeted mice, which lack the AMPA receptor subunit GluR-A and have a reduced number of functional AMPA receptors, do not exhibit deficits in their spatial learning ability when tested in a water maze.12 However, the trace eyeblink conditioning task taps the temporal, but not the spatial, domain of hippocampus-dependent memory function. Learning of the trace response may involve synaptic modifications that are different from those accompanying spatial learning.

Figure 6. Diagram illustrating the hypothetical conversion of postsynaptically silent synapses into functional synapses associated with trace eyeblink conditioning. The schematic shows axospinous synaptic contacts between a presynaptic axon terminal (AT) and postsynaptic dendritic spine (SP). A. Nonperforated axospinous synapse with a small PSD that contains only NMDA receptors (NMDAR) but lacks AMPA receptors (AMPAR). Due to this, the synapse cannot generate a postsynaptic response and is postsynaptically silent. B. The PSD of this synapse is increased in size as a consequence of trace eyeblink conditioning. This structural modification is hypothesized to reflect the insertion of AMPAR, which makes the synapse functional.

Figure 6. Diagram illustrating the hypothetical conversion of postsynaptically silent synapses into functional synapses associated with trace eyeblink conditioning. The schematic shows axospinous synaptic contacts between a presynaptic axon terminal (AT) and postsynaptic dendritic spine (SP). A. Nonperforated axospinous synapse with a small PSD that contains only NMDA receptors (NMDAR) but lacks AMPA receptors (AMPAR). Due to this, the synapse cannot generate a postsynaptic response and is postsynaptically silent. B. The PSD of this synapse is increased in size as a consequence of trace eyeblink conditioning. This structural modification is hypothesized to reflect the insertion of AMPAR, which makes the synapse functional.

Our hypothesis helps to explain why the enlargement of the PSD is a selective process that is characteristic only of nonperforated axospinous synapses and that does not involve any other synaptic subtype.4 Especially relevant to this question are the data demonstrating that the ratio of AMPA to NMDA receptors is directly proportional to the PSD size in axospinous synapses from the rat CA1 stratum radiatum and that the AMPA receptor number regresses to zero when a PSD diameter is smaller than 180 nm.90,97,113 These observations strongly suggest that the pool of silent axospinous synapses lacking AMPA receptors consists primarily of those synaptic junctions that have a relatively small, nonperforated PSD. Accordingly, our data show that only the smallest nonperforated PSDs, which probably lack AMPA receptors, are increased in their area by trace eyeblink conditioning (Fig. 7). Such a change may result from the insertion of AMPA receptors that are rapidly delivered to spines in response to synaptic NMDA receptor activation.7 ,105 Provided that the spine delivery and insertion of AMPA receptors are triggered by associative learning, the observed enlargement of the smallest nonperforated PSDs may represent a structural correlate of the conversion of silent synapses into functional ones. Further ultrastructural studies using double labeling of AMPA and NMDA receptors are required to verify the validity of this supposition.

Figure 7. Comparison ofpseudoconditioned and conditioned rabbits in terms of the distribution of nonperforated axospinous synapses according to the size of their PSD area (data from ref. 34). The total number of nonperforated synapses that had PSDs belonging to the smallest size category (PSD area < 20 nm2 x 103) was decreased in the conditioned group. Nonperforated synapses with PSDs that fell into all larger size categories were increased in number after conditioning. These data indicate that only the smallest nonperforated PSDs are enlarged in their area in conditioned rabbits.

Figure 7. Comparison ofpseudoconditioned and conditioned rabbits in terms of the distribution of nonperforated axospinous synapses according to the size of their PSD area (data from ref. 34). The total number of nonperforated synapses that had PSDs belonging to the smallest size category (PSD area < 20 nm2 x 103) was decreased in the conditioned group. Nonperforated synapses with PSDs that fell into all larger size categories were increased in number after conditioning. These data indicate that only the smallest nonperforated PSDs are enlarged in their area in conditioned rabbits.

Conclusions

Quantitative electron microscopic studies of the vertebrate brain have provided evidence that alterations in synaptic ultrastructure associated with learning and memory include both synaptogenesis and structural remodeling of synapses. A learning-induced net synaptogenesis manifested by an increase in the number of all synaptic contacts in a given brain region does not appear to be a ubiquitous phenomenon because it was detected in some studies, but not in others. Do these data tell us that only some varieties of learning and memory promote a net gain in synapse number or that such a gain takes place only during a certain phase of the acquisition/consolidation process?

Additionally, there are two known kinds of specific synaptogenesis, each one generating a special subset of synaptic connections that may be necessary for establishing a memory trace. One of these is confined to multiple-synapse bouton formation, which may complement a net synaptogenesis or occur independently of it. The addition of multiple-synapse boutons following learning may involve spine motility. Recent findings indicate that dendritic spines are highly dynamic structures capable of rapid motility. The neurobiological significance of this phenomenon remains unknown. Our models of multiple-synapse bouton formation raise the possibility that some existing or newly formed spines may relocate to single-synapse boutons activated by conditioning stimulation in order to synapse with them. The other kind of specific synaptogenesis, which may accompany learning-related adult neurogenesis, has not been directly demonstrated so far. Do both kinds of specific synaptogenesis represent a generalized phenomenon or are they unique for some particular forms of learning and memory?

Synapses have also been reported to undergo a learning-induced structural remodeling. The most demonstrative example of this is an enlargement of the PSD. Such a change was shown to selectively involve nonperforated axospinous synapses that had the smallest PSDs, usually lacking AMPA receptors, which probably made them postsynaptically silent. The increase in nonperforated PSD area was postulated to reflect the insertion of AMPA receptors and to represent a structural correlate of learning-associated conversion of postsynaptically silent synapses into functional ones. Is this hypothesis valid?

A learning-related increase in the proportion of perforated axospinous synapses is another alteration in synaptic ultrastructure resulting from structural remodeling of synapses: this alteration is completed within minutes, while the assembly of excitatory synapses takes at least 1 hour. It has been postulated that perforated axospinous synapses may evolve from nonperforated ones and then undergo a further restructuring. This would culminate in the formation of the perforated synaptic subtype distinguished by multiple transmission zones completely separated by spine partitions. The addition of synapses belonging to the latter subtype is known to occur early after the induction of hippocampal LTP, and it is possible that the same structural synaptic modification may be also associated with behavioral learning. Does the number of perforated synapses with multiple transmission zones increase soon after acquisition of new behaviors and is this change accompanied by a corresponding loss of nonperforated synapses?

The various learning-induced alterations in synaptic ultrastructure reported in the literature and reviewed here may be events that are interrelated or independent from each other. There are arguments in favor of the former possibility. For example, the enlargement of nonperforated PSDs promoted by learning may represent a step in the transformation of nonperforated axospinous synapses into perforated ones. Additionally, it is conceivable that specific synaptogenesis leading to MSB formation may be due to the splitting of axospinous synapses with multiple transmission zones. Such synapses transiently increase in number after LTP induction, and the reversal of this change coincides with MSB formation. Does behavioral learning elicit a similar sequence of structural synaptic alterations? The challenge of future research will be to address the questions posed above and to provide a better understanding of whether learning-associated net synaptogenesis, specific synaptogenesis and structural synapse remodeling are the links in a common chain of consecutive processes that underlie memory formation, consolidation and long-term storage.

Acknowledgements

Supported by grant 1 RO1 AG17139 from the National Institute on Aging.

References

1. Altman J, Das GD. Autoradiographic and histologic evidence of postnatal neurogenesis in rats. J Comp Neurol 1965; 124:319-335.

2. Alvarez-Buylla A, Kirn JR, Nottebohm F. Birth of projection neurons in adult avian brain may be related to perceptual or motor learning. Science 1990; 249:1444-1446.

3. Ambrogini P, Cuppini R, Cuppini C et al. Spatial learning affects immature granule cell survival in adult rat dentate gyrus. Neurosci Lett 2000; 286:21-24.

4. Andersen P, Soleng AF. Long-term potentiation and spatial training are both associated with the generation of new excitatory synapses. Brain Res Rev 1998; 26:353-359.

5. Angevine JB. Time of neuron origin in the hippocampal region. An autoradiographic study in the mouse. Exp Neurol Suppl 1965;2:1-70.

6. Bailey CH, Kandel ER. Structural changes accompanying memory storage. Annu Rev Physiol 1993; 55:397-426.

7. Barnea A, Nottebohm F. Seasonable recruitment of hippocampal neurons in adult free-ranging black-capped chickadees. Proc Natl Acad Sci USA 1994; 91:11217-11221.

8. Bayer S. Changes in the total number of dentate granule cells in juvenile and adult rats: a correlated volumetric and thymidine study. Exp Brain Res 1982; 46:315-323.

9. Black JE, Isaaks KR, Anderson BJ et al. Learning causes synaptogenesis, whereas motor activity causes angiogenesis, in cerebellar cortex of adult rats. Proc Natl Acad Sci USA 1990; 87:5568-5572.

10. Bliss TVP, Collingridge GL. A synaptic model of memory: long-term potentiation in the hippocampus. Nature 1993; 361:31-39.

11. Bourgeois J-P, Goldman-Rakic PS, Rakic P. Formation, elimination, and stabilization of synapses in the primate cerebral cortex. In: Gazzaniga MS, ed. Cognitive Neuroscience. A Handbook for the Field. 2nd Edition. Cambridge: MIT Press, 1999; 23-32.

12. Buchs P-A, Muller D. Induction of long-term potentiation is associated with major ultrastructural changes of activated synapses. Proc Natl Acad Sci USA 1996; 93:8040-8045.

13. Cameron HA, McKay RD. Restoring production of hippocampal neurons in old age. Nature Neurosci 1999; 2:894-897.

14. Cameron HA, Woolley CS, McEwen BS et al. Differentiation of newly born neurons and glia in the dentate gyrus of the adult rat. Neurosci 1993; 56:337-344.

15. Carlin PK, Siekevitz P. Plasticity in the central nervous system: do synapses divide? Proc Natl Acad Sci USA 1983; 80:3517-3521.

16. Cohen RS, Siekevitz. Form of the postsynaptic density. A serial section study. J Cell Biol 1978; 78:36-46.

17. Dailey ME, Smith SJ. The dynamics of dendritic structure in developing hippocampal slices. J Neurosci 1996; 16:2983-2994.

18. Del Bigio MR. Proliferative status of cells in adult human dentate gyrus. Microsc Res Tech 1999; 45:353-358.

19. Desmond NL, Weinberg RJ. Enhanced expression of AMPA receptor protein at perforated axospinous synapses. NeuroReport 1998; 9:857-860.

20. DeVoogt TJ, Nixdorf B, Nottenbohm F. Synaptogenesis and changes in synaptic morphology related to acquisition of new behavior. Brain Res 1985; 329:304-308.

21. Disterhoft JF, Coulter DA, Alkon DL. Conditioning-specific membrane changes of rabbit hippoc-ampal neurons measured in vitro. Proc Natl Acad Sci USA 1986; 83:2733-2737.

22. Doubell TP, Stewart MG. Short term changes in the numerical density of synapses in the intermediate and medial hyperstriatum ventrale following one-trial passive avoidance learning. J Neurosci 1993; 13:2230-2236.

23. Dunaevsky A, Blazeski R, Yuste R et al. Spine motility with synaptic contact. Nature Neurosci 2001; 4:685-686.

24. Dunaevsky A, Tashiro A, Majewska A et al. Developmental regulation of spine motility in the mammalian central nervous system. Proc Natl Acad Sci USA 1999; 96:13438-13443.

25. Dyson SE, Jones DG. Synaptic remodelling during development and maturation: junction differentiation and slitting as a mechanisms of modifying connectivity. Dev Brain Res 1984;13:125-137.

26. Edwards FA. LTP—a structural model to explain the inconsistencies. Trends Neurosci 1995; 18:250-255.

27. Engert F, Bonhoeffer T. Dendritic spine changes associated with long-term synaptic plasticity. Nature 1999; 399:66-70.

28. Eriksson PS, Perfilieva E, Björk-Eriksson T et al. Neurogenesis in the adult human hippocampus. Nature Med 1998; 4:1313-1317.

29. Federmeier K, Kleim IA, Anderson BJ et al. Formation of double synapses in the cerebellar cortex of the rat following motor learning. Soc Neurosci Abstr 1994; 20:1435.

30. Fiala JC, Allwardt B, Harris KM. Dendritic spines do not split during hippocampal LTP or maturation. Nature Neurosci 2002; 5:297-298.

31. Fischer M, Kaech S, Knutti D et al. Rapid actin-based plasticity in dendritic spines. Neuron 1998; 20:847-854.

32. Friedman HV, Bresler T, Garner CC et al. Assembly of new individual excitatory synapses: time course and temporal order of synaptic molecule recruitment. Neu

Was this article helpful?

0 0
Unraveling Alzheimers Disease

Unraveling Alzheimers Disease

I leave absolutely nothing out! Everything that I learned about Alzheimer’s I share with you. This is the most comprehensive report on Alzheimer’s you will ever read. No stone is left unturned in this comprehensive report.

Get My Free Ebook


Post a comment