R E V I E W S NATURE REVIEWS | NEUROSCIENCE VOLUME 6 | APRIL 2005 | 277 More than a century ago, soon after Ramon y Cajal first described the dendritic spine1, it was proposed that the spine is likely to be the locus of long-term synaptic plas- ticity that is associated with the storage of memories in the brain. Since its conception, this dogma has been supported by indirect, correlative and anecdotal obser- vations for instance, an ENRICHED ENVIRONMENT enhances spine formation2, and mentally retarded children express immature spines3.As excitatory synapses reside on spine heads, and synapses have been shown to express plastic properties similar to those that are thought to take place during learning and the forma- tion of memory, it was natural to assume that spine morphology is correlated with memory storage. But what are the long-term morphological changes that the spine undergoes to ���store���memories? Over the past century, extensive research has been conducted in this field. Initially, experiments were carried out with brain slices that underwent patterned stimulation or animals that performed a learning task. The brain tissue was then fixed and compared with control tissue ��� a tedious process that required the measurement of hundreds or thousands of dendritic spines using three-dimensional electron microscopy (EM) to reconstruct dendritic segments4,5 (FIG. 1). In the past decade, the more sensitive single- and TWO-PHOTON MICROSCOPY has been used in conjunction with fluores- cent molecular tools, which has allowed live dendritic spines to be imaged at high resolution, and with little bleaching or photodynamic damage6,7. The same dendritic segments can be compared before and after exposure to a conditioning treatment, thereby reducing the need to collect large amounts of data from different populations of neurons7���11. The downside of these methods is that only a small number of spines are studied, which can lead to a sampling bias and might not be representative of the entire population of spines. Nevertheless, riding on a new wave of enthusiasm, these recent reports have revolutionized the view that the spine is a stable structural extension of the synapse and a possible locus of stable memories.At the spine,receptor turnover can be rapid, which allows fast changes in spine function on a minute-to-minute basis however, this brings into question the traditional view of the spine. Although new evidence has accumulated in recent years and new hypotheses are emerging, it is not clear what takes place in the spine while a ���memory��� is established. In this review, we first identify the area of interest (that is, spine changes that accompany functional plasticity in adult animals) and discuss relevant issues in the study of spines. We then formulate criteria for establishing a relationship between a ���memory��� and a dendritic spine. Finally, we critically examine the results of recent studies to clarify and unify our understanding of the roles of dendritic spines in neuronal plasticity. Intentionally, we avoid discussing the progress that has been made recently in the understanding of the overall molecular construction of the spine as this is beyond the scope of this review. DENDRITIC SPINES AND LONG-TERM PLASTICITY Menahem Segal Abstract | A recent flurry of time-lapse imaging studies of live neurons have tried to address the century-old question: what morphological changes in dendritic spines can be related to long- term memory? Changes that have been proposed to relate to memory include the formation of new spines, the enlargement of spine heads and the pruning of spines. These observations also relate to a more general question of how stable dendritic spines are. The objective of this review is to critically assess the new data and to propose much needed criteria that relate spines to memory, thereby allowing progress in understanding the morphological basis of memory. ENRICHED ENVIRONMENT Growing a young mammal in an environment that is enriched with stimuli and motor demands has been shown to enhance the cognitive skills of that animal, as well as the complexity of its neurons. TWO-PHOTON MICROSCOPY A form of microscopy in which a fluorochrome that would normally be excited by a single photon is stimulated quasi- simultaneously by two photons of lower energy.Under these conditions,fluorescence increases as a function of the square of the light intensity,and decreases as the fourth power of the distance from the focus. Because of this behaviour,only fluorochrome molecules near the plane of focus are excited, greatly reducing light scattering and photodamage of the sample. Department of Neurobiology, The Weizmann Institute, Rehovot,76100 Israel. e-mail: menahem.segal@ weizmann.ac.il doi:10.1038/nrn1649

FILOPODIUM A highly motile cytoplasmic extension of an axon or dendrite that is 2���10 ��m long and less than 1 ��m thick. It is assumed to serve as a sensing element in the formation of synaptic contacts with adjacent neurons. 278 | APRIL 2005 | VOLUME 6 www.nature.com/reviews/neuro R E V I E W S close to the dendritic shaft,from which it later pulls out a spine. In mature neurons, for which variations in spine density in the order of ��30���40% have been reported16, few filopodia are found, indicating that the mechanism for producing spines in mature neurons might differ from that in developing neurons13. Spine motility. Spines contain actin,a contractile molecule that binds to a family of adaptor and linking molecules, and connects them to the postsynaptic density and its associated receptor molecules.Filamentous actin (F-actin) is highly unstable, with a short half-life (~40 s (REF. 17)) and is thought to consume considerable cellular energy18. Local spine motility,or ���morphing���19 (BOX 1),which might be relevant to the role of spines in memory formation, has been detected in dendritic spines using high-resolu- tion time-lapse imaging of green fluorescent protein (GFP).The functional significance of spine morphing is not known, but the authors of a recent study proposed that it might be involved in the diffusion of molecules through the plasma membrane into the spine20. Such a mechanism might also be important for the fast delivery of receptors into the synapse,a process that is likely to be accelerated during the acquisition of a memory. Spine pruning. Spine pruning is a natural process in adult animals there is a striking reduction of about 30% in spine density in female rats during oestrus16 and in hibernating animals21. Spine pruning seems to be an active process, which requires the activation of NMDA (N-methyl-D-aspartate) receptors22, and takes place alongside spine formation, during which NMDA recep- tors are acutely activated10.Activation of neurons with glutamate results in the rapid disappearance of spines, which reappear at the same locations when the stimulus is removed23.So, studies that are based on counting fixed spines might underestimate these dynamic changes, because a strong stimulus can cause simultaneous spine formation and pruning24, thereby reducing the total estimated spine changes. If spine pruning is an active, ongoing process, what happens to the presynaptic terminal and postsynaptic specialization when a spine is pruned? The presynaptic terminal might either detach and disappear,as suggested by the authors of a recent in vivo imaging study7,or main- tain its connection with the pruned spine.The latter pos- sibility is more likely,as the synaptic junction is enriched with cell adhesion molecules,which make the synapse a bond that cannot be easily detached. In fact, synapto- somal preparations take advantage of this property and isolate the pre- and postsynaptic membrane as one unit. If the presynaptic terminals are still connected after spine pruning,does the synaptic potential that is produced by a pruned spine differ from the synaptic response of an intact one? Does pruning affect the behaviour of the neuron? Several studies are beginning to address these issues.For example,neurons that were transfected with a constitutively active form of Rho GTPase lost their spines, but the presynaptic terminals remained attached to the dendritic shaft and the neurons still expressed MINIATURE EXCITATORY POSTSYNAPTIC CURRENTS (mEPSCs), similar to Crucial issues in dendritic spine research Spine formation. The great heterogeneity among differ- ent spines on the same dendritic segment raises ques- tions about the mechanisms of spine formation and the functional significance of this heterogeneity.It is not clear how spines are formed ��� a process that has important implications for understanding the functions of spines in memory formation. Spines can be formed de novo or converted from an existing shaft synapse.In the former case, a FILOPODIUM shoots out from the dendrite, forms a synapse with the presynaptic terminal and then collapses to become a short,1���2 ��m,���mature���spine12,13.The mole- cular mechanisms that lead to the formation of a mature, functional spine might be different for the two cases,as might the involvement of adhesion molecules and post- synaptic density protein 95 (PSD-95)14,15. The de novo formation of spines is a favoured topic of developmental neurobiologists.A newly differentiated, immature neu- ron has dendritic filopodia that seem to be searching for presynaptic partners13. An intermediate process has been proposed, in which a filopodium searches for a presynaptic partner, makes contact and pulls the axon a b 0.2 ��m 0.2 ��m Figure 1 | Three-dimensional reconstructed electron microscopy picture of a dendrite, two spines and an associated axon. a | A single electron microscope section of a dendrite and two spines that are associated with a passing axon. The axon makes a mature synapse with one of the spines. The cell was transfected with green fluorescent protein (GFP), imaged with a confocal laser scanning microscope, fixed, stained, sectioned and imaged using an electron microscope. b | The dendrite was then reconstructed (shown at a different angle), to show the types of connections that are formed in hippocampal cultures. Images courtesy of K. Braun, Otto-von-Guericke-University, Magdeburg, Germany. Box 1 | Spine motility Local spine motility has been detected by Matus and collaborators in neurons transfected with green fluorescent protein (GFP).Both young- and mature-looking spines have been studied,some of which were already innervated by a presynaptic terminal73.Spine motility declines with age74.These observations indicate that there are continuous changes in spine shape and size,and cast doubt on the significance of persistent changes in spine shape (for example,changes in the length of the postsynaptic density34) in long-term synaptic plasticity,especially if the changes are small (~20���30%).Nonetheless,more recent observations indicate that spine motility ceases in response to the application of the glutamate agonist AMPA (��-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid)75. Similarly,it has been suggested that spine motility is inversely correlated with the activity of the synapse74,and that blockade of activity with tetrodotoxin (TTX) enhances spine motility74.Furthermore,among the many molecules that control the shape,size and density of spines,��N-catenin (a cadherin-associated protein) regulates spine motility such that spines that lack ��N-catenin are more motile15.As TTX blocks the invasion of ��N-catenin into spines,this might be the mechanism through which TTX controls spine motility.