Main

Dendritic spines are morphological specializations that protrude from the main shaft of neuronal dendrites. Typically 0.5–2 μm in length (but up to 6 μm in the CA3 region of the hippocampus)1,2, dendritic spines are found at a linear density of 1–10 spines per μm of dendritic length in mature neurons3. Most excitatory synapses in the mature mammalian brain occur on spines, and a typical mature spine has a single synapse located at its head. So, dendritic spines represent the main unitary postsynaptic compartment for excitatory input. Most principal neurons of either glutamate-releasing (for example, pyramidal neurons) or GABA (γ-aminobutyric acid)-releasing (for example, Purkinje neurons) type bear dendritic spines, but many classes of neuron do not (for instance, most GABA-releasing interneurons). Spiny neurons are rarely found in lower organisms (for example, Drosophila melanogaster and Caenorhabditis elegans), indicating that spines evolved to accommodate the more complex functions of 'advanced' nervous systems, such as the mammalian brain. In this review, we cover the essential background to dendritic spines, but focus on recent advances in our understanding of spine plasticity, and the molecular mechanisms that regulate spine formation and shape.

The structure of dendritic spines

Dendritic spines come in a wide variety of shapes and sizes, ranging in volume from less than 0.01 μm3 to 0.8 μm3 (reviewed in Refs 1,4). On the basis of detailed anatomical studies of fixed brain tissue, dendritic spines have been classified by shape as thin, stubby, mushroom- and cup-shaped5,6,7 (Fig. 1). However, the arbitrary classification of spines into these four categories underestimates the great heterogeneity of spine morphology, which is apparent even on a single dendrite8,9. It can also bias us towards a misleadingly static view of spine morphology. Live imaging studies have revealed that spines are remarkably dynamic, changing size and shape over timescales of seconds to minutes and of hours to days (see below and Fig. 1). So, fixed structures seen under the electron microscope are probably snapshots of spines in morphological transition. Two-photon time-lapse imaging of motile spines showed that, in developing neurons, ∼50% of observed spine-like protrusions maintained their morphological classifications over several hours, whereas the other half switched classes10. As spine motility is developmentally regulated11, fewer transitions between categories presumably occur in mature neurons.

Figure 1
figure 1

Morphological classification of dendritic spines.

Full size image

In addition to varying in shape and size, spines also differ in their content of organelles and specific molecules. In general, large spines have proportionately larger synapses and contain a greater diversity of organelles (Fig. 2). The postsynaptic density (PSD) is an electron-dense thickening of the membrane that is found at the synaptic junction, which is usually located at the head of the spine. The PSD occupies ∼10% of the surface area of the spine, and is exactly aligned with the presynaptic active zone. As the size of the spine head is proportional to the area of the PSD, to the number of postsynaptic receptors12 and to the number of presynaptic docked vesicles13, the growth of the spine head probably correlates with a strengthening of synaptic transmission.

Figure 2: Structure of the dendritic spine.
figure 2

A mushroom-shaped spine is depicted, containing various organelles, including smooth endoplasmic reticulum (SER), which extends into the spine from SER in the dendritic shaft9. SER is present in a minority of spines, correlating with spine size. The SER in spines functions, at least in part, as an intracellular calcium store from which calcium can be released in response to synaptic stimulation24,84. In some cases, SER is seen to move close to the postsynaptic density (PSD) and synaptic membrane, perhaps by specific protein–protein interactions between PSD proteins Shank and Homer, and the inositol-1,4,5-trisphosphate receptor (InsP3R) of the SER82,83. Particularly common in larger spines is a structure known as the spine apparatus, an organelle characterized by stacks of SER membranes surrounded by densely staining material. The role of the spine apparatus is unknown, although it might act as a repository or a relay for membrane proteins trafficking to or from the synapse. Vesicles of 'coated' or smooth appearance are sometimes observed in spines (particularly in large spines with perforated PSDs16), as are multivesicular bodies, all consistent with local membrane trafficking processes. Coated vesicles (CV) are found not only within the spine head, but also close to, or apparently fusing with, the synaptic membrane9,16,85. Polyribosomes have been detected in dendritic shafts, often at the base of spines, and occasionally extending into spines, indicating that protein translation might occur within the immediate postsynaptic compartment86. The enlarged box illustrates specific proteins and protein–protein interactions within the PSD. GRIP (glutamate-receptor-interacting protein), and CASK (calcium/calmodulin-dependent serine protein kinase) and synbindin, are PDZ-containing scaffold proteins that bind to AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid) receptors and syndecan 2, respectively. AMPAR, AMPA receptor; F-actin, filamentous actin; GKAP, guanylate-kinase-associated protein; Kali-7, Kalirin-7; mGluR, metabotropic glutamate receptor; NMDAR, N-methyl-d-asparate receptor; SPAR, spine-associated RapGAP.

Full size image

Most synapses have a single, continuous ('simple') PSD per spine, but some PSDs are seen to be discontinuous or 'perforated'14 when viewed by single-section electron microscopy. Perforated PSDs can be further categorized as 'fenestrated', 'horseshoe' or 'segmented' (completely partitioned PSDs on one spine) on more-detailed three-dimensional analysis (Fig. 3). Perforated PSDs might reflect the growth of synapses, perhaps representing an early phase of synapse duplication and spine division15,16. Others have proposed that perforated synapses are the morphological correlate of enhanced receptor turnover at the PSD, such as might occur during LONG-TERM POTENTIATION (LTP)16,17. The PSD contains glutamate receptors of the AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid) and NMDA (N-methyl-d-aspartate) types, and more AMPA receptor immunoreactivity has consistently been found to be associated with perforated than with non-perforated PSDs12,18. Perforated PSDs might reflect enhanced AMPA receptor insertion into the postsynaptic membrane, which occurs during synaptic potentiation19,20,21,22 (Fig. 3).

Figure 3: A model of changes in spine and PSD morphology after LTP.
figure 3

Schematic depiction of changes in dendritic spine morphology associated with long-term potentiation (LTP), adapted from the work of Muller and colleagues15,16. Electron-microscopic analysis of putatively activated spines indicates that LTP induction results in a transient increase in 'perforated' postsynaptic densities (PSDs), followed by an increase in the number of multiple spine boutons (MSBs). Normal spines have 'simple', continuous PSDs (left). At 30 min after LTP induction in hippocampal slice cultures, activated spines (middle) showed an increase in the proportion of perforated PSDs (particularly of the segmented type), and a corresponding reduction in the number of simple PSDs. The perforated synapses are associated with larger spine heads, expanded PSD area, and a higher frequency of coated vesicles in the spine. Segmented PSDs in particular showed spinules (finger-like protrusions of the postsynaptic membrane into the presynaptic terminal). By 2 h after LTP (right), the percentage of perforated PSDs, and the average sizes of PSD and spine heads, have returned to normal, but the proportion of MSBs (presynaptic terminals contacting more than one spine) is increased. Most of these MSBs arose from contacts of one terminal with multiple spines from the same dendrite, and had smaller PSDs than average, indicating the splitting of a pre-existing single spine. Overall, these results are consistent with the idea that synaptic potentiation might involve an activity-dependent metamorphosis from continuous PSDs in small spines, to expanded, segmented PSDs in enlarged spines, to bifurcation of spines contacting the same presynaptic terminal. However, this model has been inferred from static electron-microscopic images taken at different time points after LTP induction. The sequential progression of the proposed stages of this model remains to be shown.

Full size image

Function of dendritic spines

What is the significance of dendritic spines for synaptic transmission? The fact that each dendritic spine usually accommodates a single synapse indicates that the significance of spines might relate to the creation of a local synapse-specific compartment, rather than the mere expansion of postsynaptic surface area23. The prevailing opinion is that the primary function of spines is to provide a microcompartment for segregating postsynaptic chemical responses, such as elevated calcium24. Spines can act as semi-autonomous chemical compartments, because they are separated from the dendritic shaft by a neck that is often thin and up to a few micrometres in length. The geometry of the spine neck might control the kinetics and magnitude of postsynaptic calcium responses. Calcium responses in spines with long necks have a shorter latency and slower decay kinetics than those in short-necked spines25,26,27. Moreover, changes in the length of the spine neck during spine motility correlate with altered diffusional coupling between the dendrite and the spine, and with calcium kinetics within the spine28,29. Spines can therefore compartmentalize calcium, and this function is affected by the morphology of spines. However, it remains to be determined whether dynamic changes in spine shape actually are significant for regulating the communication between synapses and dendrites in vivo.

Another useful feature of the spine is the relatively small volume of the spine head, which allows large changes in intraspine calcium levels in response to the opening of a small number of receptors or channels. For example, it is estimated that individual spines contain only 1–20 voltage-sensitive calcium channels, depending on their size24,30. Furthermore, as several different types of calcium-permeable receptor/channel are colocalized in spines, the spine head can act as an efficient integrator of different postsynaptic signals. However, it should be emphasized that, despite the widespread acceptance that spines can segregate and integrate synaptic signals, the physiological significance of spines for brain function is still not clear.

Development of spines

FILOPODIA rapidly protrude and retract from dendrites, especially during early stages of synaptogenesis8,31,32. With a transient lifetime of minutes10,33, and sometimes bearing synapses in vivo8, dendritic filopodia are widely believed to be the precursors of dendritic spines. Filopodia are most abundant in the brain during the first postnatal week in vivo, but are subsequently replaced by shaft synapses and stubby spines. With further development, shaft synapses and stubby spines decrease in number, and synapses on thin and mushroom-shaped spines predominate in the adult rat brain8. The sequential appearance of the various spine types in developing brain led Harris and colleagues to propose that dendritic filopodia draw the presynaptic contact to the dendrite, leading to the formation of shaft synapses from which mature spines subsequently emerge4. This hypothesis is supported by the observation that filopodia in cultured hippocampal neurons actively 'initiate' physical contact with nearby axons32. Two recent time-lapse studies, in which spine morphogenesis and PSD95–green fluorescent protein (GFP) clustering (as a marker of PSDs) were imaged simultaneously, also indicate that synapses initially form on dynamic filopodia-like spines. However, these filopodia can convert directly into stable spines, coincident with formation of the postsynaptic specialization34,35. Curiously, the emergence of stable spines from shaft synapses was observed in one of these studies35, but not in the other34. It has also been shown by TWO-PHOTON TIME-LAPSE MICROSCOPY that stubby spines and other types of spine can originate from filopodia in developing hippocampal neurons10. Interestingly, the opposite transformation (spines turning into filopodia) was also observed in the same study. Together, the data do not point to a simple developmental relationship between filopodia and spines. It seems that filopodia can transform into spines without first resorbing into the shaft. Moreover, the filopodia-to-spines transition is unlikely to be a predestined process, but instead one that is reversible and regulated by local factors, such as synaptic activity.

Activity-dependent regulation of spines

Regulated changes in spine morphology and number might reflect mechanisms for converting short-term changes in synaptic activity into lasting alterations in the structure, connectivity and function of synapses. Because spine number and shape probably relate directly to synaptic transmission, there is great interest in the activity-dependent regulation of spine morphology (see Ref. 36 for a recent review). Spine formation and spine density can be affected by activity over both short and long timescales.

Earlier electron-microscopic studies that examined stimulated brain tissue gave mixed results. LTP-inducing stimulation has been correlated with spine enlargement that lasts for hours, together with a shortening of the spine neck37,38, and with increased spine size, synapse area and frequency of U-shaped/spinule-containing ('concave') spines39 (but see Refs 5,40 for dissimilar findings). Andersen and colleagues found a ∼30% increase in spine density, and a large increase in the number of bifurcated spines in potentiated rat dentate gyrus in vivo, but no significant change in spine dimensions41. By contrast, Sorra and Harris failed to find changes in spine number after LTP stimulation in the CA1 region of hippocampal slices42.

A disadvantage of the above studies was having to compare two separate populations of spines/synapses in which inter-spine variability was large, the subset of potentiated synapses small, and the time window of spine modification potentially narrow. In more recent ultrastructural studies that focused on a subset of putative activated spines (labelled by a calcium-precipitation technique), Muller and colleagues obtained evidence for a rapid morphogenetic sequence of events after LTP that includes PSD segmentation and spine duplication15,16 (Fig. 3).

Time-lapse imaging of living tissue is appropriate for studying structures as dynamic as dendritic spines, even though it affords less spatial resolution than electron microscopy. Using this approach, the growth of a subpopulation of small spines was detected in hippocampal slices after chemically induced LTP43. More strikingly, new spine/filopodia-like structures appeared specifically in the area of stimulation in electrically stimulated hippocampal slice cultures44,45. The emergence of these protrusions required NMDA receptor activation and correlated with LTP, but was not obvious until ∼30 min after stimulation. So, the early LTP that occurs within minutes of tetanus cannot be explained by the formation of new spines/filopodia.

Changes in spine density have also been observed in vivo, correlating with environmental factors that affect brain activity (such as visual deprivation, hibernation and the oestrus cycle; Table 1). In humans, abnormal spine density or shape is associated with many nervous system disorders (for example, mental retardation, Down's syndrome, FRAGILE-X SYNDROME and epilepsy), indicating at least an indirect link between spine morphogenesis and disease (Table 1). In general, spines seem to be maintained by an 'optimal' level of synaptic activity, with overall spine density increasing when there is insufficient activity, and decreasing when stimulation is excessive. The dynamic nature of spines could offer a morphological substrate for neurons to adjust constantly the number of axospinous synapses, allowing them to maintain excitatory homeostasis.

Table 1 Environmental and disease factors that affect spine density
Full size table

Glutamate receptors and spines

Recent real-time imaging44,45 and ultrastructural studies focused on activated spines15,16 have provided compelling evidence that the number and configuration of spines can change in response to potentiating stimuli. Although it is still a matter of faith that these morphological changes have a functional role with respect to synaptic plasticity, it has become imperative to pursue the molecular mechanisms that underlie the activity-dependent regulation of spines. In this context, it is natural to start with glutamate receptors, molecules that 'report' synaptic activity and are concentrated in the PSD of spines.

Spine morphology is profoundly influenced by the activity of glutamate receptors. NMDA application causes an acute collapse of dendritic spines, and a loss of spine actin in cultured neurons46. Inhibition of calcineurin, a calcium/calmodulin-dependent phosphatase that is stimulated in response to NMDA receptor activation, attenuated the NMDA-induced loss of spine actin, indicating a possible role for calcium and calcineurin in regulating spine stability46.

Low levels of AMPA receptor activation (such as might be afforded by spontaneous neurotransmitter release) are required to maintain spines in organotypic cultures of the hippocampus47. On a much shorter timescale, AMPA receptor stimulation can also 'freeze' spines by stimulating calcium influx48. Conversely, release of intracellular calcium by caffeine has been reported to stimulate spine elongation49. One way to reconcile such results is to propose a bimodal relationship between calcium concentration and spine growth50. Moderate levels of intraspine calcium (such as those provided by release from the smooth endoplasmic reticulum, or by AMPA-receptor-mediated depolarization and influx through voltage-gated calcium channels) promote spine stability/growth, whereas high levels of calcium (such as those observed after a prolonged activation of NMDA receptors) induce shrinkage or collapse. A similar dichotomy in the effects of postsynaptic calcium elevation has been invoked to explain the different calcium requirements of long-term depression (LTD) and LTP.

Genetic evidence for a role of glutamate receptors or synaptic activity in spine morphogenesis is largely lacking. An interesting exception is the knockout mouse that lacks the NR3A subunit of the NMDA receptor, which showed enhanced NMDA responses associated with increased spine density during early postnatal development51. However, spine density seemed normal in mice that lacked the NMDA receptor subunit NR1 in the CA1 region of the hippocampus52. More detailed studies of spine morphology are needed in knockout mice that are deficient in glutamate receptors or in other proteins involved in synaptic transmission.

Rapid spine motility

Distinct from the morphological plasticity that occurs over hours or days, dendritic spines also show rapid motility. Advances in video microscopy and GFP technology have revealed that most spines can change shape over a timescale of seconds to minutes in cultured neurons, brain slices and the intact brain11,33,53,54. The shape change involves remodelling of the actin cytoskeleton in the spine, and actin-based protrusive activity from the spine head48,53. Although the underlying molecular mechanisms are unknown, spine motility is more pronounced during the critical period of development, and wanes with neuronal maturation11,33,55.

Whether the rapid actin-based motility can be controlled by synaptic activity is still under debate. The activation of either AMPA or NMDA receptors strongly inhibited spine actin dynamics and the actin-based protrusive activity from the spine head, causing spines to become more rounded and regular in shape48. The inhibition of spine motility by AMPA receptors was dependent on postsynaptic membrane depolarization and the influx of calcium through voltage-activated channels48. In accord with an activity-dependent suppression of spine movement, the motility of dendritic spines was found to be inversely correlated with developmental age and contact with active presynaptic terminals, and was stimulated by TETRODOTOXIN55. Intriguingly, volatile anaesthetics have also been found to arrest rapid spine motility56. By contrast, Dunaevsky et al. failed to detect a change in spine motility after blockade or stimulation of neuronal activity11, or in correlation with presynaptic contact57. This discrepancy could be explained by the different tissue preparations used: dispersed hippocampal cultures48,55,56 versus hippocampal slices11,57. Interestingly, a recent in vivo study in the barrel cortex found that spine motility is sensitive to input deprivation, but only during a brief critical period of development33.

Back-propagating action potentials have been shown to produce a tiny and rapid 'twitch' of the spine, coinciding with a transient increase in intraspine calcium58. In contrast to the previously described movements of the spine, this fast spine contraction is independent of the age of the neuron, and is found in spines that are contacted by an active presynaptic terminal58. It should be emphasized that although constantly moving spines are a captivating phenomenon, the functional significance of spine motility occurring over seconds or minutes is completely obscure at present.

Molecular mechanisms regulating spines

Actin-binding proteins and small GTPases. In recent years, considerable progress has been made in identifying the molecules that control spine growth and maturation (Table 2). Presumably, the cytoskeleton of spines is crucial for their development and stability. Filamentous actin (F-actin), consisting of β- and γ-isoforms of actin59, is highly concentrated in dendritic spines, whereas microtubules are generally sparse or missing1. A high-resolution photoconversion method revealed that actin filaments are particularly enriched at the PSD and around the internal membrane system of the spine60. The shape and stability of spine head and neck are likely to be determined largely by the actin cytoskeleton61. Proteins that bind to and modify the actin cytoskeleton are prime candidates for regulators of spine morphogenesis.

Table 2 Molecular pathways that affect dendritic spines
Full size table

An expanding set of actin-binding and actin-regulatory molecules has been detected in dendritic spines. Some of them are specifically enriched in this subcellular compartment (for example, α-actinin, drebrin, spinophilin/neurabin II, adducin, spine-associated RapGAP (SPAR) and cortactin62,63,64,65,66,67,68). Overexpression of the actin-binding protein drebrin induced the elongation of a subset of spines in cultured cortical neurons63. In spinophilin-deficient mice, spine density was increased in young animals, but returned to normal in adults69. Moreover, cultured cortical neurons from spinophilin knockout mice showed more filopodia or spine-like protrusions per length of dendrite. However, the exact mechanism by which drebrin or spinophilin might regulate actin dynamics in filopodia/spines remains to be determined.

Perhaps the best-known regulators of the actin cytoskeleton are the small GTPases of the RHO/RAC/CDC42 family. Transgenic mice that expressed constitutively active Rac1 developed supernumerary dendritic spines of much smaller size than normal in cerebellar Purkinje cells70. Similarly, constitutively active Rac1 disrupted the normal spine morphology of pyramidal neurons in slice culture, causing a net increase in dendritic protrusions (filopodia-like processes and LAMELLIPODIA-like ruffles)71,72. On the other hand, DOMINANT-NEGATIVE Rac1 caused a progressive reduction in spine number71, indicating that Rac1 activity is important for the maintenance of spine density. Support for this model came from the finding that overexpression of Kalirin-7, a guanine nucleotide exchange factor (GEF) for Rac1, increased the number of dendritic spine-like protrusions and the size of spines in cortical neurons, whereas a Kalirin-7 mutant that lacked GEF activity reduced the number of spines73. Kalirin-7 is enriched in the PSD, perhaps by binding to PSD95 and other PDZ-DOMAIN-containing proteins73.

Compared with Rac, the effects of RhoA on dendritic spines are less consistent. Overexpression of constitutively active RhoA strongly reduced the number of spines, but only in a subset of neurons72. Inhibition of Rho activity in some cases resulted in supernumerary spines, and in other cases in elongated spine necks. On the basis of these data, it is possible that Rac and Rho signalling might act antagonistically in spine formation/growth72. The third member of this family of GTPases — Cdc42 — seems to have little effect on spine size or density72.

The small GTPase RAS, although better known in the context of receptor tyrosine kinase signalling and cancer, has also been implicated in neuronal spinogenesis. Filopodia can be induced in cultured neurons by multiple depolarizing stimuli; this effect depends on activation of the Ras/mitogen-activated protein kinase pathway74. As it is present in the PSD75 and activated by NMDA receptor stimulation, Ras seems to be well positioned to participate in activity-dependent spine morphogenesis. A closely related GTPase — Rap — is also present in the NMDA receptor protein complex75. SPAR, a GTPase-activating protein (GAP) for Rap, is enriched in spines through binding to PSD95 (Ref. 67). SPAR interacts with F-actin and profoundly reorganizes the actin cytoskeleton in heterologous cells. Overexpression of SPAR in cultured hippocampal neurons caused the enlargement and elaboration of spine heads, making spines more complex ('thorny' and 'multilobed') in appearance. SPAR-enlarged spines were frequently associated with multiple synaptic contacts, and many of the irregular-shaped spines appeared to be branched, indicating that these spines might be dividing. By contrast, a dominant-negative mutant of SPAR that lacks RapGAP activity caused an elongation and thinning of spines, some of which resembled filopodia67. These results indicate that active (GTP-bound) Rap might stimulate spine elongation, whereas the RapGAP SPAR stimulates spine head growth and maturation. Intriguing in this context is the homology between Rap and Bud1, a small GTPase that controls the site of bud formation in yeast.

Receptors and scaffold proteins. Despite their small size, dendritic spines contain an amazing variety of surface receptors, scaffold proteins and signalling molecules, many of which are specifically enriched in spines. Several of these molecules have been implicated in spine morphogenesis. The cell-surface HEPARAN-SULPHATE proteoglycan syndecan 2 is concentrated in the PSD and spines, and binds to cytoplasmic PDZ-containing proteins through its carboxyl terminus76,77. Overexpression of syndecan 2 in hippocampal neurons accelerated the maturation of spines, an effect that depended on an intact carboxyl terminus77. PDZ-domain-containing scaffold proteins that bind to the carboxyl terminus of syndecan 2 (Refs 76,78) presumably recruit a protein complex that promotes spine development.

PDZ-domain-containing scaffold proteins, such as PSD95 and Shank, are believed to organize the PSD and to represent a molecular interface between glutamate receptors in the synaptic membrane and the spine cytoskeleton79,80 (Fig. 2). Overexpression of PSD95, which binds directly to NMDA receptors, increased the number and size of spines in cultured neurons81. Overexpression of Shank, which links NMDA receptor and metabotropic glutamate receptor (mGluR) complexes through multiple protein interactions68,82, promoted the maturation of mushroom spines in developing neurons, and increased the size of spine heads in mature neurons without an effect on spine number83. The enlargement of spines by Shank was dependent on and cooperative with Homer, a protein that also binds to mGluRs and inositol-1,4,5-trisphosphate receptors (InsP3R) (Fig. 4). Indeed, Shank and Homer seem to mediate the recruitment of InsP3R (and presumably smooth endoplasmic reticulum) into dendritic spines83. Dominant-negative mutants of Shank reduced spine density, possibly by decreasing the stability of affected spines or by inhibiting spine formation. As with PSD95, postsynaptic overexpression of Shank/Homer caused a significant enhancement of presynaptic function in addition to spine enlargement81,83, emphasizing the close functional relationship between the two sides of the synapse. Although overexpression of PSD95 and Shank/Homer can enlarge spines and boost synaptic transmission, it remains to be seen whether these proteins direct the maturation and/or plasticity of dendritic spines in vivo.

Figure 4: Shank and Homer are targeted to spines and cause dendritic spine enlargement.
figure 4

A cultured hippocampal neuron overexpressing Shank and Homer shows the accumulation of Shank protein (stained in green) in the heads of dendritic spines, which are greatly enlarged.

Full size image

Concluding comments

Even in these early days of the molecular exploration of spines, it is clear that multiple structural proteins and signalling pathways are involved in spine morphogenesis. This is perhaps not surprising, given the complexity of spine structure and its dynamic regulation over both short and long timescales. In principle, molecular insights into spine formation should enable us to manipulate dendritic spines in vivo by genetic approaches, allowing us to approach the long-standing question: what are spines good for in terms of brain plasticity and behaviour? So far, however, the molecular mechanisms that have been characterized are not necessarily dedicated to dendritic spine regulation, so phenotypes of mouse knockouts will be hard to interpret as a specific consequence of spine alteration. In this context, it will be crucial to identify the primary factors that determine the formation of spines, either intrinsic (the set of 'spine-enabling' genes expressed in spiny neurons that distinguishes them from non-spiny neurons) or extrinsic (the putative secreted molecules that induce the emergence of filopodia/spines in response to synaptic activity). With the field appreciating more and more the dynamic behaviour of spines, it will be important to understand the relationship between the structural plasticity of spines, and the movements of molecules and membranes into and out of this postsynaptic compartment. As the technologies for studying such mechanisms and processes become increasingly sophisticated, the next few years promise to be particularly exciting for the many neuroscientists who are interested in dendritic spines.