Among the largest cells in the body, neurons possess an immense surface area and intricate geometry that poses many unique cell biological challenges. This morphological complexity is critical for neural circuit formation and enables neurons to compartmentalize cell–cell communication and local intracellular signalling to a degree that surpasses other cell types. The adaptive plastic properties of neurons, synapses and circuits have been classically studied by measurement of electrophysiological properties, ionic conductances and excitability. Over the last 15 years, the field of synaptic and neural electrophysiology has collided with neuronal cell biology to produce a more integrated understanding of how these remarkable highly differentiated cells utilize common eukaryotic cellular machinery to decode, integrate and propagate signals in the nervous system. The present article gives a very brief and personal overview of the organelles and trafficking machinery of neuronal dendrites and their role in dendritic and synaptic plasticity.
A major challenge for all cells is to traffic integral membrane proteins, secreted factors and lipids to appropriate subcellular domains in the right amount at the right time. With surface areas up to 10000-fold greater than typical mammalian cells, and dendrites and axons that extend for hundreds of microns to metres from the cell body, neurons are at the pinnacle of complexity among metazoan cells. Pathways for membrane traffic originally emerged in simpler and smaller cells, and a central question in neuronal cell biology is how such evolutionarily conserved pathways became specialized to subserve the unique dimensions and segregated signalling that is the hallmark of neurons. Although intracellular organelles for membrane trafficking were observed in neurons more than a century ago , the mechanisms governing membrane transport to local dendritic domains and the role such transport plays in neuronal development, signalling, plasticity and morphology have emerged only recently.
Most neurons possess highly branched dendrites contacted by hundreds of synapses. This complex geometry allows for spatial segregation and integration of diverse signals , and it is well established that the functional properties of dendrites are highly compartmentalized. Individual dendritic segments possess distinct molecular composition, synapatic inputs and electrophysiological properties [2–5]. The partitioning of dendrites into functionally distinct segments requires exquisite control over receptor and ion channel distribution [6,7], in large part by control over local cargo transport to and from the dendritic membrane. In the present article, I review a select set of work that has revealed the presence of specific organelles and trafficking machinery in neuronal dendrites that mediate dendritic growth and synapse plasticity. I concentrate on the organelles of the early secretory pathway, endocytic pathway and exocytosis machinery that together regulate the molecular composition of dendritic domains and the function of synapses. I discuss the relationship between dendritic trafficking and forms of neuronal plasticity involved in learning, memory and brain development.
Early secretory trafficking in dendrites
As in non-neuronal cells, the neuronal ER (endoplasmic reticulum) is distributed in the cytoplasm and extends throughout dendrites and the soma [8–11] where it functions both as an intracellular Ca2+ signalling compartment  and as the site of membrane protein and lipid synthesis . Although SER (smooth ER) appears most abundant in dendrites, ER-bound ribosomes and translocon subunits are likewise found in distal dendrites [8,11,14,15], indicating the presence of RER (rough ER). The latter is significant, as local dendritic synthesis of membrane proteins such as neurotransmitter receptors has been proposed to mediate spatially delimited modification of dendritic segments [16–18], but, unlike the case for cytosolic proteins, such local synthesis and secretory processing requires a local complement of secretory organelles.
Progression of cargo through the secretory pathway requires proper folding of newly synthesized proteins in the ER and concomitant post-translational modifications, including N-glycosylation and disulfide bond formation. Following quality control checks, cargo concentrates at specialized ERESs (ER exit sites)  present on the ER membrane throughout the neuronal soma and dendrites [20,21]. Cargo emerges from the ER as COPII (coatamer protein II)-coated vesicles which then merge with ERGICs (ER–Golgi intermediate compartments)  that are also widely distributed throughout dendrites [9,10,23]. ER exit is typically the rate-limiting step for the biosynthesis of integral membrane proteins, including critical postsynaptic neurotransmitter receptors such as AMPA (α-amino-3-hydroxy-5-methylisoxazole-4-propionic acid)-type glutamate receptors [24,25] and NMDA (N-methyl-D-aspartate)-type glutamate receptors, which contain ER retention and ER export signals that regulate surface expression [26–31]. Furthermore, by up-regulating surface expression, chaperoning and ER exit has been proposed as a mechanism for the long-term action of drugs on their pharmacological targets, including the CNS (central nervous system) actions of nicotine on neuronal nicotinic acetylcholine receptors [32–35].
As secretory cargo progresses through the cis, medial and trans compartments of the GA (Golgi apparatus), further glycosylation and proteolysis occurs . Such modification requires multiple enzymes whose localization and enzymatic activity in the GA is regulated by the low intraluminal pH of the Golgi lumen . Indeed, loss of GA acidification disrupts GA morphology [38,39], impairs glycosylation and processing of lipids and protein [39–42], and alters cargo sorting [43,44]. Recent work has shown that the intraluminal pH of the GA is regulated by the ubiquitin ligase Ube3a , whose expression in the brain is required for dendrite growth , spine formation [47,48] and synaptic plasticity [48–50]. Accordingly, brain-specific loss of Ube3a occurs in the severe neurodevelopmental disorder Angelman syndrome .
In the absence of Ube3a, neurons display dramatically distended Golgi cisternae because of marked under-acidification of the GA . Elevation of Golgi pH leads to cisternal swelling [39,52–55] and disrupted protein glycosylation [54,56–61]. Regarding the latter, the enzymatic activity, trans-Golgi accumulation and heteromeric assembly of sialyltransferases are highly pH-sensitive [59,61,62]. Correspondingly, surface sialylation is significantly reduced in Ube3a-deficient cells and in the brains of Angelman syndrome model mice . Sialylation modulates interactions among proteins and lipids, thereby regulating cellular adhesion and cell signalling [63–65], including regulation of nervous system development and function [63,66–71]. Golgi ion homoeostasis and maintenance of pH-dependent functions of the neuronal GA, including sialylation, may thus contribute to dendritic growth and neuronal plasticity. Indeed, Golgi dysfunction and sialylation deficiency is associated with disorders of brain development and function including language impairment  and epilepsy .
In most mammalian cells, the ER, ERESs and ERGIC elements are distributed throughout the cell, whereas the GA is situated in the perinuclear area close to the MTOC (microtubule-organizing centre) . This distribution of intracellular compartments necessitates an inward then outward directionality to pre- and post-Golgi trafficking respectively. This arrangement is quite different in neurons as the neuronal GA consists of both classic perinuclear Golgi stacks and discrete structures termed GOs (Golgi outposts) dispersed in select dendrites [9,10,20,75–79]. The monitoring of early secretory cargo emerging from the dendritic ER has shown that many post-ER carriers traffic from the dendrite to the somatic GA, whereas a subset of post-ER carriers utilize dendritic GOs [20,79]. Both modes of early secretory trafficking are biologically notable: the former for the long-range transport of post-ER carriers to reach the GA, and the latter for the use of non-canonical Golgi compartments separated spatially from the somatic GA (Figure 1).
ER–Golgi trafficking in neuronal dendrites
In dendrites, GOs have been defined by classic cisternal stack ultrastructure visible by EM as well as by immunoreactivity for markers such as GM130 (cis-Golgi matrix protein of 130 kDa) [9,20,77]. However, not all dendrites contain morphological or molecular markers of the GA [9,10,20,77]. In the absence of bona fide GOs, how is early secretory cargo trafficked and processed to reach the PM (plasma membrane)? One possibility is long-range transport of post-ER carriers to the soma . Another possibility is that local dendritic trafficking of nascent secretory cargo bypasses classical Golgi , perhaps utilizing distinct post-ER carriers . In PC12 cells, Rab1-containing ERGICs and ERESs, but not Golgi membranes, are selectively distributed to the neurites and growth cones of polarized cells , indicating that satellite secretory systems devoid of generic Golgi membranes are formed during neuritogenesis. One possibility is that dendrites utilize ERGIC rather than classical GA for cargo maturation before local PM delivery (Figure 1). Given the known impact of Golgi-associated glycosylation on ion channel distribution , stability  and biophysical properties , it is tempting to speculate that bypassing the GA may diversify the properties of dendritically synthesized neurotransmitter receptors and voltage-gated ion channels, and hence affect dendritic excitability. At this point, it is not known whether or how cargo carriers emerging from the dendritic ER are selected for long-range or local trafficking to subsequent secretory organelles.
Before exiting the ER in discrete carriers, newly translated polypeptides reside in the lipid bilayer of the ER membrane. The ER itself exists as a continuous anastomosing network in dendrites [8,11,85,86]. For numerous synaptic receptors, ER export is rate-limiting for PM delivery, and multisubunit channels and receptors can dwell in the ER for various periods up to several hours [24,87–91]. This suggests that lateral diffusion within the ER determines the ultimate location of ER exit and hence the spatial range of new cargo transport to the PM . Consistent with this notion, quantitative photobleaching, super-resolution imaging and ultrastructural analyses indicate that newly synthesized membrane proteins, including AMPA receptors, rapidly diffuse within the continuous network of the dendritic ER . Nascent cargo diffusion is inhomogeneous in dendrites, with zones of diffusional confinement corresponding to local regions of increased ER complexity at dendritic branch points, at dendritic spines, and near GOs  (Figure 2).
Local zones of ER complexity confine secretory cargo and direct dendrite growth
In non-neuronal cells, ER morphology is controlled by diverse mechanisms including atlastins [92,93], reticulons  and Rab10 . In dendrites, the spatial range of receptor lateral mobility in the ER becomes progressively limited over neuronal development and is rapidly restricted by group I mGluR (metabotropic glutamate receptor) signalling through a mechanism involving PKC (protein kinase C) and the microtubule-binding ER protein CLIMP63 (cytoskeleton-linking membrane protein of 63 kDa)  (Figure 2B). CLIMP63 is an ER integral membrane protein whose binding to microtubules is regulated by PKC-dependent phosphorylation [96–98]. When bound to microtubules, CLIMP63 promotes elongation of ER tubules to the cell periphery along microtubule tracts [96,98]. Upon mGluR–PKC activation, site-specific phosphorylation of one or more N-terminal serine residues disrupts CLIMP63 microtubule binding, and thus increases ER complexity, thereby trapping nascent membrane proteins within smaller domains of diffusional confinement. Manipulations that increase ER complexity promote AMPA receptor surface expression and increase synaptic strength , perhaps by directing a higher proportion of newly synthesized receptors for local forward trafficking. Together, these findings suggest tight coupling between the local exit of ER cargo and insertion at the PM.
Post-Golgi trafficking and asymmetric dendrite growth
The secretory pathway is the primary site of lipid biosynthesis and membrane addition and thus contributes to polarized cell growth. In neurons, secretory trafficking is required for dendrite growth and maintenance. Disrupting Golgi function with BFA (brefeldin A) or by overexpressing inhibitory mutants of PKD (protein kinase D) or Arf1 (ADP-ribosylation factor 1) to block post- and intra-Golgi trafficking respectively blocks dendrite growth  (Figure 3). Moreover, such manipulations cause a dramatic simplification of established dendrites within hours, indicating a large-scale requirement for ongoing secretory trafficking in maintaining dendritic morphology . In Drosophila, mutations in genes encoding the critical early secretory trafficking proteins Rab1, Sar1 and Sec23 were identified in a screen for dendritic morphology defects, and the dendritic defects seen in these mutants correlated with disrupted Golgi membranes . In mammalian cortical and hippocampal pyramidal neurons, the somatic Golgi is oriented towards the large apical dendrite, and post-Golgi cargo flux is strongly biased towards the large apical dendrites . Golgi polarity precedes dendrite polarity and predicts the formation of the longest dendrites [77,99]. Fragmentation and dispersion of the somatic Golgi by overexpressing GRASP65 (Golgi reassembly stacking protein of 65 kDa) prevents the specification of the apical dendrite without affecting overall total dendrite growth . Notably, in the reeler mutant mouse lacking reelin expression, neurons in the cerebral cortex display misoriented apical dendrites. The secreted factor reelin promotes translocation of the GA into developing apical dendrites via a mechanism requiring an LKB1/Stk25/GM130 signalling pathway , and the Cdc42/Rac1 guanine-nucleotide-exchange factor αPIX/Arhgef6 . Thus polarized post-Golgi trafficking determines the generation of an apical dendrite and sustains asymmetric dendrite growth (Figure 3).
Asymmetric dendritic growth is controlled by polarized secretory trafficking
Organelles of the secretory pathway concentrate at dendritic branch points. GOs are stably positioned at a subset of branch points, where they both receive and transmit secretory cargo . In Drosophila sensory neurons, focal laser disruption of cellular domains containing GOs arrests local branch extension and retraction . Even earlier in the secretory pathway, local zones of ER complexity and ER export localize to dendritic branch points, and such zones predict the site of new dendritic branch formation . Manipulations that increase ER complexity promote dendritic branching, whereas simplification of the ER reduces dendritic branching , suggesting that local post-ER trafficking directs membrane cargo for dendrite formation and growth (Figures 2 and 3). Together, these observations indicate that dendrites contain satellite secretory systems that provide an elegant means by which cargo can be directed towards one dendritic branch or another. By spatially targeting the addition of lipid membrane and integral membrane proteins, such distributed biosynthetic machinery allows neurons to locally regulate dendritic morphology and molecular composition, ensuring compartmentalized function of dendritic segments.
Exocytosis at the dendritic PM
Following processing in the secretory pathway, intracellular cargo must be exocytosed at the dendritic PM to reach the surface of dendrites. Fusion of intracellular membrane stores with the PM governs cell-surface composition, regulates cell shape and allows the release of soluble factors. Whereas constitutive exocytosis maintains the complement of PM proteins and lipids, many forms of exocytosis are regulated by molecular or electrical stimuli. The most intensely studied form of regulated exocytosis is depolarization-induced neurotransmitter release at presynaptic terminals, for which the molecular machinery for vesicle docking and fusion has been delineated in exquisite detail [102,103]. From the number and positioning of voltage-gated Ca2+ channels [104,105], to the full molecular composition of synaptic vesicles , to the biophysics and dynamics of SNARE (soluble N-ethylmaleimide-sensitive fusion protein-attachment protein receptor)-dependent fusion , to the Ca2+ sensing of synaptotagmins , few subcellular events in all of cell biology have been defined in as much detail. In contrast, much less is known about exocytosis on the other side of the synapse, and it is only in the last 15 years or so that it has been recognized that exocytosis occurs from postsynaptic compartments. It has become increasingly clear that specific forms of dendritic exocytosis regulate synapse function, plasticity and neuronal morphology (Figure 4).
Organelles for exocytosis in dendrites
Dendritic release of neurotransmitters and neuropeptides
EM studies of several neuron types have indicated the presence of both classic and dense core secretory vesicles in dendrites that contain glutamate, GABA (γ-aminobutyric acid), dopamine and neuropeptides. In size, shape and clustered distribution at presumptive fusion sites, these vesicles bear close resemblance to presynaptic vesicles [109–115]. In brain areas such as the olfactory bulb, dendritic release sites are in contact with other dendrites that likewise contain clustered vesicles for neurotransmitter release. The reciprocal nature of these connections has been confirmed by electrophysiological analysis, demonstrating functional release of both excitatory and inhibitory neurotransmitters at dendrodendritic synapses, with a classic example being the synapse between granule and mitral cell dendrites in the main olfactory bulb [116–119].
In the olfactory bulb, dendritic release of GABA from granule cells requires intracellular Ca2+, similar to release from axons and suggesting the presence of Ca2+- regulated exocytic release machinery [116,120]. Whereas granule cell dendrites release GABA, contacting mitral cell dendrites release glutamate. Immunolabelling studies have shown that glutamatergic postsynaptic scaffolding proteins PSD (postsynaptic density)-93 and PSD-95 are present at mitral–granule cell synapses . Notably, activation of NMDA receptors is sufficient to trigger dendritic GABA release from granule cells [122–124] by coupling depolarizing NMDA receptor-induced current to activation of voltage-gated N- or P/Q-type Ca2+ channels . Much like in presynaptic terminals, fast Ca2+ chelators [e.g. BAPTA (1,2-bis-(o-aminophenoxy)ethane-N,N,N′,N′-tetra-acetic acid)] block GABA release from granule neurons, whereas slow Ca2+ chelators (e.g. EGTA) do not , consistent with tight coupling of Ca2+ entry to vesicle release machinery . Dendrodendritic synapses represent an exception to the classic features of neuronal polarity, but highlight the ability of dendrites to exhibit highly regulated spatially restricted exocytosis.
In midbrain dopamine neurons of the SN (substantia nigra) and VTA (ventral tegmental area), axonal release of dopamine occurs in the projection fields of the striatum, whereas release from dendrites occurs locally in the SN and VTA. Initial experiments measured [3H]dopamine release in the SN after KCl-induced depolarization . Microdialysis and voltammetry measurements in the SN and VTA confirmed local midbrain release of dopamine, presumably from dendrites [127–129]. Dopamine is also released from dissociated dopamine neurons in primary culture  and exogenous expression of VMAT2 (vesicular monoamine transporter 2) in cultured hippocampal neurons permits dendritic dopamine release from these non-dopaminergic cells . The precise mechanism of dendritic dopamine release has not yet been fully elucidated. Some experiments have suggested that dendritic release occurs by reversal of the dopamine transporter rather than vesicular fusion . However, other studies have found that SNARE protein-dependent membrane fusion is required for exocytosis of dendritic dopamine [130,133,134]. As release from axons and dendrites occurs on to different neurons in different brain areas, these modes of neurotransmitter release have very different functions, and may reflect release from different organelles. Indeed, dopamine has been observed in LDCVs (large dense-core vesicles), small synaptic vesicles and tubulovesicular structures resembling SER or endosomes [135,136].
Many neurons contain LDCVs that store neuropeptides which are released from dendrites. A well-studied example comes from magnocellular neurons of the supraoptic nucleus of the hypothalamus. These neurons have dendrites situated in and receiving input from the CNS, whereas their axons project into the peripheral hypophyseal portal circulation outside the blood–brain barrier. This unique anatomical arrangement means that axonal exocytosis confers peripheral release of neuropeptide which, for oxytocin, mediates uterine contractions and milk ejection associated with reproductive physiology, whereas central release from dendrites mediates adaptive social behaviours associated with reproductive and affiliative states. Presumptive fusion of LDCVs at the dendritic membrane of magnocellular neurons has been directly observed by EM . The dendritic release of oxytocin is Ca2+-dependent and can be triggered by NMDA receptor activation in the absence of action potentials . In granule neurons of the hippocampus, dynorphin peptides are exocytosed from dendrites where they retrogradely inhibit neurotransmitter release from perforant path terminals through presynaptic κ-opioid receptors . Whereas axonal release of dynorphin is blocked by inhibitors of N-type Ca2+ channels, dynorphin-mediated depression at perforant path synapses is blocked by antagonists of both N-type and L-type Ca2+ channels , once again suggesting the presence of distinct dendritic release machinery that responds to Ca2+ from different sources to mediate exocytic release.
Retrograde synaptic signalling by dendritic exocytosis
One of the most obvious functions of dendritic exocytosis is retrograde signalling to presynaptic terminals. Among such retrograde signals are released growth factors and neurotrophins including BDNF (brain-derived neurotrophic factor). Secreted BDNF controls a variety of cellular functions including synaptic transmission, synapse differentiation, morphological plasticity and gene transcription by binding and activating TrkB (tropomyosin receptor kinase B) receptors [141–145]. In addition to its well-known release from axon terminals [146–148], it has become clear that BDNF is also released from dendrites in an activity-dependent manner. When expressed in cultured hippocampal neurons, GFP-tagged BDNF localizes to punctate vesicular structures in dendrites [149–153]. Following electrical stimulation or K+-induced depolarization, BDNF–GFP puncta disappear over several seconds . Dendritic BDNF–GFP vesicle fusion requires CaMKIIα (Ca2+/calmodulin-dependent protein kinase type IIα)  and occurs upon dendritic depolarization by back-propagating action potentials . The specific vesicular structures containing BDNF are not well-defined, and the degree to which the sorting of endogenous BDNF differs from heterologous expressed BDNF–GFP remains an area in need of additional study. DCVs (dense-core vesicles) are thought to harbour BDNF at presynaptic terminals, but ultrastructural analysis reveals only sparse DCVs in the dendrites of many CNS neurons thought to release BDNF [85,150,152]. An alternative possibility is that BDNF is directly secreted to the PM in post-Golgi carriers or through a Golgi–endosome pathway.
In presynaptic terminals, synaptic vesicle fusion is triggered by the influx of Ca2+ ions, which bind to C2 domains on synaptotagmin 1, resulting in conformational rearrangement of core SNARE machinery that directly couples elevated Ca2+ to SNARE-mediated exocytosis . A distinct synaptotagmin family member, Syt4 (synaptotagmin IV), regulates activity-dependent release of BDNF from cultured hippocampal neurons and localizes to BDNF-containing vesicles in dendrites . Exocytosis of pHluroin-tagged Syt4 was visualized directly in hippocampal neuron dendrites and increased significantly upon depolarization . Neurons from Syt4-knockout mice display increased BDNF release relative to wild-type neurons, suggesting, perhaps counterintuitively, that Syt4 plays a negative role in postsynaptic exocytosis. Syt4 also regulates retrograde signal-mediated plasticity at the Drosophila neuromuscular synapse. Specifically, high-frequency stimulation of muscle cells increases the release probability of contacting presynaptic boutons. In the absence of Syt4, this form of retrograde plasticity is absent, but can be rescued by expression of Syt4 in muscle, suggesting that Ca2+ influx is coupled to postsynaptic vesicular trafficking via Syt4 [154,155]. Syt4 may thus form part of an evolutionarily conserved machinery for postsynaptic exocytosis.
Exocytosis in dendrite growth and morphological plasticity
In addition to the release of neurotransmitters and neuropeptides for interneuronal signalling, exocytosis contributes to the growth and morphogenesis of elaborate dendritic arbours. Originating as small roughly spherical cells upon terminal asymmetric division of neuronal precursors, neurons increase their total membrane area by many orders of magnitude, a process requiring huge amounts of membrane synthesis directed to growing dendrites and axons. As discussed above, disruption of the ER–Golgi secretory pathway in developing neurons prevents dendritic outgrowth in both mammals and fruitflies [77,78] (Figure 3). Interestingly, TeNT (tetanus toxin), a protease that cleaves most VAMPs (vesicle-associated membrane proteins), has limited impact on dendritic arbour development when added to cultured hippocampal neurons, even for days or weeks . Moreover, neurons from animals lacking VAMP2 or SNAP-25 [25 kDa SNAP (synaptosome-associated protein)] have normal morphology [157,158], raising the question of which SNARE proteins are mediating membrane fusion that contributes to dendrite growth. One possibility is that growth-related membrane fusion occurs via the toxin-insensitive VAMP family member Ti-VAMP (TeNT-insensitive VAMP)/VAMP7, which is required for neurite outgrowth in differentiating PC12 cells [159,160]. Although it is clear that local zones of ER complexity and ER export predict the site of new dendritic branch formation and areas of dendritic exocytosis , it is not yet known how spatial positioning of these early secretory compartments is coupled to specific exocytic machinery at the dendritic PM.
It is now clear that many forms of synaptic plasticity require exocytosis at or near the postsynaptic membrane (Figures 5 and 6). Initial studies found that postsynaptic infusion of various agents that disrupt SNARE-mediated membrane fusion via a recording pipette rapidly block LTP (long-term potentiation) at Schaffer collateral-CA1 synapses . Emerging from such observations has been a prevailing model where intradendritic vesicles harbouring AMPA receptors fuse with the PM upon LTP induction, and this increase in AMPA receptor number in turn increases synaptic strength [162,163]. In addition to such functional plasticity, postsynaptic exocytosis mediates morphological plasticity at glutamatergic synapses [164–168]. Specifically, dendritic spines increase their volume ~2-fold following stimuli that induce LTP . A variety of manipulations that prevent SNARE-mediated membrane fusion prevent LTP-induced spine growth [165,167,168], suggesting that spine exocytic events couple functional plasticity (more AMPA receptors), structural plasticity (more membrane and other molecules) and retrograde signalling (released factors).
Spine machinery for local exocytosis and endocytosis
Endosomal sorting of dendritic cargo
Membrane compartments for dendritic exocytosis
What is the identity of intracellular membrane stores involved in postsynaptic vesicle fusion? Serial EM reconstruction studies have demonstrated the presence of numerous membrane-bound organelles, including REs (recycling endosomes), in dendrites and dendritic spines [85,167]. These studies, together with the early observations that AMPA receptors are endocytosed and reinserted upon synaptic activation , suggested dendritic recycling endosomes as the primary internal membrane compartment mobilized to the PM in response to LTP-inducing stimuli [169–172] (Figures 5 and 7). Indeed, blocking RE transport using dominant-negative versions of Stx13 (syntaxin 13) and the Eps15 homology domain protein Rme1/EHD1 blocked LTP, providing direct evidence for RE involvement in synaptic plasticity . A noteworthy early observation was that stimuli that trigger synaptic potentiation not only increase the surface levels of AMPA receptors, but also increase the release of previously endocytosed transferrin, the classic recycling endosome cargo . This observation suggests that LTP-stimulated membrane fusion is not selective for specific cargo, such as specific subtypes of AMPA receptors, but instead may act through conserved endosomal machinery independent of receptor content present on the exocytosing vesicle. Such a model is consistent with recent observations identifying synaptic potentiation in response to LTP-inducing stimuli independent of specific receptor subtype . The promiscuity of LTP with respect to receptor subtypes suggests that signalling pathways may act on conserved cellular transport and exocytosis machinery rather than specific cargo molecules.
The first optical demonstration of activity-triggered exocytosis in dendrites relied on the styryl dye FM1-43 and showed destaining of FM1-43-labelled compartments in minutes upon neuronal stimulation [174,175]. More recently, exocytosis of cargo tagged with the pH-sensitive SEP (superecliptic pHluorin) has been visualized directly. SEP-labelled AMPA receptors have been used in several studies to visualize postsynaptic exocytosis [164,165,176–183]. To selectively visualize newly exocytosed SEP–GluA1 (glutamate A1 receptor), Petrini et al.  monitored SEP–GluA1 signal after bleaching a large dendritic region. Small dendritic regions at the boundary of the bleached region were repeatedly bleached to create an optical barrier to exclude fluorescent signal from SEP–GluA1 diffusing laterally in the PM. In this paradigm, approximately 20% of the SEP–GluA1 signal recovered over 20 min indicating constitutive endocytic cycling of AMPA receptors within dendritic segments . Whereas a chemical LTP stimulus of 0 Mg2+/glycine increases the frequency of SEP–GluA1 exocytic events , glutamate receptor blockers and TTX (tetrodotoxin) decreases the frequency of SEP–GluA1 exocytic events . Simultaneous imaging of SEP–GluA1 and the recycling endosome marker TfR (transferrin receptor)–mCh (mCherry) has shown the abrupt appearance of SEP–GluA1 signal in TfR–mCh-positive spines after stimulation with bicuculline/glycine . Newly inserted SEP–GluA1 either quickly diffused out of the spine (38% of events) or remained in the spine head (62% of events). In all cases, co-localized TfR–mCh signal diffused out of the spine immediately following the rapid burst of SEP–GluA1 fluorescence even if newly inserted SEP–GluA1 was retained near the site of spine fusion . Thus, although multiple classes of cargo can be delivered to the spine membrane by all-or-none endosomal fusion events, the fate of cargo after exocytosis can vary on the basis of the molecular identity of the cargo and the state of the synapse. Taken together, a suite of strong evidence supports the notion that paradigms of synaptic potentiation trigger exocytosis of AMPA receptors.
Exocytic domains and molecular machinery for cargo delivery to dendrites
Where is cargo exocytosed in relation to the postsynaptic specialization? In the case of SEP–GluA1, studies have reported exocytic events throughout the somatodendritic compartment, but not in dendritic spines proper [179,180,182]. SEP–GluA1 inserted at the subjacent dendritic shaft can diffuse into spines , and may become trapped at the PSD by scaffold interactions [184–186] or local endocytic recapture [183,187]. More localized activity manipulations such as two-photon glutamate uncaging have demonstrated exocytosis of SEP–GluA1 in dendrites near and within activated spines [180,181]. Notably, the amplitude of uEPSCs (uncaging-induced excitatory postsynaptic currents) increases first in spines and then in the adjacent dendritic shaft following LTP induction , consistent with insertion of glutamate receptors directly in activated spines.
REs containing AMPA receptors are mobilized by MyoVb for postsynaptic plasticity
One limitation of using a specific receptor cargo as an optical reporter is that the cargo molecule may occupy more than one defined intracellular membrane compartment. Visualization of a tagged fusion of the TfR (TfR–SEP), a classic RE-specific marker, showed that exocytosis of AMPA receptor-containing endosomes occurs within spines, immediately adjacent to the PSD . SEP–GluA1 cargo within TfR-positive endosomes is retained in spines for at least several minutes, whereas co-exocytosed TfR from the same endosome quickly diffuses out of the spine within seconds , demonstrating selective AMPA receptor retention at or near synapses [184–186]. Within spines, exocytosis of REs occurs at microdomains enriched for Stx4 (syntaxin 4) , one of the four members of the syntaxin family that localizes to the PM. Disruption of Stx4 blocks spine RE fusion and impairs LTP , indicating that Stx4 defines an exocytic domain in dendritic spines for synaptic plasticity and forms part of the core SNARE machinery for spine exocytosis (Figure 5).
The other SNARE proteins and adaptors that partner with Stx4 for spine exocytosis have not been determined. Interestingly, SNAP-25 participates in exocytosis of NMDA receptor in dendrites [188,189] and GluA1 insertion is sensitive to botulinum toxin A which selectively cleaves SNAP-25 . A different SNAP family member, SNAP-23 (23 kDa SNAP), is enriched in dendritic spines and localizes at or near the PSD , much like Stx4 , suggesting that SNAP-23 contributes to the exocytic machinery of dendrites with its known SNARE partner Stx4 . Furthermore, a VAMP family member is likely to be involved in spine exocytosis since postsynaptic infusion of botulinum toxin B or tetanus toxin blocks LTP [161,192]. As discussed above, the synaptotagmin family member Syt4 is an attractive candidate for participation in spine-localized SNARE-mediated exocytosis, although this has not been formally demonstrated. Finally, components of the exocyst have been found in spines, and disrupting Sec8 or Exo70 prevents surface trafficking of AMPA receptors, which instead accumulate in intracellular compartments within spines .
Before frank membrane fusion, REs are mobilized into spines by the actin-based motor protein MyoVb (MyoV is myosin V)  (Figure 7). MyoVb, and the related MyoVa and MyoVc, are Ca2+-sensitive motors whose three-dimensional conformation changes significantly with elevated Ca2+ [195,196], such that in low Ca2+ MyoVb adopts a closed conformation bound to actin, but disassociated from endosomes. As the Ca2+ concentration rises, MyoVb adopts a more extended conformation, allowing physical association with the endosomal adaptor protein Rab11-FIP2 (Rab11 family-interacting protein 2) [194–199]. Hence, upon NMDA receptor-induced Ca2+ influx, MyoVb is recruited to REs and mobilizes them into actin-rich spines where they are positioned to efficiently undergo fusion at Stx4-positive exocytic domains (Figures 5 and 7). In hippocampal slices, LTP is acutely blocked by selective chemical-genetic inhibition of MyoVb, demonstrating a requirement for MyoVb-based mobilization of REs for synaptic potentiation . Expression of a dominant-negative version of the related MyoVa or siRNA against MyoVa similarly blocks LTP , although synaptic transmission and plasticity was found to be normal in mice lacking MyoVa , suggesting differential effects of class V myosins.
In summary, it is now clear that exocytosis is a central aspect of the cell biology of dendrites and plays a critical role in both functional and structural plasticity of synapses. A combination of molecular, biochemical, imaging and electrophysiological approaches have provided a much more detailed view of the unique features, spatial regulation, protein machinery and function of exocytosis in dendrites and spines.
Postsynaptic endocytosis and endosomal transport
After delivery to specific surface domains of dendrites, the removal of dendritic membrane cargo relies on endocytosis. Endocytosis is, of course, the ubiquitous mechanism by which all cells internalize nutrients, remove membrane from the cell surface and control the level of cell-surface constituents. For many ion channels and postsynaptic receptors, endocytosis and post-endocytic trafficking in dendrites is a primary mechanism controlling their levels and location. In recent years, endocytosis and endosomal sorting of glutamate receptors have emerged as central mechanisms responsible for diverse forms of synaptic plasticity [169–172,183,187,202–207]. Endocytic machinery physically associates with protein complexes of the PSD at excitatory glutamatergic synapses, a finding that initially suggested that endocytosis takes place near the PSD. For example, the GluA2 (glutamate A2 receptor) AMPA receptor subunits and the GluN2B (glutamate N2B receptor) NMDA receptor subunit both interact with the AP-2 (adaptor protein 2) clathrin adaptor complex [207–209], and the PSD adaptor Homer directly binds dynamin-3 [187,210], a member of the dynamin family of large GTPase that mediates vesicle fission . Interactions between postsynaptic receptors and endocytic machinery are tightly regulated. GluA2 forms a Ca2+-dependent complex with AP-2 through the Ca2+-binding protein hippocalcin required for LTD (long-term depression) . On the other hand, AP-2 binding to GluN2B is prevented by the binding of PSD-95 to an adjacent domain of the receptor  as well as by tyrosine phosphorylation of GluN2B by Fyn . This phosphorylation-dependent stabilization can be reversed via tyrosine dephosphorylation by STEP (striatum-enriched phosphatase), which promotes the endocytosis of Glu2B-containing NMDA receptors  and is itself subject to complex regulation via phosphorylation, ubiquitination and proteolytic cleavage .
Initial evidence for dynamic glutamate receptor trafficking demonstrated that AMPA receptors undergo continual endocytosis and exocytosis [169–172,205,215,216]. Endocytosis of glutamate receptors occurs primarily through a dynamin- and clathrin-dependent pathway [171,172,187] and most dendritic spines contain the full assembly of clathrin coat proteins and adaptors that capture, remove and recycle glutamate receptors from the cell surface [183,187,217–219]. In dendritic spines, clathrin puncta are found adjacent to, but not overlapping, the PSD where clathrin coats repeatedly assemble and disassemble during ongoing endocytosis  (Figure 5). EM has shown the presence of coated structures representing all phases of endocytosis in dendritic spines [218,220], with immunolabelling for AP-2, clathrin and dynamin localized just lateral to the edge of the PSD . The stable lateral positioning of clathrin EZs (endocytic zones) in spines provides a mechanism to fine-tune glutamate receptor levels on a spine by spine basis.
How does the endocytic machinery target to spines and what does it do? Studies in hippocampal neurons has shown that dynamin-3 is selectively enriched at the postsynaptic specialization and is the only dynamin isoform known to directly bind to the PSD adaptor protein Homer [187,210,221–224]. Homer proteins assemble as tetramers and form extended protein networks with members of the Shank family of multidomain PSD scaffolding proteins [225–228]. Disrupting dynamin-3 or preventing its interaction with Homer uncouples the EZ from the PSD, leading to a decrease in the number of PSDs that possess an adjacent EZ . This spatial uncoupling of the EZ depletes synaptic AMPA receptors and decreases synaptic strength, supporting a model whereby the EZ ensures spine-localized endocytic cycling and serves as a trap to recapture and maintain the extrasynaptic pool of AMPA receptors, thereby limiting the escape of AMPA receptors from spines . Indeed, single-particle tracking and high-resolution live-cell imaging have indicated that both EZs and local receptor recycling maintain a mobile pool of AMPA receptors that is required for synaptic potentiation .
Endocytosis in dendrites is regulated by the activity-regulated gene product Arc/Arg3.1. Transcription of the Arc gene, transport of its mRNA into dendrites and local translation of Arc protein are all strongly up-regulated by neuronal activity [229,230]. Once up-regulated, Arc binds to the endocytic adaptor endophilin-2 and stimulates the endocytosis of dendritic and postsynaptic cargo, most notably AMPA receptors and the amyloid precursor protein [231,232] (Figure 5). This activity-induced internalization of AMPA receptors provides a homoeostatic mechanism to dampen excitatory synaptic transmission following bouts of strong neuronal activity . The precise mechanisms by which Arc stimulates endocytosis are not clear, but one possibility is that Arc directs subsets of cargo at specific endocytic sites for specific endosomal sorting pathways . Although Arc expression is limited to neurons, specialization of endocytic sites occurs for specific cargo and controls receptor desensitization and signalling in diverse cell types. For GPCRs (G-protein-coupled receptors), arrestin recruitment to activated receptors is well known to link cargo to clathrin-mediated endocytosis and downstream signalling [234,235]. Interestingly, dendrites possess specific modes of endocytic recycling for GPCRs that differ in spatial and kinetic features [236,237], and clathrin puncta themselves become increasingly stable on dendrites as neurons mature in culture .
Following endocytosis, dendritic cargo such as AMPA receptors and NMDA receptors are sorted in endosomes for either recycling or lysosomal degradation (Figure 6). For AMPA receptors, agonist binding alone leads to rapid down-regulation and lysosomal degradation, whereas activation of NMDA receptors initially causes AMPA receptor endocytosis that can be followed by subsequent recycling back to the cell surface . Such endosomal sorting corresponds with the phosphorylation state of the GluA1 subunit. Dephosphorylation of Ser845 of GluA1 occurs during LTD  and may be the sorting signal that leads to lysosomal degradation , whereas phosphorylation of GluA1 at Ser845 accompanies reinsertion at the dendritic membrane via REs [169,240,241]. Coincident with dephosphorylation of GluA1 at Ser845, LTD-inducing stimuli trigger Ca2+-dependent activation of Rab5 which leads to the endocytosis and early endosomal transport of GluA1 . In the case of NMDA receptors, internalized receptors are routed along degradative and recycling pathways on the basis of specific motifs in GluN1 (glutamate N1 receptor), GluN2A (glutamate N2A receptor) and GluN2B subunits [203,243].
Internal membrane compartments resembling endosomes have been observed in dendrites. These include coated and uncoated vesicles, tubules and MVBs (multivesicular bodies) [85,167]. Approximately 70% of endosome-like structures are within or at the base of dendritic spines, and approximately one-third of CA1 hippocampal spines are associated with endosomes, which suggests that the same endocytic organelles are shared by multiple spines . Initial studies showed that Stx13, an endosomal SNARE protein, is found in dendritic tubulovesicular compartments, where it co-localizes with TfR, a protein known to be recycled through the endosomal pathway . Similarly, Neep21 (neuron-enriched endosomal protein of 21 kDa) is a binding partner of Stx13 that localizes to and defines a population of early dendritic endosomes [245–247]. Disruption of either Stx13 or Neep21 decreases the recycling of GluA1 and GluA2 after endocytosis and impairs synaptic potentiation [166,248].
Whereas early and recycling endosomes are abundant in dendrites, degradative compartments such as late endosomes, MVBs and lysosomes are more sparse [85,169,249]. MVBs have been noted in association with postsynaptic sites [85,148,250–253]. Lysosomes generally concentrate in the soma and proximal dendrites of neurons [254,255], although ultrastructural and biochemical studies have long supported the presence of lysosomes and lysosomal degradation at synapses . In non-neuronal cells, the peripheral versus the perinuclear positioning of lysosomes is controlled by regulated association with kinesin and dynein motors [257–262], and destabilization of microtubules causes accumulation of lysosomes in dendrites [254,255], suggesting active microtubule-based transport from the somatodendritic compartment into axons. Interestingly, the proximity of lysosomes to early endosomes can determine the propensity of internalized cargo to transit recycling versus degradative pathways , suggesting that the presence or absence of lysosomes can control dendritic signalling. Indeed, internalized AMPA receptors can be degraded by lysosomes [169,204] and stimuli that induce LTD up-regulate autophagy and presumably subsequent autophagolysosome-mediated degradation  (Figure 6). Many neuronal membrane proteins are long-lived with half-lives of days to weeks . It remains an open question how endocytic cargo destined for destruction in distal dendrites is transported to lysosomes, or, conversely, how lysosomes are generated and distributed upon demand through the dendritic arbour. Given the important role of lysosomes in protein clearance, and their dysfunction in neurodegenerative disorders , the local regulation of dendritic lysosomes remains an important topic for future study.
Conclusions and future directions
It has been over 100 years since Golgi first described intracellular organelles in neurons . The last 15 years have witnessed a remarkable convergence between neuronal membrane trafficking, brain development and synaptic plasticity. Previously separated as classic disciplines of anatomy and neurophysiology, it is now clear that a full understanding of neural circuits in health and disease requires dissection of the subcellular and organellar mechanisms of neuronal trafficking. For, at their core, many questions of synapse and circuit plasticity are issues of cell biology. From the modification and maintenance of individual synapses, to the integrative properties of dendritic branches, to the distribution of channels and receptors that determines the computational features of neurons, emergent properties of cells and circuits rely on the remarkable ability of neurons to compartmentalize. As a result, neurons have adapted conserved cellular trafficking machinery for highly specialized functions.
Beyond a role in normal brain development and physiology, many of the most prevalent and devastating neurological disorders are now understood to be disorders of neuronal membrane trafficking. From the production, release and clearance of β-amyloid and tau in Alzheimer's disease, to the abnormal accumulation and transport of α-synuclein in Parkinson's disease, to the trafficking defects associated with abnormal huntingtin in Huntington's disease, core aspects of neurodegeneration impinge directly on the organelles for dendritic trafficking. In addition, it is now clear that the most common psychiatric disorders, including schizophrenia, autism, bipolar disorder and ADHD (attention-deficit hyperactivity disorder), originate during brain development where the growth and development of dendrites and synapses requires exquisite regulation of dendritic trafficking. Targeting these disorders with next-generation therapeutics will require a better understanding of how receptors and ion channels are trafficked, how synapses are modified and how deficits in neuronal trafficking can be corrected.
Many questions remain. What are the full set of molecular sensors that connect neural activity to dendritic membrane trafficking? What controls the biogenesis, localization and transport of dendritic organelles themselves? How do neurons maintain their size, shape and composition in the face of ongoing macromolecular turnover? How do genetic programmes that direct neuronal subtype differentiation impinge on trafficking machinery? How does local dendritic trafficking influence post-translational modification of signalling molecules, channels and receptors? What are the molecular mechanisms by which development, aging and disease alter dendritic transport? Although much work remains, the fusion of fields at the interface of membrane trafficking and neuronal plasticity continues to bud new insight into the inner workings of Nature's most complex cells.
Thudichum Medal Lecture Delivered at the University of Bath on 24 June 2013, as part of the 5th Conference on Advances in Molecular Mechanisms Underlying Neurological Disorders Focused MeetingMichael Ehlers
adaptor protein 2
ADP-ribosylation factor 1
brain-derived neurotrophic factor
cytoskeleton-linking membrane protein of 63 kDa
central nervous system
ER exit site
ER–Golgi intermediate compartment
glutamate A1 receptor
glutamate A2 receptor
glutamate N2B receptor
cis-Golgi matrix protein of 130 kDa
Golgi reassembly stacking protein of 65 kDa
large dense-core vesicle
metabotropic glutamate receptor
neuron-enriched endosomal protein of 21 kDa
protein kinase C
protein kinase D
Rab11 family-interacting protein 2
soluble N-ethylmaleimide-sensitive fusion protein-attachment protein receptor
vesicle-associated membrane protein
ventral tegmental area
I thank Cyril Hanus, Matthew Kennedy, Mikyoung Park and Zhiping Wang for assistance with the Figures. I thank Soheil Aghamohammedzadeh, John Allen, Cyril Hanus, Tom Helton, Juliet Hernandez and Matthew Kennedy for a critical review of the paper before submission. I thank all members of the Ehlers laboratory past and present who made receipt of the Thudichum Medal possible.
Work in my laboratory is supported by Pfizer, Inc.