|Home | About | Journals | Submit | Contact Us | Français|
Dynamin, the founding member of a family of dynamin-like GTPases (DLPs) implicated in membrane remodelling, has a critical role in endocytic membrane fission events. The use of complementary approaches, including live cell imaging, cell free-studies, X-ray crystallography and genetic studies in mice has greatly advanced our understanding of the mechanisms by which dynamin acts, its essential roles in cell physiology and the specific function of different dynamin isoforms. In addition, several connections between dynamin and human disease have also emerged that highlight specific contributions of this GTPase to the physiology of different tissues.
Endocytosis, the process through which cells internalize portions of their plasma membrane along with extracellular material, is fundamentally important in cell physiology. Endocytosis counterbalances the continuous delivery of membrane to the cell surface by exocytosis, regulates the plasma membrane abundance of proteins, controls the signalling output of receptors, mediates cellular uptake of nutrients and is also exploited by pathogens to enter cells. To support these various functions, multiple molecularly and morphologically distinct forms of endocytosis, both constitutive and regulated, operate in all cells 1-4.
Generation of an endocytic vesicle requires the recruitment of various proteins from the cytosol that orchestrate inward plasma membrane bending to form a deeply invaginated bud and subsequently promote its fission. One such protein that directly participates in the fission reaction is the GTPase dynamin, the founding member of a family of GTPases that have diverse roles in membrane remodelling events throughout the cell (Figure 1). The role of dynamin in endocytosis has now been investigated for more than 20 years 5-8. Considerable evidence indicates that dynamin assembles into helical polymers at the necks of budding vesicles and that its GTP-hydrolysis dependent conformational change promotes fission of the underlying tubular membrane to generate a free endocytic vesicle (Figure 2). However, several aspects of its function, its precise mechanisms of action and specific roles of multiple dynamin isoforms in mammalian cells have remained elusive. Furthermore, although multiple dynamin-dependent and dynamin-independent endocytic pathways have been described, we only have a good understanding of how dynamin is recruited and then acts at fission sites for clathrin-mediated endocytosis. In this review, we discuss how complementary experimental approaches have greatly increased our understanding of dynamin's action. We also highlight how this information has been further enhanced by new insights into the function and mechanitic action of dynamin-like proteins.
Dynamin was originally identified as a GTPase that co-purified with brain microtubules 6, 9. Its role in endocytosis was revealed only subsequently, when the mutations responsible for the temperature-sensitive paralytic phenotype of Drosophila melanogaster shibire mutants were mapped to the dynamin gene 7, 8. In these mutants, paralysis results from the neuronal activity-dependent depletion of synaptic vesicles that is accompanied by the accumulation of arrested ‘collared’ endocytic pits at the presynaptic plasma membrane 5 Subsequent studies revealed that similar collared pits formed at mammalian presynaptic plasma membranes upon exposure to GTPγS, a slowly hydrolyzable GTP analog, and that they contain oligomeric assemblies of dynamin 10. The assembly of helical dynamin polymers either in solution 11 or on membrane templates was additionally reconstituted with the purified protein 11 (Figure 2c). Further investigations in non-neuronal cells showed that dynamin is a general component of clathrin-coated endocytic pits and that its GTPase activity is important for endocytosis: expression of dynamin mutants with an impaired GTP binding (such as the commonly used K44A mutant) and/or hydrolysis cycle have dominant-negative effects on endocytosis 12-15.
Mammalian genomes contain 3 dynamin genes 16. The proteins encoded by these genes share the same domain organization and an overall 80% homology, but have distinct expression patterns. Dynamin 1 is selectively expressed at high levels in neurons and is generally not present in non-neuronal tissues 17, 18, although it can be detected in many cultured cell lines 19, 20. Dynamin 2 is expressed ubiquitously 17, 21. Dynamin 3 is found predominantly in the brain (at much lower levels than dynamin 1) and testis, and at lower levels in some tissues such as the lung 16, 17, 22. Dynamin diversity is compounded by the existence of multiple splice variants for each of the three dynamins 16.
Invertebrates such as C. elegans and D. melanogaster possess only a single dynamin gene 7, 23. So, the existence of three dynamin genes in mammals could in principle reflect differences between the housekeeping role of dynamin in clathrin-mediated endocytosis, mediated by dynamin 2, and specialized forms of endocytosis in cells that additionally express dynamin 1 and/or 3. In addition, this triplication of the dynamin gene during evolution may be partly explained by a need to fine tune overall dynamin levels in specific tissues. Different dynamin isoforms have some unique protein-protein interactions24-27. However, most differences between isoforms are quantitative rather than qualitative; these include affinities for SH3-domain-containing proteins 22, rates of GTPase activity, oligomerization efficiency and lipid binding properties28. Nonetheless, a recent report that differences in the lipid binding characteristics of dynamin 1 and 2 cause robust differences in their membrane fission activity 28 indicates that this remains an area of active research and additional studies are required to fully address the issue of redundancy between dynamin isoforms.
Dynamins are the founding members of a family of GTPases known as dynamin-like proteins (DLPs), which participate in membrane remodelling (Figure 1 and Table 1) and share basic features of domain organization (Box 1). They also have similar catalytic mechanisms and modes of action that differ from those of the Ras superfamily of regulatory GTPases. The GTPase domains (G domains) of DLPs undergo nucleotide-dependent dimerization and dimer-dependent GTP hydrolysis 29-32. G domain dimerization reciprocally induces structural rearrangements in the catalytic centre of the partner DLP which stimulate GTP hydrolysis 33. In contrast to small regulatory GTPases of the Ras family, DLP family GTPases have low affinity for nucleotides and do not require separate guanine nucleotide-exchange factors (GEFs) or guanine nucleotide-activating proteins (GAPs), as they load spontaneously with GTP and have intrinsic mechanisms for stabilizing the transition state during GTP hydrolysis 29. In the case of dynamin, this is thought to be supported by a sodium or potassium ion within the active site 30. In other DLPs, divergent mechanisms are used to stabilize the transition state. However, the lack of requirement for additional GAP proteins and the importance of G domain dimerization is conserved within the DLP family.
Furthermore, whereas the GTP binding and hydrolysis cycle of regulatory GTPases mediates recruitment and release of effector proteins, the cellular function of DLPs is tightly coupled to the dimerization of their G-domains and, at least for some of them, to their polymerization29, 31, 32, 34, 35. The conformational changes produced by GTP binding and hydrolysis in the G domain of DLPs is transduced into a movement relative to adjacent domains 36-38 that when propagated along the polymer is expected to produce a force on membranes 31, 32. Thus, for dynamin, the cycle of GTP loading and hydrolysis is intimately coupled to its membrane fission activity (Figure 2b). This explains why mutations in dynamin designed to lock it into a GTP-bound state do not result in constitutive activation of endocytosis 15, in contrast to the persistent activity exhibited by regulatory GTPases that harbour equivalent mutations.
On the basis of its primary sequence, dynamin, a cytosolic protein, has been typically described as comprising: an N-terminal GTPase or G domain; a ‘middle’ or ‘stalk’ region; a pleckstrin homology (PH) domain; a GED domain, so called because its interactions with the GTPase domain had suggested a function as a GTPase effector domain 39; and a proline-rich C-terminal region, typically referred to as the proline-rich domain (PRD, Figure 2a)16. More recently, information about the structure of dynamin and DLPs 30, 34, 36, 37, including the crystal structure of dimers of nearly full length dynamin (lacking only the PRD)31, 32 have allowed for a modified definition of dynamin domains that better reflects the predicted three-dimensional hairpin-like folding of the full-length protein (Figure 2a and b).
The G domain sits upon a helical bundle, also known as the bundle signalling element (BSE)30 or neck 40, which is formed by 3 helices derived from sequences at the N and C terminal sides of the G domain and from the C-terminal region of the GED domain, respectively (Figure 2). Consistent with the idea that this region of the GED domain is in close physical proximity with, and functionally linked to, the GTPase domain, a screen for suppressor mutations of a mutation in the G domain of Drosophila dynamin identified a mutation within the GED C-terminus 41. The BSE is followed by a stalk that is composed of helices from the middle domain and the N-terminal region of the GED 34, 42, 43; and a PH domain 44 that forms the vertex or ‘foot’ of the stalk hairpin and binds membranes. The PRD, expected to be unfolded, emerges at the boundary between the BSE and the G-domain, most likely projecting away from the membrane, where it might interact with other proteins.
The stalk of dynamin dimerizes in a cross-like fashion (Figure 2b) to yield a dynamin dimer in which the two G-domains are oriented in opposite directions 31, 32, 34. This dimer represents the basic dynamin unit, and is different from the additional dimerization interface that is generated by the interaction of two G domains, which will be discussed below.
The PH domain binds acidic phospholipids in the cytosolic leaflet of the plasma membrane and PI(4,5)P2 in particular via a positively charged surface at the foot of the dynamin hairpin 44, 45. PH domain mutants that impair phosphoinositide binding exert dominant-negative effects on clathrin-mediated endocytosis 46, 47. The binding between the isolated dynamin PH domain and phosphoinositides is of very modest affinity (>1mM) but this membrane interaction is strengthened by charge-dependent association with other negatively charged phospholipids and by avidity afforded by dynamin polymerization 45-48. A hydrophobic loop emerging from the PH domain may promote membrane interaction and may also have curvature generating or sensing properties 28, 49.
The PRD contains an array of PxxP amino acid motifs that interact with many SH3 domain containing proteins (Table 2) to localize dynamin at endocytic sites and coordinate dynamin's function with these other factors during endocytosis50-53. Accordingly, dynamin lacking the PRD cannot rescue endocytic defects in dynamin KO fibroblasts 19. The PRD-SH3 interactions are typically of moderate affinity but the presence of multiple SH3-binding motifs in the PRD and multiple SH3-domain-containing proteins at endocytic sites, and the polymeric state of these proteins results in a significant avidity effect that enhances the ability of such interactions to concentrate dynamin. At least some interactions of the dynamin PRD are regulated by phosphorylation 50.
Purified dynamin spontaneously polymerizes into rings and helices when incubated in low ionic strength solutions 11 or in the presence of narrow negatively charged tubular templates (such as membrane tubules, microtubules or actin bundles) 6, 54-56. It can also tubulate membrane bilayers under appropriate conditions by forming a continuous membrane coat around them 10, 57-59. Dynamin polymerization resulting from the side-by-side apposition of dimers via stalk tip interactions (Figure 2b, interfaces 1 and 3) occurs an angle that determines the diameter of the ring 31, 32. The tetrameric form of dynamin, which can be abundant in solution 60, may represent an intermediate in higher order assembly 34, 38. Thus, the stalks form the core of the ring, whereas the BSE and G-domains of each dimer project towards the adjacent rungs of the dynamin helix 31, 32 (Figure 2b). It follows that G-domain dimerization, which is critical for GTP hydrolysis and for dynamin function, can only occur across adjacent rungs (Figure 2b), which explains the robust stimulation of dynamin's GTPase activity upon polymerization 56. A similar overall domain organization and mode of polymerization is predicted to be shared by all DLPs implicated in membrane fission 61. However, the module that binds the membrane in these other proteins is divergent from dynamin and the PRD is replaced by sequences with different binding properties. Such divergence likely reflects the different mechanisms that recruit different DLPs to membranes and may speak against a key role of the PRD and the PH domain in the mechanics of fission.
The mechanisms by which dynamin drives membrane fission have been the subject of intense debate, and have been analysed in living cells 12, 15, 46, 56, broken cell preparations 10, 62 and minimal systems based on purified dynamin and artificial lipid bilayers 57, 58, 63-65. Models derived from these experiments have suggested that once dynamin has polymerized around the neck of an endocytic bud (or a tubular membrane template in vitro), its GTP-hydrolysis-dependent structural reorganization triggers constriction (twistase and/or constrictase models) or stretching (poppase model) to promote membrane fission (Fig. 2b). Importantly, cell-free studies have shown that purified dynamin alone can cut synthetic lipid tubules in the presence of GTP 57, 58, 63, 65. Thus, while other factors may assist the action of dynamin in vivo, dynamin is sufficient to mediate membrane fission.
Insights obtained from recent crystallographic and cryo-EM studies of dynamin and DLPs have offered important new clues about how dynamin works. First, the demonstration that the G domain of dynamin belongs to the superfamily of GTPases that undergo GTP-dependent homodimerization (GADs) 29 strongly argues against the possibility that GTP-bound dynamin may function by recruiting another effector 39. Second, as G domain dimerization is critical for GTPase activity 29, 30, 61, 63, 66 and requires interactions between adjacent rungs of the dynamin helix (Figure 2b) 30, dynamin's action requires at minimum a polymer that wraps around a membrane template to connect to the next rung. Third, studies of dynamin and of the DLP atlastin suggest that GTP hydrolysis by a G domain dimer leads to a prominent lever-like movement of the adjacent domain, the BSE in the case of dynamin 36, 37,38. Such a movement could constrict the dynamin helix when propagated along its subunits. As this movement is triggered by the interaction of G domains on adjacent rungs of a helix, it could produce a coordinated rotational sliding of one rung on the next, which is consistent with the reported twisting of dynamin-coated tubules during GTP hydrolysis 57. However, in principle, such movement could also simply constrict the tubule by producing a structural change in the ring without any rotation43.
Surprisingly, however, if a dynamin helix has been assembled in the absence of hydrolysable GTP, subsequent GTP loading and hydrolysis does not necessarily trigger fission, possibly because a long dynamin coat can function as a stabilizing scaffold 57, 63, 64. In cell-free systems, efficient fission of lipid tubules occurs optimally under conditions where dynamin polymerization occurs in the presence of GTP, allowing rapid coupling between dynamin assembly, GTP hydrolysis and polymer disassembly 58, 63, 65. Under these conditions, fission is proposed to result from the membrane destabilization that is predicted to occur when the constricted ring disassembles (Figure 2b). A long continuous dynamin helix may be less efficient in promoting fission because it lacks a constriction focal point or has less efficient subunit dissociation. In fact, taking into consideration bulk cytoplasmic GTP concentrations of ~2 mM versus the binding affinity of dynamin for GTP (Kd=~1μM) 67, dynamin oligomerizing into helices should be in a predominantly GTP-bound state and be poised to hydrolyse GTP and disassemble quickly once G domains of one turn can dimerize with the G domains of the next turn. Under some physiological conditions, local regulation of GTP levels probably contributes to proper dynamin function as mutations in the nucleoside diphosphokinase gene, which helps maintain such levels, make synapses more sensitive to dynamin mutations 68.
Although there have been important advances in understanding dynamin structure, we have much to learn about how conformational changes produced by the GTP hydrolysis cycle of dynamin relate to dynamin-mediated membrane fission. For example, the relative contributions of G domain dimerization versus subsequent GTP hydrolysis in the constriction and fission of membranes remain undefined. Furthermore, GTP hydrolysis may not only produce a powerstoke but also contribute to membrane destabilization by inducing helix disassembly..
Although dynamin alone can constrict and cut lipid tubules, the constriction of the dynamin ring in vivo is likely to be assisted or regulated by other factors to achieve fission (Figure 3). These include: membrane tension generated by actin, either via its polymerization or via myosin motors 57, 63, 69, 70; a heterogeneous distribution of lipids on the two sides of the constriction contributing to line tension 71; enzymatic degradation of PI(4,5)P2 leading to catastrophic dissociation of dynamin and other endocytic factors after GTP hydrolysis 72; or the partial destabilization of the bilayer by neighbouring lipid-binding proteins, such as proteins containing BAR domains 73. The potential role of BAR domain proteins has acquired interest with the realization that dynamin GTPase activity involves interactions between dynamin rungs, because the intercalation of BAR domains between rungs could hinder such an interaction 73-75, and thus potentially inhibit dynamin's action. Cryo-EM studies aimed at elucidating the precise organizational relationship between dynamin rings and BAR domain proteins should help to address this question.
Dynamin may contribute to multiple forms of endocytosis, but its action is better understood in the context of clathrin-mediated endocytosis 12-14, 17, 19, 20, 22, 69, 76 (Figure 3). During formation of endocytic buds (Figure 3), a subset of scaffold proteins (such as FCHO1, Eps15 and intersectin)69, 77 and clathrin adaptors (for example the AP-2 complex, AP180/CALM and epsin) are first recruited to the PI(4,5)P2-rich plasma membrane, coincident with the binding of some of these proteins to endocytic sorting motifs of integral membrane proteins1, 78. Such components cluster cargo, induce membrane curvature and also have actin nucleating properties 1, 78. The coat subsequently grows through the assembly of the clathrin lattice, which, through positive feedback, recruits additional cargo adaptors and endocytic factors 1, 78, 79. Dynamin also slowly accumulates around the growing pit69. Deep invagination of the bud and formation of a narrow neck, which is often assisted by a burst of actin polymerization19, 80, involves the recruitment of BAR-domain-containing proteins, several of which bind dynamin as well as the PI(4,5)P2 phosphatase synaptojanin (Box 2, Figure 2 b, c, d, f and g) 19, 74, 75, 81. It is at this point that dynamin rapidly accumulates at bud necks to mediate fission (Figure 3a and c)69, 82.
The enrichment of dynamin at clathrin-coated endocytic pits has been extensively demonstrated by fluorescence microscopy12, 53, 69, 82 including super-resolution methods 62 and electron microscopy 10, 83. Furthermore, live imaging of cells expressing fluorescent dynamin shows that dynamin levels at the pits are low during early stages of clathrin-coated pit maturation (Figure 2 a and c), most likely reflecting low affinity interactions with other endocytic factors andrise markedly during later stages - reflecting polymerization at the bud neck - to reach a peak that coincides with membrane fission (Figure 3c) 69, 82, 84. Factors that cooperatively, but redundantly, regulate dynamin assembly here include: PI(4,5)P2 and other acidic phospholipids present in the plasma membrane 85, 86; the tubular membrane template at the bud neck55; proteins containing dynamin binding SH3 domains (primarily BAR domain containing proteins 74, 75 whose polymerization forms a powerful high avidity affinity matrix 73, see Table 2); and actin, which can directly bind dynamin 87 (Figure 3). The coinciding accumulation of BAR proteins, actin and dynamin at late stages of endocytosis 69, 82 is particularly striking (Figure 3c), consistent with feed-forward mechanisms.
The sequence of events underlying clathrin-mediated endocytosis is evolutionarily conserved, as this process shows mechanistic similarity to endocytosis at actin patches in yeast 88, 89. Surprisingly, however, there are mixed reports as to whether the closest yeast homologue of dynamin, Vps1, does 90, 91 or does not 88, 92 participate in actin patch endocytosis, questioning how essential dynamin is for clathrin-mediated endocytosis. It is possible that unique properties of the yeast plasma membrane, for example tension and/or the presence of a cell wall, explains this difference93.
Endocytosis can also occur without clathrin, and dynamin has been implicated in some clathrin-independent endocytic pathways (Figure 1). For example, dynamin has been implicated in caveolin-dependent internalization and both dynamin and actin were transiently detected at caveolae, coincident with their fission from the plasma membrane 94, 95. However, the dynamics and function of dynamin at caveolae remains largely undefined. Surprisingly, the absence of dynamin is accompanied by a loss of caveolae and of their core component caveolin1 rather than by the increase in caveolae abundance that might be expected if dynamin were required for their fission19. Thus, it is not yet clear whether dynamin has a key role during endocytic fission of caveolae.
With respect to clathrin- and caveolin-independent endocytic pathways, evidence for a role of dynamin is in many cases primarily dependent on dominant-negative mutant approaches (which may exert indirect effects) or pharmacological inhibition with drugs (which may have off-target or non-physiological effects; Box 3). Thus, definitive answers about how dynamin affects such pathways will require a combination of methods and should include loss of function studies. As robust bulk endocytosis still occurs in fibroblasts lacking dynamin 19 and in the presence of dominant-negative dynamin mutants 12, dynamin cannot have a universal role in endocytic vesicle fission.
Vesicle budding, and thus membrane fission, occurs at multiple points throughout the secretory and endocytic pathways. Although dynamin does not affect most of these events, several experimental approaches that it might contribute to the fission of clathrin-coated vesicles that bud from the trans-Golgi network (TGN) 96 as well as of other vesicles that bud from either the Golgi complex or endosomes 20, 97-99. The yeast dynamin homologue Vps1 has also been implicated in retrograde traffic between endosomes and the Golgi complex100. However, there was no obvious accumulation of clathrin-coated pits at the trans-Golgi network (TGN) of fibroblasts lacking dynamin19, suggesting that dynamin is not universally required for clathrin-mediated budding. It is also important to consider that some defects observed at non-endocytic sites when dynamin is inhibited may arise indirectly through clathrin sequestration at the cell surface, perturbation of membrane traffic, effects on signalling pathways and/or cytoskeletal dynamics.
Dynamin interacts both directly and indirectly with the cytoskeleton. The relation between these interactions and the endocytic function of dynamin represents one of the most interesting and poorly understood aspects of dynamin biology.
A major property of dynamin, and one likely to be fundamentally important to its action, is its link to the actin cytoskeleton. A link between dynamin and actin at endocytic sites is not surprising, consistent with the importance of actin for many endocytic events, including at least a subset of clathrin-mediated endocytosis19, 69, 70, 101, 102. However, immunofluorescence studies and live cell analysis show that dynamin also colocalizes prominently with actin meshworks that are nucleated by the actin regulators N-WASP/WAVE and Arp2/3 at several other sites, including lamellipodia 103, membrane ruffles (including circular dorsal ruffles 104 and ruffles that mediate phagocytosis 105), invadopodia 106, podosomes 107, actin comets 108, 109 and the actin pedestals that can be associated with bacterial cell entry110 (Figure 1 and and3).3). At these sites, dynamin is not restricted to the membrane interface, but is present in the actin meshwork itself (Figure 1). This colocalization is consistent with the PRD-mediated binding of dynamin to numerous actin regulatory proteins, including proteins that contain Cdc42 interacting or regulatory domains (summarized in Table 2). Indeed, dynamin can bundle around actin filaments assembled in the presence of its partner cortactin, which can simultaneously bind F-actin54, 104. The stalk region of dynamin itself can also bind F-actin directly87. This raises the question of whether dynamin can directly influence actin dynamics and may have a role in actin function independent of endocytosis. However, as membrane remodelling is the shared property of dynamin and other DLPs, and is thus likely to be the most critical function of dynamin, links between dynamin and actin may predominantly reflect its role in membrane remodelling during endocytosis. Studies in yeast emphasize the critical role of actin in endocytosis, further supporting this possibility 88, 92, 111.
The interaction of dynamin with actin and its regulatory proteins may help position dynamin at endocytic sites (Figures 1 and 3a,b). For example, a role of actin in directing the localization of dynamin at endocytic sites is supported by studies of mammalian cells that lack dynamin, which have placed F-actin upstream of dynamin at clathrin coated pits (Figure 2d and g)19.. Actin may also generate membrane tension at the neck of endocytic buds, thus synergizing with dynamin during fission. Conversely, perturbation of dynamin function in living cells (by the expression of mutant dynamin or via pharmacological inhibition) can perturn actin dynamics87, 103, 108, 109, 112, 113. However, in cells that lack dynamin, stress fibers do not seem to be affected19. The most obvious change observed in these cells is the accumulation of Arp2/3-dependent actin meshworks around the necks of arrested endocytic clathrin-coated pits (Figure 3) 19. There is also evidence of alterations to other Arp2/3-dependent actin-based structures in these cells, such as circular membrane ruffles and invadopodia/podosomes (Shawn M. Ferguson, Olivier Destaing, Roland Baron and Pietro De Camilli, unpublished results) 106, 107, 113. However, it is difficult to discriminate between direct and indirect effects in these experiments, as disrupted endocytosis may alter signalling pathways (see below).
The recent crystallographic studies of dynamin 30-32 have raised new questions about potential non-endocytic effects of dynamin on actin dynamics. The helical organization of dynamin polymers at the bud neck appears to be optimally suited to allow an interaction in trans between dynamin G domains to promote GTP-hydrolysis (Figure 2B. However, purified dynamin can also polymerize into helices around actin bundles and microtubules. It will be important to determine whether such a transient helical arrangement occurs physiologically at non-endocytic sites; these structures would be expected to produce detectable bursts of fluorescence in dynamin-GFP-expressing cells ) but have not yet been observed.
Although dynamin interacts with microtubules in vitro 6, dynamin only shows strong colocalization with microtubules in cells when its PH domain is mutated 114. Moreover, disruption of dynamin function or lack of dynamin both result in a robust increase in the level of acetylated tubulin 19, 114, a modification associated with stable microtubules. This effect may be directly related to the ability of dynamin to bind microtubules, or to altered microtubule dynamics that result from alterations in the structure and functional state of the cell cortex (for example the great abundance of endocytic clathrin coated pits and of a robust accumulation of actin around them), where the stability of the “plus” end of microtubules is regulated.
Dynamin 2 concentrates at sites of abscission and was implicated in the completion of cytokinesis 115-117. Given the role of both microtubules as well as membrane traffic during abscission 118-120 and the connections of dynamin to each of these pathways, the precise role of dynamin in cytokinesis remains unclear. Interestingly, dynamin 2 is phosphorylated during mitosis by Cyclin dependent kinase 1 (Cdk1), and its dephosphorylation by calcineurin at these sites is required during cytokinesis 116. This phosphorylation cycle of dynamin 2 is reminiscent of the phosphorylation cycle of dynamin 1 in nerve terminals, which is mediated by cyclin-dependent kinase (Cdk5) and calcineurin (Figure 4d) 121. The broad use of this signaling pathway may be indicative of its fundamental importance. Indeed, this extends to Cdk1 phosphorylation of DRP1 that promotes mitochondrial fission during mitosis122.
As clathrin-mediated endocytosis is the major pathway of receptor internalization, dynamin regulates the plasma membrane abundance, and thus signalling output, of diverse receptors including growth factor receptors, GPCRs, ionotropic neurotransmitter receptors, ion channels, cell adhesion proteins and signalling proteins such as Notch 123-126. This was first emphasized by the major neurogenic developmental effects of the Drosophila shibire mutation 127 that arise from inhibition of Notch signaling (a phenotype also observed when dynamin is lost126). In addition to these indirect effects of dynamin on signalling, dynamin also interacts directly with a large collection of signalling proteins, (Table 2). These interactions may help coordinate signalling from receptors internalized by clathrin-coated pits with the dynamics of the pits and resulting vesicles and probably reflects an inseparable partnership between endocytosis and signalling. The broader effects of endocytosis on signalling have been extensively studied and reviewed elsewhere 3, 123, 128.
A full and systematic assessment of how dynamin affects signalling has yet to be performed. However, dynamin loss significantly increases phosphorylation and activation of the tyrosine kinase Ack, that interacts with clathrin and other factors implicated in clathrin-mediated endocytosis 129. These findings suggest that dynamin may limit Ack signalling by promoting clathrin coated pit turnover.
Gene knockout studies in mice 17, 19, 22 and the cells derived from them have provided numerous insights into dynamin function and the specific roles of the three dynamin isoforms. Dynamin 2 KO mice die early in embryonic development, consistent with the ubiquitous expression and housekeeping functions of this isoform19. Conditional dynamin 2 KO cells showed transient defects in clathrin-mediated endocytosis. However, the severity of this phenotype was reduced by dynamin 1 upregulation, demonstrating that these two isoforms have at least some overlapping functions 19, 20.
A double conditional KO of dynamin 1 and 2 in mice19 and analysis of derived fibroblasts confirmed that dynamin has an essential role in clathrin-mediated endocytosis but also yielded unexpected results. Surprisingly, dynamin 1, 2 double knockout fibroblasts survived for several weeks in culture without proliferation 19, which precluded analysis of the proposed role for dynamin in cytokinesis130. Most strikingly, dynamin 1 and 2 double KO cells accumulated endocytic clathrin-coated pits connected to the plasma membrane by long tubular necks (Figure 3a) 19, similarly to the phenotype produced by some dynamin mutants83 or to the phenotype observed in cells recovering from a temperature-induced block in endocytosis 131. These tubular necks, which were highly dynamic had an outer diameter of approximately 36 nm 19, consistent with that of tubules coated by BAR-domain-containing proteins wich narrow curvature73. Accordingly, fluorescent imaging analysis shows that a variety of BAR domain proteins such as endophilin, SNX9, amphiphysin and tuba are present in these tubules 19, 129, as were N-WASP, components of the Arp2/3 complex, other actin regulatory proteins and F-actin (Figure 3a and d and ref. 19).
Thus, although dynamin may modulate early stages of clathrin-coated pit formation, it is essential only for a late step of clathrin-mediated endocytosis, when membrane fission occurs. Moreover, the key function of dynamin is to terminate the formation of a deeply invaginated pit that is orchestrated by other factors. These studies have also helped elucidate the sequence of events that occur upstream of dynamin, by allowing otherwise transient intermediates to accumulate (Figure 3a). They showed that the actin cytoskeleton has a primary role in elongating the tubular necks of the coated pits and that the recruitment of BAR proteins on tubules, which is mediated to some extent by their curvature-sensing properties may help shape, stabilize and elongate the tubules in tight coordination with actin.
Endocytosis has a housekeeping role in neurons as in other cells. However, endocytosis, and clathrin-mediated endocytosis in particular, also has a specialized role in nerve terminals, mediating the local recycling of synaptic vesicle membranes after exocytosis (Figure 4a) 132. Consistent with this, disruption of synaptic transmission is the most dramatic and rapid effect produced by the temperature-sensitive inactivation of dynamin in Drosophila 5, 7, 8. Furthermore, perturbing dynamin impairs synaptic vesicle recycling, and thus neuronal communication in several experimental models 5, 17, 22, 23, 133-138.
The fact that all three dynamins are expressed in neurons, with both dynamin 1 and 3 being expressed primarily in the nervous system, raises questions about the specific roles that each of the three dynamins has here 17, 21. On the basis of KO studies, only dynamin 2 is essential for the development of the nervous system 19. Its conditional KO using the Cre/Lox system in early neuronal progenitors139, prior to the onset of expression of dynamins 1 and 3, drastically impairs brain development (S.M.F. and P.D.C., unpublished observations). However, conditional deletion of dynamin 2 in neurons at an early postnatal stage140 does not affect mouse viability or cause any discernible neurological phenotype (S.M.F. and P.De C., unpublished observations). This indicates that, once dynamins 1 and/or 3 are expressed, they can replace the function of dynamin 2.
Dynamin 1 is present at massive concentrations in the nervous system, is concentrated at synapses, and shows markedly increased levels during synaptogenesis 17, 18. Additionally, dynamin 1 and several other endocytic proteins collectively called ‘dephosphins’ 141-143 are constitutively phosphorylated in resting synapses. This dynamin phosphorylation depends at least in part on Cdk5 121 and, upon nerve stimulation, dynamin is rapidly dephosphorylated by the Ca2+ and calmodulin-dependent phosphatase calcineurin (Figure 4d)141; this is predicted to facilitate dynamin interactions with endocytic proteins to promote endocytosis50. The direct interaction of a dynamin 1 splice variant (the ‘b’ C-terminal variant) with calcineurin, via a PxIxIT-like calcineurin binding motif (PRITIS), further emphasizes the potential physiological importance of this Ca2+-dependent dephosphorylation reaction for synaptic function (24, 25 and Silvia Giovedi’, S. M. F. and P. De C., unpublished observations). On the basis on these considerations, dynamin 1 was expected to be essential for synaptic vesicle endocytosis.
Surprisingly, dynamin 1 is dispensable for basic functions of the nervous system 17. However, dynamin 1 KO mice rapidly develop a severe neurological phenotype, fail to thrive and die within two weeks of birth 17. Thus, dynamin 1 is not unique among the three dynamins in its ability to support synaptic vesicle recycling. However, the massive increase in dynamin 1 expression during the first post-natal weeks is critical to sustain the increase in synaptic activity that occurs as the nervous system matures. Consistent with this, direct measurements of endocytic rates at the calyx of Held synapse in the mouse brain stem show that the endocytic rate in the dynamin 1 KO is normal after small stimuli but rapidly saturates and fails to scale with exocytosis as stimulus strength increases 138.
Dynamin 3 was proposed to have a specific postsynaptic role, , forming specialized endocytic sites that locally recycle AMPA receptors in dendritic spines144. Consistent with this, dynamin 3 binds the EVH1 domain of the post-synaptic protein Homer via a PxxF motif that is not present in dynamin 1 and 2 and this property appeared consistent with the reported strong localization of dynamin 3 in dendritic spines27. However, this dendritic enrichment of dynamin 3 was not confirmed by another study that instead emphasized an overlapping role with dynamin 1 presynaptically 22. Thus, it is unclear whether any one dynamin isoform shows preferential enrichment or function within the postsynaptic compartment, although strong evidence indicates that the dynamins plays an important role in the endocytosis of neurotransmitter receptors 145, 146.
Dynamin 3 KO mice do not show any obvious pathological phenotype 22. By contrast, dynamin 1, 3 double KO mice have a more severe phenotype than the dynamin 1 single KO 22. They develop normally in utero but, although they maintain a limited degree of synaptic transmission, they die within several hours after birth. This result supports a partially overlapping function of the two neuronally enriched dynamins in synaptic transmission.
Electrophysiological recordings from cultured dynamin 1 single and dynamin 1, 3 double KO neurons confirmed that synaptic transmission can still occur in the absence of these ‘neuronal’ dynamins, although the ability to release neurotransmitter and to sustain release upon sustained stimulation is strongly reduced 17, 22. Likewise, these mutants show a delay in compensatory endocytosis in response to stimulus17, 22. The main ultrastructural defect observed is reduced synaptic vesicle number and a massive accumulation of endocytic clathrin coated pits located along deep invaginations of the plasma membrane (Figure 4a and b)17, 22, 147. Dynamin 1 has been implicated in bulk endocytosis that helps to recover synaptic vesicle membrane in response to strong stimuli 148. However, a form of bulk endocytosis still occurs in response to massive stimulation in both dynamin 1 KO 138, 147 or dynamin 1, 3 double KO neurons (Yumei Wu, S. M.F. and P. De C., unpublished observations).
Overall, the results of these mouse KO studies demonstrate that even the combined absence of dynamin 1 and 3 does not abolish synaptic vesicle recycling. Most likely, dynamin 2 is sufficient to support synaptic vesicle endocytosis, albeit at a slower rate. This conclusion is supported by the ability of dynamin 2 overexpression to at least partially rescue synaptic vesicle recycling defects in dynamin 1 KO neurons 17. However, the additional occurrence of dynamin-independent synaptic vesicle recycling cannot be excluded. These findings support a model wherein the occurrence of 2 neuronally enriched dynamin isoforms (dynamin 1 and 3) and the extremely high levels of one of them (dynamin 1), are primarily needed not to support a specific form of endocytosis, either pre- or postsynaptically, but to allow clathrin-mediated endocytosis to operate over a very broad range of neuronal activities. Nonetheless, unique properties of dynamin 1 and 3 (and their splice variants) may fine tune their functions in neurons.
Roles of abnormal dynamin function in genetic disease have begun to emerge. Whereas dynamin 2 mutations show links to tissue-specific diseases, dynamin 1 mutations affect selectively the nervous system..
Multiple unique missense mutations, or short deletions, within the middle, PH and stalk domains of dynamin 2 have been identified in patients with two autosomal dominant genetic conditions, intermediate Charcot-Marie-Tooth disease 149 and centronuclear myopathy 150. Charcot-Marie-Tooth disease is a peripheral neuropathy characterized by muscular weakness of the extremities and defects in neuronal axon conduction. Centronuclear myopathy is a congenital disease characterized by muscle weakness. The dominant mode in which these disorders are inherited supports the idea that the mutant dynamin acquires a toxic gain of function (Box 3).
How perturbed dynamin 2 function results in these two conditions is unclear. There is a lack of clear correlation between either of the diseases and the mutation sites in the dynamin structure, making mechanistic interpretations difficult. It is interesting to note, however, that PH domain mutations in Charcot-Marie-Tooth disease are more frequently located in the N-terminal portion of the domain whereas mutations in centronuclear myopathy are more frequently located in its C-terminal portion 151. Structural analysis shows that, although disease-linked mutations in the PH and stalk domains are distant within the primary sequence of dynamin, they actually cluster at an interface between these two domains and so can potentially influence dynamin oligomerization 31. It additionally remains unknown why altering the function of dynamin 2 so selectively affects certain tissues. Mutations in amphiphysin 2, a major binding partner of dynamin in brain and muscle, can also give rise to an autosomal recessive form of centronuclear myopathy 152. The muscle-specific isoform of amphiphysin 2 is expressed at very high levels, and affects muscle function153, raising the possibility that muscle pathology in these centronuclear myopathies r involves a dynamin interaction with amphiphysin 2.
So far, no human disease has been mapped to the dynamin 1 or dynamin 3 genes. Nonetheless, given the prominent expression and functions of these proteins in neurons, mutations that produce neurological phenotypes are likely. Supporting this possibility, a spontaneous mouse mutation - the fitful mutant mouse – that causes seizures and hearing impairment maps to the dynamin 1 gene 154. This missense mutation (A408T) occurs in an alternatively spliced region within the middle domain of dynamin 1 and may affect its oligomerization. Electrophysiological recordings from neurons of these mice demonstrated defects in GABAergic neurotransmission in response to prolonged stimulation 154, a defect similar to that of dynamin 1 KO mice 17, 147. Thus, subtle perturbations of dynamin 1 function that are otherwise compatible with life can enhance seizure susceptibility and dynamin 1 should be considered as a candidate gene in idiopathic cases of human epilepsy. It is of further interest that KO mice for three dynamin-binding proteins; amphiphysin 155, syndapin 156 and endophilin 157 also have seizures. Additionally, mutations in human synapsin 1, another protein implicated in synaptic vesicle recycling, have been identified in patients with epilepsy 158. Most likely, impairing synaptic vesicle recycling lowers seizure threshold through greater effects on inhibitory neurons that have a high rate of synaptic vesicle turnover and thus a high endocytic demand.
Another spontaneous animal model with impaired dynamin 1 function is Exercise Induced Collapse, a recessive condition that has been observed in dogs 159. These otherwise healthy animals display acute and severe muscle weakness resulting in a life threatening collapse in response to intense exercise or excitement. The underlying missense mutation, R256L, is located within the dynamin 1 G domain. The functional consequence of this mutation on GTPase activity remains uncharacterized.
The function of dynamin has been extensively investigated both in vitro and in vivo, and this has provided powerful insights into its cellular roles. Studies of dynamin have also helped identify factors that deform membranes and/or sense membrane curvature and that participate in the control membrane fission. Cryo-EM and crystallographic studies of dynamin and other members of the dynamin superfamily have provided insight into how it mediates membrane fission. However, major questions remain open.
The precise mechanism through which dynamin achieves fission remains to be elucidated. The similarities and evolutionary relationship with DLPs that mediate membrane fission (such as dynamin and Drp1) and membrane fusion (such as atlastin, OPA1 and mitofusin) are intriguing and unexpected because it suggests analogies between the mechanisms that mediate these seemingly opposite membrane-remodelling processes. Elucidating the basis of this similarity will be important.
Likewise, it will be interesting to determine why DLPs are needed for only a subset of membrane fission reactions. Other GTPases have been implicated in some membrane budding reactions that do not involve dynamin or other DLPs: for example, the Arf1 GTPase affects COP1-dependent budding from the Golgi complex and the Sar1 GTPase regulates COPII-dependent budding from the endoplasmic reticulum 160. However, both Arf1 and Sar1 are classic regulatory GTPases that function by recruiting downstream effectors, indicating that they use a fundamentally different mechanisms to that of dynamin. The two intracellular fission reactions besides endocytosis that clearly require DLPs are the fission of mitochondria 161-164 and the fission of chloroplast (in plants). The endosymbiotic origin of these organelles implies that their outer membrane is ancestrally derived form the plasma membrane, possibly explaining why fission of their outer membrane requires a dynamin.
Other exciting areas that need to be explored include possible non-endocytic functions of dynamin, particularly those involving the cytoskeleton, and the actin cytoskeleton in particular. The possibility that dynamin could be a target for cancer therapy because of its role in either cytokinesis 130, 165 or cell migration and tissue invasion 166, deserves consideration. Interestingly, the invasive activity of pancreatic ductal carcinoma was associated with elevated dynamin 2 expression 166. Are the potential non-endocytic actions of dynamin mediated by its polymerization in a manner that is analogous to that at the neck of clathrin coated pits? Structural studies indicate that a helical polymerization of dynamin around tubular templates 30-32 is optimally suited to promote G domain dimerization and thus GTP hydrolysis. However, perhaps other structural configurations are possible.
It can be anticipated that the spectrum of genetic diseases resulting directly from mutations in dynamin genes will expand. Indeed, several genes implicated in endocytosis, including PICALM and genes that encode the dynamin-binding proteins CD2AP and amphiphysin 2 have been linked to Alzheimer's disease by genome-wide association studies 167, 168. Dynamin 2 also mediates internalization and infectivity of bacterial 169 and viral pathogens including HIV 2, 26. In addition to these possibilities, as dynamin-dependent endocytosis is implicated in multiple processes that are relevant to disease, polymorphisms within dynamin genes may contribute to susceptibility to many other specific diseases.
Twenty years after the discovery that dynamin has a critical role in endocytosis, the function and mechanisms by which this protein acts remains a most important and timely field of investigation.
Dynamins are the founding members of a family of GTPases known as dynamin-like proteins (DLPs), which share basic features of domain organization and roles in membrane remodelling (Figure 1). Within this family, some proteins act similarly to dynamin in mediating membrane fission. For example, Drp1 (Dynamin related protein 1) is important for fission of mitochondria and peroxisomes 161-164. Others are more distant members and mediate homotypic membrane fusion 161-164. These include, for example, mitofusin which controls fusion of mitochondrial outer membranes, OPA1 (optic atrophy 1 gene) which is important for inner mitochondrial membrane fusion, and atlastin which mediates fusion of endoplasmic reticulum membranes 161-164, 170, 171. Also included in this family on the basis of their structure and assembly properties, are the anti-viral, interferon-inducible myxovirus resistance (Mx) proteins 9, 172 and guanylate-binding proteins (GBPs) 173 whose precise functions remain unknown. Plant DLPs control membrane traffic during cell plate formation in cytokinesis 174 and mediate chloroplast division 175, 176. A bacterial DLP implicated in membrane remodeling has also been identified and structurally characterized 40, 175. Additionally, although differing from the DLPs in that they are ATPases rather than GTPases, the EHD proteins share structural similarities and are also implicated in membrane remodelling 177.
Mutations in DLPs other than dynamin, and more specifically in DLPs that affect mitochondrial and endoplasmic reticulum dynamics, have been identified as causes of inherited diseases. These include mutations in OPA1 (optic atrophy type 1) 178, MFN2 (Charcot-Marie-Tooth Type 2A) 179, atlastin (hereditary spastic paraplegia) 180 and a DRP1 (lethal perinatal defects in mitochondria and peroxisome fission 181.
Some proteins that function at the membrane interface bind preferentially to curved membranes. This can result from: the intrinsic curvature of the membrane binding surface of these proteins; their propensity to oligomerize into curved scaffolds; or the presence of amphipathic helices whose partial insertion into the bilayer is facilitated by the loose packing of phospholipids at sites of high positive curvature. The same proteins can also force the membrane to curve, and thus induce, propagate or stabilize bilayer curvature 73, 182-185. The ability of these proteins to function primarily as curvature sensors or curvature inducers depends on various factors, such as regulated affinity for the membrane (through changes in phosphorylation, folding state or surface charge of the membrane) and local concentration 182. Dynamin itself has important curvature sensing and generating properties 55, owing to its polymerization into rings 11 and to the presence of a hydrophobic loop emerging from its PH domain 49.
Bar domains. Many proteins that bind dynamin's PRD via an SH3 domain also contain BAR domains, protein modules with curvature generating and sensing properties (Table 2) 73, 81, 186. BAR domains comprise amino-acid sequences that fold into coiled-coils and dimerize into elongated curved structures optimally suited to bind the negatively charged cytosolic leaflet of the plasma membrane 187, 188. The presence of amphipathic helices or lipid-binding modules that flank the BAR domains and their propensity to polymerize further enhance the bilayer-molding properties of these proteins 73. In BAR domain proteins that bind dynamin, the bilayer binding surface is typically concave (with shallow curvature in F-BAR domains such as those of syndapin/pacsin and of FBP17/Toca1/CIP4 73, 188, 189, and narrow curvature in classic BAR domains such as those of endophilin, amphiphysin and Snx9/18 73) and thus optimally suited to bind endocytic intermediates 73. BAR domain proteins are proposed to coordinate the progressive acquisition of bilayer curvature at the necks of coated pits with the recruitment of factors that drive constriction, fission and immediate post-fission events (Figure 3). Dynamin binding to the SH3 domain of BAR or F-BAR domain proteins also relieves autoinhibitory intramolecular interactions that limit the membrane-binding activity of the BAR or F-BAR domain 189, 190.
This work was supported in part by the Howard Hughes Medical Institute, the G. Harold and Leila Y. Mathers Charitable Foundation, National Institutes of Health grants (R37NS036251, DK45735 and DA018343), the W.M. Keck Foundation and a NARSAD distinguished Investigator Award to P.D.C.. We wish to thank Hongying Shen, Michelle Pirucello, Andrea Raimondi and Oliver Daumke for their helpful suggestions and insights and Michelle Pirucello and Andrea Raimondi for assistance with preparation of figures.
Shawn M. Ferguson is an assistant professor in the Department of Cell Biology at Yale University. He received his PhD from Vanderbilt University with Randy Blakely and pursued postdoctoral research with Pietro De Camilli at Yale. His laboratory investigates mechanisms controlling lysosome homeostasis and the impacts of this process on cancer and neurodegenerative diseases.
Pietro De Camilli is a Professor in the Department of Cell Biology at Yale University, an Investigator in the Howard Hughes Medical Institute and a founding Director of the Yale Program in Cellular Neuroscience, Neurodegeneration and Repair. He received his MD from the University of Milano and was a postdoc with Paul Greengard at Yale. His research focus on mechanisms in membrane traffic, with emphasis on membrane traffic at synapses.