In this paper, we examined the dynamic events underpinning the initiation and subsequent maturation of activity-dependent bulk endocytosis at the amphibian NMJ. Our results confirm that initiation of bulk endocytosis takes place during stimulation 
. Importantly, we provide evidence that a delayed form of maturation of bulk endocytosis supports the coupling between exo- and bulk endocytosis – a process differentially controlled by actin and dynamin.
Our data demonstrate that bulk endocytosis is triggered shortly after the onset of stimulation, a finding that is consistent with previous studies performed in the central nervous system: cerebellar granule neuronal cultures 
and the calyx of Held 
. Our stimulation protocol, aimed at depleting synaptic vesicles, was stronger than that used in these previous studies. However, our results confirm that internalization of FM1-43-labeled presynaptic plasma membrane occurs during such stimulation and is initiated through the formation of FM1-43 hotspots on the presynaptic membrane that evolve into fluorescent balls and tubules. We found that this network of internalized membrane continued to grow after the cessation of stimulation, a process consistent with a form of maturation of the bulk endosomal compartment that has not been previously described. This maturation process might be triggered by the longer duration of our stimulation (20 Hz for 10 min) intended to deplete both the RRP and RP of synaptic vesicles 
. Recently, similar donut-like structures were detected in presynaptic nerve terminals treated with the stimulatory neurotoxin glycerotoxin 
Our study provides quantitative analysis of the percentage of recycling (photoconverted) vesicles associated with bulk endosomes. The size and shape of the bulk cisternae and the halo of fluorescence surrounding the limiting membrane of the endosomes, further supported by our photoconversion EM analysis, suggest that recycling vesicles still associated with plasmalemmal invaginations (fluorescent balls) closely follow the maturation of invagination into bona fide bulk endosomes. The majority of recycling vesicles (~84%) were clustered in the immediate vicinity of the bulk endosomes. Other recycling vesicles (~16%) found outside that zone were reminiscent of the photoconverted vesicles previously described 
. Whether the omega-shaped structures detected represent fusion or fission of recycling vesicles with the cisternae is unknown and is beyond the scope of this study. It is worth noting that due to the high density of small recycling vesicles surrounding each bulk endosome, the actual percentage of internalized membrane found on bulk endosomes was negligible (<1%) when compared to that of recycling vesicles.
The observed significant correlation between the amount of bulging/collapsing and internalized membrane surface at high-frequency stimulation indicates that this form of plasticity is essential to prepare the plasma membrane for the development of bulk endocytosis and points to a sensing role of the bulging phase in determining how much vesicular membrane was incorporated into the presynaptic membrane during the exocytic phase. The increase in membrane surface area expected to result from high levels of exocytic fusion should be paralleled with a decrease in cytoplasmic volume due to the loss of volume previously occupied by synaptic vesicles 
. In conditions of high stimulation, an increase of nerve terminal surface area has previously been described in bipolar nerve terminals 
. This was linked to the increase in plasma membrane surface area resulting from the exocytic fusion of synaptic vesicles. To the best of our knowledge, our study is the first to describe changes in nerve terminal volume associated with activity. At this stage, we can only invoke the possibility of a compensatory mechanism such as an inward hydrodynamic flow to balance the reduced cytoplasmic volume perhaps associated with dynamic changes in the cytoskeleton. More work is required to assess these possibilities.
The lack of correlation between the amount of bulging/collapsing and internalized membrane surface observed at 5 Hz stimulation suggests that at this frequency, other forms of endocytosis such as clathrin-mediated single vesicle recycling may also be contributing to the regeneration of synaptic vesicles pools. At the amphibian NMJ, a switch from single vesicle to bulk endosomal synaptic vesicle recycling has been reported 
. Our findings are consistent with such a switch in the mode of endocytosis, from single vesicle to bulk endosomal, occurring at higher frequency stimulation.
Our data point to a critical role of actin and dynamin during the maturation of bulk endocytosis. Our results show that the kinetics of presynaptic bulging exhibits two distinct phases: a slow period followed by a fast phase reaching a plateau just preceding the collapse. Together with molecular motors such as myosins 
, actin has previously been reported to act in tension sensing 
. It is tempting to speculate that, at the plateau, the actomyosin machinery involved in bulging has reached a point at which the membrane resistance opposes its forces. Actin-binding proteins promoting fast actin depolymerization 
could be involved in the collapsing phase. The parallel internalization, formation of bulk endosomes and generation of surrounding recycling vesicles is likely to involve polymerization and branching of actin as previously proposed 
, perhaps via a tread-milling effect 
Our data are in good agreement with previous studies 
regarding the role of actin polymerization in the recovery of neurotransmitter release after synaptic depletion at the NMJ. The slow nature of functional recovery further suggests that actin driven bulk endocytosis is primarily responsible for replenishing the reserve pool of synaptic vesicles as previously proposed 
Dynamin inhibition completely blocked bulk endocytosis under high frequency stimulation conditions. The detection of branched tubulation networks in nerve terminals treated with various dynamin inhibitors and the altered distribution of dynamin-1 immunoreactivity in dyngo-4a-treated stimulated nerve terminals argue for an effect of dynamin on the maturation of bulk endocytosis. A key assumption of this hypothesis is that the observed tubules are still connected with the plasma membrane, a situation supported by our confocal imaging and electron microscopy data. The molecular mechanisms underpinning the formation of these branched tubules following dynamin inhibition are unknown. However, BAR domain proteins, such as amphiphysin and endophilin, could be involved in the early membrane-deforming stages of endocytosis 
. Furthermore, BAR proteins have been shown to be part of an actin regulatory network and, upon overexpression, to induce tubular invagination of the plasma membrane 
. Perhaps not surprisingly, several studies have demonstrated both a direct interaction between dynamin and actin 
and an indirect interaction via proteins that bind both dynamin and actin 
. Moreover, there is increasing evidence for a functional link between endocytosis and actin dynamics, with actin being involved in specific steps of the endocytic process 
. In view of the role of dynamin in regulating the actin cytoskeleton 
, disruption of dynamin function and subsequent dysregulation of actin function are consistent with the formation of extensive tubulation of the plasma membrane observed in our imaging experiments. Our results are also in good agreement with the electron microscopy data generated with the dynamin-1 knockout mouse 
and with the temperature-dependent phenotype of the shibire Drosophila
mutant of dynamin 
. This is suggestive of a role for dynamin in shaping the recycling vesicle membrane accompanied by a defect in the GTPase scission activity produced by the dynamin inhibitor 
In close agreement with studies based on the use of another dynamin inhibitor, dynasore, we found that dyngo-4a had a limited but significant effect on phasic quantal neurotransmitter release at low frequency stimulation as previously described 
. This fits well with the FM1-43 destaining data generated by treatment of hippocampal neurons and sympathetic neurons with dynasore 
. At higher frequency stimulation, dyngo-4a did not significantly affect the rate of synaptic depression. Two possible scenarios can account for this lack of effect: (i) a dyngo-4a-insensitive endocytosis underpins fast recycling of synaptic vesicles or (ii) dyngo-4a-mediated inhibition of endocytosis taking place during synaptic depletion does not have time to generate fully functional synaptic vesicles and therefore does not significantly impact on the rate of synaptic depression. The latter scenario is more likely as dynasore has been shown to affect all forms of synaptic vesicle endocytosis 
. Importantly, the lack of effect on synaptic depression is in good agreement with evidence from cortical neurons from dynamin-1 knockout mice 
In summary, our study provides evidence demonstrating that both actin and dynamin are differentially required for the development and maturation of activity-dependent bulk endocytosis. We show that bulk endocytosis is initiated during stimulation but the maturation process only occurs after the end of the stimulation. We propose that it may contribute to a mechanism by which nerve terminals couple exocytosis and bulk endocytosis, by sensing how much presynaptic membrane to retrieve in response to sustained stimulation. We also demonstrate that these dynamic presynaptic events are functionally linked to the recovery of synaptic transmission following synaptic depression.