Introduction

Chemical synapses exhibit striking dynamic changes in connection strength during repetitive use. The changes are termed short-term plasticity or frequency dynamics, and are thought to play an important role in how information is processed (Tsodyks and Markram, 1997; Abbott and Regehr, 2004; Buonomano and Maass, 2009). Multiple presynaptic vesicle trafficking mechanisms are involved, but the identity of the mechanisms, how they interact, and the implications for biological computation are not understood (Neher, 2015).

Most attempts at a detailed understanding begin with the premise that presynaptic vesicles are segregated into multiple pools, including at least one readily releasable pool and one reserve pool. Readily releasable vesicles are thought to be docked to release sites embedded within the active zone of the plasma membrane of synaptic terminals, whereas reserves reside in the interior. Nevertheless, pools are typically not defined by morphological criteria but instead by the timing of mobilization, which is a general term for the full sequence of events required for constituent vesicles to undergo exocytosis. Based on this, the readily releasable pool has been divided into fast- and slow-releasing subdivisions at a wide variety of synapse types; slow-releasing readily releasable vesicles are often termed reluctant (Wu and Borst, 1999; Sakaba and Neher, 2001; Moulder and Mennerick, 2005). Reserve pools are less studied, but have likewise been divided into quickly and slowly mobilized pools at some synapse types (Neves and Lagnado, 1999; Rizzoli and Betz, 2005; Denker et al., 2011).

A widespread premise has been that the pools are connected in a series where vesicles are transferred from slowly to quickly mobilized pools as diagrammed in Figure 1A (Pieribone et al., 1995; Hilfiker et al., 1999; Denker and Rizzoli, 2010; Rothman et al., 2016; Miki et al., 2016; Doussau et al., 2017; Milovanovic et al., 2018; Lin et al., 2022). If so, reluctant vesicles might be docked to the same type of release site as fast-releasing vesicles, but in an immature priming state that could then transition to the mature, fast-releasing state. We refer to models with this premise as homogeneous release site models.

Figure 1.

However, a growing body of evidence suggests that reluctant vesicles are fully primed for release at a different kind of release site that is inherently inefficient at catalyzing exocytosis. First, to our knowledge, none of the homogeneous release site models proposed so far can account for a series of electrophysiological experiments at calyx of Held synapses where: (1) the reluctant subdivision was only minimally depleted during moderately intense stimulation that exhausted the fast-releasing subdivision; and (2) abruptly increasing the intensity could then drive exocytosis of the remaining reluctant vesicles directly, without first transitioning to a fast-releasing state (Mahfooz et al., 2016 and Figure5-figure supplement 1 of Raja et al., 2019). And second, recent optical imaging and molecular studies have confirmed substantial heterogeneity among release sites, even within the same synapse, at a wide variety of synapse types (Hu et al., 2013; Müller et al., 2015; Böhme et al., 2016; Akbergenova et al., 2018; Maschi and Klyachko, 2020; Li et al., 2021; Karlocai et al., 2021; Gou et al., 2022).

To account for this, one of the starting premises of our working model is that fast-releasing and reluctant subdivisions of the readily releasable pool operate in parallel, which could be as in Figure 1B if based solely on the studies referenced so far. Figure 1C depicts a different possibility, tested here, where quickly and slowly mobilized reserve pools are likewise arranged in parallel, and independently supply vesicles to corresponding fast-releasing and reluctant subdivisions of the readily releasable pool.

No such fully parallel arrangement has been proposed previously, at least not explicitly. However, an additional premise of our working model - not related to distinctions between subdivisions of the readily releasable pool - is that each readily releasable vesicle is physically tethered to a short chain of reserve vesicles, which serve as replacements after the readily releasable vesicle undergoes exocytosis (Figure 1D; Gabriel et al., 2011; see Wesseling et al., 2019 for supporting ultrastructural evidence).

If so, reserve vesicles chained to either type of readily releasable vesicle would only advance after the readily releasable vesicle had undergone exocytosis. The reluctant vesicles would, on average, create space for advancement more slowly than fast-releasing vesicles during continuous stimulation because they would undergo exocytosis less frequently. As a direct consequence, reserves chained to reluctant vesicles would be mobilized more slowly than reserves chained to fast-releasing vesicles, and would behave as if constituting a slowly mobilized pool. Reserves chained to fast-releasing vesicles would be mobilized more quickly, and would be processed in parallel. In sum, the combination of two of the premises of our working model predict the fully parallel arrangement outlined in Figure 1C.

Key for testing this idea: reserve vesicles chained to reluctant and fast-releasing readily releasable vesicles would only be mobilized with measurably different timing when the frequency of stimulation is low enough that individual reluctantly releasable vesicles remain in the readily releasable pool for extended periods of time before undergoing exocytosis. The reasoning is given below, in the Results section, where needed to explain the design of key experiments.

Here we begin by demonstrating that reserve vesicles at hippocampal synapses are segregated into multiple pools that can be distinguished by the timing of mobilization during low frequency stimulation. We then show further that the two types of reserves do not intermix, even when the frequency of stimulation is high, confirming that the two types are processed in parallel.

Results

We originally developed our working model to account for results from both primary cell culture and ex vivo slices (Stevens and Wesseling, 1999b; Garcia-Perez et al., 2008; Gabriel et al., 2011; Mahfooz et al., 2016; Raja et al., 2019). We chose to use cell cultures for testing the prediction that quickly and slowly mobilized reserve pools are processed in parallel because cultures are better suited for staining and destaining synaptic vesicles with FM-dyes, which could be used to distinguish between reserve pools.

In a first set of experiments, diagrammed atop Figure 2, we began by staining vesicles within synapses with 60 s of 20 Hz electrical stimulation (1200 pulses) during extracellular bath application of FM4-64. We then removed the FM4-64 and washed with Advasep-7 or Captisol.

Figure 2.

Advasep-7 and Captisol are closely related β-cyclodextrin derivatives that facilitate dye clearance from membranes (Kay et al., 1999). The stain followed by wash procedure is thought to stain nearly all of the vesicles within synaptic terminals that can recycle, and eliminate most of the background fluorescence caused by dye bound to non-vesicular membrane (Betz et al., 1992; Ryan and Smith, 1995; Chi et al., 2001; Gaffield and Betz, 2006). The vesicles then retain the dye until undergoing exocytosis, after which they likely destain completely as the dye is washed away into the extracellular space (Figure 2–Figure Supplement 1).

The idea was to use the time course of destaining to measure mobilization of stained vesicles. To asses the timing during low frequency stimulation, we monitored destaining in the continued presence of the β-cyclodextrin while stimulating at 1 Hz for 25 min (1500 pulses; blue in Figure 2A-B). And finally, after a 4 min rest interval, we completed destaining to a low level with 100 s of 20 Hz stimulation (2000 pulses; red in Figure 2A-B; see also Figure 2–Figure Supplement 2).

Note that the concept of mobilization used here is not the same as pool depletion because mobilized stained vesicles are continuously replaced by unstained vesicles, either reconstituted from recycled vesicular membrane or recruited from other sources. Pool depletion during intense stimulation - particularly readily releasable pool depletion - was assessed in the electrophysiological studies that motivated the working model (reviewed in Wesseling, 2019), but not directly in the present study.

Destaining time courses were only accepted for further analysis when baseline fluorescence loss was ≤ 1.5 %/min. Baseline loss could have been caused by multiple factors including spontaneous exocytosis and focal drift of the microscope, but likely not by photobleaching (Figure 2–Figure Supplement 3).

The results were not compatible with straightforward models involving only a single pool containing vesicles that intermix quickly. That is, quickly mixing single pool models would predict that destaining at any point in time during 1 Hz stimulation would be a constant fraction of the amount of stain remaining within the pool at that time, meaning:

which is equivalent to:

where k is a constant - equal to the fractional destaining (at any time) per unit of time - F(t) is fluorescence intensity of the dye within the pool at time t, and dF(t)/dt is the slope of the destaining time course at that time. However, the time course of destaining during 1 Hz stimulation was not compatible with Eqn 1 because fractional destaining was not constant. Instead, fractional destaining decreased substantially, by a factor of 2.7 ± 0.1 over the 25 min (n = 7 preparations; Figure 2C, left panel, magenta line is the prediction if fractional destaining was constant).

For the analysis: F(t) was estimated as the median fluorescence intensity of punctal regions of interest (ROIs; see Figure 2–Figure Supplement 4) after subtracting the signal remaining after the final 20 Hz train. Fractional destaining was then estimated from short time intervals by dividing the slope of straight line fits of F(t) during each interval (magenta lines in Figure 2B) by F(t) at the start of the same interval (green circles). Further analysis showed that the decrease evident in Figure 2C was not caused by technical errors related to baseline fluorescence loss seen in some preparations, or by flaws in the premise that the final 20 Hz train fully destains all recycling vesicles. Specifically: Correcting for the baseline fluorescence loss decreased the estimate of fractional destaining for later intervals (e.g., minute 21 - 25) more than for earlier intervals (e.g., minute 0 - 1.5), and, as a consequence, indicated an even greater decrease in the estimate over time; i.e., estimates of fractional destaining decreased by a factor of 4.4 ± 0.4 during the 25 min of 1 Hz stimulation after correcting (Figure 2C, right panel). Likewise, correcting for any error in the premise that the final 20 Hz train eliminated dye from all recycling vesicles could increase the factor even more, although any correction for this sort of error would be small.

Next, an analysis of the individual ROIs showed that deviations from Eqn 1 occurred at almost all individual synapses despite variation between individuals in the details (Figure 2–Figure Supplement 5; see also Waters and Smith, 2002). That is, measurement noise prevented applying the analysis in Figure 2B-C to many of the individuals. However, Eqn 1 is mathematically equivalent to the single exponential decay:

where F(0) is the fluorescence intensity at the start. And, the best fitting version of Eqn 2 for each individual could be compared to the full time course of each, acquired during the full 25 min of 1 Hz stimulation. This allowed a higher resolution analysis because the full time courses consisted of more data points than the short intervals used to calculate fractional destaining in Figure 2C, making them less sensitive to measurement noise. Statistically significant deviations of p < 0.05 were detected for >90 % of individuals (5638 of 6252) in an analysis of residuals where k was allowed to vary freely between individuals (Figure 2D; see Figure 2–Figure Supplement 6 for methodology, and Figure 2E for the analogous analysis at the level of preparations). This analysis shows that deviations from Eqn 1 occur at the level of individual synapses and rules out the concern that the decrease in fractional destaining evident in Figure 2C might have been caused by heterogeneity among the individuals.

Most individual ROIs and the collective behavior of populations could be fit with the weighted sum of two exponentials (Figure 2–Figure Supplement 7), although weighted sums of three or more exponentials and a variety of other functions could not be excluded. The result is consistent with models containing two or more reserve pools where the decrease in fractional destaining in Figure 2C would be caused by selective decrease in the number of stained vesicles in reserve pools that are mobilized quickly, leaving the remaining dye predominantly trapped in vesicles that are mobilized slowly. Box 1 illustrates this point with a model where the reserve pools operate in parallel (right panel, quickly mobilized vesicles in pathway a vs slowly mobilized in pathway b). However, further experiments, below, were still required to rule out alternatives where the reserve pools are instead connected in series, and a variety of other explanations for the decrease in fractional destaining that do not involve multiple reserve pools.

No recovery of fractional destaining during rest intervals

A key point is that mixing between the reserve pools would have to be slow during 1 Hz stimulation, if it occurs at all. Otherwise, mixing would have caused the stained vesicles in quickly and slowly mobilized reserve pools to equilibrate over the 25 min of 1 Hz stimulation, which would have prevented the decreases in fractional destaining. And indeed, further experiments showed that the decreases in fractional destaining did not reverse during long rest intervals (Figure 3).

Figure 3.

Evidence against selective depletion of dye from readily releasable pools

One alternative to multiple reserve pools that needed to be ruled out was the possibility that the decrease in fractional destaining during 1 Hz stimulation in Figure 2B-C was caused by selective dye loss from readily releasable pools. Selective dye loss from readily releasable pools without multiple reserve pools seemed unlikely because spontaneous mixing between readily releasable and reserve pools is thought to occur on the order of one minute (Murthy and Stevens, 1999), which is fast compared to the destaining time course during 1 Hz stimulation in Figure 2B-C. Nevertheless, to test this possibility, we stained synapses as above with FM4-64, but this time destained first with a 20 Hz train of 80 pulses followed immediately by 25 min of 1 Hz stimulation (diagram atop Figure 4A). The 80 pulses at 20 Hz is enough to exhaust the readily releasable pools at these synapses (Stevens and Williams, 2007; Garcia-Perez et al., 2008), but did not greatly alter fractional destaining measured during the subsequent 1 Hz stimulation (Figure 4B, compare blue circles to green rectangles). This result confirms that the decrease in fractional destaining seen during long trains of 1 Hz stimulation was not caused by selective depletion of dye from readily releasable pools.

Figure 4.

Box 1.

Multiple reserves can explain decrease in fractional destaining

Comparison of models with one reserve pool (left) and with two parallel reserve pools (right) that feed separate subdivisions of the readily releasable pool; wide arrows signify fast-releasing and narrow arrows signify reluctant. For both models, the number of stained vesicles within the readily releasable pools (RRPs) decreases over time of stimulation, accounting for the decrease in the absolute amount of destaining for each unit of time. However, with a single reserve pool, the ratio of stained fast-releasing to stained reluctant vesicles never changes, and, as a consequence, the fractional destaining at any point in time (see Results) does not change. In contrast, with two reserves, vesicle mobilization in path a is faster than in path b, causing path a to become exhausted sooner. As a consequence, the ratio of stained fast-releasing to stained reluctant vesicles decreases over time, resulting in decreased fractional destaining.

For these experiments, we included a second 1 Hz train after a rest interval of 4 min (diagram atop Figure 4A) to test if fractional destaining would recover during rest intervals after being driven to the minimum value. No recovery was seen (compare minute 29 - 33 to minute 21 - 25 in Figure 4B, either panel). This result confirms the conclusion from Figure 3 that decreases in fractional destaining induced by long trains of 1 Hz stimulation and measured during subsequent 1 Hz trains do not reverse quickly during rest intervals, if at all.

Evidence against explanations that do not involve multiple reserve pools

Next we rule out alternative explanations for the long-lasting decreases in fractional destaining that do not involve pools at all.

Long-term presynaptic depression

In principle, low frequency stimulation can induce long-term depression in presynaptic function that might cause fractional destaining to decrease even if there were only a single reserve pool. However, no long-term presynaptic depression was seen during 1 Hz stimulation when synaptic vesicle exocytosis was measured using vGlut1-synaptopHluorin fluorescence instead of FM4-64 (Figure 5).

Figure 5.

The absence of long-term depression in vGlut1-synaptopHluorin fluorescence is compatible with the decrease in fractional destaining measured with FM4-64 because, unlike FM4-64, synap-topHluorin does not disassociate from vesicle membranes after exocytosis and therefore tracks presynaptic function through multiple rounds of exo/endocytosis (Miesenböck, 2012). Instead, the result argues against long-term presynaptic depression as a cause for the decrease in fractional destaining measured with FM4-64. We note, however, that the result does not contradict previous reports of presynaptic long-term depression where induction required postsynaptic depolarization (Goda and Stevens, 1996) because postsynaptic depolarization was likely prevented in the present study by glutamate receptor antagonists.

Long-lasting switch to kiss and run

Alternatively, fractional destaining might decrease if endocytosis switched from the standard mode to a faster kiss-and-run mode, which sometimes is too fast to allow complete clearance of FM4-64 from the membrane of individual vesicles (Klingauf et al., 1998). This explanation seemed unlikely because the switch in timing of endocytosis would have to persist for minutes during rest intervals to account for the results in Figure 3 and Figure 4. Nevertheless, to test this, we conducted experiments similar to those documented in Figure 2, except using the FM2-10 dye, which dissacociates from membranes more quickly than FM4-64, allowing faster clearance (Klingauf et al., 1998).

Despite the faster clearance, the decrease in fractional destaining during 1 Hz stimulaton was not altered (Figure 6, compare blue circles to green rectangles). The result does not rule out changes in the timing of endocytosis, but does argue that clearance was fast enough throughout our experiments to allow complete clearance of FM-dyes, even if changes occured. As a consequence, the result argues against changes in the timing of endocytosis as the cause of the decrease in fractional destaining seen above. Notably, the results are not directly comparable to earlier studies conducted in the absence of β-cyclodextrins (see Figure 2–Figure Supplement 1).

Figure 6.

The results in this section all support the hypothesis that the decrease in fractional destaining seen during long trains of 1 Hz stimulation is caused by selective depletion of a quickly mobilized pool of reserve vesicles because they rule out alternative explanations that have arisen. Remaining doubt is addressed, below, with affirmative evidence that multiple reserve pools are indeed present.

Faster destaining when staining is induced with low frequency stimulation

As a first step, we began by reasoning that vesicles reconstituted from recycled membrane during 1 Hz stimulation would have to be predominantly targeted back to the quickly mobilized reserve pool. Otherwise, targeting to the slowly mobilized reserve pool would have displaced already stained vesicles from the slowly to the quickly mobilized reserve pool, which would have prevented the decreases in fractional destaining seen above.

To test this, we stained vesicles during stimulation with 240 or 1200 pulses at 20 Hz or 240 pulses at 1 Hz in interleaved experiments. We then compared fractional destaining during subsequent 1 Hz stimulation. The experimental protocol is diagrammed in Figure 7A.

Figure 7.

Fractional destaining was greater when vesicle recycling had been driven by 1 Hz stimulation during the staining phase of the experiment (compare bar c to a and b in Figure 7B), as expected if a larger fraction of the stained vesicles were targeted to a quickly mobilized reserve pool. Further experiments diagrammed in Figure 7C showed that 240 pulses at 20 Hz and at 1 Hz stain synapses to equivalent levels (Figure 7D-F), indicating that the greater fractional destaining seen after selectively staining vesicles that recycle during 1 Hz stimulation was not because fewer vesicles had been stained. Taken together, these results suggest that vesicles reconstituted from membrane that is recycled during ongoing 1 Hz stimulation are targeted predominantly to a quickly mobilized reserve pool. In contrast, vesicles reconstituted during 20 Hz stimulation are targeted to both quickly and slowly mobilized reserve pools. The information allowed us to design a method to stain quickly and slowly mobilized reserve pools with different colors.

Two color staining of separate reserve pools distinguished by mobilization timing

That is, we next confirmed the presence of multiple reserve pools with the experiment diagrammed atop Figure 8. The idea was to stain the quickly and slowly mobilized pools with different colored dyes, and then monitor destaining of each color separately. To achieve this, we first stained all recycling vesicles with FM4-64 (red) using the standard 60 s of 20 Hz stimulation. Next, we partly destained with 1 Hz stimulation for 10 min. We reasoned that 1 Hz for 10 min would nearly completely destain any quickly mobilized reserve pool because it was enough to drive fractional destaining to close to the minimum value in Figure 2B-C, and elsewhere. We then partly re-stained the synapses with FM1-43 (green), but this time by stimulating at 1 Hz, which we reasoned would predominantly re-stain quickly mobilized reserves based on the results in Figure 7B. For this second staining phase, we chose a duration of 5 min, which was calculated from experiments, above, so that the number of vesicles stained with the new (green) stain would be similar to the number, stained in red, remaining from the first staining phase. Finally, we destained both colors: first with 1 Hz stimulation to measure fractional destaining; and then with 20 Hz stimulation to fully destain both quickly and slowly mobilized reserves as usual.

Figure 8.

As predicted, the green stain applied during 1 Hz stimulation destained >4-fold faster than the red stain applied during the initial 20 Hz stimulation (Figure 8A). The results were reversed when vesicles were stained first with green at 20 Hz and second with red at 1 Hz (Figure 8B). And, finally, Figure 8C confirms that the number of stained vesicles at the start of the destaining phase of the experiment was similar for the two colors.

These results show that reserve synaptic vesicles are segregated into at least two pools that can be distinguished by the timining of mobilization during 1 Hz stimulation. The one remaining caveat is that the results so far do not rule out last-in/irst-out models where reserve vesicles are stored in a single pool with constituents that inter-mix slowly, if at all, during 1 Hz stimulation (Rizzoli and Betz, 2004; Kamin et al., 2010; see Figure 8–Figure Supplement 1). However, all serial models - which include single pool last-in/first-out models - would require fast mixing either between or within reserve pools during 20 Hz stimulation to account for the results in Figure 7B and Figure 8. And, this is ruled out, below.

Mixing continues to be slow/absent near body temperature

The experiments above were conducted at room temperature, and there is evidence that vesicles in the interiors of synaptic terminals are more motile at body temperature (Westphal et al., 2008; Kamin et al., 2010; Lee et al., 2012; Park et al., 2012). It is not known if the motility is related to rate-limiting steps in synaptic vesicle trafficking that would influence the timing of mobilization. However, the decrease in fractional destaining during 1 Hz stimulation and the absence of mixing during rest intervals were both preserved at 35 C (Figure 8–Figure Supplement 2 and Figure 8–Figure Supplement 3).

Parallel mobilization

To distinguish between serial and parallel processing of the two types of reserves, we begin by showing that decreases in fractional destaining were absent when stimulation was 20 Hz (Figure 9A and bars labeled b-d in Figure 9C). And, 25 min of 1 Hz stimulation had almost no impact on the timing of destaining when subsequent stimulation was 20 Hz (Figure 9B, and bars labeled f and g vs bars b and c, respectively, in Figure 9C). The absence of a large impact was striking because fractional destaining would have been decreased by a factor of ∼4 if measured instead during 1 Hz stimulation, as shown in multiple experiments above. These results suggest that the distinction between quickly and slowly mobilized reserves seen during 1 Hz stimulation is no longer evident when stimulation is 20 Hz.

Figure 9.

Box 2.

Different rate-limiting mechanisms at 1 vs 20 Hz

Wide arrows signify fast-releasing and narrow arrows signify reluctant readily releasable vesicles. The absence of rundown in fractional destaining when stimulating is 20 Hz (right panel) is in-line with parallel models, as diagrammed above, because: (1) 20 Hz stimulation is intense enough to quickly drive both fast-releasing and reluctantly-releasing components of the readily releasable pool (RRP) to a near-empty steady state; after which, (2) exocytosis is rate-limited by the timing of recruitment of reserve vesicles to vacant space within the RRPs (Wesseling and Lo, 2002; Garcia-Perez and Wesseling, 2008; Garcia-Perez et al., 2008; Raja et al., 2019) instead of by mechanisms that determine probability of release of already releasable vesicles; and, finally (3) the timing of recruitment is the same for fast and reluctantly releasing subdivisions (Garcia-Perez and Wesseling, 2008; Mahfooz et al., 2016; see Wesseling, 2019 for a discussion of discrepancies reported for other synapse types).

To test this, we compared destaining during 1 Hz stimulation following conditioning with either 10 min of 1 Hz stimulation (Figure 10A), or 20 s of 20 Hz stimulation (Figure 10B). Both conditioning trains destained the synapses by about 50 %, but only the 1 Hz train caused a decrease in fractional destaining measured during the subsequent 1 Hz stimulation (insets of Figure 10A & B, quantified in Figure 10C). Taken together, these results demonstrate that vesicles in the two types of reserve pools are mobilized with equivalent timing when stimulation is 20 Hz, whereas vesicles in quickly mobilized reserve pools are mobilized >4 fold faster than vesicles in slowly mobilized reserve pools when stimulation is 1 Hz. The finding explains why short trains of 20 Hz stimulation stained both quickly and slowly mobilized vesicles whereas 1 Hz stimulation selectively stained quickly mobilized vesicles in Figure 7 and Figure 8.

Figure 10.

The finding fits with core principles of parallel models, which predict that the decrease in fractional destaining would be occluded within seconds after the onset of stimulation at 20 Hz because the rate-limiting step in vesicle mobilization would shift upstream from catalysis of exocytosis to recruitment of reserve vesicles to the newly vacant readily releasable pool (see Box 2). However, an additional experiment was still needed to rule out serial models where 20 Hz stimulation activates a mechanism that quickly mixes together vesicles in the quickly and slowly mobilized reserve pools (or within a single last-in/first-out pool).

Such a mixing mechanism seemed unlikely because it would have to be complete within a few seconds to account for the absence of a noticeable rightward shift in the time course of destaining during 20 Hz stimulation when initiated after 1 Hz trains had already selectively destained the quickly mobilized reserve vesicles (i.e., left panel Figure 9–Figure Supplement 1). To test for a mixing mechanism in a way that would be conclusive, we destained synapses by: first stimulating at 1 Hz for 10 min to selectively destain the quickly mobilized vesicles; followed by 12 s at 20 Hz to activate any mixing mechanisms; and then again at 1 Hz to measure any effect on fractional destaining (see diagram atop Figure 10D). Figure 10–Figure Supplement 2 confirms that fractional destaining caused by 12 s of 20 Hz stimulation was not altered by long trains of 1 Hz stimulation, confirming that any mixing mechanism would have to have been fully active. Nevertheless, fractional destaining measured during the second 1 Hz train remained reduced, with no indication of any recovery (Figure 10E). This combination of results rules out activity-dependent mixing between quickly and slowly mobilized reserves, and, as a consequence, rules out any of the remaining serial models (i.e., any that were not already ruled out by the results in Figures 2 - 9).

In contrast, parallel models do predict the absence of recovery in fractional destaining in Figure 10D-E because the mechanism that causes the decrease during 1 Hz stimulation - i.e., selective depletion of dye from the quickly mobilized reserve - would quickly become fully relevant again as soon as the readily releasable pool had been replenished (i.e., within tens of seconds; Garcia-Perez and Wesseling, 2008). Taken together, these results demonstrate that quickly and slowly mobilized reserve pools are processed in parallel.

Modeling

Finally, our working model, which is a specific type of parallel model, could fit the full spectrum of results in the present report using previously estimated values for rate-limiting vesicle trafficking parameters (see Appendix 1). The exercise confirmed that parallel models can account for experimental details throughout the present study, but, given the number of parameters with otherwise unconstrained values, did not by itself provide new support for any specific parallel model. And indeed, the generic parallel model diagrammed in Box 1 and Box 2 could fit the results equally well.

Discussion

Synaptic vesicle trafficking in presynaptic terminals is central to brain function and is an increasingly important target of anti-epilepsy medicines (Lyseng-Williamson, 2011). A more detailed understanding might generate insight into the first principles underlying biological computation, and might aid second generation drug discovery (García-Pérez et al., 2015). Here we show that quickly and slowly mobilized reserve pools of vesicles are present in hippocampal synapses, and that the two types are processed in parallel, rather than in series as previously assumed.

The experiments were designed to test predictions of our working model of synaptic vesicle trafficking in presynaptic terminals (Figure 1D), which emerged from two separate lines of inquiry. The first line argued against the concept that mass action of reserve vesicles influences the timing of vesicle recruitment to the readily releasable pool (Stevens and Wesseling, 1999b; Garcia-Perez et al., 2008; see Miki et al., 2020 for a recent complementary type of evidence), while nevertheless supporting the often-linked concept that depletion of reserve pools is one of the mechanisms that causes short-term synaptic depression during extensive stimulation (Gabriel et al., 2011). The second line supported the concept that fast- and reluctantly-releasing readily releasable vesicles differ because they are docked to distinct types of release sites rather than because they are at distinct stages of biochemical priming (Wesseling and Lo, 2002; Garcia-Perez and Wesseling, 2008; Mahfooz et al., 2016; Raja et al., 2019; see also Hu et al., 2013; Böhme et al., 2016; Akbergenova et al., 2018; Maschi and Klyachko, 2020; Li et al., 2021; Karlocai et al., 2021; Gou et al., 2022).

A key point is that the present conclusion that quickly and slowly mobilized reserve vesicles are processed in parallel does not depend on either line of inquiry - or on our working model - and the evidence against serial models would continue to be equally strong even if doubts arose about the previous conclusions. Nevertheless, we do view the present result as intriguing support for our working model because: (1) the opposite conclusion - that reserves are processed in series as widely assumed until now - would have caused us to reject the model; and, (2) no alternative parallel model has yet been proposed.

Absence of mixing

Notably, our working model contains a slow undocking mechanism operating continuously at ∼1 /min that would have been consistent with slow mixing between quickly and slowly mobilized reserves if the undocked chains of vesicles were to mix freely (Gabriel et al., 2011). Because of this, the model did not anticipate the complete absence of recovery in fractional destaining over 8 min-long rest intervals in Figure 3. However, the absence of mixing does not argue against the model (see Appendix 1) because chains of vesicles might be attached to a stable cytoskeletal scaffold that would be preserved after undocking, or might reside within a liquid phase gel that prevents intermixing (e.g., Siksou et al., 2007; Fernández-Busnadiego et al., 2010; Cole et al., 2016; Milovanovic et al., 2018).

Relationship to a deep reserve

The present results do not rule out the possibility that other types of synapses additionally or instead harbor a variety of reserve pools that are connected in series (Neves and Lagnado, 1999; Richards et al., 2003). And indeed, the present results only pertain to the vesicles that are mobilized during 60 s of 20 Hz stimulation, whereas many vesicles within hippocampal presynaptic terminals are mobilized much more slowly, if at all during this time (Harata et al., 2001). We refer to the vesicles that are not mobilized in the time frame of minutes as the deep reserve. One possibility is that deep reserve vesicles do mix slowly with one or both of the quickly and slowly mobilized reserves identified above (Denker et al., 2011). This might account for the timing information in Rey et al. (2015) where vesicles that were recycled more recently were mobilized more quickly over 10 s of min, although very slow mixing between the quickly and slowly mobilize reserve pools identified above cannot be ruled out either.

Relation to multiple classes of vesicles

The experiments in the present study were designed to measure vesicle exocytosis involved in action potential evoked release, whereas transmitter released spontaneously is thought to be stored in a different class of vesicles (Kavalali, 2015).

The present results are consistent with the possibility that the decision about whether a recycling vesicle will ultimately become reluctant or fast-releasing upon entering the readily releasable pool is made earlier, at the time of entry into the reserve pool. Alternatively, the parallel mobilization of slow and fast reserves might be analogous to current ideas about parallel cycling of vesicles involved in spontaneous release, which would then require multiple classes of vesicles (Raingo et al., 2012). However, adding the concept of multiple classes of vesicles could not replace the requirement for multiple classes of release sites because, otherwise, the readily releasable pool would eventually fill with reluctantly releasable vesicles during low frequency stimulation, which would result in long-lasting depression that was not seen in experiments designed to test this (see Figure 5).

Multiplexed frequency 1ltering

Finally, our working model explains slowly mobilized reserves as a logical consequence of multiple classes of release sites, with no functional significance of their own. However, the variation among synapses documented in Figure 2–Figure Supplement 5 might nevertheless be relevant for understanding biological computation.

That is, it has already been noted that inefficient release sites that engender reluctantly releasable components of the readily releasable pool would function as high-pass/low-cut frequency filters when transmitting information encoded within presynaptic spike trains, whereas efficient release sites engendering fast-releasing components would function as low-pass/high-cut filters (Mahfooz et al., 2016). The presence of multiple types of release sites might therefore endow individual synapses with a mechanism for selectively transmitting multiple types of frequency information while filtering out other types, which is analogous to the concept of multiplexing in digital information technology. The variation among synapses in Figure 2–Figure Supplement 5 suggests that synapses in vivo likely contain the machinery for modulating multiplexing, which might provide a means for storing more information than modulating only the synaptic connection strength evident during low frequency use (Bartol et al., 2015). Notably, we previously observed extensive variation between calyx of Held synapses in ex vivo tissue in the ratio of fast-releasing to reluctant readily releasable vesicles, which, when taken together with the present results, suggests that mechanisms for modulating multiplexing might be present at a wide range of synapse types (Mahfooz et al., 2016).

Methods and Materials

Cell culture and imaging methods were similar to Raja et al. (2019).

Imaging

Imaging was performed 11-21 days after plating at 0.25 Hz, using a CCD camera (Photometrics CoolSnap HQ), and 25X oil objective (Zeiss 440842-9870), except where indicated. Focal drift was avoided in most experiments by clamping focal distance by feeding back the signal from a distance sensor attached to the objective (Physik Instrumente D-510.100, ∼1 nm resolution) to a piezoelectric objective drive (MIPOS250, PiezosystemJena). For most experiments, FM4-64 was imaged with a green LED (530 nm; 50 ms exposures) via the XF102-2 filter set from Omega. When used in combination with FM1-43, FM4-64 was instead imaged with an amber LED (590 nm; 1 s exposures) via a custom set containing band pass excitation filter FB600-10, dichroic DMLP638R, and long-pass emission filter FELH0650, all from Thorlabs. FM1-43 was imaged with a blue LED (470 nm; 200 ms exposures) via the XF100-2 filter set from Omega and FM2-10 was imaged with the same blue LED (200 ms exposures), but with the XF115-2 filter set from Omega. vGlut1-synaptopHluorin (Voglmaier et al., 2006) was expressed by infecting at day 7 after plating with an AAV1 construct and imaged with the blue LED via the XF116-2 filter set from Omega; exposure length was 200 ms except for Figure 5A where exposure length was 300 ms and the objective was 60X (Olympus UPlanFL N) rather than 25X. LEDs were all Luxeon Star and were driven with 700 mA of current. Bathing solution was exchanged continuously during imaging at 0.2 − 0.5 ml/min within a sealed chamber holding ∼35 µL. Heating to 35 C was monitored with a bead thermistor (Warner, TA-29) built into the chamber, and exposed to the extracellular solution. Electrical stimulation was bipolar (0.5 ms at - 30 V then ms at + 30 V) via two platinum electrodes built into the chamber. Neurotransmitter receptors were blocked with (in µM): picrotoxin (50); DNQX (10); and DL-APV (50). Other solutes were (in mM): NaCl (118); KCl (2); Ca²⁺ (2.6); Mg²⁺ (1.3); Glucose (30); and HEPES (25). FM4-64 and FM1-43 were used at 15 µM, and FM2-10 at 100 µM. Advasep-7 and Captisol were purchased from Cydex Pharmaceuticals or graciously provided as samples and used at 1 mM during the destaining phase of FM-dye experiments and throughout experiments using vGlut1-synaptopHluorin.

Processing

Time lapse images were aligned and Regions of Interest (ROIs) identified as described in Raja et al. (2019) (see Figure 2–Figure Supplement 4). Between 175 and 1773 ROIs were detected for each field of view. For summary statistics, the median values from a single field of view/experiment were counted as n = 1 unless otherwise indicated.

Normalization

For comparing images across preparations, median or individual ROI values were: divided by the mean value of the background region and then normalized by the baseline signal, except where indicated.

Curve fitting

The Matlab code for identifying the best fitting single exponential was:

function k = GetBestSingleExp(DcyDat, dpmin) s = fitoptions(‘Method’,’NonLinearLeastSquares’,… ‘Lower’, 0, ‘Upper’, 0.3, ‘Startpoint’, -0.002); f = fittype(‘exp((k*-1)*t)’,’options’,s, ‘independent’, ‘t’); Time = (0:(length(DcyDat)-1))/dpmin; cfun = fit(Time’,DcyDat,f); k = cfun.k;

Funding

This work was funded by: the Spanish Ministry of Science (SAF2013-48983R, BFU2016-80918-R, and PID2019-111131GB-I00) and the Unión Temporal de Empresas (UTE) project at the Centro de Investigación Médica Aplicada of the Universidad de Navarra. The funders had no role in study design, data collection and analysis, or preparation of the manuscript.

Acknowledgements

We thank Dr. Silvio Rizzoli for help understanding relationships between previous models of synaptic vesicle cycling, and Drs. Artur Llobet, Francisco Martini, Donald Lo, William Wetsel, Robert Renden, Jay Coggan, and Ana Gomis for advice about the writing.

Appendix 1

Destaining time courses fit with working model in Figure 1D

Appendix 1 Figure 1.

Parameter values specified by the results from the original electrophysiological experiments see Gabriel et al. (2011) - were the length of docked and undocked tethering units (4 vesicles, including the readily releasable vesicle when docked), the timing of recruitment of a vesicle from a docked tether unit to the release site when vacant (0.13 s⁻¹ at rest, accelerating to 0.22 s⁻¹ during 20 Hz stimulation - alpha in the Matlab code below), and the timing with which tethering units are replaced (0.017 s⁻¹ at rest, accelerating to 0.025 s⁻¹ during 20 Hz stimulation gamma in Matlab code). However, the original results only specified parameters that are rate-limiting for neurotransmitter release during intense stimulation, and there were a variety of additional parameters - relevant to vesicle trafficking, but not rate-limiting for release - that were relevant to FM-dye destaining, especially during low frequency stimulation.

The most relevant additional parameters pertained to the release sites and included: (1) the number of types of release sites; (2) the probability with which each type catalyzes exocytosis after single action potentials (i.e., p_rs for probability of release for the release site when occupied by a readily releasable vesicle, which is equivalent to p_v,hi and p_v,lo in Mahfooz et al., 2016, except with the possibility for more than two types of release sites); and, (3) the relative amount of each type.

In addition, we fixed the number of non-docked tethering units that can exchange with each docked unit at 1, meaning that each release site would process vesicles from two tethering units (8 vesicles). The value was chosen because depletion of the readily releasable pool destained synapses by slightly more than 1/8 (0.15 ± 0.01 in Figure 4). This implies that a typical synapse with 7 release sites would contain 7 × 8 = 56 recycling vesicles, which is in-line with results in Ryan et al. (1997); Harata et al. (2001); Schikorski and Stevens (2001).

Next, for fitting destaining during 20 Hz stimulation, it was necessary to allow >5-fold facilitation, at least for the release sites with the lowest values for p_rs, because otherwise 20 Hz stimulation would not exhaust the readily releasable pool: exhaustion during 20 Hz stimulation is verified experimentally in Figure 2 of Garcia-Perez et al. (2008); and >5-fold facilitation is verified in Figure 2 of Stevens and Wesseling (1999a). We had facilitation increase with a single exponential with rate parameter of 30 action potentials to maintain consistency with Figure 2 of Stevens and Wesseling (1999a), but the precise timing was not a key factor because of the flexibility provided by the three release site parameters.

Finally, we reasoned that previously docked tethering units would need to have the capacity to re-dock to the same release site. Otherwise: Either a large number of dye-stained vesicles would be trapped within synaptic terminals during long trains of low frequency stimulation; or quickly and slowly mobilized reserves would mix on the time course of minutes. A large number of trapped vesicles is ruled out in Figure 2A and elsewhere, and mixing is ruled out in Figure 3 and Figure 8, and elsewhere.

FM-dye destaining curves (Appendix 1 Figure 1) could be well fit by including as few as two types of release sites, but at least three types were required to match additionally the short-term depression seen in synaptic strength during 20 Hz stimulation (Appendix 1 Figure 2).

Appendix 1 Figure 2.

The Matlab code for simulating the model for a single release site was:

function [DestainTimeCourse, SynapticStrengthTimeCourse] = … DestainReleaseSite(NumberOfTetherUnits, Time_seconds, NumberVesiclesPerUnit, … prs_VsTime, alpha_VsTime, gamma_VsTime, … ActionPotentials_VsTime) DestainTimeCourse(length(Time_seconds)) = nan; SynapticStrengthTimeCourse(length(Time_seconds)) = 0; TetherUnits(1:NumberOfTetherUnits, 1:NumberVesiclesPerUnit) = 1; %All vesicles in all tether units are stained to start with %Each space on tether unit can be 1 (full, stained) -1(full, unstained) %or 0 (vacant) DeltaTime_seconds = Time_seconds(2)-Time_seconds(1); DockedUnit = 1; for i = 1:length(Time_seconds) DestainTimeCourse(i) = sum(TetherUnits(:)==1); %fluorescence is equivalent to number of stained vesicles if (TransitionP(gamma_VsTime(i), DeltaTime_seconds)) [DockedUnit, TetherUnits] = … SwitchDockedUnit(DockedUnit, 1:NumberOfTetherUnits, TetherUnits); end if (TetherUnits(DockedUnit,end)==0) if (TransitionP(alpha_VsTime(i),DeltaTime_seconds)) TetherUnits = AdvanceTetherUnit(DockedUnit, TetherUnits); end end if (ActionPotentials_VsTime(i) && … (TetherUnits(DockedUnit,end)∼=0) && … (prs_VsTime(i)) > rand()) TetherUnits(DockedUnit,end) = 0; SynapticStrengthTimeCourse(i) = 1; end end function P = TransitionP(DeltaTime_seconds, rateconstant) P = ((1-exp(-rateconstant*DeltaTime_seconds)) > rand()); function [NewDocked, TetherUnits] = … SwitchDockedUnit(CurrentlyDocked, AllChoices, TetherUnits) AllChoices(AllChoices==CurrentlyDocked) = []; NewDocked = randsample(AllChoices, 1); TetherUnits(NewDocked, TetherUnits(NewDocked, :)==0) = -1; % The preceding line replaces vacant spaces on tethers with unstained vesicles function TetherUnits = AdvanceTetherUnit(DockedUnit, TetherUnits) if (TetherUnits(DockedUnit,end) ∼= 0) %Only advance if vacant (error check) warning(‘Trying to advance a chain with no vacancy at end’); else for i = length(TetherUnits(DockedUnit,:)):-1:2 TetherUnits(DockedUnit, i) = TetherUnits(DockedUnit,i-1); end TetherUnits(DockedUnit, 1) = 0; %This makes the end vacant end

Simulations were repeated 1000 times for each type of release site for Appendix 1 Figure 1 and 100 000 times for Appendix 1 Figure 2 before averaging and then calculating the weighted sum across types of release sites.

Figure 2–Figure supplement 1.

Figure 2–Figure supplement 2.

Figure 2–Figure supplement 3.

Figure 2–Figure supplement 4.

Figure 2–Figure supplement 5.

Figure 2–Figure supplement 6.

Figure 2–Figure supplement 7.

Figure 7–Figure supplement 1.

Figure 7–Figure supplement 2.

Figure 8–Figure supplement 1.

Figure 8–Figure supplement 2.

Figure 8–Figure supplement 3.

Figure 9–Figure supplement 1.

Figure 9–Figure supplement 2.

Figure 10–Figure supplement 1.

Figure 10–Figure supplement 2.