Interpretation of presynaptic phenotypes of synaptic plasticity in terms of a two-step priming process

Quantitative analysis of neurotransmitter release has its roots in the early work of Sir Bernhard Katz and collaborators. Studying the stochastic nature of quantal release of acetylcholine at the neuromuscular junction, they noticed that the statistics of release is not quite compatible with the expectation for mutually independent release of synaptic vesicles (SVs) from an unlimited reservoir. Rather, they concluded that their data can best be described by a binomial distribution. Thus, they argued that there must be a limiting resource of SVs, which prevents statistical fluctuations of release from being larger than what they called the number of “units available,” and introduced the concept that release is proportional to the total number of available units, n, and a mean probability of release, $p¯$⁠, of such units (Del Castillo and Katz, 1954). Originally, n was assumed to be constant and often referred to as the readily releasable pool (RRP) of SVs (but see notable exceptions, such as Worden et al. [1997], Vere-Jones [1966], and Quastel [1997]). The classical studies on the kinetics of release and its role in short-term plasticity (STP) up to 2002 have been extensively and competently reviewed by Zucker and Regehr (2002). The priming process is seen as the tethering and docking of SVs to specialized release sites, as well as the build-up of macromolecular complexes, which mediate Ca²⁺-dependent triggering of exocytosis. Priming occurs on the tens of milliseconds to seconds timescale, while action potential (AP)-triggered SV fusion happens within tens to hundreds of microseconds.

Classical release models often assumed a homogeneous pool of primed SVs, limited by the number of release sites, which is assumed to be constant during the time of a typical electrophysiological experiment (tens of minutes). Many classical studies assumed priming to be a one-way process, such that in the absence of stimulation, eventually all release sites fill up with SVs. Consequently, the resting RRP has a fixed size, equal to the number of release sites (Liu and Tsien, 1995; Dittman and Regehr, 1998; Hallermann et al., 2010). Recently, however, evidence has been accumulated that the priming process is dynamic and reversible, such that the RRP may fluctuate in size, and even at rest a significant fraction of release sites may remain empty (reviewed by Neher and Brose [2018]). Indeed, variance–mean analysis of glutamatergic transmission at cerebellar parallel fiber to molecular layer interneurons showed that only about 45% of release sites are occupied at rest (Miki et al., 2016; Malagon et al., 2020). Further electrophysiological evidence demonstrated that synapses may transiently overfill the RRP beyond its resting state after high-frequency activity (Junge et al., 2004) and that the expression of a particular combination of synaptic proteins, comprising at least one isoform of CAPS, Munc13-1, and Munc18-1, is favorable for establishing a large stable pool at rest (Jockusch et al., 2007; He et al., 2017). Correspondingly, electron tomography demonstrated that a similar combination of proteins is required for the presence of tightly docked SVs at active zones (Imig et al., 2014). Using high-pressure quick-freeze technology together with precisely timed stimulation of synapses, it could be demonstrated that in spite of consumption of release-ready SVs during stimulation, the number of tightly docked SVs increases shortly after stimulation and reaches a peak when quick-freezing is performed about 10 ms after stimulation. However, this increase is only transient and decays back to baseline within about 100 ms (Watanabe, 2016; Chang et al., 2018; Kusick et al., 2022). Using in vitro secretion assays, Prinslow et al. (2019) demonstrated that partially assembled molecular complexes are prone to disassembly by the action of NSF. Together with mutagenesis studies of Munc13, these findings have led to a molecular scheme of step-wise and reversible buildup of the release machinery with a pivotal role of Munc13 and its regulatory domains (Neher and Brose, 2018; Dittman, 2019).

Several recent electrophysiological studies have accommodated such evidence by proposing a two-step priming process under the restrictive assumption of only one fully mature state of the release machinery, which is release-competent (Doussau et al., 2017; Miki et al., 2018; Pulido and Marty, 2018; Bornschein et al., 2020; Kobbersmed et al., 2020; Lin et al., 2022; Weingarten et al., 2022). These models thus recapitulate the step-wise assembly and maturation of the release machinery, as discussed above. Numerical simulations based on such priming schemes are capable of reproducing a variety of electrophysiological findings. In reference to the results from electron tomography, the two states in question may be called “loosely docked” (LS) and “tightly docked” (TS) (Neher and Brose, 2018). It is an open question, however, whether the postulated functional states of the model actually correspond to the two morphologically defined states or else represent some other steps in the buildup of the molecular release machinery.

Alternatively, kinetic models have been developed, which assume that SVs migrate between or dock at different release sites (Grande and Wang, 2011), from which release occurs in parallel with different release probabilities, p_r (high and low), and different priming rates (fast and slow) (Hallermann et al., 2010; Delvendahl et al., 2013; Mahfooz et al., 2016). Many electrophysiological findings, including biphasic depression and recovery kinetics, can readily be explained both by parallel and sequential models. It is obvious, though, that molecular interpretations of such features, as well as of the changes observed after genetic or pharmacological manipulations, may differ substantially. Specifically, many features, which on the basis of parallel or classical single-pool models are explained as differences and variations in release probability, can very well also be explained assuming constant release probability combined with changes in the resting equilibrium and dynamics between priming states (Pulido and Marty, 2018; Lin et al., 2022). In the following one variety of a sequential model, the LS/TS model (Neher and Brose, 2018; Lin et al., 2022) is shortly explained and a few examples are given for differences in the interpretation of published data when viewed in terms of classical pool models or else as a reversible sequence of priming steps during the buildup of a fully functional release machinery.

As detailed above, the kinetic scheme of the LS/TS model was inspired by the finding that the priming process is reversible (Smith et al., 1998; Murthy and Stevens, 1999; He et al., 2017) and by the evidence from electron microscopy that under conditions of intact priming a subset of SVs near the plasma membrane is tightly docked within 2 nm distance (Imig et al., 2014; Maus et al., 2020). Considering such findings, Neher and Brose (2018) proposed a model (Fig. 1 A) with a fixed number, N, of release sites, which at any given time can either be empty (state ES) or occupied by an SV in one of two states: the TS state or the LS state. The model furthermore assumes that docked vesicles can reversibly change states with rate constants in the forward priming direction, k₁ and k₂, being accelerated by activity (and more specifically by cytosolic [Ca²⁺]) and with backward rate constants b₁ and b₂ (Fig. 1, A and B). The latter were assumed to be constant for simplicity. Importantly, SV fusion was allowed to occur only from the fully primed state TS with a probability (per AP) of p_fusion. Surprisingly, short-term plasticity patterns recorded at the calyx of the Held synapse in response to stimulation up to 20 Hz could be simulated by this relatively simple kinetic scheme (Lin et al., 2022). As described in more detail below in the context of short-term depression and frequency-invariant transmission, the assumption of priming rates linearly increasing with cytosolic [Ca²⁺] leads to a particularly simple scenario with a plateau in the relationship between the level of synaptic depression and stimulation frequency in the range 5–20 Hz.

For higher frequencies (>20 Hz), three observations require extensions of the model: (1) the steady-state quantal content of release declines progressively with increasing frequency, reflecting limitations of vesicle supply; (2) many synapses exhibit prominent paired-pulse facilitation (PPF); and (3) p_fusion increases, in part due to Ca²⁺-current facilitation (Borst and Sakmann, 1998; Cuttle et al., 1998; Müller et al., 2008; Lin et al., 2012). The first observation, a limited rate of supply of SVs, can conveniently be implemented by assigning a Michaelis–Menten type saturation to the Ca²⁺-dependence of one or both of the forward priming rate constants. Alternatively, so-called “release-site clearing” may be introduced as a process limiting the rate of SV supply. In that case, a site being vacated by SV fusion is assumed to reside in a refractory state for a certain time before being able to accept a new vesicle (state “ERS,” Fig. 1 C). The second feature, PPF observed at frequencies ≥50 Hz, depends on several aspects of the model (see below). In particular, it requires an extension, which specifically simulates the discovery by electron tomography of a transient increase of tightly docked SVs after APs. This transient tight docking lasts for about 50–100 ms (Kusick et al., 2022), which is similar to the time of decay of PPF (Chang et al., 2018). It is implemented in the model as a fusion-competent “labile tight state” (TSL; Fig. 1 C; Lin et al., 2022). TSL vesicles are assumed to have the same p_fusion as TS vesicles but transition back to LS with a rate constant much faster than the backward rate constant b₂ by which the TS vesicles convert back to LS (see more details in the paragraph of PPF, below).

The LS/TS model is not only able to reproduce several experimentally observed aspects of short-term plasticity quite faithfully but it can also account for the large heterogeneity in synaptic strength which is observed at the calyx of Held and several other glutamatergic synapses (typically varying over a factor of 10). By means of varying the relative abundance of fully primed TS vesicles in resting synapses, the LS/TS model offers a potent mechanism for adjusting synaptic strength without the need to postulate changes in release probability and total number of release sites. Similar conclusions were reached by Pulido and Marty (2018) using a slightly different two-step priming model. Furthermore, incompletely primed, but docked vesicles were recently postulated for hippocampal synapses between CA1 pyramidal cells and oriens lacunosum-moleculare (O-LM) interneurons by Aldahabi et al. (2022).

Before discussing changes in pool size and release probability, as observed during experimental manipulations, some clarification is required regarding the methods that are commonly used for the estimation of these quantities.

Most traditional methods of quantal analysis implicitly assume the classical model of a single homogenous RRP (see Neher [2015] and Thanawala and Regehr [2016] for review). Common to these methods is the application of a strong stimulus (high-frequency train of APs, hypertonic stress, or direct presynaptic depolarization under voltage-clamp) to deplete the RRP within a time short enough, such that pool replenishment during the stimulus is negligible, or else can be corrected for. The RRP is then calculated as the replenishment-corrected cumulative release. Subsequently, a quantity commonly termed release fraction (F), fusion efficiency (f_e), or vesicular release probability (p_vr) can be obtained as the fraction of the RRP released by a single AP. For a parallel model, assuming two or more subpools of fusion-competent SVs, this fraction represents an average of release probabilities of contributing subpools, weighted by the respective subpool sizes at rest. For the sequential (LS/TS) model, this fraction reports the “true” release probability (p_fusion) of TS vesicles only, if the RRP estimate corresponds to just the TS subpool. However, modeling of data obtained from the calyx of Held synapses indicates that typical pool-depleting stimuli deplete both the LS and TS subpools (as discussed in Lin et al. [2022] and Neher and Taschenberger [2021]). Thus, the quantity reported as RRP is the sum of LS and TS subpools present at stimulation onset, and the fraction of the RRP released by a single AP represents an apparent release probability, which depends on the ratio of subpool sizes. Correspondingly, changes in apparent release probability upon molecular perturbations may well represent changes in the LS/TS ratio at constant p_fusion. Thus, a role in the process of SV fusion might be erroneously assigned to a protein under study.

The analysis of mean and variance of quantal release, elicited by repetitively applied stimuli, is often used to determine RRP size, the number of release sites, N, and release probability (Silver, 2003). For a homogeneous vesicle pool of size N, which refills completely in pauses between individual trials, variance is given by σ² = N · q² · p_r · (1 − p_r), where q is the quantal size and p_r is the probability of release. A plot of variance versus mean, measured under several different conditions, such as different concentrations of extracellular Ca²⁺, can be fitted by a parabola. This curve starts near the origin with low values of both mean and variance from measurements at low [Ca²⁺] with small p_r. With increasing p_r, variance reaches a maximum at p_r = 0.5 and subsequently decreases due to the term (1 − p_r) in the equation for variance. The parabola approaches 0 when p_r approaches 1, which means that there are no more trial-to-trial fluctuations since the same number of SVs (= N) is released at each trial. The intersection of the parabola with the abscissa is therefore given by the product of N*q. Quantal amplitude, q, is given by the initial slope of the parabola. Thus, the number of release sites, N, can be calculated. Importantly, this holds under the assumption of a single homogeneous pool, which fills up completely between stimuli.

However, for reversible and dynamic priming, responses fluctuate even for p_fusion = 1 due to the stochastic nature of the priming process. Thus, p_r in the equation for variance has to be replaced by the product of p_fusion and p_occupancy, where p_occupancy represents the fraction of release sites occupied by fusion-competent SVs (Vere-Jones, 1966; Quastel, 1997; Scheuss and Neher, 2001). Therefore, the parabola in its descending part approaches 0 only when both p_fusion and p_occupancy approach the value of 1. This is the case only when at the time of stimulation almost all sites are occupied (p_occupancy = 1) and all fusion-competent SVs are released (p_fusion = 1). Unfortunately, such conditions are difficult to obtain (Malagon et al., 2020), requiring the strongest stimulation and conditions of enhanced priming. In addition, saturation and desensitization of postsynaptic receptors may distort the descending branch of the parabola (Meyer et al., 2001) unless quanta can be counted directly, as is the case for some types of synapses in the cerebellum (Malagon et al., 2020; Silva et al., 2021) and hippocampus (Tanaka et al., 2021). As a consequence, the parabola may include unreliable data points on its descending branch and the extrapolation toward the intersection with the abscissa may be inaccurate. Provided N can be estimated reliably, one can calculate p_r for a given condition as the fraction of N released by a single stimulus. However, there is still the problem of splitting p_r into its factors p_fusion and p_occupancy, which is not possible on the basis of mean and variance alone.

In the absence of readily available methods to separate priming and fusion by statistical methods, several procedures have been described to address the multiplicity of pools and functional states directly. Next for optimization of model fits, assuming pools with different topologies and kinetics (Hallermann et al., 2010; Delvendahl et al., 2013; Doussau et al., 2017) special properties of postulated subpools have been exploited to determine their properties. For instance, the assumption of a subpool of “superprimed” SVs with high release probability, dependent on the presence of at least one isoform of Rab3, has been used for its isolation by subtraction of responses recorded in synapses of quadruple Rab3 knockout mice from those recorded in control synapses (Schlüter et al., 2004). Subtraction of control responses from those measured after induction of LTP has been used to demonstrate that LTP increases a component of release with high p_fusion and slow recovery kinetics (Weichard et al., 2023). Special statistical properties of so-called “pre-primed” SVs have been used to estimate their contribution to overall release at hippocampal synapses (Gustafsson et al., 2019). Likewise, the contribution of SVs occupying “docking sites,” as opposed to those occupying “replacement sites,” has been estimated by detailed model fitting and statistical analysis of quantal responses to repetitive, high-frequency stimulation (Miki et al., 2016, 2018; Silva et al., 2021). A variant of the technique of non-negative tensor factorization (NTF) has been developed to determine the release probability of TS vesicles in a way that is consistent with the LS/TS model and with the variation of synaptic strength between synapses (Neher and Taschenberger, 2021; Lin et al., 2022).

Unfortunately, it is still unproven whether any of these approaches allow an unambiguous separation between the effects of a given manipulation on vesicle priming from those on vesicle fusion or else an unequivocal differentiation between parallel and sequential models. However, as pointed out by Miki et al. (2016), variance analysis of cumulative release at cerebellar synapses is more readily explained by sequential than by parallel models. Nevertheless, one should be aware that alternative interpretations are possible, which lead to quite different conclusions in terms of molecular mechanisms when either a parallel or else a sequential scheme is assumed. Some examples will be discussed here.

Markram and Tsodyks (1996) studied changes in STP following the induction of presynaptic LTP in layer 5 neocortical synapses. They observed that single EPSPs were potentiated significantly. However, the sum of EPSPs during short bursts of activity was hardly changed because the increase in the first EPSP of a burst was compensated by decreases in subsequent EPSPs. They called this phenomenon “Redistribution of Synaptic Efficacy” and pointed out its high relevance for information processing since APs occur often in short bursts in the brain area studied.

Standard quantal analysis within the classical model can readily explain this finding by assuming a change in release probability and a limited RRP, which is full at rest and refilled relatively slowly. In this scenario, presynaptic AP bursts would progressively empty the RRP and this would happen faster at a higher release probability such as after induction of LTP. The first response would be larger and subsequent ones smaller due to the limitation of RRP sizes. Thus, a reasonable conclusion from these findings is that induction of this kind of presynaptic LTP causes an increase in release probability. One might therefore argue that this change is primarily mediated by proteins regulating exocytosis, such as SNAREs, synaptotagmin 1 and 2, Ca²⁺ channels, and/or affecting the coupling of Ca²⁺ channels to release sites.

However, as shown by Lin et al. (2022), a shift in the balance between LS and TS subpools at rest in favor of the latter might equally well cause the observed redistribution of synaptic efficacy, resulting in larger initial amplitudes and faster fatigue of responses during bursts. As suggested before (Neher and Brose, 2018), such a shift may well be caused by the priming protein Munc13, the function of which is regulated by second messengers such as [Ca²⁺], diacylglycerol, and phosphoinositides. Thus, considering two-step priming schemes opens up completely different options for interpretation. In fact, a recent study by Weichard et al. (2023) could describe changes in synaptic strength and STP following LTP induction equally well with release models based on either parallel or sequential priming schemes. Importantly, increased synaptic strength and slower recovery from synaptic depression following LTP induction were found to arise primarily from a larger proportion of fully primed slowly recovering SV in the absence of a change in release probability.

Synaptic strength of glutamatergic synapses of a given type often varies by about an order of magnitude (Harris and Stevens, 1989; Debanne et al., 1996; Dobrunz and Stevens, 1997; Losonczy et al., 2002; Taschenberger et al., 2016). Presynaptic parameters contributing to this heterogeneity may of course be differences in the size and morphology of terminals, giving rise to varying numbers of active zones and release sites (Harris and Stevens, 1989; Grande and Wang, 2011). Very often synaptic strength is negatively correlated with “paired pulse ratio” (PPR), the ratio of second over first responses to a pair or burst of stimuli. A change in PPR is generally considered an indication of a change in release probability. Thus, as in the case of induction of LTP discussed above, differences in PPR between synapses may well be interpreted as differences in release probability. However, as shown by Lin et al. (2022), this feature can again be explained by variation in the TS/LS ratio at rest. In this context it should be noted that Munc13, a prime candidate for regulation of the TS/LS ratio (Betz et al., 1998; Rhee et al., 2002; Junge et al., 2004; Shin et al., 2010; Neher and Brose, 2018; Dittman, 2019; Rizo, 2022), is expressed in at least two isoforms in the brain, which may confer different degrees of stability to vesicles in the fully primed state (Rosenmund et al., 2002; He et al., 2017), but see Holderith et al. (2022). Thus, rather than differences in the release mechanism, differential expression of isoforms of priming proteins might be the cause of the diversity of synaptic strength among synapses of one type, between synapses of different types, or after activity-induced long-term changes in synaptic strength including LTP and LTD (Weichard et al., 2023). Furthermore, Munc13 has regulatory binding sites for [Ca²⁺], calmodulin, and diacylglycerol (Betz et al., 1998; Rhee et al., 2002; Lipstein et al., 2013), providing options for regulation spanning a time range from milliseconds to hours and days (Tewson et al., 2012; Kruse et al., 2016; Schuhmacher et al., 2020). The distribution of synaptic strengths may well reflect the history of modulatory influences, both long- and short-term, that a given synapse was exposed to.

It should be noted that in the classical view of a single homogenous vesicle pool and a non-reversible priming process, only the number of release sites and the release probability can be invoked as presynaptic parameters describing diversity in synaptic strength since all release sites will eventually be fully occupied at rest. Reversibility of vesicle priming allows for partial occupancy of release sites at rest and, therefore, provides options for dynamic changes in both directions, facilitatory and depressing (Pulido and Marty, 2018). The repertoire of modulatory influences on neurotransmitter release is thus expanded, possibly comprising the whole range of second messenger-mediated signaling.

Depression of synaptic responses during train stimulation can have pre- and postsynaptic causes (Zucker and Regehr, 2002). Depletion of the RRP due to limited resupply of readily releasable vesicles is generally considered to be the most relevant presynaptic contribution to depression (Elmqvist and Quastel, 1965; Betz, 1970; Rosenmund and Stevens, 1996; Tanaka et al., 2021). During repetitive stimulation, the response amplitude reaches a steady state that represents the balance between the mean release rate (the product of release probability, stimulation frequency, and remaining RRP) and the mean rate of vesicle resupply. Notably, it was often found that the rate of resupply necessary to account for the observed steady-state release during stimulation was substantially higher than that estimated from the time course of recovery of the RRP after stimulation. Thus, it was postulated that the vesicle priming rate must be higher during stimulation than during recovery at rest. Ca²⁺-dependent enhancement of vesicle supply was therefore postulated (Dittman and Regehr, 1996; Stevens and Wesseling, 1998; Wang and Kaczmarek, 1998; Dittman et al., 2000; Hosoi et al., 2007; Hallermann and Silver, 2013; Mahfooz et al., 2016; Weingarten et al., 2022). Recent studies suggest synaptotagmin 3 (Weingarten et al., 2022), synaptotagmin 7 (Liu et al., 2014; Chen et al., 2017; Tawfik et al., 2021; Kusick et al., 2023), and the C2B domain as well as the calmodulin-binding domain of Munc13 (Lipstein et al., 2013, 2021) to be candidates for the Ca²⁺-sensor mediating this type of stimulus-dependent enhancement of the priming process. The Ca²⁺-dependence of the resupply of release-ready SVs may lead to biphasic recovery from depression if stimulus trains elicit long-lasting increases in presynaptic residual [Ca²⁺].

Biphasic depression and recovery time courses, typically observed at various glutamatergic synapses, have been simulated by models assuming two populations of vesicles or release sites (parallel model): a high-p population with slow priming and a low-p population with fast priming (Ritzau-Jost et al., 2018). However, the LS/TS model with only a single type of release site can also reproduce all aspects of depression and its recovery studied at the calyx of Held (Lin et al., 2022). The model assumes forward priming rate constants to be described by the sums of their basal values at rest plus components, which rise linearly with cytosolic [Ca²⁺]. This allows one to distinguish two frequency ranges for describing responses to repetitive stimulation: a low-frequency range (0.1–2 Hz) in which steady-state release is determined by both forward and backward transitions and a mid-frequency range (5–20 Hz) in which forward transitions dominate over backward transitions. In the low-frequency range, steady-state release decreases with increasing frequency because vesicle resupply during long interstimulus intervals (ISIs) is primarily determined by the basal rate constants at rest. Resupply is less for shorter ISIs (higher frequencies). In the mid-frequency range, however, basal priming during short ISIs is negligible in comparison to the accelerated priming during AP-induced Ca²⁺ transient. Each transient shifts certain fractions of SVs from their pre-AP state to the corresponding downstream state depending on the time integral over the Ca²⁺ transient and the steepness of the Ca²⁺-dependence of priming. The fractions transferred between the first and second priming steps will be referred to as s₁ and s₂ in the following. Provided that AP-induced Ca²⁺ transients do not summate significantly in this frequency range, APs act nearly “autonomously,” advancing such fractions of SVs irrespective of the ISI duration. Likewise, p_fusion is constant in this range of stimulation frequencies. This results in steady-state levels of site occupancies, which are almost independent of frequency since both release and resupply increase nearly linearly with stimulation frequency. Steady-state release then depends mainly on three parameters: s₁, s₂, and p_fusion (Lin et al., 2022). Model synapses display depression or facilitation at low frequencies depending on whether their steady-state TS occupancy is lower or higher than resting TS occupancy. The latter is determined by four other model parameters, namely the resting values of forward and backward priming rate constants (k_1,rest, k_2,rest, b₁, and b₂). Thus, one can adjust TS occupancies at rest and at steady-state independently from each other, a consideration that will be relevant when discussing facilitation at low frequencies and frequency-invariant transmission below.

At the calyx of Held, the plateau level of steady-state synaptic depression is located at about 35% of the initial EPSC (see Fig. 2 A and Fig. 3 A, gray traces). Thus, these synapses display depression in the low-frequency range. At frequencies >20 Hz, the steady-state release per AP deviates from the midfrequency plateau declining progressively due to a saturation of vesicle supply (see below).

Some types of synapses show facilitation instead of depression at low frequencies, often referred to as FF or augmentation (Zucker and Regehr, 2002). Common to these types of short-term plasticity is a very large dynamic range (an increase up to 10-fold) and an onset already at stimulation frequencies as low as 1 Hz (Toth et al., 2000; Nicoll and Schmitz, 2005; Chamberland et al., 2014; Midorikawa and Sakaba, 2017). At intermediate frequencies (10–100 Hz) and low extracellular [Ca²⁺], FF is hard to distinguish from PPF, another type of facilitation to be described below. Therefore, the two types of facilitation are often conceived together as short-term facilitation. Implicitly, the same “Ca²⁺-sensor” is assumed to mediate Ca²⁺ sensitivity (Jackman et al., 2016). The two-step priming scheme, however, suggests a very different mechanism for facilitation in the frequency of 1–20 Hz in comparison with that at 100–200 Hz. As detailed above, the LS/TS model, as applied to the calyx of Held data, predicts that steady-state release is relatively constant during repetitive stimulation at 5–20 Hz and that this synapse shows depression since the resting occupancy of TS is higher than that at steady state. The model prediction of depression can readily be converted into facilitation by assuming a lower occupancy of the TS state at rest. Consequently, the first EPSC in a stimulus train will be small. During the train, release will increase toward the steady-state level, as determined by parameters s₁, s₂, and p_fusion. Fig. 2 A (red traces) shows predictions for release time courses in response to 5 Hz stimulation for different relative occupancies of TS at rest. Note that all release time courses are calculated using the same p_fusion = 0.39, which was assumed to be constant throughout trains. Individual traces start at different initial quantal content (m₁) depending on the chosen value for k_2,rest and converge toward similar steady-state quantal contents (m_ss). It turns out that a value of 0.044 (priming events per site and second) for the priming rate constant k_2,rest results in the release that is almost constant during the entire stimulus train (Fig. 2 A; open circles). Fig. 2 B shows the ratio “steady-state release over first release” as a function of the LS/TS ratio. It is seen that “facilitation,” defined this way, reaches high values for LS-dominated synapses (large LS/TS ratio), whereas this ratio is <1 for TS-dominated synapses. In this sense, facilitation and depression at low frequencies reflect the same process and FF may be considered “negative depression,” not involving changes in p_fusion.

Models for FF with similar “stimulus-dependent mobilization” were actually proposed by Worden et al. (1997) and Pan and Zucker (2009) for describing this phenomenon at the lobster neuromuscular junction. FF at cerebellar granule to Purkinje cell synapses (Doussau et al., 2017) was also postulated to be the result of a stimulus-dependent shift between two states of priming. The important point, shown in Fig. 2, is that one can change a model of low-frequency depression, as observed in the calyx of Held, into one for FF by just changing one parameter, the resting rate constant of priming. In this view, the Ca²⁺ sensor for FF is the same as the one that enhances priming at elevated cytosolic [Ca²⁺], candidates for which are Munc13 (Lipstein et al., 2021), synaptotagmin 7 (Turecek and Regehr, 2018), and/or synaptotagmin 3 (Weingarten et al., 2022). Enhancement of p_fusion is not required for simulating FF at low frequencies.

The calyx of Held, a synaptic terminal in the auditory pathway, is often considered a simple one-to-one relay for converting presynaptic into postsynaptic spike trains, irrespective of frequency. This was shown to be actually the case for calyx synapses of adult mice (Lorteije et al., 2009). Such frequency invariance is also observed in some other types of synapses involved in sensory signal processing and motor control (MacLeod et al., 2007; Bagnall et al., 2008; Turecek et al., 2016, 2017). Synaptotagmin 7 is required to obtain frequency invariance in these synapses through a facilitating mechanism compensating for short-term depression (Turecek et al., 2017). In contrast to the situation for adult animals, these synapses and the calyx of Held synapse display pronounced short-term depression during early development, which gradually diminishes with age. For P13–P15 calyx synapses, which is the developmental stage studied by Lin et al. (2022), a plateau region is observed in the relationship between steady-state synaptic strength and stimulation frequency (see above and Fig. 3 A, trace calyx of Held). It may be speculated that this plateau is an early sign of frequency invariance, as it develops. If so, the mechanism discussed above in the context of depression and frequency facilitation may give a hint on how synapses can achieve frequency invariance over a wider frequency range.

Parameters for the resting occupancy of TS can be chosen in ways that the plateau level of release is approached during a stimulus train either from above (depression) or from below (FF; see above). Likewise, parameters can be chosen to obtain an initial release matching the plateau level. Then TS-occupancy and steady-state release are frequency invariant from very low frequencies up to 20 Hz without the need to invoke a balance between changes in p_fusion and those in priming (see Fig. 3 A, trace ‘low-frequency invariant’). At the high-frequency end, the simple model, as discussed so far, has to be extended by several features, such as a rise in p_fusion, saturation of vesicle priming, and release from a labile tight state, TSL (see above and the section on PPF). Such effects dominate steady-state release at a frequency >50 Hz. Thus, one can conceive several mechanisms on how to extend the range of frequency invariance to higher frequencies, for instance by reducing saturation of priming (see Fig. 3 A, trace ‘extended’ for an example). This change extends the frequency range of near-constant steady-state release up to 100 Hz with remaining variations confined to about ±10%. Importantly, such adjustments involve changes in p_fusion only for the highest frequencies (100 and 200 Hz). In contrast frequency invariance at synapses for sensory signal processing and motor control was explained as a balance between depression and facilitation, the latter being simulated as an increase in release probability (Turecek et al., 2017).

In spite of the constancy of steady-state release over a wide frequency range, the model proposed here does predict dynamic changes in release during high-frequency trains or when abruptly changing frequency. Fig. 3, B and C, shows examples for such time courses. Similar transient dynamic changes were also reported for frequency-invariant synapses between Purkinje cells and deep cerebellar nuclei and vestibular synapses (Turecek et al., 2017).

PPF, as discussed here, is one of the most studied forms of facilitation. It can be observed during pairwise stimulation at short interstimulus intervals in many types of glutamatergic synapses (Atluri and Regehr, 1996; Zucker and Regehr, 2002) and also at the neuromuscular junction (Katz and Miledi, 1968). Typically, only the second and/or third response in a stimulus train is augmented and subsequent responses diminish rapidly. The same types of synapses often show pronounced depression at lower frequencies (Varela et al., 1997; Sun et al., 2005). At the calyx of Held, indications of PPF are observed in a fraction of synapses, if stimulated at frequencies ≥50 Hz (Lin et al., 2022; see Fig. 2 A, gray trace, open circles). The extent of PPF strongly depends on residual Ca²⁺. Several studies, therefore, tried to identify “the Ca²⁺-sensor” of PPF (Zucker and Regehr, 2002; Jackman et al., 2016), often assuming that this sensor, when Ca²⁺-bound, enhances release probability. However, in the framework of the LS/TS model in its extended form (Lin et al., 2022), Fig. 1 C, PPR depends on the interplay of at least four factors (Lin et al., 2022):

• (1)

  Depletion of the RRP by the release during the first stimulus depending on p_fusion.

• (2)

  Facilitation of presynaptic Ca²⁺-current (Borst and Sakmann, 1998; Müller et al., 2008), increasing p_fusion of the second response.

• (3)

  Ca²⁺-dependent enhancement of vesicle replenishment (depending on s₂).

• (4)

  A transient boost of replenishment of fusion-competent SVs by the labile state, TSL.

Three of these factors contribute to an increase in the second response in a pair of stimuli, while the first factor counteracts such facilitation. The labile state (TSL; Fig. 1 C) was introduced to simulate the transient docking of SVs after an AP, as observed by “flash and freeze” electron microscopy (Kusick et al., 2022). Its contribution to release was found to be essential for an adequate simulation of PPF as observed at the calyx of Held synapses (Lin et al., 2022). Specifically, about half of the depletion of the RRP due to release is compensated by the recruitment of TSL vesicles, which transiently contribute to the release of the second, but not to that of the first stimulus in a train. The full contribution of TSL vesicles is obtained only for frequencies of 100 Hz and higher, given the assumption of the model that the increase of tightly docked SVs after an AP is transient and decays with a time constant of about 50 ms (Kusick et al., 2022). In contrast, the contribution to PPF by the regular refilling process at lower frequencies, represented by the Ca²⁺-dependent priming rate constant k₂ in the model, is minor, although this process is enhanced by ([Ca²⁺]) (Lin et al., 2022).

Commonly, PPF is called and discussed as such only, if the net effect results in PPR >1. In order for the calyx model to achieve such net facilitation, the release probability of the second stimulus needs to be increased by about 20% (half of the 40% reduction by depletion). Ca²⁺ current facilitation and saturation of Ca²⁺ buffers may contribute to this effect, as shown by detailed studies of these mechanisms (Borst and Sakmann, 1998; Cuttle et al., 1998; Rozov et al., 2001; Müller et al., 2008; Lin et al., 2012; Eisner et al., 2023). The finding of only a limited contribution of increased release probability to PPF is in line with a recent study at cerebellar synapses in which short-term plasticity at 100 Hz could fully be explained by a two-step priming model without the assumption of an increase in p_fusion (Miki et al., 2016). Common to both models of release at the cerebellar and calyx of Held synapse is the assumption that a large part of PPF is mediated by a change in occupancy of the fusion-competent vesicle state—in contrast to the dominating view that PPF mainly reflects p_fusion.

Taken together, the sequential two-step priming scheme provides options for the explanation of seemingly very different phenomena, such as short-term depression, frequency facilitation, and frequency invariance based on properties of the vesicle priming process, without having to postulate changes in the release probability of fully primed vesicles (Fig. 4). This does not exclude, however, that such changes may be responsible for certain plasticity features or else contribute to their fine-tuning.

Conceiving the priming of synaptic vesicles as a reversible, dynamic, and regulated multistep process opens up novel options for the interpretation of the role of presynaptic proteins in short-term plasticity and in generating diversity of synaptic strength among synapses. In particular, many changes in synaptic function induced by molecular perturbations or pharmacological manipulations of synaptic proteins that previously were interpreted as changes in release probability can be understood—at least in part—as changes in the balance between fully primed (TS) vesicles and immaturely primed (LS) vesicles. This balance, in turn, is determined, next to release, by activity-dependent enhancement of priming. Several features of synapse heterogeneity and short-term plasticity can be explained on this basis, without postulating differences or changes in release probability. Unfortunately, distinguishing between this “priming-centered” view and alternative classical models by electrophysiological experiments alone is difficult, involving detailed statistical analysis of responses to stimulation trains (Miki et al., 2016; Lin et al., 2022) or consideration of specific properties of certain types of synapses (see above). In line with the arguments about the importance of priming, as discussed here, a recent study, using novel methods of fluctuation analysis, concluded that short-term depression at the hippocampal mossy fiber–CA3 interneuron synapse is the result of a reduction in site occupancy rather than in the probability of exocytosis (Tanaka et al., 2021). Thus, a deeper understanding of the processes leading to and controlling neurotransmitter release will certainly benefit from a combination of molecular, optical, and functional studies (Blanchard et al., 2020). It is expected that this understanding will uncover the roles of several proteins in vesicle priming, which so far have been associated with the process of vesicle fusion.

Joseph A. Mindell served as editor.

The author thanks Alain Marty, Holger Taschenberger, Kun-Han Lin, and Nils Brose for their excellent advice on the manuscript. A particular thank you to Holger Taschenberger and Hartmut Sebesse for help on figures and illustrations.

Open Access funding provided by the Max Planck Society.