ReviewSynaptic Transmission

A Unified Model of the Presynaptic and Postsynaptic Changes During LTP at CA1 Synapses

See allHide authors and affiliations

Science's STKE  10 Oct 2006:
Vol. 2006, Issue 356, pp. re11
DOI: 10.1126/stke.3562006re11


Long-term potentiation (LTP) has been studied extensively at CA1 synapses of the hippocampus, and there is evidence implicating both postsynaptic and presynaptic changes in this process. These changes include (i) addition of AMPA channels to the extrasynaptic membrane and diffusional equilibrium of extrasynaptic receptors with synaptic receptors, (ii) sudden addition of AMPA channels to the synapse in large groups, (iii) a change in the mode of glutamate release (presumably from kiss-and-run to full fusion), and (iv) a delayed increase in the number of vesicles released. However, it remains unclear whether (or how) these changes work together. We have incorporated all of these processes into a structural model of the synapse. We propose that the synapse is composed of transsynaptic modules that function quasi-independently in AMPA-mediated transmission. Under basal conditions, synapses are partially silent; some modules are AMPA-silent (but contribute to NMDA-mediated transmission), whereas others are functional (and contribute to both AMPA- and NMDA-mediated transmission). During LTP, there is both a rapid change in the mode of vesicle fusion and a rapid insertion of a postsynaptic complex (a hyperslot) containing many proteins (slots) capable of binding AMPA channels. The combined effect of these pre- and postsynaptic changes is to convert AMPA-silent modules into functional modules. Slot filling is transiently enhanced by a rapid increase in extrasynaptic GluR1, a form of the AMPA-type receptor. A slower transsynaptic growth process adds AMPA-silent modules to the synapse, enhancing the number of vesicles released and thereby enhancing the NMDA response. This model accounts for a broad range of data, including the LTP-induced changes in quantal parameters. The model also provides a coherent explanation for the diverse effects of GluR1 knockout on basal transmission, LTP, and distance-dependent scaling.


CA1 hippocampal synapses have been a major model system for understanding basal synaptic transmission and synaptic plasticity. These synapses are glutamatergic, have vesicle-mediated transmission, and demonstrate long-term potentiation (LTP), an activity-dependent process that leads to long-lasting changes in synaptic strength. The properties of these synapses have been investigated in hundreds of studies and much has been learned about their properties, both in the basal state and after LTP (Appendix 1).

Glutamate binds to postsynaptic N-methyl-D-aspartate (NMDA)–type receptors (NMDARs) and AMPA (α-amino-5-hydroxy-3-methyl-4-isoxazole propionic acid)–type receptors (AMPARs), ligand-activated channels that underlie the excitatory postsynaptic potential (EPSP). Various presynaptic and postsynaptic processes have been proposed to account for the enhanced transmission seen after induction of LTP. These include different types of postsynaptic processes that result in the addition of the AMPA type of glutamatergic channel to the synapse, as well as different types of presynaptic processes that change the mode or probability of vesicle release. It remains unclear how these processes could work together to produce LTP. Furthermore, there are several puzzling observations that seem difficult to reconcile with each other (see below). Thus, many investigators view the experimental results as internally contradictory.

We have assumed that all of the proposed presynaptic and postsynaptic processes indeed occur and have attempted to incorporate them into a unified model, making a minimal number of additional assumptions. Because the data set is so extensive, many models that we explored could eventually be ruled out. The model that we present here is consistent with the broad range of data listed in Appendix 1 and provides a satisfactory resolution of puzzling aspects of the data set.

Our model provides an explanation of how LTP changes the quantal properties of synaptic transmission, namely, quantal size (the synaptic response to a single vesicle) and quantal content (the number of effective vesicles released in response to a nerve impulse). Both quantal size and quantal content are increased during early LTP (14), whereas at later times (>30 min) it appears that LTP is due primarily to a change in quantal content (5).

Our model is the first comprehensive model for LTP that has been put forward, but it is desirable for competing models of this kind to be developed, prompting experiments to distinguish among them. In Appendix 2, we list specific predictions of the model proposed here. Establishing a successful model of the processes involved in LTP will provide a sound framework for identifying the molecular basis of each process.


The basic properties of the model are shown in Figs. 1 and 2. Figure 1 deals with the properties of basal synaptic transmission, in particular the basis of distance-dependent scaling (whereby average synaptic strength increases with distance from the soma, thereby compensating for electrotonic decrement and equalizing the effectiveness of synapses at depolarizing the soma). The figure illustrates the modular structure of a partially silent synapse (see caption for definitions). Figure 2 shows how the synapse changes during the two phases of LTP.

Fig. 1.

Model of distance-dependent scaling at partially silent distal and proximal synapses. The synaptic structure shown here consists of two transsynaptic modules (blue boxes). As shown, each module has one vesicle release site on the presynaptic side. On the postsynaptic side, the module can accommodate a single hyperslot (group of green and blue slot proteins; only three of ~20 are shown). Blue slot proteins bind GluR2/3 AMPA channels; green slot proteins bind GluR1+ AMPA channels. NMDA channels are held in the synapse by purple slot proteins. Glutamate in the cleft is shown as local within the cleft to indicate that only AMPA channels near the release site will be activated. NMDA channels, which are more sensitive to glutamate, are activated wherever release occurs. In each diagram of the synapse, the left module is "silent" both because glutamate is released slowly (small fusion pore) and because the module lacks a hyperslot (green and blue slot proteins). The right module is AMPA-functional. GluR1+ slots in this module are variably filled, depending on a diffusional equilibrium with extrasynaptic GLuR1+ receptors. The GluR2/3 slots are constitutively filled. (A) In the wild type (wt), more GluR1+ slots are filled distally because of the higher extrasynaptic GluR1+ concentration. This results in a higher local AMPA channel density and therefore larger quantal size. (B) In the GluR1 knockout (KO), extrasynaptic receptors are eliminated. Now distal and proximal transmission will have equal quantal size because transmission relies on equal numbers of constitutively filled GluR2/3 slots.

Fig. 2.

Model of early and late phases of LTP (see Fig. 1 for definition of structural elements) at a partially silent synapse. In the early phase of LTP, the AMPA-silent module is made functional by a coupled process in which there is both a change in vesicle release mode (from kiss-and-run to full fusion mode) and the addition of a hyperslot containing many AMPA channels to the postsynaptic membrane. This mode change is represented by a change in the size of the fusion pore; the higher cleft concentration of glutamate achieved after full fusion is indicated by making the glutamate red. The coupled presynaptic and postsynaptic changes convert an AMPA-silent module into a functional one and thereby increase the apparent quantal content. By a different process, LTP induces a transient elevation of extrasynaptic GluR1+, leading to a transient increase in the number of filled GluR1+ slots and therefore to a transient increase in quantal size. In the slower phase of LTP, the synapse grows through a transsynaptic process that adds an AMPA-silent module to the synapse.

In what follows, we first discuss different observations that have been difficult to reconcile with each other and review relevant properties of quantal transmission. We then present our model in five sections. We have tried to keep the model as simple as possible, but it is clear that additional complexities exist (these are described in Appendix 3 and will be referred to as they become relevant).

Puzzling Aspects of the Data Set

A successful model for LTP should provide explanations for the following puzzling aspects of the data set:

1) Although it is clear that AMPA channels are added to the synapse during LTP (6), why is it that there is no persistent increase in quantal size (5), the standard measure of postsynaptic responsiveness?

2) Both early and late phases of LTP are associated with a substantial increase in apparent quantal content. It has been argued that such changes may be due to postsynaptic changes, reflecting the addition of AMPA channels at silent synapses. However, it is problematic that similar increases in quantal content are observed in adults, where silent synapses are thought to be rare (7).

3) The AMPA/NMDA ratio of evoked and spontaneous currents does not vary over a wide range from synapse to synapse (leaving aside silent synapses) (8, 9). How can this be reconciled with the large variation of the AMPA/NMDA channel ratio, as determined by immuno–electron microscopy (EM)? Studies using this method show that AMPAR content increases linearly with synapse size (10); NMDAR content is highly sublinear (10, 11).

4) Knockout of GluR1 abolishes distance-dependent scaling, abolishes extrasynaptic receptors (12), and selectively reduces the early phase of theta-burst LTP (1315). Why should such diverse processes be affected?

5) At individual synapses, LTP occurs as a large "all-or-none" increase in AMPA-mediated current (16). How can this be reconciled with structural data indicating that LTP produces growth in the size of the synapse (17) and that the distribution of synapse size is graded over a very large range (more than an order of magnitude) (18)?

Quantal Transmission at CA1 Synapses

The quantal response at hippocampal CA1 synapses does not involve the opening of all the channels at the synapse (1921). Quantal size is about 10 pA (22) (as measured by distance between evenly spaced peaks in a histogram of sucrose-evoked currents), and the amplitude of miniature excitatory postsynaptic currents (mEPSCs) (Fig. 3, A1 and A2) is of the same order (both measured dendritically to minimize the effects of cable filtering). Given that single AMPA channels generate 0.5 to 1 pA (23), the quantal response is generated by ~20 channels. This is far fewer than the ~100 AMPA channels contained within the typical CA1 synapse, which can produce more than 100 pA in response to applied glutamate (19, 24) (Fig. 3B).

Fig. 3.

Properties relevant to distance-dependent scaling. (A1) mEPSCs evoked by local application of sucrose have a larger locally measured current at a distal site than at a proximal site (insets show recording/stimulation sites). (A2) Distributions of mEPSC amplitude. [From (19)] (B) Current evoked by local glutamate uncaging at individual spines is larger at larger spines and is larger at distal spines than proximal ones of the same size. [From (19)] (C1) Current evoked by glutamate application to excised extrasynaptic membrane is larger distally than proximally and is virtually abolished in the GluR1 knockout. (C2) mEPSC amplitude is normally higher distally than proximally, but amplitudes become equal in the GluR1 knockout. [From (12)]

Recent modeling studies of glutamate diffusion and AMPA channel activation in realistic synaptic geometries indicate that the channels that open are not uniformly distributed within the synapse (25). The measured rise time of the quantal response is very fast (~100 μs) (Fig. 4, A1) (22). Simulations that correctly predict this fast rise time indicate that the subset of channels that open at the peak of the response are localized in a hotspot (~100 nm radius) near the location at which the vesicle is released (Fig. 4, A2) (25). This hotspot represents a small fraction of large synapses, which can be up to 1 μm in diameter (18). Only near the site of release is the cleft glutamate concentration in the millimolar range that is required to effectively load the multiple glutamate binding sites on the multimeric AMPARs. The channel then opens efficiently because the opening rate constant is higher than the rate of desensitization (a glutamate-stimulated process in which the channel enters a state that cannot open). Outside the hotspot, opening is inefficient because site occupancy is low, and the opening rate constant is therefore slower than the desensitization rate (25). If desensitization is blocked with cyclothiazide, channels outside the hotspot can open, albeit slowly, and enhance the tail of the miniature mEPSCs (19). The role of desensitization in spatially restricting channel opening will be important in later considerations of the effects of cyclothiazide.

Fig. 4.

Properties relevant to quantal transmission. (A1) Rapid rise time of locally recorded mEPSCs (black) and prediction of computational model (red). (A2) Model shows hotspot of high AMPA channel open probability centered on site of vesicle release at synapse of average size (rectangle). [From (25)] (B) Synapse stained to reveal both presynaptic active zone and postsynaptic density shows alignment of the edges of the two structures. [From (52)] (C1) AMPA channel number, as determined by immuno-EM of serial sections, varies linearly with synapse area of CA1 synapses. (C2) Same for mossy fiber synapses in CA3. [From (10)] (D) Amplitude histograms of evoked responses (minimal stimulation) before LTP induction and shortly afterward. There is reduction in failure probability and increase in quantal size, as indicated by the increase in distance between evenly spaced peaks (note scale change). [From (2)] (E) Quantal analysis at times later than in panel (D) show increase in response (E) and quantal content (m), but no change in quantal size (v). [From (5)]

This understanding of how the quantal response is generated requires a revision of how quantal amplitude is interpreted. Because the quantal response is generated by channels within the hotspot, quantal size is relatively insensitive to the total number of channels at the synapse [except for the smallest synapses, which have the dimension of the hotspot (18)]. Thus, if synapses grow by addition of channels to the edge, quantal size would not substantially increase. For this reason, quantal size cannot be equated with postsynaptic strength, as in the traditional interpretation. Rather, quantal size is dependent on the local density of AMPA channels in the region of the synapse where a vesicle has been released (25, 26). It is necessary to use the word "local" because of the possibility that density may not be uniform within a synapse, as has been found at GABAergic synapses (27).

The lack of importance of total AMPA channel number in determining quantal size at glutamatergic synapses is nicely illustrated at the mossy fiber input to CA3 pyramidal cells. A mossy fiber input occurs on a single giant spine that contains many synapses. These have nearly identical AMPA channel density (Fig. 4, C2) but highly variable dimensions and AMPA receptor content (10). Thus, if quantal size is determined by receptor density, quantal variance should be small; if quantal size is determined by receptor number, quantal variance should be high. The experimental data show low quantal variance (28), consistent with the proposal that receptor density rather than receptor number is what determines quantal size.

The other critical quantal parameter is quantal content, the number of effective quanta released. According to the classical rules of quantal transmission, as developed for the neuromuscular junction, changes in quantal content are due exclusively to changes in presynaptic vesicle release. In contrast, at CA1 synapses, released vesicles may be ineffective because the synapse lacks AMPA channels (a silent synapse) (2931). An increase in apparent quantal content could thus be due to a postsynaptic process that adds AMPA channels to a silent synapse. However, because the activation of AMPA channels is so local, another possibility is that synapses could be "partially silent"; that is, they could have subregions that lack AMPA channels. Quantal content could thus increase if these subregions were "AMPAfied."

NMDA channels bind glutamate much more strongly than do AMPA channels and do not undergo rapid desensitization. Thus, the opening of NMDA channels is relatively insensitive to the position of vesicle release within the synapse and the rate at which glutamate exits from the vesicle (32). Vesicles released within subregions of a synapse that lack AMPA channels will therefore produce an NMDAR-mediated response but no AMPAR-mediated response. Evidence supporting the heterogeneity of transmission within single synapses is discussed below.

Properties of Basal Transmission: The GluR1 Subunit in Distance-Dependent Scaling

Review of the data

EPSPs generated at distal dendritic synapses (up to 250 μm from the soma) undergo attenuation as they spread to the soma (22). A compensation for this attenuation, known as distance-dependent scaling, makes the somatic EPSP amplitude independent of distance. This occurs because the synaptic conductance of distal synapses is higher than that of more proximal ones (Fig. 3, A1, A2, and C2). The higher synaptic AMPA conductance is due to enhanced quantal size rather than to increased quantal content (19). Because quantal size is determined primarily by the density of AMPA channels (see above), one would expect that the density is higher distally than proximally. Consistent with this idea, two-photon uncaging of glutamate produces larger locally measured AMPA currents at distal spines than at proximal spines of the same dimension (Fig. 3B) and the AMPA channel density, as measured by immuno-EM, is higher (19, 33).

AMPARs can potentially be composed of four different types of subunits (GluR1 to GluR4) and are believed to be tetramers comprising subunits arranged in a dimer-of-dimers manner. A clue about the type of subunit involved in distance-dependent scaling was provided by the discovery that the density of functional extrasynaptic AMPARs (analyzed by applying glutamate to patches excised from dendritic membranes) increases with distance from the soma (Fig. 3, C1) (34). The GluR1 subunit of AMPARs appears to be an obligatory component of extrasynaptic receptors, because these extrasynaptic receptors are almost totally absent in the GluR1 knockout mouse (Fig. 3, C1) (12). Surprisingly, the synaptic response, although reduced in size in the knockout, remains substantial. Distance-dependent scaling is not present in these residual responses (Fig. 3, C2), indicating the importance of GluR1 in this form of scaling.

Distance-dependent scaling in the model

Figure 1 shows how the model accounts for distance-dependent scaling. For reasons explained below, we propose that synapses are composed of transsynaptic modules into which a multimolecular AMPAR-binding complex can be inserted postsynaptically. This complex is termed a hyperslot because it contains many slot proteins (35), each of which can bind an AMPAR. We propose that basal transmission is mediated by two types of AMPAR; thus, hyperslots have two types of slots for AMPAR. One of these binds receptors rich in GluR1 (termed here GluR1+); the other binds GluR2/3. We use the designation GluR1+ because the detailed subunit composition is unclear [these receptors may be heteromers of GluR1 and GluR2, but there may also be GluR1 homomers (36)]. We further assume that the GluR2/3 slots are constitutively filled, whereas GluR1+ slots are variably filled in a way that depends on a regulated diffusional equilibrium with extrasynaptic receptors [the higher the extrasynaptic concentration, the greater the filling (Appendix 3, comment 1)]. Because the extrasynaptic GluR1+ concentration is higher distally than proximally, more of the distal slots will be filled and the total density of AMPA receptors will therefore be higher, as previously proposed (34). In the GluR1 knockout, extrasynaptic receptors are absent and the GluR1+ slots will be empty. Because the remaining Glu2/3 slots are constitutively filled, there will still be considerable residual transmission in the GluR1 knockout and transmission will now be equal (as measured locally) at proximal and distal synapses. The model thus accounts for the larger quantal size at distal synapses in the wild-type mouse and the absence of scaling in the GluR1 knockout. The model is also consistent with other lines of evidence regarding the dual nature of basal AMPAR-mediated transmission (Appendix 3, comments 2 and 3).

LTP and Postsynaptic Insertion of AMPA Channels

Review of the data

To account for the changes in quantal transmission produced by LTP, we have incorporated into our model two types of postsynaptic changes and two types of presynaptic changes. Here, we discuss the form of postsynaptic change that leads us to postulate a hyperslot.

Experiments with electrophysiological tagging, in which receptors have an altered electrophysiological signature (Fig. 5, A1 and A2), provide direct evidence for the insertion of GluR1 receptors into the synapse during LTP (6). The first indication that channels might be inserted in large groups came from experiments done on small populations of synapses (16, 37, 38). These experiments showed that LTP develops as a sudden unitary increase in transmission that occurs with a short, random delay (around 1 min) after induction. The increase in synaptic current is about 8 pA (averaging over the 12 examples shown in the three studies cited above; measured in soma). This increase is much larger than can be accounted for by a single AMPA channel (<1 pA; measured in soma). Indeed, the jump in current is similar to quantal size [5.7 pA, averaging over 26 examples shown in (2, 3); measured in soma], an event due to about 20 channels. Recent work using glutamate uncaging has demonstrated similar unitary increases, proving that they are postsynaptic in origin (Fig. 5, B1 and B2). It is noteworthy that these increases were found at synapses that were not initially silent (37). Taken together, these results suggest that AMPA channels can be added to the synapse in groups and that multiple groups can be accommodated within a single synapse.

Fig. 5.

Evidence for postsynaptic processes that contribute to LTP-induced change in synaptic transmission. (A1) LTP in cells infected (inf) with altered GluR1 increases rectification, indicative of the synaptic insertion of GluR1 channels altered to show strong rectification. (A2) Traces showing current at holding potential of either +30 mV or –60 mV demonstrate stronger rectification in infected cells after LTP (left) than before (right). [From (6)] (B1) LTP induction by pairing (at arrow) causes a sudden increase in the response evoked by glutamate uncaging at an individual spine. The sudden increase occurs at different times in different experiments. (B2) The development of the increase is always sudden. [From (37)] (C1) Noise analysis used to estimate number and conductance of AMPA channels in the extrasynaptic membrane before LTP. (C2) Same after LTP. (D1 and D2) Summary data show that LTP produces an increase in channel number but not in conductance. [From (50)]

Addition of groups of AMPA slots in the model

To account for the large and sudden increase in the current generated by applied glutamate, we assume that groups of AMPA channels are held together by a hyperslot (Figs. 2 and 3) that contains multiple slots for both GluR1+ and GluR2/3 channels (Appendix 3, comment 4). Thus, when a hyperslot inserts into an empty "module," it brings many additional channels into the synapse and can therefore substantially enhance the EPSC. Because the magnitude of the sudden current increase is of the same order as the quantal response, the hyperslot (and thus the module) is presumably of the same dimension as the hotspot of channel activation that underlies the quantal response (Fig. 4, A2), in other words, about 100 nm in radius.

Although the sudden jumps in synaptic current must have a postsynaptic origin (37), presynaptic changes may also occur during early LTP. To account for data regarding changes in the mode of vesicle release (see below), we have incorporated a mode change into our model. Thus, as illustrated in Fig. 2, LTP converts a silent module into an AMPA-functional module by a coupled process that involves both the addition of a hyperslot to the postsynaptic membrane and a change in the mode of vesicle release within the same module.

LTP and Presynaptic Mode of Vesicle Release

Review of the data

The most direct evidence (39, 40) for an LTP-induced change in the mode of glutamate release comes from studying the effectiveness of a low-affinity competitive antagonist of NMDA channels [L(+)-2-amino-5-phosphonovaleric acid (L-AP5)]. NMDA channels will be similarly activated by a process that rapidly releases glutamate from a vesicle, yielding high cleft glutamate concentrations, or by a slower release process that produces lower cleft glutamate concentrations for a longer time (32). However, a competitive antagonist will be much more effective in reducing the NMDA response if the glutamate concentration is low. Thus, the finding (Fig. 6, A1 to A3) that L-AP5 greatly reduces the NMDA response before LTP but not after LTP suggests that LTP enhances the concentration of cleft glutamate. The amplitude of the NMDA response is not itself affected after induction of LTP, implying no change in the number of vesicles released (at these early times after LTP induction). These findings therefore point to a process by which the same amount of glutamate is released more rapidly from the vesicle, leading to higher concentration of glutamate but for a shorter time. A further indication of a more rapid glutamate release is the faster time to peak of the NMDA response. All these findings have been adequately accounted for by a computational model (40) in which it was postulated that the specific alteration involves a change from a kiss-and-run mode of vesicle release, in which the fusion pore conductance is small (slow release), to a full fusion mode in which the fusion pore conductance is large (fast release). Recent work provides direct evidence that both modes of release occur at CA1 synapses (4143) and that the ratio of vesicles released by the two modes can be modulated (44, 45).

Fig. 6.

Evidence for presynaptic processes that contribute to LTP-induced change in synaptic transmission. (A1) Before LTP induction, L-AP5 abolishes the NMDA EPSC, indicating low cleft glutamate concentration. (A2) LTP induced by a pairing protocol produces a very large increase in AMPA current. (A3) After LTP induction, the NMDA EPSC is no longer strongly inhibited by L-AP5, suggesting the presence of a higher cleft glutamate concentration. [From (39)] (B) Bottom: NMDA EPSCs were measured at +40 mV; then, voltage was changed to –60 mV and an increase in the probability and amplitude of evoked AMPA-mediated responses was observed in response to cyclothiazide. Returning voltage to +40 mV showed that the NMDA response was not increased, indicating that vesicle release was not enhanced by cyclothiazide. Top: Traces at times indicated below. [From (40)] (C1) Direct measurement of vesicle release using activity-dependent destaining of FM 1-43 shows faster destaining after LTP induction, indicating enhanced probability of vesicle release. FM 1-43 labels synaptic vesicles. (C2) Statistics before and after LTP induction or control. [From (57)]

Whereas NMDA channels can be activated either by long low-concentration pulses or by short high-concentration pulses, AMPA channels can be activated only by the latter (25). Consistent with this finding, experiments at lamprey reticulospinal synapses directly show that changing from full fusion mode to kiss-and-run mode greatly reduces the amplitude of the AMPA-mediated response (46). Thus, similar changes caused by LTP induction could contribute to the enhancement of AMPA-mediated transmission.

Experiments with cyclothiazide, which blocks AMPA channel desensitization, provide additional evidence consistent with a change in the mode of glutamate release from vesicles. Cyclothiazide increases both the frequency of spontaneous events (47) and the probability of evoked responses (40, 48) (Fig. 6B). After LTP, cyclothiazide is much less effective at enhancing the evoked response. In the hippocampus, these effects are unlikely to be presynaptic (49) because the evoked NMDA response was not increased (39). The proposed explanation (39) is that blocking desensitization allows the slow mode of glutamate release to become effective at activating AMPA channels [see (25) for a variant]. After LTP, the effect of cyclothiazide is reduced because now all events are mediated by the fast mode of release, which is capable of activating AMPA channels.

Two observations suggest that the mode of vesicle release within individual synapses is heterogeneous. First, even at the same synapse, some evoked events contain only NMDA responses, whereas others have both NMDA and AMPA components (32). Second, the kinetics of the rising edge of the AMPA response in cyclothiazide has fast and slow components (39), the ratio of which varies from trial to trial.

Changing mode of vesicle release in the model

Our model incorporates the proposal that LTP involves presynaptic changes in the mode of vesicle release (39) and indicates how these presynaptic changes may be coupled to local postsynaptic changes. As shown in Fig. 2, we propose that AMPA-silent modules are silent both because they lack AMPA channels and because they use the slow mode of glutamate release from vesicles. Under basal conditions, synapses contain both functional and silent modules. This heterogeneity explains why the AMPA/NMDA ratio and the kinetics of the rising edge of the NMDA response will depend on whether a vesicle happens to be released within a functional or silent module. Cyclothiazide allows the glutamate released within silent modules to activate AMPA channels in nearby functional modules, thereby increasing the apparent probability of transmission and the rate of spontaneous mEPSPs.

During LTP induction, silent modules are rapidly made functional by changing the mode of the vesicle release to full fusion and by adding a postsynaptic hyperslot and the associated AMPA channels. The result will be an increase in the number of vesicles that are effective in activating AMPA channels, thus increasing the apparent quantal content and reducing the probability of failure, two of the major changes associated with early LTP (14). We next consider the basis for the increase in quantal size that is also associated with the early phase of LTP.

LTP and Changes in Extrasynaptic GluR1 Concentration

Review of the data

A second postsynaptic process has been implicated in the control of AMPA channels. The number of functional channels in a patch of extrasynaptic membrane can be determined by applying glutamate to outside-out patches excised from dendritic membrane. Experiments of this kind (Fig. 5, C1, C2, D1, and D2) indicate that LTP causes a substantial increase in the concentration of extrasynaptic AMPA receptors but not in their single-channel conductance (50). These receptors are presumably GluR1+ (as discussed above).

Extrasynaptic GluR1 affects the filling of synaptic slots in the model

Because extrasynaptic GluR1 appears to be in equilibrium with synaptic receptors (51), the observed increase in extrasynaptic receptor concentration will increase the filling of the GluR1+ slots that are empty under resting conditions (Figs. 1 and 2). This will increase the total density of AMPA receptors and therefore enhance quantal size. We propose that this is the primary explanation for the increase in quantal size that occurs shortly after LTP induction, although other factors might also contribute to this increase (Appendix 3, comment 4).

The available data indicate that late-phase LTP is not associated with an increase in quantal size (5)—in other words, that the LTP-induced increase in quantal size is transient. We infer that the LTP-induced increase in extrasynaptic GluR1 is transient (see below).

LTP Affects Synaptic Growth and the Probability of Vesicle Release

Review of the data

CA1 synapses differ in size by more than an order of magnitude (18). Their structure appears to be governed by a transsynaptic growth process, because the edges of the postsynaptic density and the presynaptic active zone are exactly aligned (Fig. 4B) (52). Spines grow rapidly after LTP induction (53), but it cannot be assumed that synapses necessarily grow with the same kinetics as spines. The available data on synapse growth come from EM reconstruction studies, which indicate that synapses grow in size within less than 1 hour (probably 30 min) after LTP induction (17, 5456).

Over the same period of synaptic growth, there is an increase in the probability of vesicle release, as supported by four lines of evidence: (i) The most direct evidence is based on FM dyes, which can be loaded into vesicles and are then released from the presynaptic terminal when vesicles fuse with the presynaptic membrane. Enhanced release has been observed after LTP induction using a theta burst protocol (which involves bursts of stimulation delivered at 5 Hz), after multiple 200-Hz tetani (Fig. 6, C1 and C2), and after multiple 100-Hz tetani, but not after a single 100-Hz tetanus (5759). This increase in vesicle release develops slowly, being apparent at 30 min and larger at 1 hour (59). No change in the number of releasing boutons was observed, which suggests that presynaptic modulation at this time scale is due to changes at existing synaptic terminals (57). (ii) If a larger number of vesicles is released, the NMDA component of the EPSC should become larger. This has been observed after induction of LTP by a tetanus (6063). The variance of the NMDAR-mediated current decreases after LTP, as expected if there is an increase in the number of vesicles released (64). (iii) There is an increased rate of NMDA channel block by the open channel blocker MK801 (64), as expected if there is an increase in release probability. (iv) The probability of a synaptically induced spine Ca2+ signal is increased after LTP. This is consistent with enhanced release, although the complexity of the Ca2+ release process in spines leaves open other interpretations (65).

The above changes in the NMDA response would not be expected if the only presynaptic change was in the mode of glutamate release, a change to which NMDA channels are insensitive. Overall, the data thus point to two forms of presynaptic change: a rapid change in the glutamate release mode from vesicles, and a slower increase in the number of vesicles released.

Note that all the positive evidence for an increase in vesicle release was obtained under conditions in which LTP was induced using high-frequency stimulation. In contrast, several of these same tests gave negative results under conditions where LTP was induced using a low-frequency pairing protocol. The importance of this methodological difference is dealt with below.

Coupling LTP-induced growth and release probability in the model

We propose that LTP-induced synapse growth is due to the addition of transsynaptic modules (Fig. 2). Because modules are transsynaptic structures, it naturally follows that the active zone and postsynaptic density will grow together and remain aligned at their edges, as observed (Fig. 3, C1 and C2). Because each module contains one or more vesicle release sites, the probability of vesicle release will increase as growth occurs. This will not, however, increase the AMPA response, because the newly added modules lack AMPA receptors and release will be in the "slow" mode. However, the gradual increase in the probability of release will produce a gradual enhancement of the NMDA component, as observed. We have assumed that LTP does not change the number of NMDA channels even during synaptic growth. The basis of this assumption is the finding that the number of NMDA channels varies only weakly with synapse size (7, 11, 66) and that the NMDAR-mediated current evoked by glutamate application is nearly independent of spine size (67, 68).

Summary, Implications, and Predictions

We have considered the large body of evidence regarding the properties of basal synaptic transmission and the changes that occur during LTP (Appendix 1). At the start of this effort, it was unclear whether a coherent model that incorporated all of this evidence could be formulated. Indeed, this body of work is often regarded as presenting incompatible views of how LTP occurs. A central conclusion of our work is that the four processes that have been proposed to enhance transmission are not incompatible and can be incorporated into a compact and sensible model that is consistent with the data set of Appendix 1. We emphasize that although all four processes work together in our model, it is possible that particular LTP induction protocols may evoke only a subset of these processes.

Accounting for the LTP-induced change in quantal parameters

Early LTP (Fig. 4D) is associated with a decrease in probability of failures, an increase in apparent quantal content, and an increase in quantal size (14) (but see Appendix 3, comment 9). The increase in quantal size appears to be transient, as late LTP is not associated (Fig. 4E) with an increase in quantal size (2, 5).

We explain these findings as follows: AMPA silent modules are rapidly (within minutes) made AMPA-functional by two types of changes: (i) The mode of glutamate release from the vesicle changes from a kiss-and-run mode, which is too slow to activate AMPA channels, to a full fusion mode that can activate AMPA channels. (ii) Hyperslots are added to the postsynaptic side of the module, providing multiple slot proteins that can anchor AMPA channels within the module. These two processes account for the increase in apparent quantal content and the reduction in probability of failure. Quantal size increases at early times after LTP induction because GluR1 is rapidly added to the extrasynaptic membrane, resulting in enhanced filling (by diffusional equilibrium) of GluR1+ slots at the synapse. The density of AMPA channels will be elevated and quantal size will therefore increase (Appendix 3, comment 5). Because the increase in quantal size is transient, we surmise that the LTP-induced increase in extrasynaptic receptors is transient. There is not yet any experimental examination of this issue, but a transient change makes sense because it would restore the basal levels appropriate for distance-dependent scaling.

Although the evidence for the lack of any long-term change in quantal size is based on only a single report (5), this finding is consistent with immuno-EM analysis of AMPAR content. This work shows that under basal conditions there is little variability in AMPA channel density (Fig. 3, C1), the main determinant of quantal size, and thus also argues that any LTP-induced change in quantal size must be transient.

Properties of the AMPA/NMDA ratio

Our model predicts that during early LTP, there will be a selective increase in the AMPA component of the EPSP because the number of NMDA channels is constant and the change in vesicle release mode affects only the AMPA response. As silent modules are slowly added to the synapse, the increase in vesicle release will increase the NMDA component (these will release glutamate in the slow mode). Thus, the AMPA/NMDA ratio will rise initially and then relax toward its original level. Two major studies of the time course of the AMPA/NMDA ratio report results consistent with this prediction (69, 70) (Appendix 3, comment 6). A further prediction is that although the ratio of AMPA and NMDA channels varies markedly with synaptic size (10, 11), the AMPA/NMDA ratio of synaptically evoked responses will be nearly invariant with size (except at the smallest synapses, which are completely silent). The experimental results are consistent with this prediction (71).

The coupling of local presynaptic and postsynaptic changes

This idea that the LTP-induced increase in quantal content is due to coupled presynaptic and postsynaptic changes is novel and deserves special comment. Because the proposed coupling is local—it takes place within the individual module—it requires a position-specific retrograde signal. Thus, the addition of a postsynaptic hyperslot may directly influence the mode of vesicle release, possibly through adhesion molecules.

Such local coupling makes energetic sense. The slow mode of glutamate release is thought to occur because the fusion pore is small and associated with incomplete fusion of the vesicle with the plasma membrane (40). This kiss-and-run mode is energy efficient because it allows for recycling of the empty vesicle back into the cytoplasm. In contrast, the fast release that can activate AMPA channels requires a large fusion pore and appears to lead to full fusion of the vesicle with the plasma membrane. This must be followed by energy-consuming processes: endocytosis to form a vesicle and reinsertion of vesicle proteins into this membrane. Because the slow release is adequate to activate NMDA channels (and thereby to activate plasticity), it makes energetic sense to use the slow mode when AMPA channels are not present in a module and to switch to the fast mode only after AMPA channels are added to the module.

Although we propose that the conversion of silent modules to functional ones occurs through both presynaptic and postsynaptic changes, this process could work nearly as well with only postsynaptic changes. We emphasize this point because the evidence for changes in release mode remains inferential and awaits confirmation by more direct methods.

Partially silent synapses

An additional novel aspect of the proposed model is that synapses can be "partially silent." Early in development, many synapses may be completely silent (7, 72). However, as animals mature, AMPA-silent synapses disappear. We propose that these nonsilent synapses may nonetheless have one or more silent modules in which release does not lead to a measurable AMPA response. The existence of both silent and functional modules within the same synapse provides a simple explanation for why individual EPSCs generated at the same synapse can have different AMPA/NMDA ratios (32) and for many surprising effects of cyclothiazide, most notably the increased probability of evoked and spontaneous responses seen in the presence of this drug. In addition, the concept of a partially silent synapse resolves a difficulty with an alternative interpretation of the LTP-induced increase in quantal content. According to an older view, this increase arises from conversion of totally silent synapses to functional ones. However, this theory suffers from the problem that the increase in quantal content has been observed at ages when totally silent synapses have largely disappeared (7). Finally, the idea that the AMPA response is generated primarily with a module means that the quantal amplitude is not an appropriate measure of the postsynaptic strength. Thus, addition of postsynaptic channels to silent modules can be a critical aspect of synaptic strengthening without an accompanying change in quantal size.

Explanation of the diverse effects of GluR1 knockout

The proposal that a subset of slots bind to GluR1+ receptors and are filled by an equilibrium with extrasynaptic receptors that are almost exclusively GluR1+ leads to an explanation of several phenomena. Because extrasynaptic GluR1+ is higher distally than proximally (50), distal GluR1 slots will be filled more than proximal ones, accounting for distance-dependent scaling. In the absence of GluR1, scaling is abolished because the residual transmission is due to GluR2/3 slots, which contribute equally at all distances. One effect of LTP induction is to flood the extrasynaptic membrane with GluR1+ (50, 73), thereby increasing quantal size during early LTP. However, because it appears that this change in quantal size is transient, the increase in the number of extrasynaptic receptors is also likely to be transient, returning (by some unknown process) to a level appropriate for distance-dependent scaling. According to this view, the early phase of LTP should be preferentially reduced in the GluR1 knockout. Evidence consistent with this prediction has been reported (13, 14).

Gradedness of synaptic strength: Growth primes subsequent LTP

EM data indicate that a substantial increase in synaptic area occurs within an hour after LTP induction (17, 56). We propose that this growth occurs by addition of transsynaptic modules. Such transsynaptic coordination ensures that the presynaptic active zone and the postsynaptic density are always in exact register, as the anatomical data show (52). We have attributed the enhanced vesicle release to the growth of the active zone because it is the simplest model we could propose, but there may well be additional modulatory processes that affect release (74, 75).

We have assumed that the new modules added to synapse during growth are AMPA-silent. Thus, glutamate release in these modules is slow and will not affect the AMPA response. The NMDA response, however, will be increased, perhaps explaining the slowly developing component of the LTP induced in the NMDA response reported in some studies (69, 70). An important consequence of the addition of silent modules is the potential for converting silent modules to functional ones during a subsequent period of LTP induction. This would explain the experimental finding that within several hours after LTP saturation, additional LTP can be induced (76). We thus view synaptic growth not as strengthening per se, but as a priming mechanism for future strengthening.

Although we have incorporated "all-or-none" transitions (16) into our model (hyperslot addition), this does not imply that synapses are binary. Indeed, there is direct evidence for at least three states of synapses (77), and we suggest that many potentiated states may arise as additional modules are added (for additional considerations, see Appendix 3, comments 7 and 8). The overall picture is that synapses start small, with perhaps only a single silent module (i.e., a silent synapse). As LTP occurs over multiple bouts, the synapse gradually grows as more modules are added, accounting for the enormous variation in the dimensions of synapses and their AMPA channel content (10).

Status of evidence against presynaptic changes during LTP

Several lines of evidence have argued against any changes in presynaptic release during LTP [summarized in (78)]. If the LTP field is to converge, some understanding of these results must be achieved. One line of investigation that argues against presynaptic changes is the lack of any LTP-induced change in the response of glial glutamate transporter currents to synaptic stimuli. However, these experiments were done at short times after induction (79), a time at which our model would predict no change. This is because the presynaptic changes at early times would be due to changes in the mode of release and would not involve changes in the amount of glutamate release (if vesicles empty totally). It would be useful (Appendix 2) to use glial transporter currents to measure release at the late times when an increase in the number of vesicles released is postulated.

An additional argument against any change in release is that the LTP induced by low-frequency pairing produces no change in the NMDA current (80) and no change in the use-dependent block of NMDA channels, a block dependent on the rate of vesicle release (81). In this pairing protocol, cells are depolarized by voltage clamp of the postsynaptic cell to activate NMDA channels, but presynaptic frequency is left at the test frequency (generally <0.1 Hz). However, contrary results have been consistently obtained when LTP is induced using high-frequency (generally 100 Hz) tetanic stimulation; with such stimulation, many laboratories report an increase in the NMDA component (60, 63, 64, 70, 82, 83). It is thus possible that the increase in vesicle release requires substantial high-frequency stimulation and is simply not evoked by the low-frequency pairing protocol.

Concluding remarks

In an attempt to understand the presynaptic and postsynaptic processes involved in LTP, we have developed a model that is consistent with a large body of evidence (Appendix 1). Our model shows how this data set can be accounted for within a unified framework. Specific predictions of the model are given in Appendix 2.

Appendix 1: Experimental Findings That Should Be Accounted for by a Unified Model

Properties of basal transmission and distance-dependent scaling

1) Quantal size (~15 pA), locally measured, is larger at distal synapses than at proximal ones (22). The average coefficient of variation of proximal and distal synapses is the same.

2) Current evoked by AMPA uncaging is higher distally than proximally at similar-sized spines, indicating that AMPA channel density in the synapse is higher distally than proximally (19).

3) GluR1 knockout reduces quantal size at distal synapses, making it equal to that at proximal synapses (12).

4) GluR1 knockout abolishes dendritic extrasynaptic receptors (12).

5) Knockout of GluR2 and GluR3 produces a 70% reduction in basal transmission, but LTP is not strongly affected (84).

6) GluR1 receptors are added to the extrasynaptic dendritic membrane; GluR2 is added by a mechanism that acts closer to the synapse (85, 86).

7) At least some forms of extrasynaptic AMPAR are in a diffusional equilibrium with synaptic receptors (51).

8) At the same synapse (in culture), some responses are NMDA only, whereas others are AMPA plus NMDA (32).

9) The AMPA/NMDA ratio of mEPSCs does not vary substantially from synapse to synapse (i.e., by a factor of 2 or less) (8, 9).

Cyclothiazide and L-AP5 effects

10) Cylothiazide produces an increase in the frequency of mEPSCs (47).

11) Cyclothiazide can increase the apparent probability of the evoked response (39, 48).

12) Cyclothiazide does not increase the amount of glutamate released by single stimuli, as measured by glial currents (39).

13) Cyclothiazide does not affect the NMDA response to single stimuli, which indicates that its effect on postsynaptic responses is due to postsynaptic action (47, 87).

14) Cyclothiazide affects mEPSC kinetics similarly at proximal and distal synapses (19).

15) In slices from young animals, cyclothiazide produces a slow component of the rising edge of the EPSC, but a fast component that is evident before exposure to cyclothiazide remains apparent (39).

16) The slow component of the rising edge of the EPSC in cyclothiazide (see above) is eliminated by LTP (39).

17) LTP induced by a pairing protocol (early phase) strongly reduces the ability of L-AP5 to block the NMDA response in young slices; this cannot be attributed to a change in vesicle release because the NMDA component is not affected (39).

Properties of LTP of AMPAR-mediated transmission

18) LTP decreases the probability of failures of evoked transmission (2, 88).

19) LTP measured at early times (20 min) is often associated with both an increase in quantal size and an increase in quantal content (13).

20) LTP measured at late times is not associated with an increase in quantal size, but is associated with an increase in quantal content (5).

21) LTP (early phase) is associated with an increase in the response of a synapse to applied glutamate (delivered either though focal uncaging or iontophoretically) (24, 37).

22) LTP produces an increase in the abundance of extrasynaptic AMPARs, with no change in their single-channel conductance (50, 53).

23) LTP is associated with entry of GluR1 into the synapse, as measured by electrophysiological or optical tags (6, 53, 89).

24) LTP is sometimes associated with an increase in the single-channel conductance of AMPA channels at the synapse (23).

25) LTP induced by synaptic stimulation develops through a discrete enhancement of current much larger than the current through a single channel, indicating the involvement of a process that coordinately regulates 10 to 20 AMPA channels (16, 38).

26) The discrete enhancement described above can be seen when LTP is evoked by glutamate uncaging and can be seen at synapses that are not initially silent (37).

27) GluR1 knockout rather selectively reduces potentiation at early times (20 min; referred to as short-term potentiation). In these experiments, LTP was induced by theta burst protocol or a tetanus; this could be observed even in adults (13, 14).

Properties of LTP of NMDAR-mediated transmission

28) NMDA currents evoked by glutamate uncaging are independent of spine size (67).

29) NMDA channel number, as determined by immuno-EM, varies very weakly with synapse size and does not change during development (7, 66).

30) The NMDA component of the evoked response is increased after the induction of LTP by a tetanus; this increase has a slowly developing component (20 min), so that LTP of the NMDA component eventually equals LTP of the AMPA component (70). LTP of the NMDA component is not seen after induction by low-frequency pairing (80).

31) LTP of the NMDA component of mEPSCs (in cortex) develops slowly (20 min) after induction of chemically induced LTP (69).

32) LTP induction by a tetanus produces an increase in the coefficient of variation of the NMDA component (64).

33) The rate of MK801 block of NMDA channels is increased after LTP induction by a tetanus (64).

34) The time to peak of the NMDA response is decreased by LTP induction (39).

Measurements of vesicle release properties

35) The probability of release, as inferred by direct measurement of the rate of vesicle cycling by means of FM dyes, is increased after LTP induction by theta burst stimulation and 200-Hz stimulation (57, 58).

36) The early phase of LTP (10 min) is not associated with an increase in the amount of glutamate released, as measured by the activation of glial transporter currents (79).

37) The probability of release, as inferred from the probability of spine Ca2+ signals, is increased after LTP induction (>10 min) (65).

38) Optical experiments provide direct evidence for multiple modes of vesicle release at hippocampal synapses (4144, 90).

Synaptic structure, AMPA channel composition, and the properties of quantal transmission

39) mEPSP current (10 to 20 pA) is much smaller than the saturating current measured in large spines (100 to 200 pA) (22, 24).

40) Current evoked by glutamate uncaging varies linearly with spine size (which is correlated with synapse size) (19).

41) AMPA channel number (measured with immuno-EM) varies linearly with synapse size. Variation in AMPA channel density is by a factor of no more than 2 (10).

42) Glutamate uncaging produces no detectable AMPA response on thin spines (these have the smallest synapses) (24).

43) In young animals, some synapses are silent; that is, they have an NMDA component but no AMPA component (31).

44) Only the smallest synapses lack AMPA channels (silent synapses) (7, 10).

45) Quantal size is determined primarily by the density of AMPA channels and depends only weakly on total number of channels (at all but the smallest synapses). Multiple quanta add quasi-linearly (25, 26).

46) CA3 mossy synapses have uniform AMPA density, similar to CA1. Evoked transmission of a single mossy fiber onto a single CA3 cell occurs at a single spine that contains scores of synapses of highly variable size. Quantal size has low variance, which indicates that synapse density, not synapse size, determines quantal size (28).

47) The presynaptic active zone and postsynaptic density are exactly aligned, indicating that the synapse grows as a transsynaptic structural unit (52).

48) Analysis by EM indicates that by 1 hour after LTP induction there is a substantial increase in synapse size. Indeed, this change may occur within 30 min (17, 56).

49) Synapse size varies from synapse to synapse by more than an order of magnitude (18, 52).

Appendix 2: Predictions of the Unified Model

1) The modular nature of the synapse and the concept of a partially silent synapse lead to the prediction that there should be clustering of AMPA channels in subregions of the synapse (some regions should lack AMPA channels). LTP should rapidly fill in empty regions. Modules will contain 10 to 20 AMPA channels.

2) Within synapses and extrasynaptically, there should be a rapid, but transient (<1 hour) increase in GluR1 density after LTP induction. However, the number of both GluR1 and GluR2 receptors should increase persistently at potentiated synapses because of the addition of hyperslots.

3) Immuno-EM should show that the higher density of AMPA channels at distal synapses is due specifically to GluR1.

4) Quantal size, when measured over time, will show a rapid increase after LTP induction but will then decay in 20 to 30 min even though LTP is maintained. Experiments of this kind are needed to confirm separate measurements made by different groups at early or late times.

5) Direct methods of monitoring vesicle release show an increase after LTP is induced by high-frequency stimulation, but such methods should not show the increase when LTP is induced by a low-frequency pairing protocol.

6) Measurements of the amount of glutamate released (for instance, as reported by activation of the glial glutamate transporters) should show a slowly developing increase.

7) If procedures are found that block synapse growth, these would be expected to prevent the increase in probability of vesicle release, which, taking the simplest hypothesis, we view as a consequence of growth. It is, of course, possible that additional processes exist that can enhance release, independent of growth.

8) Measurements of the mode of glutamate release from vesicles have so far relied on low-affinity NMDA antagonists and were carried out in young animals. More direct methods are becoming available to measure these changes. According to our model, such changes will be a general feature of LTP at all ages. The changes should develop rapidly after LTP induction. As the synapse then grows because of the addition of silent modules, a slow release mode should again become evident.

9) If it becomes possible to identify parts of the active zone where different modes of release occur, our model predicts that the synapse will be heterogeneous in mode and that subregions with the fast release mode will be aligned with AMPAfied subregions of the postsynaptic membrane. Subregions with the slow mode will be aligned with regions lacking AMPA channels.

Appendix 3: Additional Considerations

1) The term "regulated" is used to describe the diffusional equilibrium between synaptic and extrasynaptic receptors because there appear to be processes that (73, 91) that can alter this equilibrium. Although this complexity is not incorporated into our model, these regulatory processes probably do not affect our basic assumption—that filling of GluR1+ slots is proportional to the extrasynaptic GluR1+ concentration.

2) The existence of two fundamentally different components of basal transmission has been previously established on the basis of findings that only a component of basal transmission is affected by actin inhibitors, microtubule inhibitors, and NSF inhibitory peptide (9294). Furthermore, the dual nature of basal transmission explains why knockout of GluR2 (alone or with GluR3) leaves a substantial component of transmission (84) (see comment 3 for discussion of further points regarding the subunit-specific roles). The dual nature of basal transmission probably relates to subunit-specific mechanisms for addition of AMPA channels to synapses. CA1 synapses contain at least two types of AMPA receptors, each held by a different slot protein (95, 96). Overexpression studies have pointed to a constitutive filling of slots by the delivery of GluR2 (96) through a process that occurs close to the synapse [(86), but see (97)]. In contrast, the insertion of GluR1 into the membrane occurs at sites distant from synapses (86).

3) According to one model (6), GluR1 is rapidly added to synapses during LTP but is only transiently present; basal transmission is postulated to depend solely on GluR2/3. However, the rectification assay used by these authors is not sensitive and requires that GluR1 be highly abundant; if GluR1 is dominant initially but then falls to a steady-state level comparable to other receptor types, as we have postulated, the basal level of GluR1 might well not be detected by the rectification assay. It might be argued that the findings in the GluR1 knockout (12) could be explained by a simpler model in which there is only one type of receptor slot that anchors all types of receptors. According to this view, although the total number of channels is reduced, the ones that remain become concentrated exclusively at the synapse because binding there is tight. We were not able to account for the experimental results with this type of model; tight binding would mean that in the normal case, all slots would be filled and distance-dependent scaling could not be explained.

4) The nature of the multimolecular assembly that forms the hyperslot is unclear, but it is noteworthy that one putative slot protein, PSD95, exists in a multimolecular assembly before insertion into the synapse (98).

5) Although a transient increase in the filling of GluR1 slots may be the primary reason for the observed increase in quantal size, other processes may also contribute. Although a large fraction of quantal current is generated near the site of vesicle release, a small fraction of the quantal response (20 to 30%) is generated at more distant regions of the synapse (25). This minor component would be expected to increase as remote regions become AMPAfied. In addition, an increase in single-channel conductance is sometimes observed (23) and could contribute to the enhancement of quantal size.

6) Two major studies observed a slowly developing rise in the NMDA-mediated component of the response, one using evoked responses (70), the other miniature responses (69). In contrast, other studies have observed a rapid enhancement and no slowly developing rise; we surmise that the rapid process may be an additional modulatory enhancement of the NMDA channel itself (89), the presence of which obscures the slow development. We argue that the slow component is caused by increased vesicle release. The enhancement of mEPSC size (69) could be secondary to increased release because of the enhanced probability of multiquantal mEPSCs (25).

7) Our model follows several previous proposals that postulate sequential phases of LTP (99101) [reviewed in (102)]. Recent work has shown that protein synthesis can strongly reduce the magnitude of LTP at 20 min after induction (103), much earlier than previously thought. It is thus possible that the growth-related process in our model corresponds to the protein synthesis–dependent component described in other models.

8) Additional ideas regarding synaptic weakening must be incorporated into the model to understand the reduction of late LTP by protein synthesis inhibitors. It appears that synapses have very different sensitivity to depotentiation in different states. The synapses in young animals appear particularly sensitive to weakening; functional synapses are "labile" and may become completely silent as a result of just a few low-frequency (<0.1 Hz) test pulses (104). Experiments in older animals show little effect of test pulses under basal conditions. However, after LTP induction, partial depotentiation can be caused by low-frequency test pulses, but only during the first 15 min after LTP induction (105, 106). During this same period, complete depotentiation occurs if the stimulus frequency is raised to 5 Hz (107, 108). These weakening processes require on the order of 100 stimuli. With time (15 to 30 min), a consolidation process occurs that makes the synapse relatively insensitive to such forms of activity-dependent weakening (108). In our model, the reduced late phase of LTP produced by protein synthesis inhibitors cannot be attributed to the blockage of growth because growth does not enhance AMPA-mediated transmission. A likely explanation is that the slowly developing insensitivity to activity-dependent weakening requires protein synthesis (109); in the presence of protein synthesis inhibitors, synapses become weaker over time as the result of an action of the test pulses.

9) Whereas most studies have observed an increase in quantal size during early LTP, two studies (110, 111) reported no change in quantal size. It appeared that LTP could be explained entirely by an increase in the probability of release at a single release site. We have considered whether such data can be accounted for within the framework of our model. It appears that the synapses studied were a subset selected so that the amplitude distribution of nonfailures was described by a single Gaussian function. Given that we now know that most synapses display multiquantal release and that such release does not yield distributions described by a single Gaussian, it seems likely that the selected synapses were ones in which there was single release site. These synapses may therefore have been at the small end of the size spectrum and therefore had special properties.


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
  6. 6.
  7. 7.
  8. 8.
  9. 9.
  10. 10.
  11. 11.
  12. 12.
  13. 13.
  14. 14.
  15. 15.
  16. 16.
  17. 17.
  18. 18.
  19. 19.
  20. 20.
  21. 21.
  22. 22.
  23. 23.
  24. 24.
  25. 25.
  26. 26.
  27. 27.
  28. 28.
  29. 29.
  30. 30.
  31. 31.
  32. 32.
  33. 33.
  34. 34.
  35. 35.
  36. 36.
  37. 37.
  38. 38.
  39. 39.
  40. 40.
  41. 41.
  42. 42.
  43. 43.
  44. 44.
  45. 45.
  46. 46.
  47. 47.
  48. 48.
  49. 49.
  50. 50.
  51. 51.
  52. 52.
  53. 53.
  54. 54.
  55. 55.
  56. 56.
  57. 57.
  58. 58.
  59. 59.
  60. 60.
  61. 61.
  62. 62.
  63. 63.
  64. 64.
  65. 65.
  66. 66.
  67. 67.
  68. 68.
  69. 69.
  70. 70.
  71. 71.
  72. 72.
  73. 73.
  74. 74.
  75. 75.
  76. 76.
  77. 77.
  78. 78.
  79. 79.
  80. 80.
  81. 81.
  82. 82.
  83. 83.
  84. 84.
  85. 85.
  86. 86.
  87. 87.
  88. 88.
  89. 89.
  90. 90.
  91. 91.
  92. 92.
  93. 93.
  94. 94.
  95. 95.
  96. 96.
  97. 97.
  98. 98.
  99. 99.
  100. 100.
  101. 101.
  102. 102.
  103. 103.
  104. 104.
  105. 105.
  106. 106.
  107. 107.
  108. 108.
  109. 109.
  110. 110.
View Abstract

Stay Connected to Science Signaling

Navigate This Article