PKA drives an increase in AMPA receptor unitary conductance during LTP in the hippocampus

Long-term potentiation (LTP) at hippocampal CA1 synapses can be expressed by an increase either in the number (N) of AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid) receptors or in their single channel conductance (γ). Here, we have established how these distinct synaptic processes contribute to the expression of LTP in hippocampal slices obtained from young adult rodents. LTP induced by compressed theta burst stimulation (TBS), with a 10 s inter-episode interval, involves purely an increase in N (LTPN). In contrast, either a spaced TBS, with a 10 min inter-episode interval, or a single TBS, delivered when PKA is activated, results in LTP that is associated with a transient increase in γ (LTPγ), caused by the insertion of calcium-permeable (CP)-AMPA receptors. Activation of CaMKII is necessary and sufficient for LTPN whilst PKA is additionally required for LTPγ. Thus, two mechanistically distinct forms of LTP co-exist at these synapses.


INTRODUCTION
Long-term potentiation (LTP) of synaptic function is considered the major process underlying learning and memory 1 where it is involved in synaptic engram formation 2,3 , yet the underlying cellular mechanisms remain incompletely understood. The best-characterized form of LTP occurs at the Schaffer collateral-commissural pathway (SCCP) in the hippocampus, where it is triggered by synaptic activation of NMDA (N-methyl-D-aspartate) receptors 4 and is expressed as a persistent increase in AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid) receptor-mediated synaptic transmission 5 . This modification is primarily due to a functional modulation of AMPA receptors (AMPARs), which may involve a change in the number of active channels (N) (termed LTPN) and/or their single-channel conductance (γ) properties (termed LTPγ) (e.g., [6][7][8][9] ). Whilst there is considerable evidence that LTPN is triggered by activation of Ca 2+ /calmodulin-dependent kinase II (CaMKII) 10,11 and involves exocytosis and lateral diffusion of AMPARs 12,13 , the mechanisms underlying LTPγ are largely unknown. The two most likely molecular mechanisms involve (i) CaMKII-mediated phosphorylation of Ser831 of GluA1, which can result in an increase in the time AMPARs dwell in higher conductance states [14][15][16] or (ii) the insertion of calcium-permeable AMPA receptors (CP-AMPARs), which have a higher γ than their calcium-impermeable (CI) counterparts 17,18 .
In the present study, we have tested the hypothesis that LTPγ is due to the insertion of CP-AMPARs in young adult rodents using two theta burst stimulation (TBS) induction protocols that differed only in the timing between episodes, and applied peak-scaled non-stationary fluctuation analysis (NSFA) [19][20][21] to estimate γ before and after the induction of LTP 6,15,[22][23][24][25] . We found that the compressed TBS protocol (cTBS -inter-episode interval of 10 s) resulted exclusively in LTPN, for which CaMKII was both necessary and sufficient. In contrast, a spaced TBS protocol (sTBS -inter-episode interval of 10 min) resulted in a transient increase in γ, lasting ~15 min, which was due to the insertion of CP-AMPARs and required both CaMKII and PKA. Insertion of CP-AMPARs mediates both the initial expression of LTPγ, by enhancing the net synaptic unitary conductance, and helps trigger the processes that lead to a persistent increase in synaptic efficacy that outlasts the increase in γ. Since the PKA-dependent form of LTP also requires de novo protein synthesis and has stimulation features similar to spaced behavioural learning, LTPγ is likely to underlie the formation of synaptic engrams and long-term memory.

An increase in γ is specifically triggered by a sTBS protocol
Simultaneous field excitatory postsynaptic potential (fEPSP) recordings from stratum radiatum and somatic whole-cell recordings were obtained in response to baseline stimulation of two independent SCCP inputs (Fig. 1a). TBS was delivered to one input (test), while the second input served as a control for stability and heterosynaptic effects (Fig. 1c, d). Synaptic potentiation was quantified and γ was estimated using NSFA (Fig. 1e, f), as described previously 6 . To optimise the estimates of γ we used minimal stimulation and restricted our measurements to the first 20-30 min following TBS, since γ estimates are extremely sensitive to small fluctuations in series resistance 20 . Thus, our study focused on the induction and initial expression mechanisms of LTP.
In the first series of experiments we delivered three episodes of TBS, with each episode comprising 5 shocks at 100 Hz delivered 5 times at 5 Hz (i.e., 75 stimuli in total; see Fig. 1b schematic); in interleaved experiments we either delivered these three episodes as a cTBS (10 s inter-episode interval) or as a sTBS (10 min inter-episode interval). We referred to the resultant potentiation as cLTP ( Fig. 2a-i) and sLTP ( Fig. 2j-r), respectively. In response to cTBS there was a substantial cLTP (Fig. 2a), with EPSC amplitudes increasing to 212 ± 11% of baseline, averaged over the first 10 min after induction (Fig. 2b). For 22 neurons from 15 rats (n = 22/15), we obtained γ estimates in 10 min epochs and found it to be unaltered throughout ( Fig. 2c-g).
In response to sTBS the results were strikingly different. For this set of experiments, whole-cell recordings were obtained shortly after delivery of the second TBS episode and the effects of the third TBS was evaluated (Fig. 2j). This method was necessary because of the rapid wash-out of LTP with low access whole-cell recordings. In response to the third TBS there was a substantial additional LTP, with EPSC amplitudes increasing to 177 ± 9% of baseline, averaged over the first 10 min after induction (Fig. 2k). The estimate of γ upon break in was significantly higher (6.9 ± 0.4 pS) compared to the control input (4.9 ± 0.4 pS; Fig. 2n-o; t22 = 3.22, p = 0.0039, paired Student's t-test) and this was further increased in response to the third episode of TBS to 8.4 ± 0.4 pS (LTP10'; t22 = 3.75, p = 0.0011, Fig. 2l, m, o, p; n = 23/17). However, when we quantified γ at 10-20 minutes after the last TBS, the value (5.5 ± 0.3 pS) was no longer significantly different from the control input (LTP20'; t22 = 2.01, p = 0.0570, paired Student's ttest; Fig. 2m). In contrast to the test input, sTBS did not result in a significant γ change in the control input (4.9 ± 0.4 pS, 5.4 ± 0.4 pS and 4.6 ± 0.3 pS at the corresponding time points; Fig.   2m-n). Thus, the increase in γ is specifically related to sLTP. Furthermore, this increase in γ positively correlated with the magnitude of sLTP (Fig. 2q).
Since sLTP, but not cLTP, is associated with the insertion of CP-AMPARs 26,27 these results suggest that CP-AMPARs may account for the increase in γ. CP-AMPARs have slightly faster decay kinetics (τdecay) than CI-AMPARs 25,28 , which can be detected using single exponential fits to EPSC decays. We found that cLTP was not associated with an alteration in τdecay (Fig. 2i, Table 1; t21 = 0.66, p = 0.5146, paired Student's t-test), whereas sLTP was associated with a highly significant decrease in τdecay (Table 1; p = 0.0051, t22 = 3.11, paired Student's t-test). A regression analysis showed a trend for the τdecay to be inversely related with the increase in γ ( Fig.   2r; p = 0.0712, F(1, 21) = 3.61). Therefore, the kinetic analysis provides additional support for the notion that insertion of CP-AMPARs occurs during the induction of LTP in response to a sTBS.

The role of PKA in LTP γ
It is established that elevating cAMP by, for example, use of the phosphodiesterase 4 inhibitor rolipram, enables a weak stimulus to generate an enhanced PKA-dependent form of LTP 29 .
Previously, we found that in the presence of rolipram a weak TBS, comprising one episode of TBS, generated an LTP that is largely dependent on the insertion of CP-AMPARs 26 . Here we used this same method as an independent means to investigate whether insertion of CP-AMPARs are responsible for the increase in γ. Since only one TBS is required to induce the PKAdependent form of LTP in the presence of rolipram we could make γ measurements before and after the full induction of LTP. As illustrated in Fig. 3a-b, a single episode of TBS (wTBS; comprising 25 stimuli), when delivered in the presence of rolipram (1 µM), generated a robust LTP (234 ± 14 % of baseline for test vs. 121 ± 6 % for control input). We found that this LTP was also associated with a transient increase in γ (baseline = 4.9 ± 0.4 pS, LTP10' = 8.0 ± 0.6 pS; t20 = 5.90, p < 0.0001) that returned to baseline by the second 10 min epoch (LTP20' = 5.4 ± 0.3 pS; t20 = 1.39, p = 0.1810) following the wTBS (n = 21/15; Fig. 3c, d, f, g). This potentiation required the wTBS since the control input was largely unaffected (5.1 ± 0.  Table 1). As was the case with the sLTP, the size of the change in γ correlated with the magnitude of LTP (p = 0.0024, F(1, 19) = 12.27; Fig. 3h). Additionally, there was an associated reduction in τdecay (p = 0.0007, t20 = 3.99, paired Student's t-test; Table 1) that also negatively correlated with the increased γ (p = 0.0199, F(1,19) = 6.46; Fig. 3i). These results further support the idea that insertion of CP-AMPARs mediates LTPγ.
To more specifically test the requirement of PKA for driving alterations in γ, we included the catalytic subunit of PKA (PKA Cα; 300 U/mL) in the patch solution (Fig. 4). This treatment had little effect on the control input (Fig. 4a). However, as was the case with rolipram, the wTBS in the presence of PKA Cα generated a robust potentiation (Fig. 4a) that was associated with an increase in γ (Fig. 4b, c). The levels quantified during baseline and 10 min post TBS (LTP10') were 5.2 ± 0.5 pS and 7.8 ± 0.8 pS (t16 = 5.80, p < 0.0001, paired Student's t-test; n = 17/13; Fig.   4b). Once again, the increase in γ was only transient, since estimates of γ made between 10 and 20 min following the wTBS (i.e. LTP20') were not significantly different from baseline (5.3 ± 0.5 pS; t16 = 0.37, p = 0.7163, paired Student's t-test; Fig. 4b).
In interleaved experiments ( Fig. 4d-f), we examined the effects of the CP-AMPAR blocker, IEM-1460 (30 µM, IEM). In the presence of bath applied IEM and PKA Cα in the patch pipette, the level of LTP triggered by the wTBS was significantly less than in its absence (202 ± 16% vs.  Table 1). There was a strong correlation between the increase in γ with both the magnitude of LTP ( Fig. 4i; p = 0.0021, F(1,15) = 13.72) and the decrease in τdecay ( Fig. 4k; p = 0.0117, F(1,15) = 8.24) when wTBS was delivered in the presence of PKA Cα, but there was no such correlations when IEM was also present (Fig. 4j, l).
In conclusion, we find that activation of PKA, that occurs during (i) a sTBS, (ii) a wTBS in the presence of rolipram, or (iii) a wTBS in the presence of the catalytic subunit of PKA, results in the transient insertion of CP-AMPARs and that these receptors are responsible for the increase in γ during the initial expression phase of LTP.
It has been suggested that the role of CaMKII in LTP involves an increase in γ 14,15 . To further examine the role of CaMKII in LTP we interleaved experiments where we applied either active or inactive (heat inactivated) CaMKII (250 U/mL) via the patch pipette and delivered baseline (low frequency) stimulation to monitor basal synaptic transmission. Consistent with previous reports 33, 34 , activated CaMKII, but not inactive CaMKII, was sufficient to potentiate synaptic transmission ( Fig. 5c, d, h). However, this potentiation was not associated with an increase in γ ( Fig. 5f, g, i) or a change in rise and decay kinetics ( Fig. 5e; see also Table 1). The respective γ values for baseline (i.e., first 5 min of recording) and 10-15 min of whole-cell recording were 4.7 ± 0.6 pS and 4.3 ± 0.5 pS (t14 = 0.74, p = 0.4740, paired Student's t-test; n = 15/12; Fig. 5g).

Activation of CaMKII and PKA are both necessary and sufficient for LTPγ
Since neither PKA alone nor CaMKII alone affected γ, we explored whether the combination of the two kinases may be sufficient for the effect. We, therefore, patch loaded PKA Cα (300 U/mL) with either the active or inactive forms of CaMKII (250 U/mL). In interleaved experiments, we found that PKA Cα + active CaMKII produced a robust potentiation of synaptic responses, specifically 178 ± 10% of baseline when quantified 15 min after whole-cell ( Fig. 6a, b, f). In this case, the effect was also associated with an increase in γ ( CaMKII plus PKA Cα but no such correlation was found in the presence of IEM ( Fig. 6j-k).

The proportion of synaptically-incorporated CP-AMPARs during LTP γ
Together, the previous experiments provide multiple lines of evidence that LTPγ is due to the insertion of CP-AMPARs into synapses that contain CI-AMPARs. In order to determine the relative proportions of each it was necessary to measure γ for synapses containing either just CI-AMPARs or just CP-AMPARs, under our recording conditions. To achieve this, we used lentivirus-driven CRISPR/Cas9 expression to delete GluA2 in a fraction of neurons in vivo, allowing a direct comparison between a knock-out (KO) and a wild-type (WT) neuron within each adult brain slice (Fig. 7a). When compared with uninfected neighbouring neurons, the KO cells showed a reduced AMPAR synaptic transmission (Fig. 7b) and an inwardly rectifying current-voltage relationship (Fig. 7c, d). The level of γ in KO neurons was significantly higher at 17.3 ± 1.2 pS (n = 16) compared to 4.6 ± 0.4 pS for WT neurons (n = 17; t31 = 10.09, p < 0.0001, unpaired Student's t-test, data from 12 animals; Fig. 7e-h). This increase in γ in these neurons was associated with a decreased τdecay from 6.8 ± 0.4 ms in WT to 5.3 ± 0.4 ms in KO Fig. 7i). Assuming that these EPSCs were comprised of 100 and 0 % CP-AMPARs, respectively, then the increase in γ that we observed during sLTP can be explained by CP-AMPARs comprising ~30% of the synaptic current during the first 10 min following LTP induction.

DISCUSSION
NMDA receptor-dependent LTP has been extensively studied as the primary mechanisms utilized are crucial for the formation of long-term memories. Despite many molecules being discovered and different aspects of their regulation being uncovered, there are crucial gaps in our knowledge. One relates to the fact that long-term memory requires de novo protein synthesis yet most of our mechanistic understanding of LTP has been obtained from the study of a protein synthesis-independent form of LTP. A second pertains to the fact that much of this understanding has been derived from the study of juvenile animals, where technical issues have permitted more in-depth analysis, whereas most studies of learning and memory are conducted in adult animals. In the present study, we have addressed these issues by studying LTP at CA1 synapses in young adult rodents and have compared induction protocols that are known to activate the protein synthesis-independent (cTBS) and protein synthesis-dependent (sTBS, rolipram + wTBS) forms 35,36 . Using a cTBS protocol, LTP involved the insertion of additional CI-AMPARs, for which activation of CaMKII is both necessary and sufficient. Using a sTBS there was an additional LTP component that involved the transient insertion of CP-AMPARs, for which activation of CaMKII and PKA are both necessary and, in combination, sufficient. The insertion of CP-AMPARs increases AMPA receptor γ and this underlies the initial expression of this form of LTP, which we have termed LTPγ. The insertion of CP-AMPARs is transient and is replaced by a persistent increase in the number of CI-AMPARs.

Two distinct postsynaptic forms of LTP at CA1 synapses.
The division of NMDA receptor-dependent LTP into multiple components was made on the basis of sensitivity to various pharmacological agents and substantiated by genetic studies 36 . In particular, when a single train (tetanus or TBS) is employed, the resultant LTP is independent of both PKA activation and de novo protein synthesis; this is commonly referred to as LTP1 or E-LTP 35,37 . In contrast, when multiple trains are delivered, with an interval in the order of minutes, then there is the generation of an additional PKA and de novo protein synthesisdependent component of LTP, which is commonly referred to as LTP2 or L-LTP 27,36,38,39 .
LTP2 is generally assumed to underlie long-term memory formation, that also requires de novo protein synthesis.
NMDA receptor-dependent LTP has also been divided into two distinct postsynaptic mechanisms of expression, one involving an increase in the number of AMPARs without a change in γ (LTPN) and the other involving an increase in γ (LTPγ) 6 . Here, one of our goals was to determine whether these separate expression mechanisms specifically relate to LTP1 and LTP2. We found that LTP1 never involved an alteration in γ whereas LTP2 invariably did. The increase in γ was transient, lasting between 10 and 20 min and could be fully explained by the insertion of CP-AMPARs. In terms of signaling cascades, we found that activation of CaMKII was both necessary and sufficient for LTP1 whereas both CaMKII and PKA were required, and in combination were sufficient, for LTP2 (see model in Fig. 8). Our findings do not conflict with a large body of literature regarding alterations in AMPARs underlying LTP at these synapses and the roles of both CaMKII and PKA (e.g., 26,30,[40][41][42]. Hitherto, the roles of CP-AMPARs and alterations in γ in LTP have been controversial 6,24,[43][44][45] . However, these controversies can now be reconciled on the basis of the type of LTP under investigation. We can conclude therefore that LTP1 equates to LTPN and LTP2 with LTPγ. It is important to note, however, that although a compressed induction protocol (cTBS) will ordinarily result in just LTP1 / LTPN a spaced protocol will comprise a mixture of LTP1 and LTP2 (LTPγ), since the initial train will induce LTP1 upon which subsequent trains will add LTP2 under our experimental conditions. The relative proportion of these two components will depend on a variety of conditions, including the interval between the trains (with ~10 min being optimal for the induction of LTP2) and external factors such as stress, that likely impacts LTP1 and LTP2 differently 31 .

On the mechanism of LTP γ
Previous work has shown that an increase in AMPAR γ could result from either a CaMKIIdependent phosphorylation of Ser831 of GluA1 14,16 or by the insertion of CP-AMPARs 25 , since these have a higher single channel conductance than CI-AMPARs 17,18 . Our findings have demonstrated that LTPγ can be explained exclusively by the latter mechanism, since all changes in γ were eliminated by IEM. Furthermore, we found that activation of PKA plus CaMKII increased γ whereas CaMKII alone did not, despite leading to a substantial potentiation. The failure of CaMKII alone to increase γ, which is contrary to some previous studies 14,15 , could be explained on the basis of the native AMPAR configuration since γ alterations are affected by the subunit combination and accessary protein composition of AMPA receptors 16,18 . It is worth noting, however, that whilst activation of CaMKII alone was not sufficient to induce LTPγ its activation was necessary. It is possible, therefore, that phosphorylation of Ser831 of GluA1 is necessary, but not sufficient for LTPγ.
Space precludes a detailed discussion of the underlying molecular mechanisms of LTPγ, but it is likely to be governed by AMPAR subunit-specific regulation and trafficking 46 . In brief, our data are compatible with an exchange of a subset of CI-AMPARs for CP-AMPARs. The latter could be explained by a mechanism involving the Ca 2+ sensor PICK1 47,48 , which has been shown to bind and internalize GluA2-containing AMPARs to enable the insertion of CP-AMPARs during LTP 49 . The next step involves the replacement of the newly inserted CP-AMPARs with CI-AMPARs, a process that requires baseline (low frequency) synaptic activation 26,43 and probably involves Ca 2+ permeation through the CP-AMPARs themselves 50 . The rapid replacement of CP-AMPARs with CI-AMPARs was originally described at excitatory synapses onto cerebellar stellate neurons from P18-P20 rats 51 . At this synapse, high frequency stimulation (tetanus) induces CP-AMPARs to be replaced with the equivalent number of CI forms resulting in a reduction in the synaptic current by a third, reflecting lower γ of the latter form. We observed an initial reduction in EPSC amplitude following the triggering of LTP, which might be explained, in part, by a one-to-one exchange of CP-AMPARs for CI-AMPARs.
Additionally, the transient expression of CP-AMPARs could trigger an increase in the number of AMPAR slots at synapses that enables an increase in the number of CI-AMPARs above and beyond what can occur during LTP1.
Since CP-AMPARs increase synaptic conductance why does there need to be an exchange for a greater number of CI-AMPARs to maintain the enhanced synaptic response? One possibility is that the expression of CP-AMPARs at these synapses needs to be restricted in time due to potential excitotoxicity 52 . Therefore, they can only provide a transient mechanism of expression whilst triggering the more persistent switch resulting in a larger number of CI-AMPARs.

Developmental regulation of the expression mechanisms of LTP
There is strong evidence that the expression mechanisms of LTP are developmentally regulated.
The co-existence of two mechanisms involving the insertion of CI-AMPARs and CP-AMPARs can account for the LTP at P14 6 and in young adults, as observed herein. However, at around P7, LTP is associated with a decrease in γ 22 , which is most likely explained by the replacement of CP-AMPARs with a larger number of CI-AMPARs. Early in development there is also the insertion of AMPARs into synapses that appear to lack AMPARs altogether; so-called "silent" synapses 53,54 . A potential scenario is as follows: first synapses acquire CP-AMPARs, next these are replaced by more CI-AMPARs. Thereafter LTP can increase the number of these CI-AMPARs via two mechanisms, one of which involves the transient insertion of CP-AMPARs and one that does not.
There have been far fewer studies regarding the mechanisms of synaptic plasticity in tissues obtained from adult animals compared to juvenile animals, despite most learning and memory studies are conducted in adult animals. This is a concern when attempting to relate mechanisms of synaptic plasticity to learning and memory. Our present study, conducted exclusively in tissue from young adult animals, shows that two distinct forms of synaptic plasticity can be readily induced simply by altering the patterns of activation. Our result that a cTBS protocol induces LTP that does not involve an alteration in γ is consistent with another study in adult animals 24 .
Our finding that a sTBS induces an additional component of LTP that involves an increase in γ is the first evidence that such a process occurs beyond early developmental stages.

Functional significance of two forms of LTP
This raises the question as to why there are two distinct mechanisms to increase the synaptic complement of CI-AMPARs. Previous work has shown that the insertion of CP-AMPARs is specifically associated with the PKA and protein synthesis component of LTP 27 . It is reasonable to assume, therefore, that the transient insertion of CP-AMPARs is part of the machinery that triggers de novo protein synthesis and the consequential morphological changes (spine enlargement and/or new spine formation). In contrast, in the absence of de novo protein synthesis, the increase in synaptic CI-AMPAR number can support the increase in synaptic efficacy. Although both processes can increase synaptic strength lasting many hours in vitro, it seems probably that only the protein synthesis-dependent form triggers synaptic changes that underpin long-lasting memories (lasting from days to lifetimes). Indeed, it has been shown that spaced training with access to reward enhances the persistence of memory, and treatment with rolipram after training enhances memory retention 55 . The requirement for PKA to trigger the protein synthesis-dependent form of LTP also provides the opportunity for extensive neuromodulation. Neurotransmitters, such as noradrenaline and dopamine, and stress hormones, such as corticosterone, may, via the insertion of CP-AMPARs, augment protein synthesisdependent LTP to enhance and/or prolong the persistence of the associated memory (e.g., 30,31,42,56,57 ).

Concluding remarks
We have identified the molecular basis of two independent forms of LTP that co-exist at hippocampal synapses in young adult animals, the occurrence of which is controlled by the patterns of synaptic activation during induction. The existence of these two distinct LTP mechanisms goes a long way in explaining many of the controversies that have plagued the field.
Slices were allowed to recover at 32-34°C for 30 min, and then maintained at 26-28 °C for a minimum of 1 h before recordings were made.

Field excitatory postsynaptic potential (fEPSP) recordings
The extracellular electrophysiology was performed in both interface and submerged type chambers maintained at 32°C, and continuously perfused at 2-4 mL/min with oxygenated ACSF.
The slope of evoked fEPSPs (V/s) was measured in the CA1 region of hippocampal slices and bipolar stimulating electrodes were used at a constant voltage intensity (0.1 ms pulse width) throughout the experiments. Signals were amplified using Axopatch 1D (Molecular Devices) and digitized with BNC-2110 (National Instruments) A/D board at a sampling rate of 20 kHz.
Recordings were monitored and analyzed using WinLTP 58  To ensure recording stability, extracellular field EPSPs were simultaneously monitored as described previously 26 . Peak amplitude (pA) and initial slope (V/s) of EPSCs and fEPSPs were measured, and displayed on-line, using WinLTP 58 . Whole-cell recordings were initiated following collection of at least 10 min of stable baseline assessed by extracellular recordings.

Peak-scaled, non-stationary fluctuation analysis (NSFA)
The unitary conductance (γ) of AMPA receptors was estimated using NSFA according to 6 (see also [19][20][21] ). Whole-cell responses were carefully selected for analysis using WinWCP (University of Strathclyde, Glasgow) and Mini Analysis software on the basis of the following criteria: first, precise alignment of traces on the rise phase; second, no contamination by spontaneous or polysynaptic currents; third, complete decay from the peak EPSCs. The traces were analyzed and the variance of the decay was plotted as a function of the amplitude at that time point. The x-axis was divided into 50-bins of equal current decrement from the peak. The single channel conductance was estimated by fitting the plot to a second polynomial equation, σ 2 = iI -I 2 N + b1, where σ 2 is the variance, I is the mean current, N is the number of channels activated, i is the single channel current and b1 is the background noise. In the conductance conversion (i.e. γ = i/V), the driving force (V) is the difference between the holding (-70 mV) and reversal potentials (assumed to be 0 mV).
The kinetics of the mean EPSC from each neuron was estimated in Clampfit (Molecular Devices) by measuring 20-80% rise time (τrise) and the time constant for the decay (τdecay).
Representative sample traces are the averages of all of the traces that were selected for analysis, superimposed with individual peak-scaled traces (10 successive sweeps), unless otherwise stated.
Stimulus intensity was set to obtain a sporadic observation of transmission failures but high enough to obtain a reliable estimate of γ.

Plasmid constructs and lentivirus production
The following oligonucleotide sequences were used to generate single guide RNA (sgRNA) for

In vivo stereotactic injections and dual whole-cell recordings
The surgical procedure was performed under sterile conditions in accordance with the Institutional Animal Care and Use Committee of Seoul National University. Male C57BL/6 mice (2-3 months of age) were anesthetized by intraperitoneal injection of a ketamine (130 mg/kg body weight) and xylazine (10 mg/kg) mixture. The anesthetized mice were immobilized on a stereotactic apparatus and the lentiviral medium (0.5 µL per each at a flow rate of 0.1 µL/min; 5 x 10 9 TU/ml) was bilaterally injected at CA1 area using a microinjection syringe (Hamilton).
The coordinates used were -1.7 mm posterior, ±1.2 mm lateral to bregma and -1.5 mm below the skull surface.
Following 4-6 weeks of expression, the hippocampal slices were prepared and whole-cell recordings were made as described above. EGFP-positive and neighbouring uninfected neurons were identified by epifluorescence microscopy and compared by dual whole-cell recordings.
Rectification index was measured as described in 26 . AMPAR currents were isolated using a mixture of D-AP5 (100 µM) and L-689,560 (5 µM). The index was calculated by taking the responses from -70, 0 and +40 mV of holding voltages. Following the recordings, brain slices were PFA-fixed, stained with DAPI, and imaged on a confocal microscope (Leica SP8). Ca 2+ /calmodulin-dependent protein kinase II (CaMKII, New England Biolabs).

Statistical Analysis
All treatment groups were interleaved with control experiments. Data are presented as mean ± SEM (standard error of the mean). Responses were normalised to the baseline prior to LTP induction unless otherwise stated. Statistical significance was assessed using (two-tailed) paired or unpaired Student's t-tests or one-way ANOVA as appropriate; the level of significance is denoted on the figures as follows: *p < 0.05, **p < 0.01 and ***p < 0.001.  Individual values (circles) from each neuron are connected by lines. g, Cumulative distribution of the same data set for LTP10'. Dotted lines indicate the mean values for each input. h-i, Analysis of the relationships of γ with, LTP (h) and EPSC decay time (i). Linear regression with 95% confidence intervals (shaded) for the amount of cLTP and the corresponding level of γ. j-p, Equivalent analysis for the LTP induced by sTBS (3 x TBS at inter-episode interval of 10 min; see arrows). The whole-cell recordings were obtained after the second TBS. This was necessary due to the lability of LTP washout. m-o, statistical analysis between control and test pathways (m) and within pathway analysis for control (n) and test (o) pathway reveals a time-and pathway-dependent increase in . Note that higher conductance was observed in the test input (o) compared to the control (n) under the "baseline" state, suggesting that the first + second TBS were sufficient to increase γ. The third TBS triggered a small but discernible further increase in γ (n = 23/17, t22 = 3.75, **p < 0.01, paired Student's t-test). q-r, Analysis of the relationships of γ with LTP (q) and decay time of EPSCs (r).      Under baseline conditions synaptic transmission is mediated by GluA2-containing, calciumimpermeable (CI)-AMPARs, two shown for simplicity. 2. The first theta-burst stimulation (TBS) activates NMDA receptors (NMDARs) and this drives more CI-AMPARs into the synapse by lateral diffusion from a peri-synaptic pool, via a process that involves CaMKII. PKA is also activated (via adenyl cyclase, not shown) and this induces the process of transducing GluA2-lacking calcium-permeable (CP)-AMPARs into peri-synaptic sites on the plasma membrane. 3. LTP is expressed by the increase in number of CI-AMPARs (LTPN) but synapses also become primed for LTPγ by the availability of peri-synaptic CP-AMPARs. 4. Within ~ 1 h, the peri-synaptic CP-AMPARs are removed and, presumably, degraded. 6. If a second TBS is delivered whilst the synapses are still primed (5) then NMDAR activation drives the perisynaptically located CP-AMPARs into the synapse, via a CaMKII-dependent process. This might involve an exchange of CP-AMPARs for CI-AMPARs, which are removed from the synapse via a mechanism triggered by PICK1. 7. These CP-AMPARs increase synaptic strength due to their higher single channel conductance (LTPγ). However, their dwell time in the synapses is quite short (~15 min) before they are removed. If synapses remain active, such as by basal stimulation, activation of the transiently-available, synaptic CP-AMPARs triggers protein synthesis and the insertion of more CI-AMPARs (8), which can extend the expression of LTP for long periods.