Sharp-Wave Ripples Orchestrate the Induction of Synaptic Plasticity during Reactivation of Place Cell Firing Patterns in the Hippocampus

1Synaptic plasticity is believed to mediate the encoding of memories by strengthening connectivity between co-active neurons representing constituent features of an event or environment. Recently encoded memories are liable to interference and require consolidation, a process thought to occur during rest and sleep when recently active neural ensembles are reactivated in the hippocampus. During these consolidation epochs, existing hippocampal connectivity may be refined through further plasticity and consolidated engrams subsequently integrated into neocortex for longer-term storage. This two-step model of memory formation therefore requires that long-term potentiation (LTP) is induced during both the encoding and consolidation stages.

2LTP can be induced at hippocampal synapses by intense, high-frequency stimulation of presynaptic axons, postsynaptic depolarization coupled with presynaptic stimulation, or by delivering tightly synchronized pre- and postsynaptic activity. The latter, also referred to as spike-timing-dependent plasticity (STDP), leads to LTP or long-term depression (LTD) according to the precise timing and temporal order of pre- and postsynaptic activity. Spike-timing-dependent LTP (STD-LTP) requires causal spiking to occur within a narrow temporal window, with a presynaptic spike followed by a postsynaptic spike within 30 ms. Anti-causal activity, whereby the postsynaptic neuron fires before the presynaptic neuron, can lead to STD-LTD. However, STDP rules are synapse- and developmental-stage-specific. For example, at mature Schaffer collateral (SC)-CA1 synapses, multiple postsynaptic spikes are required for STD-LTP; this is important when considering the spiking requirements for STDP between co-active neurons encoding a given memory. In vivo, tightly correlated CA1 and CA3 pyramidal cell spiking is predicted to satisfy the requirements for STDP induction at SC-CA-1 synapses. Indeed, there are defined periods during the encoding and consolidation phases of hippocampal memory processing when CA1 and CA3 pyramidal cells are coactive and STDP may occur.

3Hippocampal place cells fire in a location-dependent manner, and thousands of cells in the hippocampal CA3-CA1 network are likely to have overlapping place fields, and therefore be co-activated, within a given environment. The firing patterns of cells with overlapping place fields may satisfy the requirements for STDP to be induced during memory encoding, for example on exploration of a novel environment. In fact, these firing patterns have been shown to induce LTP at SC-CA1 synapses in vitro, but only when cholinergic receptors are also activated in a manner that may mimic the elevated cholinergic tone observed during awake behavior. This is consistent with previous evidence for induction of LTP during encoding of memories.

4The reactivation or replay of place cell firing patterns during rest or sleep is associated with transient, high-frequency network oscillations known as sharp-wave ripples (SWRs), which are necessary for normal memory consolidation. Reactivated place cell firing patterns during SWRs undergo time compression by a factor of ∼10 when measured across all place cells on a track, leading to synchronous CA3 and CA1 pyramidal cell firing that is predicted to engage STDP. Therefore, despite reduced levels of cholinergic tone in the hippocampus during rest and sleep, the reactivation of place cell firing patterns during SWRs may support plasticity and memory consolidation in the hippocampus. However, current evidence for LTP induction during memory consolidation falls short of demonstrating that replayed spike patterns induce plasticity; bicuculline-induced bursting in CA3 can induce LTP at SC inputs to CA1 in vitro, and stimulating CA1 pyramidal cells during spontaneous SWR can increase subsequent postsynaptic responsivity in vivo. Meanwhile, an alternative hypothesis suggests that LTP during sleep could be counterproductive and proposes that synaptic renormalization during sleep may be vital for learning and memory.

5Here, we directly test the prediction that reactivated place cell firing patterns induce LTP. We used natural pre- and postsynaptic spike and local field potential (LFP) patterns simultaneously recorded from CA3 and CA1, respectively, during consolidation epochs in vivo to control synaptic inputs and postsynaptic spiking in CA1 pyramidal cells recorded in vitro. We find that reactivation of place cell firing patterns during SWRs can induce LTP and demonstrate how spike timing in relation to ongoing network activity modulates plasticity.

6To address whether synaptic plasticity is induced during reactivation of memory traces, we first sampled CA3 and CA1 place cell firing patterns recorded from adult male Wistar rats in vivo during exploration and rest periods (Figure 1A). Unit and LFP activity was recorded while rats explored a familiar linear track for 10 min and were then transferred into a rest box for a 15-min period of quiescence (Figure 1B). Subsequently, a selection of spike trains recorded from CA3 and CA1 place cells during the rest box period on a single day were used to stimulate acute hippocampal slices prepared from naive, non-implanted rats (Figure 1C).

7To test the plasticity potential of reactivated place cell firing patterns, a subset of five place cells, four from CA1 and one from CA3, recorded in one animal during the first 5 min of the rest box period were selected (Figures 1D–1F). These cells all satisfied criteria identifying them as putative excitatory pyramidal neurons and were selected because they showed clearly defined place fields that were evenly distributed along the length of the track (Figures 1D and 1E). Upon transfer to the rest box, these cells showed typical activity during SWRs, when multiple cells were active within individual SWR time windows (Figure 1F). The median number of spikes per SWR fired by CA3 neurons was 0.22 (first and third quartiles 0.03 and 1.0, respectively); the representative CA3 neuron used for in vitro experiments (CA3a in Figure 1) fired an average of 0.45 spikes per SWR. In CA1, neurons fired a median of 0.28 spikes per SWR (first and third quartiles 0.06 and 1.3), with the four exemplars firing averages of 1.42, 0.75, 0.34, and 0.44 spikes per SWR (CA1b–CA1e in Figure 1). The temporal structure of these SWR-associated spiking events commonly reflected the firing sequences observed on the track (Figures 1G and 1H), consistent with reports of remote replay of recent behavioral sequences, which is proposed to be important for the consolidation of memory.

8To assess the plasticity potential of these SWR-associated reactivation events, we turned to the in vitro hippocampal slice preparation, where the strength of synaptic connections between individual place cells (i.e., CA3 and CA1 pyramidal cells) may be accurately measured. We tested whether the activity of the five place cells recorded during the resting or quiescent phase were capable of inducing plasticity had they been synaptically coupled (though we made no assumption that these particular place cells were directly interconnected in vivo). Given the estimated numbers of place cells active in any one environment and the likely numbers engaged in reactivation, coupled with the dense connectivity between CA3 and CA1 pyramidal cells, it is not unreasonable to assume that place cells with similar activity profiles to those we have recorded will be synaptically coupled in vivo.

9We made whole-cell patch-clamp recordings from CA1 pyramidal cells in acute hippocampal slices. To replicate the activity seen by synapses in vivo during reactivation events, we stimulated these CA1 cells and their SC inputs with patterns of activity recorded from CA1 and CA3 place cells, respectively, during the initial 5 min of the post-run rest period (Figure 2A), as this epoch contained the largest concentration of SWR-associated reactivation events. Timestamps marking when each cell fired during this time period were used to create four induction protocols. In each case, the CA3 spike train provided the presynaptic input and each of the four CA1 cell spike trains provided a different pattern of postsynaptic activity. Synaptic strength at two independent SC-CA1 pathways (control and test) was monitored before and after one of the induction protocols was delivered to the test pathway. Replication of in vivo reactivation events was achieved by electrically stimulating a small number of SC axons at CA3 cell spike times to evoke subthreshold excitatory postsynaptic potentials (EPSPs; corresponding to an average baseline excitatory postsynaptic current (EPSC) amplitude of 33.9 ± 5.8 pA for test pathways and 35.1 ± 6.0 pA for control pathways), while action potentials were evoked in the postsynaptic CA1 cell by a brief somatic current injection from the patch pipette at CA1 spike times (Figure 2B). Transient increases in membrane potential caused by phasic excitation that have been observed during SWRs in CA1 pyramidal cells were also modeled in the slice preparation; a third independent SC input pathway in stratum radiatum was stimulated with five pulses at 100 Hz at timestamps when SWRs had been detected in the LFP (57 detected in 300 s). The stimulation intensity of this pathway was tuned to match the depolarization envelope duration and amplitude observed in vivo (Figure 2C) through synaptic activation of CA1 dendrites.

10Replication of the CA3a-CA1b spike train combination induced test pathway-specific LTP (Figures 2D and 2E; test path, 2.19 ± 0.47; control path, 1.15 ± 0.22; test versus control pathway p < 0.05, n = 8). These two spike trains were cross-correlated during SWR-associated activity in a 200 ms time window, where peak firing of the CA1b during SWRs occurred 0–10 ms before CA3a (Figure 2F). The CA3a-CA1c combination also induced LTP (Figure 2G; test path, 2.42 ± 0.79; control path, 1.01 ± 0.12; test versus control pathway p < 0.05, n = 8). Like CA1b, CA1c firing was tightly correlated with CA3a during SWRs within a 200-ms time window, with the CA1 cell most often firing before the CA3 cell (Figure 2H). The largest change in synaptic strength occurred following stimulation with the CA3a-CA1d combination (Figure 2I; test path, 3.36 ± 0.73; control path, 1.28 ± 0.19; test versus control pathway p < 0.05, n = 8). The cross-correlated firing of CA1d and CA3a during SWRs showed greater numbers of events where the CA3 cell fired before the CA1 cell (Figure 2J). The combination of CA3a-CA1e was the only one not to induce LTP despite having the highest number of spikes occurring during SWRs (Figure 2K; test path, 0.97 ± 0.16; control path, 1.27 ± 0. 43; test versus control pathway p > 0.05, n = 9). The spiking of CA1e and CA3a was not tightly correlated during SWRs with few CA1 spikes occurring within 30 ms of CA3 spikes (Figure 2L). In all cases, LTP developed slowly, lacking a short-term facilitatory component similar to that previously described for low-frequency STDP in hippocampal slices.

11To test the importance of subthreshold depolarizations during SWRs, we repeated these experiments in the absence of SWR-associated depolarization. Spike train stimulation delivered in the absence of SWR-associated synaptic stimulation failed to induce LTP in all cases: CA3a-CA1b (Figure 3A; test path, 0.92 ± 0.19; control path, 1.08 ± 0.32; test versus control pathway p > 0.05, n = 7), CA3a-CA1c (Figure 3B; test path, 1.07 ± 0.25; control path, 1.08 ± 0.25; test versus control pathway p > 0.05, n = 7), CA3a-CA1d (Figure 3C; test path, 0.73 ± 0.19, control path, 0.74 ± 0.13, test versus control pathway p > 0.05, n = 7), and CA3a-CA1e (Figure 3D; test path, 1.13 ± 0.20; control path, 1.43 ± 0.28; test versus control pathway p > 0.05, n = 7).

12These data suggest that depolarization associated with SWRs is required to induce LTP using spike patterns recorded during rest. However, it is not clear if depolarization originating from synaptic stimulation is required or whether somatic depolarization is sufficient. To test this, we injected an artificial sine wave current at the soma to replicate the transient membrane potential deflections observed during SWRs in vivo (Figure 3E). This method of simulating SWR associated changes in somatic membrane potential failed to facilitate LTP for the CA3a-CA1b spike train combination in the same way as synaptic stimulation (Figure 3F; test path, 1.42 ± 0.40; control path, 1.24 ± 0.27; test versus control pathway p > 0.05, n = 7). Similarly, constant depolarization of the somatic membrane potential to −60 mV during presentation of the CA3a-CA1b spike train failed to facilitate LTP (Figure 3G; test path, 1.03 ± 0.20; control path, 1.29 ± 0.38; test versus control pathway p > 0.05, n = 9). These results indicate that dendritic rather than somatic depolarization during SWRs is the critical factor for LTP induction.

13As well as the location of SWR-associated depolarization, the timing of SWR-associated depolarization is also likely to impact the induction of synaptic plasticity. To test this, we artificially de-coupled the timing of the reactivated spike patterns and the simulated SWR-associated synaptic stimulation.

14The in vivo data showed that at time points when SWR onsets were detected in the LFP, an increase in the spiking of all five cells used in the spike pattern stimulation experiments was observed (Figure 4A). Offsetting SWRs by shifting them 100 ms earlier relative to the spike times reduced the correlation between spikes and SWRs (Figure 4A). When slices were stimulated with the same spike trains as in Figure 3 but with SWR-associated synaptic stimulation triggered 100 ms early (Figure 4B), LTP was significantly attenuated or not induced at all. Pathway-specific LTP was induced following stimulation with CA3a-CA1b and offset SWRs (Figure 4C; test path, 1.61 ± 0.24; control path, 1.13 ± 0.15; test versus control pathway p < 0.05, n = 9) but the change in synaptic strength was significantly less than that observed with the correct SWR times (relative change in synaptic strength correct versus offset SWRs p < 0.05). Likewise, LTP was induced following stimulation with CA3a-CA1d (Figure 4E; test path, 1.66 ± 0.24; control path, 1.15 ± 0.15; test versus control pathway p < 0.05, n = 8) but this was also less than that observed with the correct ripple times (relative change in synaptic strength correct versus offset SWRs p < 0.05). LTP was not induced following stimulation with CA3a-CA1c (Figure 4D; test path, 1.61 ± 0.34; control path, 1.34 ± 0.32; test versus control pathway p > 0.05, n = 7) or CA3a-CA1c (Figure 4F; test path, 1.52 ± 0.58; control path, 1.45 ± 0.31; test versus control pathway p > 0.05, n = 7). The reduction in LTP suggests the timing of spikes within SWRs is critical for LTP induction.

15To probe the relationship between spike and SWR timing with higher temporal resolution, we investigated whether the timing of SWR-associated synaptic stimulation could modulate synaptic plasticity induced by artificial spike timing protocols (Figure 5A). In agreement with previous studies, we found that one subthreshold EPSP followed by one action potential (AP) 10 ms later, repeated 300 times at 5 Hz, did not induce LTP at SC-CA1 synapses (Figure 5B; test path, 1.04 ± 0.12; control path, 1.19 ± 0.21; test versus control pathway p > 0.05, n = 7). Delivering the same pairing 13 ms after the onset of an SWR-associated synaptic stimulation induced pathway specific LTP (Figure 5C; test path, 2.34 ± 0.43; control path, 1.28 ± 0.16; test versus control pathway p < 0.05, n = 9). However, delivering the same pairing 40 ms later (53 ms after the onset of the SWR-associated synaptic stimulation) did not result in pathway-specific LTP (Figure 5D; test path, 1.37 ± 0.33; control path, 1.04 ± 0.36; test versus control pathway p > 0.05, n = 6). This is not simply a form of associative plasticity coupling the strong ripple pathway with the weak test pathway, as one EPSP alone delivered to the test pathway 13 ms after SWR onset failed to induce LTP (Figure 5E; test path, 1.34 ± 0.28; control path, 1.18 ± 0.26; test versus control pathway p > 0.05, n = 6). Together, these data show that the timing of coincident pre- and postsynaptic activity in relation to the SWR-associated synaptic stimulation is critical for LTP induction. Furthermore, it suggests that CA3-CA1 spike pairs in the first portion of an SWR are the most important for inducing LTP.

16We found that the spike train capable of inducing the largest change in synaptic strength (CA3a-CA1d) contained eight such plasticity-potent events. These events were classified as a CA3 spike followed less than 30 ms later by a CA1 spike or burst and occurred either just before (less than 30% of the SWR’s duration before onset time) or during the first part of the SWR duration (in the first 60% of a SWR) (Figure 6A). To test whether these events were necessary for LTP induction, we removed the ten CA1 spikes that constituted these events from spike train CA3a-CA1d (Figure 6B). No LTP was induced by this spike train following the removal of the ten CA1 spikes (Figure 6C; test path, 1.08 ± 0.17; control path, 0.99 ± 0.25; test versus control pathway p > 0.05, n = 6). Next, we tested whether these spike events were sufficient for LTP induction by delivering these events alone. Again, no LTP was induced (Figure 6D; test path, 1.51 ± 0.26; control path, 1.28 ± 0.26; test versus control pathway p > 0.05, n = 6). Nor was any LTP induced when the intact spike train, which had previously induced robust LTP (Figure 2I), was used in the presence of the NMDA receptor antagonist DL-AP5 (Figure 6E; test path, 0.97 ± 0.20; control path, 0.87 ± 0.19; test versus control pathway p > 0.05, n = 6). Hence, based on this example, spiking events such as those defined in Figure 6A are necessary, but not sufficient, for the induction of NMDA-receptor-dependent LTP.

17To test if the results from this example pair of place cells might be generalized, we analyzed the number of such necessary spike pairings within SWRs in each spike train protocol. The number of necessary spike pairings within SWRs showed a strong correlation with the change in synaptic strength induced by these spike trains (Figure 7A; r2 = 0.89), supporting a model where LTP-competent pairings have a probability of inducing stepwise changes in synaptic strength. Other factors that might also predict change in synaptic strength, such as CA1 bursts following CA3 spikes or total number of CA1 spikes, did not correlate with induced change in synaptic strength (Figures 7B and 7C). These results support the conclusion that pairs of CA3 and CA1 spikes that occur within a short time window around the start of SWRs are the predominant factor influencing LTP induction.

18Place cell firing sequences are reactivated at compressed timescales during hippocampal SWRs, generating conditions compatible with induction of STDP, and thus facilitating learning and memory. However, direct demonstration of synaptic plasticity induced by replayed activity during SWRs has not previously been provided. In this study, we have formally tested these important hypotheses and found that reactivated place cell firing patterns are able to induce LTP at SC-CA1 synapses but require the additional excitatory synaptic input that CA1 cells receive during SWRs in vivo. Causal spike pairs occurring near SWR onset times are necessary for the induction of plasticity, indicating that infra-ripple spike timing may be a critical determinant of plasticity in vivo. We hypothesize that this form of synaptic plasticity has an important function in consolidating and maintaining hippocampal representations of space.

19Of the representative spike train pairs tested here, which were simultaneously recorded from CA3 and CA1 place cells during a post-run rest period, three were capable of inducing LTP given SWR-associated synaptic stimulation. Though all the tested spike trains had tightly cross-correlated spiking, CA3a-CA1e did not have any causal events near SWR onset and did not induce LTP under any conditions, supporting the conclusion that the timing of spikes within SWRs is critical for LTP induction. This might be expected given that CA3a and CA1e had the most distant place fields for any of the CA3-CA1 pairs and therefore their reactivation is expected to span the duration of SWRs. Interestingly, even though there were plenty of acausal CA3 and CA1 spike timings, none of the tested spike trains induced pathway specific LTD. This is similar to the situation for plasticity induced by place cell firing patterns during exploration and might be explained by a dominance for LTP over LTD or the lack of STD-LTD exhibited at mature SC-CA1 synapses compared to immature synapses which exhibit presynaptically expressed STD-LTD.

20Artificially shifting the timing of SWR-associated synaptic stimulation reduced or abolished LTP, indicating an important, time-sensitive interaction between structured place cell firing patterns and SWR-associated synaptic input. Indeed, we found that SWR-associated synaptic stimulation can powerfully modulate STDP at mature SC-CA1 synapses. These findings demonstrate how the temporal structure of reactivated place cell firing patterns interacts dynamically with network oscillations to sculpt plasticity in the hippocampus, providing important data to inform models of the impact of plasticity’s impact on place cell firing patterns. Indeed, previous models have demonstrated the importance of bursts of coincident dendritic activity to induce LTP at distal synapses in the absence of strong back-propagating action potentials. Modeling studies also highlight the importance of spike timing to generate sufficient NMDAR activation and subsequent Ca2+ influx into dendritic spines, which strengthens synaptic connectivity between place cells with overlapping place fields and creates place cell assemblies. However, such models rely on experimental data to constrain the underlying STDP rules using either phenomenological or biophysical Ca2+-based models. The latter make assumptions about the relationship between total spine Ca2+ and LTP/LTD based on the Ca2+ control hypothesis, which has been challenged by recent experimental evidence suggesting that the relative timing of Ca2+ release from distinct Ca2+ sources within dendritic spines, including NMDA receptors and voltage-gated Ca2+ channels, is key to the induction of STDP. In many modeling studies, further assumptions are made for the existence of STD-LTD that are critical for the stability of the model output but as discussed above may be incorrect and therefore require reappraisal. Thus, our data represent important information to update the assumptions underlying STDP modeling for mature SC-CA1 synapses that may reveal new insights into the role of synaptic plasticity in place cell assembly formation.

21How does SWR-associated excitatory input at one synaptic locus influence the plasticity inducing potential of pre- and postsynaptic activity patterns at a synaptic connection between two place cells? One explanation is that the coincident activation of two independent synaptic inputs induces an associative form of LTP. However, this is unlikely to be the case, since the stimulation of the test and ripple synaptic inputs in the absence of postsynaptic action potentials was insufficient to induce LTP (Figure 5E). Alternatively, the additional synaptic input and resulting dendritic depolarization may increase the amplitude of back-propagating action potentials facilitating the activation of NMDA receptors and LTP induction. One implication is that SWR-associated excitatory synaptic input enhances dendritic excitability and therefore lowers the threshold for induction of plasticity. This is supported by many studies showing that dendritic depolarization facilitates LTP induced by spike pairings (reviewed in ) and that dendritic depolarization and LTP may be enhanced by the frequency of spike pairings. Furthermore, dendritic depolarization and LTP may be enhanced by coincident synaptic inputs from SC and temperoammonic pathways leading in some cases to dendritic plateau potentials, which are strong predictors of synaptic plasticity. Plateau potentials occurring during exploration have been shown to be important for shaping place cell activity in vivo, presumably via the induction of synaptic plasticity. While these plateau potentials generated in distal dendrites by temperoammonic and SC input to CA1 pyramidal neurons are critical for some forms of plasticity induced during awake behavior, we did not see plateau potentials in our recordings, and during SWRs, plateau potentials are largely absent. We conclude that an enhancement of dendritic depolarization facilitates LTP induction during SWRs but is not reliant on the generation of plateau potentials.

22In the context of dendritic depolarization, the contribution of inhibitory synaptic inputs associated with SWRs is also highly relevant as a mechanism of potentially counteracting depolarization and inhibiting LTP induction. Inhibitory inputs during SWRs are principally located on somatic rather than dendritic compartments, and our results suggest that reducing somatic excitability does not significantly alter the threshold for plasticity induced during SWRs. Furthermore, it has been shown that stimulation of CA1 pyramidal neurons during SWRs in vivo enhances subsequent CA1 excitability, suggesting that synaptic plasticity during SWRs may be induced in the presence of inhibition. However, the role of precisely targeted inhibitory input during SWRs in regulating synaptic plasticity remains to be elucidated. The enhancement of dendritic excitability during SWRs superficially predicts that late causal spiking in SWRs will be more likely to induce plasticity. One possible explanation for the importance of early rather than late causal spiking is the slow onset of voltage- and Ca2+-dependent potassium conductances that may reduce dendritic excitability toward the end of SWRs. An example is Ca2+-dependent potassium conductances (SK channels) that are present in dendritic spines where they closely regulate NMDA receptor activity.

23It has been suggested that ripple-associated replay in the hippocampus allows recently encoded spatial engrams to become consolidated though synaptic plasticity. Neurons representing multiple elements of the engram will fire together and therefore “wire together.” However, since ripples boost firing rates across much of the CA3 and CA1 pyramidal cell network, ensuring plasticity only occurs at specific synapses may be problematic. The intra-ripple timing dependent plasticity we demonstrate in this study addresses this issue. Recently active cell assemblies undergo a degree of potentiation during behavior, with enhanced connection strengths subsequently influencing replay activity during rest and sleep. In addition, the enhanced connectivity will make cells within the recently active assembly more excitable, hence more likely to fire immediately after ripple onset. Non-participating cells or cells that have distant place fields and are therefore not tightly bound into the ensemble during exploration (e.g., CA1e) may tend to fire later in the ripple oscillation and not undergo plasticity. In this way, ripples can act to promote and tune synaptic plasticity within the hippocampal network, enhancing the signal-to-noise ratio within the neural code. Previous studies have reported both forward and reverse replay during rest. Where extended replay sequences are concerned, our data predict that reverse replay would enhance the connectivity of place cells encoding proximal locations whereas forward replay would enhance connectivity between cells encoding the beginning of a trajectory. The balance of forward and reverse replay could therefore reflect task demands, with forward replay occurring after an animal leaves a reward location and reverse replay more likely when they arrive at a new one.

24Sleep has an important role in learning and memory, but the precise nature of this role is a matter of debate. Cuing the reactivation of recently acquired information during slow-wave sleep can enhance memory, suggesting that replay in the hippocampus may support memory consolidation. Others suggest that sleep provides a vital opportunity for synaptic downscaling following cumulative potentiation during wakefulness and that further potentiation during sleep could harm memory encoding. Our data suggest that brain activity during quiescence, a state somewhere between sleep and wakefulness, may enable the connectivity of specific spatial engrams to be enhanced prior to sleep, evidence that is compatible with both theories of sleep function. These engrams may be preferentially reactivated and consolidated in the cortex during sleep; if synaptic downscaling occurs, signal-to-noise ratio will be improved, and these representations will become more salient.

25In conclusion, our results show that reactivated place cell firing patterns can induce LTP when accompanied by SWR-associated synaptic input. These data confirm a widely held assumption that reactivation during SWRs promotes synaptic plasticity. They also suggest an active role for phasic excitatory input during SWRs in tuning STDP in vivo. In future studies, it will be important to investigate how SWR-dependent STDP can influence learning and memory directly.

26All procedures were conducted in accordance with the UK Animals (Scientific Procedures) Act, 1986 and with the approval of the University of Bristol ethics committee. Three adult (350–450 g) male Wistar rats (Charles River) were chronically implanted with 19 extracellular tetrode recording electrodes (8 into CA3, 8 into CA1, and 3 into the white matter of the fimbria fornix in the right dorsal hippocampus [−3.6 mm, +2.2 mm from bregma]) under isoflurane recovery anesthesia. During the 7–21 days following surgery, the independently moveable tetrodes were lowered into the brain, targeting the pyramidal cell layer in the dorsal CA1 and CA3 (verified by the characteristic burst mode of single-unit firing and the presence of large-amplitude SWR events in the LFP signal). Recordings were made using a Digital Lynx system (Neuralynx). LFPs (sampled at 2 kHz and filtered between 0.1–475 Hz) and extracellular action potentials (sampled at 30 kHz and filtered between 0.6–6 kHz) were recorded differentially using local references in the white matter overlying the hippocampus. All channels were grounded to two screws placed in the skull overlying the cerebellum. Final tetrode tip positions were verified histologically by identifying sites of electrolytic lesions (Figure 1C) made at the end of experimental procedures under terminal sodium pentobarbital anesthesia.

27Animals were trained to run back and forth on a 200 × 10 cm linear track for a small food reward for a period of 14 days prior to surgery. During these 14 days, animals were food restricted to 90% of free feeding body weight. Recording sessions began once electrodes were in position 21 days after surgery. In a familiar recording room, animals were first placed on a raised platform in a rest box for a 15 min period before being moved to the track where they were allowed to explore freely for 10 min. Animals were then placed back in the sleep box for a further 15 min period. Animals did not receive food reward on the track and were not food restricted prior to recording sessions. Animal movement and behavior was monitored continuously by video. Position on the track was tracked using light-emitting diodes attached to a powered headstage (Cheetah Software; Neuralynx).

28All data were processed in MatLab (MATHWORKS) unless stated otherwise. Single units were isolated manually offline using MClust 3.5 (A.D. Redish, available at http://redishlab.neuroscience.umn.edu/MClust/MClust.html); inclusion criteria were set to isolation distance > 15.0 and L-ratio < 0.35. Putative pyramidal cells were classified on the basis of the spike width, waveform, and mean firing rate. Ripples were detected offline in the LFP recorded on one CA1 channel. Raw LFP signal was filtered between 120 and 250 Hz, and deflections in the ripple power envelope greater than 5 SDs from the mean were classified as ripple events. Ripple start times were defined locally as when ripple power exceeded 2 SDs. Samples of raw LFP and detected ripple times were compared manually to verify detection fidelity. For place cell analysis, the track area was divided into 10 × 10 cm bins, and mean firing rates for each neuron in each bin were calculated.

29Brain slices were prepared from adult (10- to 12-week-old) male Wistar rats following a lethal dose of anesthetic (isoflurane inhalation). Brains were dissected in ice-cold cutting solution containing 119 mM NaCl, 2.5 mM KCl, 1 mM NaH2PO4, 26.2 mM NaHCO3, 10 mM glucose, 1.3 mM CaCl2, and 2.5 mM MgSO4 equilibrated with 95% O2 and 5% CO2. Coronal slices 300–400 μm thick were cut from the dorsal hippocampus using a vibratome (Leica LS1200), and slices were incubated in artificial cerebrospinal fluid (aCSF) containing 119 mM NaCl, 2.5 mM KCl, 1 mM NaH2PO4, 26.2 mM NaHCO3, 10 mM glucose, 2.5 mM CaCl2, and 1.3 mM MgSO4 at 36°C for 30 min and then stored at room temperature until use. Before being transferred to the recording chamber, a cut was made between CA3 and CA1.

30Recordings were made in a submerged chamber perfused with aCSF (as above) at 34°C with the addition of 50 μM picrotoxin to block GABAA-receptor-mediated transmission to enable accurate measurement of monosynaptic excitatory connections between hippocampal pyramidal cells. CA1 pyramidal cells were visualized using infrared DIC optics on an Olympus BX-51 microscope. Patch electrodes with a resistance of 4–5 MΩ were pulled from borosilicate filamented glass capillaries (Harvard Apparatus) using a vertical puller (PC-10, Narashige). Pipettes were filled with intracellular solution containing 120 mM KMeSO3, 10 mM HEPES, 0.2 mM EGTA, 4 mM Mg-ATP, 0.3 mM Na-GTP, 8 mM NaCl, and 10 mM KCl (pH 7.4), 280–285 mOsm.

31Recordings from CA1 pyramidal neurons were made with an Axopatch 200B or a Multiclamp 700A amplifier (Molecular Devices, USA), filtered at 4–5 kHz and digitized at 10 kHz using a data acquisition board and signal acquisition software (CED). Cells were voltage clamped at −70 mV (junction potential correction of −11 mV not accounted for). Series resistance was monitored throughout the experiments and cells that showed a >20% change were discarded.

32Synaptic responses were evoked in control and test pathways with 100-μs-square voltage steps applied at 0.1 Hz through two bipolar stimulating electrodes located in stratum radiatum. A third stimulation pathway in stratum radiatum was used to simulate SWR-associated synaptic stimulation and dendritic depolarization during plasticity induction. The three pathways were tested regularly to ensure independence by paired-pulse protocols (Figure S1). Postsynaptic action potentials were initiated through somatic current injections (2 ms duration, 2 nA amplitude).

33Small amplitude EPSCs (typically 20–40 pA) were recorded in visually identified CA1 pyramidal cells voltage clamped at −70 mV. The stimulation intensity of each input pathway was tuned to elicit sub-threshold EPSPs following a five pulses at 100 Hz stimulus prior to baseline recording. EPSCs were recorded in voltage clamp from two independent pathways for a baseline period of 5 min. Spike train stimulation and spike timing protocols were applied after the neurons were switched to current-clamp mode within 10 min of reaching the whole-cell configuration. The resting membrane potential of the neurons was −70.0 ± 0.5 mV. Following induction, responses to both test and control pathway stimulation were monitored for a further 30–34 min in voltage-clamp mode (Figure 2D).

34Measurements were made from averages of six traces to give one data point per minute. Average data are presented as mean ± SEM. Data were normalized to the average baseline response. Data comparisons were made between test and control pathways at 25–30 min after plasticity induction using Student’s paired two-tailed t test with a significance level of p < 0.05. For between-dataset comparisons of plasticity induction, relative change in synaptic strength (mean test minus mean control pathway response during the final 5 min of the experiment) was calculated for each experiment and values compared using an unpaired Student’s two-tailed t test.

35J.H.L.P.S. conducted the experiments. J.H.L.P.S., M.W.J., and J.R.M. designed and analyzed experiments and wrote the paper.