Phase-Locked Inhibition, but Not Excitation, Underlies Hippocampal Ripple Oscillations in Awake Mice In Vivo

1Sharp wave-ripple (SWR) oscillations in the hippocampal CA1 region play a key role in memory consolidation during non-rapid eye movement sleep, immobility, and consummatory behavior (,,,; reviewed by ). Despite their important network function, the underlying synaptic mechanisms remain unclear. Several mechanisms of ripple generation have been proposed, including phasic excitation from the CA3 region, phasic inhibition from GABAergic interneurons, and gap junction coupling between pyramidal neurons. Divergent results may derive from technical limitations of the experimental approaches used. Extracellular local field potential (LFP) recording in vivo can be applied to networks in awake, behaving animals, but does not allow accurate dissection of dendritic excitation and perisomatic inhibition, which will produce similar sink-source patterns. Intracellular analysis allows precise analysis of synaptic potentials, currents, and spiking, but is largely restricted to in vitro preparations, or to the analysis of synaptic potentials in vivo with limited temporal resolution. Understanding the synaptic mechanisms of SWR generation requires recording of synaptic currents in CA1 pyramidal neurons in vivo in awake animals under voltage-clamp conditions, where excitation and inhibition can be precisely dissected.

2To probe the synaptic mechanisms of SWR oscillations, we performed simultaneous recordings of excitatory and inhibitory synaptic currents (EPSCs and IPSCs) and LFPs in the CA1 region of awake mice in vivo (Figure 1). Animals were head-fixed, but fully awake, showing behaviors of grooming, whisking, and moving. We first tested whether functional characteristics of SWRs measured near the pyramidal cell layer under our experimental conditions were identical to those in awake, freely moving animals (; Figures 1A–1C). On average, SWRs occurred with a mean frequency of 0.15 ± 0.02 Hz (694 events in 17 mice) and had a mean duration of 69.4 ± 4.3 ms. Individual ripple cycles were generated at a mean frequency of 154.0 ± 1.6 Hz, and the average number of ripples per sharp wave was 10.8 ± 0.7. These properties of SWRs were very similar to those reported previously in awake, behaving rodents (; reviewed by ).

3Next, we investigated the postsynaptic currents during SWR oscillations in CA1 pyramidal neurons in the voltage-clamp configuration (Figures 1D–1F). To obtain adequate resolution, several efforts were made to optimize voltage-clamp conditions and to rigorously assess possible errors. First, we minimized series resistance (Rs), which on average was 17 ± 1 MΩ in our dataset (range, 10–27 MΩ;,, ). Additionally, we used a cesium-based internal solution to enhance steady-state voltage control in distal dendrites. Second, we performed experimental tests for adequate voltage-clamp conditions. Measured spontaneous synaptic events showed a fast time course (Figure S1, available online), and extrapolation toward the limit of small Rs revealed even faster synaptic kinetics (Figure S2). Kinetic parameters and amplitude were only weakly correlated. Both results were consistent with adequate clamp conditions (; Figure S2). Finally, we examined the extent of voltage-clamp errors for EPSCs and IPSCs in detailed cable models based on full reconstruction of in vivo-recorded CA1 pyramidal neurons (; Figures S3–S5). For proximal synapses, errors in peak amplitude, rise time, and decay time constant of somatically recorded EPSCs and IPSCs due to cable filtering were less than a factor of two (Figure S3), and corresponding postsynaptic voltage errors were <10 mV (Figure S4). Distributed pipette capacitance did not introduce additional errors (Figure S5). In conclusion, voltage-clamp errors were unavoidable, but appeared to be within an acceptable range under our experimental conditions.

4Finally, we dissected EPSCs and IPSCs during SWRs by alternatingly setting the holding potential to either −70 mV (i.e., near the reversal potential of GABAA receptors; Figure S6) or +10 mV (i.e., near the reversal potential of AMPA- and NMDA-type glutamate receptors; ). Simultaneous recording of either EPSCs or IPSCs and the LFP revealed that the frequency of both excitatory and inhibitory events increased during SWRs (Figures 1D–1F). However, the generation of IPSCs (Figure 1E) appeared to be more tightly correlated to the SWRs than the generation of EPSCs (Figure 1D).

5To quantitatively analyze the contribution of excitation and inhibition to the generation of ripples, we performed SWR-triggered averaging of EPSCs and IPSCs (Figure 2). Synaptic currents were aligned to the maximal positive ripple deflection in the band pass-filtered LFP, averaged, and converted into synaptic conductances using the values of the driving force. SWR-triggered averaging revealed that the peak amplitude of the inhibitory conductance was markedly larger than that of the excitatory conductance (Figure 2A). On average, the SWR-associated peak conductance was 1.8 ± 0.4 nS for excitation and 6.0 ± 0.8 nS for inhibition (17 cells; p < 0.0001; Figures 2A and 2B), corresponding to an inhibition-to-excitation conductance ratio of 4.1 ± 0.5. In comparison to excitation, inhibition showed a faster rise (20%–80% rise time, 64.7 ± 5.9 ms for excitation and 42.5 ± 4.8 ms for inhibition; p = 0.0021; Figures 2A and 2B), but a slower decay (decay time constant, 47.6 ± 4.6 ms for excitation and 98.3 ± 12.1 ms for inhibition; p < 0.0001; Figures 2A and 2B). Furthermore, inhibition peaked at a slightly later time (time difference, 7.3 ± 1.9 ms; p = 0.0005). Thus, synaptic inhibition, rather than excitation, dominated during the generation of SWRs. Furthermore, excitation and inhibition showed different temporal profiles.

6Previous studies revealed that the amplitude of SWRs varies from event to event by over an order of magnitude. If either excitation or inhibition was responsible for ripple generation, we would expect that the relevant conductance would positively correlate with the ripple amplitude. To test this hypothesis, we plotted the peak amplitude of excitatory and inhibitory conductance in CA1 pyramidal neurons during individual SWRs against ripple amplitude (Figures 2C and 2D). For the excitatory conductance, peak conductance and ripple amplitude were not significantly correlated (linear correlation analysis, Pearson’s r = −0.08; 238 events; p = 0.24; Figure 2C). In contrast, for the inhibitory conductance, peak conductance and ripple amplitude exhibited a strong positive correlation (Pearson’s r = 0.40; 185 events; p < 0.0001; Figure 2D). Thus, the magnitude of inhibition, but not that of excitation, was correlated with the amplitude of individual SWRs. Furthermore, excitation and inhibition were not strictly balanced during SWRs.

7To determine whether EPSCs and IPSCs were phase-locked to individual ripple cycles, we performed interleaved phase analysis for both EPSCs and IPSCs in CA1 pyramidal neurons (Figure 3). EPSC onset times were detected as minima of the first derivative of the membrane current at –70 mV, whereas IPSC onset times were determined as maxima of the first derivative at +10 mV (see Experimental Procedures; 3,566 and 3,180 events total for 25% largest derivative extrema; 17 cells; Figures 3A and 3B). Simulations using cable models based on cell reconstructions suggested that adequate event detection was achieved under these conditions (Figure S7). After computing the Hilbert transform of the band pass-filtered simultaneously recorded LFP signal (100–250 Hz; ), the onset times of EPSCs and IPSCs were individually assigned corresponding phase values, with 0° corresponding to the peak of a ripple. Whereas EPSCs were uniformly distributed (p = 0.90 for 10% largest derivative extrema; Rayleigh test; Figure 3C, left), IPSCs were significantly clustered in the late ascending phase of ripple cycles (p = 0.0007 for 10% largest derivative extrema; Rayleigh test; mean phase angle, −62.9° ± 49.6°, equivalent to 297.1°; Figure 3C, right). This conclusion was corroborated by reverse analysis using EPSC- or IPSC-triggered averaging of the LFP (Figure 3D). Synaptic event-triggered LFP averages were weakly rhythmically modulated for EPSC onsets (Figure 3D, left), but strongly modulated for IPSC onsets (Figure 3D, right). The peak-to-trough amplitude of the triggered LFP average was significantly larger for IPSCs than for EPSCs (p < 0.0001 and 0.0038). Thus, phasic inhibition, but not excitation, was phase-locked to individual cycles of ripple oscillations.

8As PV+ interneurons strongly fire during SWRs (,; see, ), they are possible candidates for the generation of the SWR-associated inhibitory conductance. To test this hypothesis, we examined the effects of optogenetic inhibition of PV+ interneurons (Figure 4). The inhibitory opsin Archaerhodopsin T (ArchT) was selectively expressed in PV+ interneurons in the CA1 region of the hippocampus (Figure 4A). Five-second laser pulses were alternated with 11 s off periods, and properties of SWRs and associated inhibitory postsynaptic conductance were compared between on and off periods (Figure 4B). Optogenetic inhibition of PV+ interneurons did not change the incidence of SWRs, but significantly reduced SWR duration (from 89.3 ± 8.5 ms to 81.0 ± 8.1 ms; 8 mice; p = 0.008; Figure 4B, right top). Furthermore, light pulses did not change the ripple frequency, but decreased the number of ripple cycles per SWR (from 12.1 ± 1.0 to 10.9 ± 1.0; p = 0.008; Figure 4B, right center). Finally, optogenetic inhibition of PV+ interneurons markedly reduced the peak amplitude of the SWR-associated inhibitory conductance (from 9.5 ± 1.9 nS to 5.6 ± 1.2 nS; 7 cells; p = 0.016) and reduced its half-duration (Figure 4B, right bottom). In conclusion, PV+ interneurons make a major contribution to the inhibitory conductance during SWRs. Furthermore, they control the duration of SWRs and the number of ripples in SWR complexes.

9If PV+ interneurons are primarily responsible for SWR generation, optogenetic inhibition of these interneurons might result in a disruption of phase locking of IPSCs. To test this hypothesis, we performed phase analysis and IPSC-triggered LFP averaging before and after optogenetic inhibition after adeno-associated virus-mediated expression of ArchT (Figures 4C and 4D). In the absence of light, IPSCs were phase-locked to the ascending phase of the ripple cycle (Figure 4C, left), similar to the results in uninfected control animals (Figure 3C, right). In contrast, in the presence of light, the phase preference was perturbed (Figure 4C, right). Furthermore, the peak-to-trough amplitude of IPSC-triggered LFP averages was markedly reduced (Figure 4D). Thus, PV+ interneurons provide a major contribution to phase-locked IPSCs during individual ripple cycles.

10The present paper provides a quantitative analysis of postsynaptic conductances during SWRs in hippocampal CA1 pyramidal neurons in awake, behaving mice in vivo. Our results indicate that during SWRs, (1) inhibition dominates over excitation; (2) phasic inhibition, but not excitation, is positively correlated with SWR amplitude; (3) phasic inhibition, but not excitation, is phase-locked to individual ripple cycles; and (4) PV+ interneurons provide a major contribution to the SWR-associated inhibitory conductance. Our findings directly reveal the current generator underlying SWRs. This was possible because our experiments provided information about both phase and location of the underlying conductance. The onset of the inhibitory conductance occurred at a phase of ∼−60°, corresponding to the ascending phase of a ripple. Furthermore, the conductance was likely to be perisomatic because it was mediated, to a large extent, by PV+ interneurons. Thus, the inhibitory conductance represents a perisomatic current source, resulting in a positive deflection in the LFP near the pyramidal cell layer.

11Furthermore, our results shed light on the function of the rhythm generator. They suggest a model in which SWRs are generated by a combination of tonic excitation from CA3 and phasic inhibition within CA1. In contrast, our results are inconsistent with models in which phasic excitation is relayed from CA3 (; see, ). Furthermore, they are incompatible with a major contribution of gap junctions. Whether ripple generation involves feedback inhibition, mutual inhibition, or both remains to be determined. The temporal sequence of events, with action potentials in CA1 pyramidal neurons at the ripple trough, PV+ interneuron firing in the early ascending phase, and onset of inhibitory conductance in the late ascending phase of the ripples (this paper), would be consistent with a feedback mechanism. On the other hand, one might expect that GABAergic interneurons receive tonic excitation during SWRs, similar to that in CA1 pyramidal neurons. This could activate mutual inhibition circuits, independent of feedback innervation.

12Finally, our findings have implications for temporal coding of information in the hippocampal CA1 region. Importantly, they explain why CA1 pyramidal neurons in vivo fire at the ripple trough. In a ripple model based on phasic inhibition, the descending phase of the ripple coincides with the decay of the inhibitory conductance. Pyramidal cells will fire at a point where inhibition has become minimal, which is near the trough in the LFP. In this scenario, fast decay of the inhibitory conductance in CA1 pyramidal neurons in vivo (Figure S1) will ensure temporally precise action potential generation in these neurons. Precise spike timing, in turn, may be critically important for the preplay or replay of temporal activity sequences in cell assemblies and for the consolidation of memory via spike timing-dependent plasticity in downstream neocortical target cells.

13In vivo whole-cell patch-clamp recordings from CA1 pyramidal neurons and simultaneous LFP recordings were performed in head-fixed, fully awake mice. All experiments were carried out in strict accordance with institutional, national, and European guidelines for animal experimentation. Protocols were approved by the Bundesministerium für Wissenschaft und Forschung of Austria (BMWF-66.018/0008-II/3b/2010, BMWFW-66.018/0007-WF/II/3b/2014). PV+ interneurons were optogenetically manipulated using adeno-associated virus. EPSCs and IPSCs were detected using template- or derivative-based detection methods. Simulations were performed on detailed cable models of biocytin-filled CA1 pyramidal neurons. For details, see Supplemental Information.

14J.G. planned experiments, performed experiments, and analyzed data, S.-m.W. performed initial experiments, A.J.P.-A. supervised initial experiments, J.C. and P.J. conceived the project, and P.J. performed simulations and wrote the paper. All authors jointly revised the paper.