Inhibition recruitment in prefrontal cortex during sleep spindles and gating of hippocampal inputs

Different Recruitment and Phase Modulation Across Layers During Spindling.

During spindling episodes, mPFC local field potentials (LFP) exhibited large amplitude oscillations identifiable as low-voltage sleep spindles, based on several physiological criteria (Fig. 1A and SI Materials and Methods). When the LFPs were recorded simultaneously in superficial and deep layer FS (in three rats), a phase inversion was observed between layers (Fig. S1). Herein, if not stated otherwise, spindle phases will refer to deep prefrontal LFPs. FS and RS cells were discriminated on the basis of the shape of their action potential (SI Materials and Methods and Fig. S2). Their responses were different depending on the layer from which they were recorded. In the example shown in Fig. 1A, two FS cells were recorded from the two layers, respectively: although the deep-layer cell showed a slight increase in firing rate during the spindling episode, the superficial-layer cell is transiently and strongly activated. Moreover, histograms of action potential phase relative to band-pass filtered LFP in the spindle band (10–15 Hz) revealed a strong phase modulation of the superficial FS cell (P < 10⁻¹⁰, κ = 6.2, Rayleigh test) and a weaker but still significant modulation for the deep FS cell (P < 10⁻¹⁰, κ = 2.6, Rayleigh test).

Although the deep-layer LFP was taken as reference, a larger fraction of superficial cells were phase-modulated than deep-layer cells, both for FS (80% vs. 55%, P < 10⁻¹⁰, binomial test) and RS (34% vs. 16%, P < 10⁻¹⁰) neurons (Fig. 1C). However, a cell can increase its firing rate during spindles without any phase-locking. Fig. 1D shows that although all cell types showed an increase of their relative firing rate, that is the difference between rates within and outside spindles divided by rates outside spindles (the latter being restricted to SWS epochs, see SI Materials and Methods), this was consistently higher for those cell types that were phase-modulated (P < 0.01 for all cell types, two-sample t test). Moreover, the same difference across layers and cell types were observed as with phase-modulation: in a given layer, FS cells tend to consistently be more recruited (P < 10⁻¹⁰, two-way ANOVA), and superficial layers exhibited a stronger activation (P < 10⁻¹⁰). However, the interaction between cell types and layers was also significant (P < 10⁻⁵), showing that the difference between cell types was more marked in the superficial layers than in the deep layers (P < 10⁻¹⁰ and P < 0.001, post hoc t test, respectively). Moreover, both FS and RS increased their firing rate more in the superficial than in the deep layers (P < 10⁻⁷ and P < 0.001, post hoc t test, respectively).

Example neurons from Fig. 1B tend to fire at the peak of the LFP. This feature was indeed present at the population level: FS cells in both layers fire preferentially at the peak of the oscillation (Fig. 1E). Superficial RS cells showed the same preferred phase. Conversely, the small percentage of significantly modulated deep-layer RS cells (16%) did not show any particular preferred phase. Nevertheless, as a population, deep RS cells tended to fire at the peak of the spindle oscillations (P < 0.001, Rayleigh test). Similar results, but in antiphase, were obtained when superficial LFPs were taken as reference (Fig. S3).

Grouping of Thalamic and Hippocampal Activity Bursts by the Slow Oscillations.

During SWS, hippocampal activity is concentrated in SPWR events, the occurrence of which is modulated by neocortical inputs (24). Similarly, during light SWS, neocortical DOWN to UP state transitions favor the recruitment of thalamic activity into spindles (3). Cortical δ-waves, large and isolated deflection of the LFP, are the hallmark of global or local DOWN states and are referred to as K-complex when followed by spindles (as indicated with a green asterisk in Figs. 1A and 2A). As suggested by the example shown in Fig. 2A, δ-waves were preceded by an increased probability of SPWR occurrence (or conversely, some SPWRs tended to take place at the end of UP states), and the stronger the deflection, the higher the propensity of SPWR occurrence (Fig. 2B, Upper, black arrow). However, δ-waves were also followed by an increased occurrence rate of SPWRs, albeit smaller than the rate right before DOWN states (white arrow). Similarly, there was a tendency for δ-waves to be followed by spindles (Fig. 2B, Lower), also in an amplitude-dependent manner. This temporal ordering was reflected in the asymmetric SPWR-triggered spindle density over timescales spanning at least four order-of-magnitude, from subsecond to minutes (Fig. 2C) (cf. 25, 26). Hippocampal burst amplitude can modulate differently PFC response, especially in the spindle band (27). Indeed, the strongest SPWRs, as measured by the amplitude of ripple oscillations (SI Materials and Methods), were followed by higher spindle peak occurrence rate than the smallest ones (Fig. S4A). However, the relationship between SPWRs and δ-waves did not show such dependence (Fig S4B).

Neuronal Sensitivity to Hippocampal Sharp-Waves Depends on Cortical State.

SPWR-associated activity directly impinges upon prefrontal neuronal activity (14, 27). However, it is not known whether cells in different cortical layers respond differentially to SPWRs, or whether this response is state-dependent. Qualitative examination of cross-correlogram of neuronal activity for the two cell types and layers (Fig. S5), within or outside spindle episodes, revealed strikingly different responses. First, SPWR that occurred during spindle episodes are associated with an ongoing state of elevated neuronal firing (suggested by the symmetric modulation in the cross-correlogram). This finding is most remarkable for superficial FS and RS cells, and is compatible with the large, UP state-related dipole measured in superficial layers (28). Indeed, in deep layer this effect is weaker, and virtually absent for RS cells. Second, during nonspindling-associated SPWR events, cortical activity displayed a transient increase, followed by a longer-lasting (1–2 s) plateau of elevated activity compatible with the relationship between SPWRs and UP state onset occurring after δ-waves (Fig. 2A) (26, 29). However, deep-layer RS cells show only a transient firing increase, and no sustained activation (Fig. S5).

Finer examination of the relationship of prefrontal spiking activity with SPWR depending on the spindling state revealed first that FS cells were the most activated. Individual and population average z-scored cross-correlograms in a ± 400-ms window (Fig. 3A) showed that their sensitivity to SPWR relatively to their ongoing firing was enhanced during non spindling states by a factor of 2 to 3 (Fig. 3B, Left) (superficial FS: P < 0.05; deep FS: P < 10⁻¹⁰, paired t test). Furthermore, only deep-, and not superficial-, layer RS cells showed a significant transient response to SPWRs (Fig. 3B, Right). However, during spindling episodes, this response vanished.

State-Dependent γ-Response.

γ-Rhythms in rodent neocortex span a wide range of frequency, from 30 to 140 Hz (30). The example in Fig. 4 shows that γ-oscillations occur in bursts of various frequencies that tended to fire at the peaks of the deep LFP spindle oscillation. We analyzed how instantaneous broadband γ-power (30–140 Hz) changed in relation with spindle and SPWRs. First, γ-power was strongly modulated in phase by spindles, as measured by the averaged z-scored cross-correlogram in ± 500-ms windows (Fig. 5A). The power spectra of instantaneous wide-band γ-power (Fig. 5A, Inset), quantifying the depth of modulation by spindles, were not different in the deep and superficial layers (P > 0.05, Jackknife). Moreover, the power spectra in the γ-band revealed an increase in all γ-frequency bands, with a stronger effect in the 30- to 50-Hz band compared with faster γ-oscillations (>70 Hz) (Fig. 5B). We then analyzed how SPWRs related to γ-oscillations. During spindling episodes, γ-oscillations were modulated by SPWRs, as well as by the spindle cycle. The effect was not different in the two layers and was at about 0.5 SD over the baseline (Fig. 5C1). However, the rythmic modulation could be a by-product of phase-locking of SPWRs to spindles (Fig. 2C) (see also refs. 25 and 26). Conversely, during nonspindling episodes, SPWRs were preceded by a state of weak γ-activity (Fig. S6)—most probably the end of the previous UP state—and provoked a rebound, especially in the deep layers (Fig. 5C2). Interestingly, the γ activation was not the same in all frequency bands (Fig. 5D2): slow γ (30–50 Hz) band power increased sustainably with a 400-ms delay after SPWR occurrence in both layers, and fast γ (>70 Hz) was evoked transiently in the deep layers, possibly caused by (or at least correlated with) spiking itself (Fig. S7). Finally, in the deep layers, wide-band γ-power at the time of SPWR occurrence increased with SPWR strength (Fig. 5 E and F). When taken individually (Fig. 5G), 61% of deep-layer LFPs showed correlated wide-band γ-response with SPWR strength (P < 0.001, n = 151 LFP channels, binomial test), whereas only 49% of superficial LFPs did so (n = 82, not significant from chance level). This increase of wide-band γ-power was specifically the consequence of increase high γ-power (Fig S4C). If SPWR strength was significantly stronger outside rather than within spindle episodes (P < 0.01, paired t test), the relative increase of ripple-frequency band power was too low (1.3% in average) to explain the observed differences in PFC activity between the two conditions.

Rhythmic Activation of Neocortical FS Cells During Spindle Oscillations.

We investigated the interaction between thalamo-cortical spindles and hippocampal SPWRs onto the mPFC. Although it is possible that the analyzed oscillations could be better classified as high-voltage spindles, this is unlikely because several physiological parameters pointed in favor of natural sleep spindles (SI Materials and Methods). These low-voltage spindle oscillations are the hallmark of thalamo-cortical assembly recruitment during the light stages of SWS (3, 31), with thalamo-cortical cells bursting and depolarizing neocortical dendrites. FS cells were strongly modulated by spindle oscillations, both in their phases and in firing rates. Whereas superficial RS cells were also recruited, deep-layer RS cells were phase-modulated poorly, if at all (Fig. 1). This population shows a slight tendency for firing preferentially in antiphase with all other cells, in particular deep FS, and would thus fire when the local inhibition is the lowest. These results are in agreement with several other previous findings. First, the higher modulation of superficial cells by spindles is consistent with the denser projection from the main projecting thalamic input, namely the medio-dorsal nucleus, to the superficial layers of the mPFC (23). Second, spindle synchronization requires the intact thalamo-cortical system (32), with a prominent synchronizing role of cortico-cortical projections in the superficial layers (33). Third, in intracellular recordings during spindle oscillations, the weak modulation in control conditions was changed to powerful bursts by using Cl-filled pipettes, revealing strong inhibition (19).

A recent study reported that a class of superficial FS cells, namely the dendritic targeting cells, fire in antiphase with local pyramidal neurons (34). No such cells were found in our dataset but this might be explained by the lack of very superficial-layer recordings where these cells were found.

Modulation of γ-Burst Occurrence by Spindle Oscillations.

In many ways, UP states are similar to waking desynchronized states, in particular because neocortical neurons display similar patterns of correlation and conductances (35), and γ-oscillations in the LFPs (36). γ-Oscillations are associated to the activation of neocortical cell assemblies (22, 37) and are probably caused by strong interactions between FS and RS cells or between FS cells only (38, 39). Here we show that mPFC γ-activity (30–140 Hz) was remarkably modulated by spindle oscillations (Fig. 5A).

Interestingly, the hippocampal θ-rhythm (5–10 Hz) of active behavior modulates γ-bursts in a similar way: local γ-bursts in CA1 (40, 41) as well as neocortical γ-bursts (30, 42) are phase-locked to the θ-oscillation. The modulation of local fast events by remote slower oscillations is thus a common mode of interstructure communication mechanisms, both between the thalamus and neocortex, and between the hippocampus and neocortex (43, 44).

Interaction Between Thalamic and Hippocampal Inputs at Multiple Timescales.

During SWS, the hippocampal influence on mFPC is mostly mediated by SPWRs, whose occurrence is in turn modulated by the slow oscillation and its nested rhythms: δ- and spindle oscillations (24–26, 29). We provided further data on how these different comodulated events may interact.

SPWR influence on mPFC activity depended on the neocortical state and in particular its oscillatory drive from the thalamus: mPFC sensitivity to hippocampal SPWRs was greater when the mPFC was not involved in thalamo-cortical coupling (Figs. 3 and 5). In particular, deep-layer RS cells relayed hippocampal inputs preferentially during nonspindle states, when they are released from thalamic drive. In this state, SPWRs were concomitant with short (<100 ms) fast γ-bursts (>70 Hz) in the deep layers and the amplitude of the response was correlated to SPWR strength (Fig. 5 E–G and Fig. S4), pointing at activation of the deep layers: fast-γ is strongly related to spiking activity (Fig. S7), possibly because of the low-frequency component of action potentials leaking in the LFP (45). In general, outside spindling episodes SPWRs tended to appear toward the end of UP states, as shown by (i) the low spiking (Fig S5 and ref. 29), (ii) low γ-activity (Fig. 5 and Fig. S6) preceding these SPWRs, and (iii) the negative time-lag shift of the maximum peak in the the correlation between δ-waves and SPWRs (Fig. 2B, black arrow). At the end of UP states, global activity decreases leading to enhanced cell input resistance (46, 47). This process may favor enhanced responsiveness of mPFC cells to hippocampal inputs. However, the SPWRs–δ-wave correlation shows another, but smaller, peak for positive time-lag (Fig. 2B, white arrow). This finding indicates that both UP-DOWN and DOWN-UP transitions could be associated to SPWR occurrence. Whether the hippocampus passively reacts to its inputs or plays an active role in neocortical-state transitions, through mechanisms similar to what shown in vitro (47), remains to be determined (24–26, 29). Note that δ-wave strength was only correlated with the occurrence rate of SPWRs preceding them, suggesting that hippocampal input may be important for complete termination of activity during DOWN states. The second peak, at the putative DOWN-UP transition, was of comparable strength for all δ-waves. Although the present dataset focused on an output structure of the hippocampus, the mPFC, it would be interesting to study how the δ-wave strength in the up-stream areas (i.e., the entorhinal cortex) would affect SPWR occurrence.

SPWRs and spindles are highly comodulated events (25, 26). The PFC's response is positively correlated with the strength of the SPWR, as measured by hippocampal population activity (27). Strong SPWR events were clearly associated to increased occurrence rate of spindle peaks (Fig S4A). Such a relationship was absent between SPWR and δ-waves (Fig S4B). One possibility would be that SPWRs can themselves recruit thalamo-cortical cells into a spindling episodes in a strength-dependent fashion, in a way similar to the neocortical synaptic barrage associated to UP state onset (3).

Furthermore, we show that the timescales of the interaction between the SPWRs and spindle events spanned several orders of magnitude (Fig. 2C). SPWRs were phase-locked to spindles (26), but at larger timescales SPWRs preceded spindle occurrence (25), possibly because of common modulation by the slow oscillation (3, 24, 26, 29), in particular its DOWN phase (Fig. 2B), and possibly to lower frequency components in the SWS dynamics.

Cortical Deafferentiation During Spindles.

Thalamo-cortical cells have roughly two modes of activity, phasic or bursting, and the latter is thought to be responsible for the disruption of relay property of thalamo-cortical cells (3). However, the present study proposes an alternative view. The monosynaptic hippocampo-prefrontal projection offers a unique chance to study the response of a cortical area to an endogenous stimulation independent of the thalamic relay. The present results demonstrate that the cortex is generally unresponsive during spindles, suggesting that thalamic inputs in burst mode recruit massive intracortical inhibition, which itself produces a local deafferentation of the cortical area concerned, regardless of the input source.

Implications for Memory Formation.

The communication between the mPFC and the hippocampus are likely to be crucial for memory formation and consolidation (48). Based on our data, some hypotheses may be drawn about the functional implications of neural dynamics in these areas. Excitatory input from thalamo-cortical cells is thought to massively depolarize neocortical cell dendrites during spindles (19). This finding was suggested to play a role in plasticity mechanisms by enabling calcium entry in dendrites and mobilizing molecular mechanisms for stable changes in synaptic weights (19, 31). On the other hand, to consolidate a given memory trace networks must avoid interference from their spontaneous activity dynamics. Perisomatic inhibition (of which we show the possible effects), would then be the key factor in permitting dendritic depolarization yet in keeping pyramidal cell activity under tight control. However, this theory does not explain which memory trace would be consolidated through this process. During wakefulness, mPFC neuronal activity self-organizes in cell assemblies that are reactivated during sleep (13), preferentially at times of hippocampal SPWRs (14). SPWRs tend to precede spindles over multiple time scales (Fig. 2C), and cell-assembly replay likely occurs at the same times (14). Thus, a likely sequence of events would see replay first (coincident with SPWRs), and then a state favoring consolidation of the recently modified synapses, during spindles.

This theory would explain why during slow oscillations, which temporally organize the chain of events described above, replay is indeed enhanced (6, 14). Thus, although at a subsecond timescale spindles do not corresponds to peaks in replay (14) or hippocampal/PFC communication (shown here), a positive correlation is found at a longer (minutes) timescale between spindling activity and replay (6). Similarly, spindling EEG power is increased with learning (5), possibly as a result of a general increase in slow oscillation activity (49).

This hypothetical gating effect of thalamo-cortical inputs between hippocampal-triggered vs. intrisically cortical consolidation processes would be mediated by the prefrontal cortex circuitry, as we show here, with distinctive roles for FS and RS cells in different layers. The functional importance of this structure will have to be investigated in future work.