Entorhinal cortical delta oscillations drive memory consolidation

SUMMARY

Long-term memories are formed by creating stable memory representations via memory consolidation, which mainly occurs during sleep following the encoding of labile memories in the hippocampus during waking. The entorhinal cortex (EC) has intricate connections with the hippocampus, but its role in memory consolidation is largely unknown. Using cell-type- and input-specific in vivo neural activity recordings, here we show that the temporoammonic pathway neurons in the EC, which directly innervate the output area of the hippocampus, exhibit potent oscillatory activities during anesthesia and sleep. Using in vivo individual and populational neuronal activity recordings, we demonstrate that a subpopulation of the temporoammonic pathway neurons, which we termed sleep cells, generate delta oscillations via hyperpolarization-activated cyclic-nucleotide-gated channels during sleep. The blockade of these oscillations significantly impaired the consolidation of hippocampus-dependent memory. Together, our findings uncover a key driver of delta oscillations and memory consolidation that are found in the EC.

Graphical Abstract

In brief

The temporoammonic pathway originates from the entorhinal cortex and innervates the hippocampal output area, but its role in memory consolidation is unclear. Haam et al. identify potent oscillatory activities of the temporoammonic pathway during anesthesia and sleep. The blockade of delta oscillations of this pathway during sleep impairs memory consolidation.

INTRODUCTION

Although the entorhinal cortex (EC) is one of the most severely damaged brain regions in Alzheimer’s disease, suggesting its critical role in memory formation,¹ its role in memory consolidation remains largely unknown. Memory consolidation refers to the process of forming long-term memories in the brain via synaptic and systems consolidation processes after the initial encoding of short-lasting memories in the hippocampus.² The consolidation process predominantly occurs during sleep, when there are limited sensory inputs and low levels of encoding.^3,4 Neural connectivity between the EC and the hippocampus⁵ strongly indicates that the EC is uniquely poised to play a critical role in memory formation. Accordingly, recent seminal findings indicate that the EC acts as a key information processing center in the memory system.^6–8 The temporoammonic (TA) pathway is an entorhinal cortical hippocampal pathway that monosynaptically projects from EC layer III (ECIII) to the CA1 region of the hippocampus. As the CA1 area is the major information integration center in the hippocampus that provides the major hippocampal output to other cortical areas, the unique anatomical structure of the TA pathway suggests its significant role in memory consolidation. Indeed, a lesion of the TA pathway interferes with the formation of long-term memories, suggestive of its role in memory consolidation,⁹ but the underlying mechanism is largely unknown.

Neural oscillations are a form of neural information that adds complexity to the fundamental neural code such as firing frequency.¹⁰ Neural oscillations are generated and governed by the relationships among the activities of multiple neurons, more specifically synchronous activities of multiple neurons.¹¹ Seminal studies have shown the specific mechanism and role of different types of neural oscillations in memory formation.^12–16 Among those, slow wave activity, which comprises slow oscillations (0–1 Hz) and delta oscillations (1–4 Hz), are neural oscillations that are a hallmark of non-rapid eye movement (NREM) sleep and have been shown to play a pivotal role in memory consolidation during sleep.³ The two primary regions that are known to drive slow wave activity are the thalamus and the neocortex.¹⁷ The thalamocortical relay neurons in the thalamic nuclei are considered pacemaker cells, as these cells intrinsically generate delta oscillations via T-type calcium and hyperpolarization-activated cyclic nucleotide-gated (HCN) channels.^17–19 The mechanism of the neocortical slow wave activity, on the other hand, is to some extent controversial, but it has been shown to depend on synaptic inputs to these neurons.^17,19

In the present study, we used cell-type-specific neural oscillation recordings to demonstrate that the TA pathway neurons in the EC exhibit potent delta oscillations during sleep that are functionally significant. Using individual neuronal activity recordings using miniature endoscopy, we also demonstrate that sleep cells of TA pathway neurons drive delta oscillations during sleep. The delta oscillations in the TA pathway neurons were generated via HCN channels, the blockade of which impaired delta oscillations and the consolidation of contextual and trace fear memories. Thus, our study identified potent oscillatory activities of the TA pathway during sleep and revealed its critical role in offline memory processing for consolidation.

RESULTS

Selective recordings of the TA pathway neurons reveal potent oscillatory activities in these neuronal populations

Oscillatory activities of neurons are traditionally recorded with electrophysiological approaches, which offer high signal-to-noise recordings of neural activities with a high temporal resolution. As a selective recording of oscillatory activities from the TA pathway neurons requires cell-type specificity, we used an alternative approach of calcium-indicator-based neural activity recordings, which employ the genetic expression of the indicator under a specific promotor. Changes in intracellular calcium levels measured by the calcium indicator GCaMP6f indicate changes in firing activities and have been widely used in measuring an increase or a decrease in neural activities.²⁰ Neural activity recordings using a calcium indicator have slower kinetics compared with those from the traditional electrophysiological method. However, given that GCaMP6f used in this study offers a high sensitivity that can differentiate signals at a resolution of one action potential and a relatively fast rise time (<0.1 s),^20,21 sequential calcium-level changes that occur even during the decay time of a previously occurred event can be reliably detected. To selectively target the TA pathway neurons, which are long-range projecting glutamatergic neurons, we used viral expression of a calcium indicator in combination with the fiber photometry approach. In this experiment, we injected a cocktail viral solution containing adeno-associated virus (AAV) carrying either a GCaMP6f or tdTomato gene under the glutamatergic-neuron-specific promoter CaMKIIα (AAV9-CaMKIIα-GCaMP6f and AAV9-CaMKIIα-tdTomato, respectively) into ECIII of adult mice (>2 months old). It has been previously reported that ECIII is comprised of long-projecting as well as local-projecting neurons.²² To record neuronal activity selectively from ECIII neurons that project to the hippocampal area CA1 (i.e., the TA pathway), optical probes were placed in the CA1 region where GCaMP6f-expressing axon terminals were located (Figures 1A, 1B, S1A, and S1B). This approach, which combines the genetic expression of the calcium indicator and circuit-specific optical recording of the neuronal terminals, ensured that the TA pathway neurons were selectively recorded without including GABAergic, local-projecting glutamatergic neurons, or neurons that project to other parts of the hippocampus (Figures S1A and S1B). To eliminate movement artifacts, the GCaMP6f signal was normalized to tdTomato signals, which were acquired using the linear spectral unmixing approach²³ (Figures S1C–S1E). Surprisingly, the selective recordings of the TA pathway neurons revealed potent oscillatory activities of these neurons when animals were under isoflurane-induced anesthesia. Under anesthesia, the TA pathway neurons exhibited low-frequency, large-amplitude calcium transients, which were alternating with a low baseline state (Figure 1C). These large-amplitude transients rapidly disappeared when the animals woke up from anesthesia, and high-frequency, small-amplitude transients with a higher baseline state appeared (Figure 1C). The photometry signals reflect populational neuronal activities from multiple neurons, and the intensity reflects the number of neurons that are active at each moment, resembling the Poisson distribution. We calculated the Fano factor, which is the ratio of variance to mean for a time window, from the populational signals (Figures S1G and S1F) and found a significantly higher Fano factor during anesthesia, indicating a higher synchrony state compared with during waking (Figures 1D and 1E). We next tested whether or not oscillatory activities would be observed in other types of cortical neurons. Thus, we measured oscillatory activities from glutamatergic neurons in different regions of the brain by injecting the same virus cocktail and placing an optical probe at various sites. In addition, we targeted GABAergic neurons in ECIII by injecting the virus cocktail containing Cre-dependent GCaMP6f and tdTomato vectors (AAV-hSyn-DIO-GCaMP6f and AAV-hSyn-DIO-tdTomato, respectively) into ECIII of GAD2-IRES-Cre mice. Optical probes were placed into the virus-injected areas to record neuronal activities from GABAergic neuronal cell bodies in ECIII. Interestingly, hippocampal area CA1 and CA3 and ECIII GABAergic neurons did not show any detectable oscillatory activities under anesthesia (Figures 1F and 1G). On the other hand, the primary visual cortical neurons showed detectable oscillatory activities, albeit at a degree that was significantly lower than what was observed in the TA pathway neurons (Figures 1F and 1G). These observations indicate that the TA pathway neurons exhibit highly synchronous activities under anesthesia while displaying asynchronous activities during waking.

Figure 1. Selective recordings of TA pathway neurons reveal potent oscillatory activities in the TA pathway neurons.

(A) Schematic showing that the temporoammonic (TA) pathway, which provides direct synaptic inputs to the hippocampal area CA1, was selectively targeted by injecting the AAVs carrying GCaMP6f and tdTomato under the glutamatergic-neuron-specific promoter CaMKIIα to ECIII. Optical probes were implanted in the CA1 area to record the activity of the TA pathway terminals.

(B) A horizontal brain section from an AAV-injected animal, showing the labeling of the perforant pathway (PP) that projects to the CA1 area and TA alvear pathway (TAAP). Scale bars, 400 mm.

(C) Under isoflurane (2%)-induced anesthesia, the TA pathway neurons exhibited highly synchronous activities, which were absent during waking.

(D) Fano factor calculated for 10 s bins averaged over 7 animals during anesthesia and waking.

(E) Summary plot showing a significant increase in Fano factor during anesthesia compared with waking (paired samples t test, n = 7 mice, p = 0.003).

(F) Neural activities from various cortical regions under anesthesia recorded with fiber photometry. A low concentration of isoflurane (0.5%–0.75%) in combination with ketamine/xylazine (100 mg/kg ketamine, 7 mg/kg xylazine) was used during the recordings.

(G) Fano factor calculated for each brain region (GABA, n = 6; TA, n = 8; CA1, n = 4; CA3, n = 4; V1, n = 4; one-way ANOVA followed by Tukey post hoc tests, TA vs. CA1, p = 0.001; TA vs. CA3, p = 0.001; TA vs. CA1, p = 0.001; TA vs. GABA, p < 0.001; TA vs. V1, p = 0.001). Shaded line graphs present mean ± SEM. Violin plots range from the 1st through the 99th percentile. Asterisks denote the means, whereas the horizontal bars denote the medians. Boxplots within the violin plots range from the 1st through the 3rd quantile. n.s. p > 0.5, *p < 0.05, **p < 0.01, and ***p < 0.001.

Recordings of neural activities from individual cell bodies of the TA pathway reveal two subpopulations that respond distinctively to anesthesia

As the fiber photometry approach measures calcium signals from the terminals of multiple TA pathway neurons, the important questions were (1) whether the cellular mechanism responsible for the oscillatory activities observed at the terminals is located at the cell body or the terminals and whether the oscillatory activities would be also observed at the cell body, and (2) how the individual TA pathway neurons generate the populational oscillatory activities. To answer these questions, we employed recordings of individual cell bodies using a retrograde AAV approach. The retrograde AAV is internalized by cell bodies like other AAV serotypes but is also internalized by axon terminals and retrogradely transported to the cell bodies.²⁴ There are two main challenges with the retrograde AAV approach. First, when the retrograde AAV is injected into the CA1 area, the neurons in the CA1 area (i.e., CA1 pyramidal neurons) can internalize the virus and express the transgene (i.e., GCaMP6f), and the expression of GCaMP6f in the terminals of these neurons can cause a high background in the entorhinal cortical region (Figure 2A). Second, a neuronal population in ECII, called island cells, has been shown to project to GABAergic neurons in the CA1 area,²⁵ which can also be unintentionally labeled with the retrograde AAV approach (Figure 2A). To exclude these two neuronal populations, we used the Cre-off expression of GCaMP6f. Both CA1 pyramidal cells and island cells express the Wolfram syndrome-1 (WFS1) gene.^25,26 To isolate the TA pathway neurons without labeling these neuronal populations, we injected Cre-off retrograde AAV (AAV-hSyn-FAS-GCaMP6f) into the hippocampal area CA1 of WFS1-Cre mice (Figure 2A). With the Cre-off in combination with the retrograde AAV approach, GCaMP6f was selectively expressed in ECIII neurons but not in the CA1 pyramidal or island cells (Figures 2B and S2A). To examine if the retrogradely labeled TA cell bodies exhibit potent oscillatory activities similar to what we observed with the populational terminal recordings, we performed photometry recordings on the labeled cell bodies. Populational activity recordings of the TA pathway cell bodies demonstrated potent oscillatory activities under anesthesia (Figures S2A and S2B), consistent with the terminal recordings. These results indicate that the oscillatory activities of the TA pathway neurons are generated at the level of the cell bodies.

Figure 2. Recordings of individual TA neurons demonstrate that a subset of the TA neurons exhibit potent oscillatory activities at the single-cell level during anesthesia.

(A) Schematic showing that the injection of Cre-off retrograde AAV (AAV-hSyn-FAS-GCaMP6f) in the hippocampal area CA1 in WFS1-Cre mice allowed specific labeling of the TA pathway neuron cell bodies in the EC.

(B) A brain section from a retrograde AAV-injected WFS1-Cre mouse showing GCaMP6f expression in ECIII. The expression of the calcium sensor GCaMP6f was absent in ECII island cells and CA1 pyramidal neurons that endogenously express WFS1. Scale bar, 200 mm.

(C) Schematic showing the miniature endoscopy recordings of the TA pathway neurons during anesthesia and waking, and offline cell extraction. Individual neurons were identified with the CNMF-E cell extraction approach. Scale bar, 100 mm.

(D) Clustering of TA pathway neurons to anesthesia oscillators (AOs) and anesthesia non-oscillators (ANOs) based on the power of 0–1 Hz during anesthesia (n = 76 total cells from two recording sessions; AOs, n = 20; ANOs, n = 56).

(E) Example traces of five individual AOs and ANOs during anesthesia and waking.

(F) The average traces of AOs (n = 20) and ANOs (n = 27) during anesthesia and waking.

(G) Average correlation coefficient of AOs (n = 20) and ANOs (n = 27) over time. The red and blue lines indicate the average correlation coefficient from AOs (190 pairs) and ANOs (351 pairs), respectively. The three time segments (divided by dashed lines) correspond to 2% isoflurane (anesthesia [Anes]), 0.5% isoflurane, and post-Anes (Post).

(H) Average correlation coefficient calculated for pairs of AOs during Anes and post-Anes (n = 76 cells; mixed ANOVA followed by Holm post hoc tests; AOs, Anes vs. Post, p < 0.001; ANOs, Anes vs. Post, p = 0.08; Anes, AOs vs. ANOs, p < 0.001; AO, Anes vs. ANO, Post, p < 0.001). Shaded line graphs present mean ± SEM. Violin plots range from the 1st through the 99th percentile. Asterisks denote the means, whereas the horizontal bars denote the medians. Boxplots within the violin plots range from the 1st through the 3rd quantile. n.s. p > 0.5, *p < 0.05, **p < 0.01, and ***p < 0.001.

We then asked how individual neuronal activities contribute to the potent populational oscillatory activities during anesthesia. To measure individual neurons in the TA pathway, we employed miniature endoscopy, which acquires a series of fluorescent images of GCaMP6f-expressing cells over time, unlike fiber photometry, which acquires a photon count. The offline cell extraction using the extended constrained non-negative matrix factorization (CNMF-E) approach^27–29 allowed us to isolate the activities of individual TA pathway neurons (Figure 2C). Interestingly, we observed two distinct populations of TA pathway neurons that can be clustered based on the oscillatory power at 0.01–1 Hz frequency during anesthesia (Figure 2D). The first subpopulation of the TA pathway neurons (~26% of the labeled TA pathway neurons), hereafter called anesthesia oscillators (AOs), exhibited rhythmic oscillatory activities under anesthesia at a single-cell level that are synchronous to each other (Figures 2E and 2F). On the other hand, we observed a second subpopulation of the TA pathway neurons (~74% of the labeled TA pathway neurons), hereafter called anesthesia non-oscillators (ANOs), which exhibited little to no oscillatory activities under anesthesia but showed increased neuronal activity during waking (Figures 2E and 2F). The AOs showed a significantly higher correlation during anesthesia compared with during waking (Figures 2G and 2H). On the other hand, ANOs did not show an increase in correlation during anesthesia (Figures 2G and 2H). These results together demonstrate that the TA pathway neurons are comprised of mixed populations, a subset of which drive potent populational oscillatory activities under anesthesia.

The TA pathway neurons exhibit delta oscillations during natural sleep

The states of light anesthesia and sleep have remarkable similarities since anesthetics modulate the neural circuits in the hypothalamus and midbrain that control sleep and waking states.^30–35 Light levels of anesthesia recapitulate similar restorative effects that natural sleep produces.^36–38 Furthermore, light anesthesia has been shown to facilitate memory consolidation similar to sleep.^39,40 However, there are distinct physiological characteristics that are only observed in anesthesia but not in slow-wave sleep.⁴¹ Therefore, we investigated if the TA pathway neurons exhibited oscillatory activities during natural sleep. As mice are nocturnal animals, we conducted 2–3 h of in vivo recordings during the early part of the light cycle (Figure 3A) and sleep classification to examine oscillatory activities that occur during a particular behavioral state. Waking and sleep states are most commonly detected by using electroencephalography (EEG) and electromyography (EMG). However, simultaneous recordings of photometry with EEG/EMG can affect the sleep state, as the two patch cables are prone to tangling, which can significantly impair animals’ mobility and natural behavior. While EEG- and EMG-based sleep detection has remained the gold standard, movement-based sleep detection methods using motion sensors and video recordings have been successfully implemented recently.^42–44 We undertook an alternative approach of a video-based sleep classification method, which was trained using EEG and EMG data (Figures S3A and S3B). The video-based sleep classification method showed over 90% accuracy in detecting sleep when compared with EEG- and EMG-based sleep detection (Figure S3C).

Figure 3. The TA pathway neurons exhibit delta oscillations during sleep.

(A) Schematic showing photometry recordings of the TA pathway with simultaneous video recordings during the light phase.

(B) Waking and sleep states classified by the ensemble learning of video data.

(C) Example filtered photometry recording trace of the TA pathway neurons show two different states: oscillatory and desynchronized states.

(D) The power spectral density function from two different states. During the oscillatory activities, high power in the delta frequency range (1–4 Hz) was observed.

(E) Spectrogram of the photometry trace shown in (B).

(F) Average delta power during waking and sleep (n = 8 mice, paired samples t test, p < 0.001).

(G) Average Fano factor during waking and sleep (n = 8 mice, Wilcoxon singed-rank test, p = 0.008). Violin plots range from the 1st through the 99th percentile. Asterisks denote the means, whereas the horizontal bars denote the medians. Boxplots within the violin plots range from the 1st through the 3rd quantile. n.s. p > 0.5, *p < 0.05, **p < 0.01, and ***p < 0.001.

Using the video-based activity detection method, we detected sleep epochs over 2 h recordings (Figure 3B). Interestingly, we observed strong oscillatory activities in the TA pathway only when animals were sleeping but not during waking (Figure 3C). Frequency domain analysis revealed that the oscillatory activities shown during the sleep period were within the delta oscillation frequency, as shown in the power spectral density plot and spectrogram (Figures 3D and 3E). The power of delta oscillations in the TA pathway neurons was significantly increased during sleep epochs compared with during wake epochs (Figures 3D–3F). Notably, delta oscillations are brain oscillations that are well known in the thalamic regions during slow-wave sleep,^45–47 but delta activity during sleep has not been extensively studied in the entorhinal region. The potent delta oscillations shown with our photometry recordings indicate that cell-specific oscillatory activity recordings isolate the potent oscillatory activities of the TA pathway neurons at the delta frequency, which would have been difficult to detect with the recording of mixed neuronal populations. The Fano factor during sleep was significantly higher than that during waking (Figure 3G), suggesting that the TA pathway neurons exhibit a higher degree of populational synchrony during sleep compared with during waking.

Sleep cells of the TA pathway drive delta oscillations during natural sleep

We then asked how individual TA pathway neurons work together to generate delta oscillations during natural sleep. To answer the question, we conducted miniature endoscopy recordings of retrogradely labeled TA pathway neurons at the beginning of the light phase in combination with simultaneous video recordings for the classification of sleep and wake states (Figure 4A). Interestingly, we observed heterogeneous subpopulations of the TA pathway neurons; a subpopulation that exhibited an increase in neuronal activities during sleep, hereafter called sleep cells (~15% of the total TA pathway neurons); another subpopulation that showed an increase in neuronal activities during waking, hereafter called wake cells (~76% of the total TA pathway neurons); and lastly, ones that did not show any preference (no difference [n.d.] neurons) (Figures 4B–4D). We then examined the populational activities of the sleep cells (i.e., an average of all sleep cell activities) and found high delta power during sleep but not during waking. The populational activities of wake cells, on the other hand, did not show any increase in delta power during either the waking or the sleep state (Figure 4F). When the entire TA pathway populational activities were measured using photometry, retrogradely labeled TA pathway neurons exhibited an increase in delta power during sleep (Figures S2C–S2F), similar to photometry recordings of terminals. An important question was if individual sleep cells exhibit delta oscillations at the single-cell level or if only the collective signal would exhibit oscillatory activities. Spectrograms of individual neurons, recorded with miniature endoscopy, suggest that sleep cells indeed exhibit oscillatory activities at delta frequency during sleep but not during waking (Figures 4G, 4H, and S4B). On the contrary, wake cells showed an increase in activity during waking but did not show an increase in delta power during sleep either individually or collectively (Figures 4F–4H and S4A). We then asked how individual neuronal activities are synchronized to collectively manifest as populational delta oscillations. To understand the difference in populational activity between wake and sleep cells due to phase synchrony, we performed a phase synchrony analysis. Sleep cells showed higher phase synchrony compared with wake cells during waking, and the synchrony was further increased during sleep in sleep cells (Figure 4I). These data collectively suggest that populational synchrony, as well as individual neuronal activities, shifts during sleep, which accounts for the manifestation of delta oscillations. When the cell bodies of the ECIII neurons were directly labeled by injecting the virus carrying the calcium indicator under the glutamatergic-neuron-sensitive promoter (AAV9-CaMIIα-GCaMP6f) to the EC, this neuronal population, which includes local projecting neurons in addition to the TA pathway neurons, contained a subpopulation that exhibited elevated delta activity and phase synchrony during sleep (Figures S5A–S5C), consistent with the findings from the retrogradely labeled TA pathway recordings.

Figure 4. Sleep cells of the TA pathway drive delta oscillations during sleep.

(A) Schematic showing miniature endoscopy recordings of TA pathway neurons during sleep.

(B) Two distinct populations of the TA pathway neurons were identified based on the neural activity levels during sleep and waking sections. Example histograms of activity levels from wake (left) and sleep (right) cells during waking (cyan) and sleep (magenta). The dashed lines represent the median values of the activities during waking (blue) and sleep (red).

(C) Among the total of 195 cells (from 5 recording sessions from two animals), sleep cells account for 15.38% of the TA pathway neurons (30 cells), whereas wake cells account for 75.9% of the population (148 cells). No difference (n.d.) cells account for 8.72% of the population (17 cells).

(D) Example traces of sleep and wake cells during sleep (magenta shaded) and waking (white) epochs. S1 through S5 are the traces of individual sleep cells. W1 through W10 are the traces of 10 randomly selected wake cells.

(E) Activity levels from wake and sleep cells during waking and sleep (n = 149 wake cells, n = 30 sleep cells; two-way mixed ANOVA followed by Holm post hoc tests; wake cells, waking vs. sleep, p < 0.001; sleep cells, waking vs. sleep, p < 0.001; sleep, wake cells vs. sleep cells, p < 0.001; waking, wake cells vs. sleep cells, p = 0.009).

(F) (Top) Example spectrograms plotted for the average trace of wake and sleep cells from a recording session showed an increase in the delta in sleep cells but not in wake cells. (Bottom) The time course of delta power shows an increase in delta oscillations in sleep cells during sleep.

(G) Summary graph showing that sleep cells exhibit an increase in delta power during sleep (n = 149 wake cells, n = 30 sleep cells: Mann-Whitney test, p < 0.001).

(H) Scatterplot showing the activity levels and delta power during sleep from wake (green) and sleep (blue) cells. All values were normalized to that during waking.

(I) The phase synchrony among wake and sleep cells during the two different states. Sleep cells exhibit an increase in global phase synchrony during sleep, whereas wake cells show a decrease in phase synchrony during sleep (n = 149 wake cells, n = 30 sleep cells, 274 sleep epochs, 497 wake epochs; two-way mixed ANOVA followed by Holm post hoc tests; wake cells, waking vs. sleep, p < 0.001; sleep cells, waking vs. sleep, p < 0.001; waking, wake cells vs. sleep cells, p < 0.001; sleep, wake cells vs. sleep cells, p < 0.001; wake cells, sleep vs. sleep cells, waking, p < 0.001; wake cells, waking vs. sleep cells, sleep, p < 0.001). Violin plots range from the 1st through the 99th percentile. Asterisks denote the means, whereas the horizontal bars denote the medians. Boxplots within the violin plots range from the 1st through the 3rd quantile. n.s. p > 0.5, *p < 0.05, **p < 0.01, and ***p < 0.001.

Delta oscillations in the EC are critical for memory consolidation

Our next question was what generates delta oscillations in the TA pathway neurons and whether the oscillatory activities during sleep are indispensable for memory consolidation. Delta oscillations that we observed in the TA pathway neurons resemble delta oscillations in the thalamic neurons, which are considered intrinsic rhythm generators during sleep. Delta oscillations in the thalamus are intrinsically driven by the interplay between the T-type calcium and HCN channels.^17–19 We then asked if the delta oscillations in the TA pathway are driven by T-type calcium or HCN channels. To answer this question, we infused selective channel blockers locally to ECIII through chronically implanted cannulas at the beginning of the sleep cycle (Figures 5A and S6A). Infusion of the T-type calcium channel blocker NNC 55–0396 (0.25 mg) did not have a significant effect on delta oscillations during sleep (Figures S6B–S6E), indicating that T-type calcium channels do not significantly contribute to the generation of delta oscillations in the TA pathway neurons. It has been shown that HCN channels are expressed in ECIII pyramidal neurons, which control the excitability of these neurons.^48,49 Interestingly, the blockade of HCN channels using ZD7288 (1 μg) abolished delta activity in the TA pathway during sleep (Figures 5B and 5C). The blockade of HCN channels in ECIII using ZD7288 occluded an increase in delta power and Fano factor during sleep (Figures S6F and S6G). Infusion of ZD7288 did not have a significant effect on the power of theta oscillations (7–12 Hz) of TA pathway neurons (Figure S6H). These results indicate that delta oscillations in the TA pathway neurons during sleep are driven by HCN channels. The important question was whether the HCN-channel-mediated delta oscillations during sleep are indispensable for the memory consolidation process. To investigate the role of HCN-channel-mediated delta oscillations in memory consolidation, we used the contextual and cued fear conditioning test. For cued fear conditioning, we used trace conditioning, which is known to be dependent on the hippocampus, with the tone and shock presentations separated by 20 s.^50,51 To block delta oscillations in the TA pathway during consolidation, ZD7288 or vehicle (i.e., saline) was infused into the ECIII immediately after the acquisition of fear memory but before sleep (Figure 5D). Before the infusion, the ZD7288 group exhibited comparable levels of freezing during the fear memory acquisition compared with the control group (Figures 5E and 5F). Interestingly, compared with the control group, ZD7288-infused mice showed significantly reduced freezing levels when introduced to the same context where fear memory was formed, indicating its impairment in the consolidation of contextual fear memory (Figure 5G). Furthermore, the ZD7288-infused mice showed significantly reduced freezing during the cued conditioning test, where a conditional stimulus (tone) was presented without shocks in a novel context (Figure 5H). The ZD7288-infused mice showed a tendency for reduced freezing during the 2 week context test, but the difference did not reach statistical significance (Figure 5I; p = 0.059). During the 2 week cue test, the ZD7288-infused mice showed significantly reduced freezing, indicating reduced memory consolidation with the blockade of HCN channels (Figure 5J). We then asked if the effect of ZD7288 on the EC is permanent and if it would prevent new acquisition or consolidation after a recovery. When the ZD7288-infused mice were subject to the retraining of contextual memory 4 weeks after the infusion, the mice showed a significant increase in freezing during acquisition, indicating that these mice can encode fear memory (Figure S7A and S7B). Furthermore, the previously ZD7288-infused mice showed a significant increase in freezing during the context and cued conditioning tests after retraining (Figures S7C and S7D), indicating that the effect of ZD7288 is not permanent; the ZD7288-infused mice were able to consolidate fear memory during retraining once significant time had lapsed (i.e., 4 weeks) since the ZD7288 infusion.

Figure 5. Blockade of HCN channels blocks the TA delta and impairs memory consolidation.

(A) Schematic showing microinfusion before sleep recordings. Mice were infused either with the vehicle (saline) or the HCN channel antagonist ZD7288 (6.6 mM) just before recordings during the light phase.

(B) Example time course of behavioral state (S, sleep; W, waking) and spectrogram from recordings after saline or ZD7288 injection.

(C) The average delta power after saline or ZD7288 injection (paired samples t test, p = 0.033, n = 3).

(D) Schematic showing microinfusion after the fear conditioning acquisition but before consolidation during sleep.

(E) Percentage of freezing during baseline (BL), intertrial interval 1 (ITI1), ITI2, and ITI3 of saline- and ZD7288-infused mice on the training day. The black and blue lines indicate the average percentage of freezing of saline-infused mice (n = 9) and ZD7288-infused mice (n = 11), respectively (two-way mixed ANOVA followed by post hoc tests with Holm corrections; BL, saline vs. ZD, p = 1.000; ITI1, saline vs. ZD, p = 1.000; ITI2, saline vs. ZD, p = 0.339; ITI3, saline vs. ZD, p = 0.295).

(F) The average freezing during ITI1, ITI2, and ITI3. The white and blue bars indicate the average freezing of saline-infused mice (n = 9) and ZD7288 (ZD)-infused mice (n = 11), respectively (independent samples Student’s t test, p = 0.127).

(G) Average percentage of freezing during the context test (saline, n = 9 mice; ZD, n = 11 mice; independent samples Student’s t test, p = 0.032).

(H) Average percentage of freezing during the cue test (saline, n = 9 mice; ZD, n = 11 mice; independent samples Student’s t test, p = 0.025).

(I) Average percentage of freezing during the 2 week context test (saline, n = 9 mice; ZD, n = 11 mice; independent samples Student’s t test, p = 0.059).

(J) Average percentage of freezing during the 2 week cue test (saline, n = 9 mice; ZD, n = 11 mice; independent samples Student’s t test, p = 0.023). Data are presented as mean ± SEM. Violin plots range from the 1st through the 99th percentile. Asterisks denote the means, whereas the horizontal bars denote the medians.

(K) Percentage of freezing during BL, ITI1, ITI2, and ITI3 of saline- and XE991-injected (1 mg/kg) mice on the training day. The black and blue lines indicate the average percentage of freezing of saline-infused mice (n = 8) and XE991-injected mice (n = 7), respectively (two-way mixed ANOVA followed by post hoc tests with Holm corrections; BL, saline vs. XE991, p = 1.000; ITI1, saline vs. XE991, p = 1.000; ITI2, saline vs. XE991, p = 1.000; ITI3, saline vs. XE991, p = 1.000).

(L) The average freezing during ITI1, ITI2, and ITI3. The white and blue bars indicate the average freezing of saline-infused mice (n = 8) and XE991 (XE)-infused mice (n = 7), respectively (independent samples Student’s t test, p = 0.462).

(M) Average percentage of freezing during the context test (saline, n = 8 mice; XE, n = 8 mice; independent samples Student’s t test, p = 0.032).

(N) Average percentage freezing during the cue test (saline, n = 8 mice; XE, n = 8 mice; independent samples Student’s t test, p = 0.025).

To examine whether the effects of ZD7288 are specific to hippocampus-dependent memory, we conducted delay fear conditioning, during which a foot shock is presented during the last 2 s of the 20 s tone presentation. This delay fear conditioning protocol tests hippocampus-independent memory, unlike trace conditioning.^50,51 Microinfusion of ZD7288 into ECIII caused no significant effect on delay fear conditioning performance (Figure S8), suggesting that the effect of ZD7288 is selective to hippocampus-dependent memory.

To examine whether the blockade of other types of channels would result in similar cognitive deficits, we conducted trace fear conditioning experiments with the blockade of the voltage-gated Kv7 channels. Kv7 channels are non-inactivating potassium channels expressed in the EC,⁵² and the activation of these channels generates the M-current, modulating the firing properties and oscillatory activities of entorhinal cortical neurons.^53,54 To examine whether the blockade of Kv channels affects memory consolidation, the Kv7 blocker XE991 (1 mg/kg), which can cross the blood-brain barrier, or saline was injected intraperitoneally immediately after the training session of trace fear conditioning. The XE991-injected mice exhibited comparable levels of freezing during acquisition compared with the saline-injected mice (Figures 5K and 5L). Interestingly, during the context test, these XE991-injected mice showed improved performance compared with the saline-injected control with a significantly higher rate of freezing (Figure 5M). During the cue test, XE991-injected mice showed comparable levels of freezing compared with the saline-injected controls (Figure 5N), indicating that blockade of the M-current did not have a significant effect on hippocampus-dependent cued memory.

These results collectively demonstrate the functional role of HCN-channel-mediated delta oscillations in the consolidation of hippocampus-dependent memory.

DISCUSSION

Recent advancements suggest that the EC plays a critical role in memory formation.^6–8 The important yet remaining question is what functional role the EC plays in memory consolidation. Given that memory consolidation is a crucial step that forms long-term memories, elucidating the memory consolidation mechanism would be critical for a better understanding of the memory formation mechanism. Here, we reveal potent oscillatory activities of entorhinal cortical neurons during sleep and the mechanism and functional role of the oscillations in memory consolidation.

In this study, we demonstrate that the TA pathway neurons generate delta oscillations during sleep, using cell-type- and circuit-specific recordings. Previous studies have shown that ECIII neurons show persistent firing activity during neocortical up-down states.⁵⁵ However, the question of whether TA pathway neurons participate in oscillatory activities during natural sleep at a single-cell and/or populational level has remained unanswered. We demonstrated that the TA pathway neurons exhibit behavioral-state-dependent and functionally significant delta oscillations that have not been reported previously. The TA pathway neurons also exhibited robust oscillatory activities under anesthesia, the degree of which was significantly higher compared with neurons in the primary visual cortex (at an approximate magnitude of 20–80 times). This is surprising considering that the primary visual cortical neurons have shown robust slow-wave activity under anesthesia.^56,57 This discrepancy can be explained by the following reasons. Firstly, populational activity recordings using photometry isolate neural activities of a specific cell type, whereas local field potential recordings measure activities from the vicinity of the electrode and may have included signals from other types of cells that do not show strong oscillatory activities. Secondly, oscillatory activities measured with GCaMPs highly reflect neurons’ spiking activities, whereas population activities recorded with the electrophysiological approach (e.g., local field potential recordings) mainly reflect synaptic inputs to the neurons and are minimally influenced by short-lasting spiking activities in the neurons.⁵⁸ Thirdly, oscillatory activities recorded at the single-cell level may not necessarily manifest at the populational levels. Oscillatory spiking activities of primary visual cortical neurons under anesthesia have been previously shown with single-unit recordings.^56,57 On the contrary, our calcium-sensor-based photometry recordings reflect the spiking activities of a population of neurons. As populational spiking activities reflect not only the average rate of individual neuronal activities but also the degree of synchrony among these activities, oscillatory activities shown at the single-cell level may not generate populational-level oscillations unless there are sufficient levels of synchrony within the population.

We found that delta oscillations of the TA pathway neurons depend on HCN channels. This finding indicates the TA pathway neurons generate rhythms by intrinsic membrane properties rather than being driven by synaptic inputs. This is more analogous to thalamic neurons, which can generate delta rhythms even at a single-cell level as pacemaker cells, than the neocortical slow oscillations that rely on synaptic interactions among neocortical neurons.⁵⁹ We demonstrate using miniature endoscopy that individual TA pathway neurons exhibit oscillatory activities at a single-cell level during anesthesia and sleep.

From the measurements of individual neuronal activities using miniature endoscopy, we found that delta oscillations during sleep are mediated by a subpopulation of the TA pathway neurons (i.e., sleep cells). Unlike sleep cells, wake cells exhibited an increase in activity during waking. The existence of two functionally different subpopulations (i.e., wake and sleep cells) in the TA pathway neurons suggests that the TA pathway neurons may play two distinct roles during memory encoding and consolidation. This is consistent with a previous study that showed the role of the TA pathway in the encoding of spatial, object, and temporal association memories.^60–63 Given the interlaminar connectivity found in the EC, the two distinct subpopulations not only contribute to the division of labor in learning processes but also may be critical for the crosstalk between encoding and consolidation processes. In addition to the difference in activities during sleep and waking, the modes of how sleep and wake cells operate seem to be distinct. Sleep cells have a higher correlative activity among the groups regardless of the sleep and waking states. On the other hand, wake cells have significantly lower phase synchrony compared with sleep cells and show even a further decrease in phase synchrony during sleep.

Animals have two distinct sleep states, NREM and REM sleep. In this study, we used an EEG/EMG-trained video-based machine learning algorithm to detect sleep and waking states in photometry studies. While this technique allowed us to record neural activities without using additional patch cables for EEG/EMG recordings, which are prone to tangle with the photometry cable, limiting animals’ natural sleep patterns, this sleep detection algorithm cannot differentiate NREM from REM sleep. However, all in vivo recordings were performed in the early phase of the light cycle, during which NREM sleep is most dominant with only brief bouts of REM sleep. Our EEG/EMG data suggest that NREM accounts for more than 90% of sleep (92.45% ± 0.97%) (Figures S9A and S9B). In our photometry studies, we sometimes observed that the power of delta decreases at the very end of the sleep bout, during which REM sleep typically occurs. These results suggest that delta oscillations of the TA pathway most likely occur during NREM sleep. Future studies will investigate whether these entorhinal cortical delta oscillations specifically occur during NREM sleep.

In this study, we demonstrated that delta oscillations of the TA pathway neurons are critical for the consolidation of newly encoded memory. When delta oscillations in the TA pathway were blocked immediately after the encoding of new fear memory but before sleep-mediated consolidation, consolidation of fear memory was significantly impaired. It remains unclear whether the TA delta oscillations modulate synaptic or systems consolidation. Given the fact that we observed the effect of ZD7288 on a relatively short timescale (1–2 days), it is most likely that the entorhinal cortical delta activity contributes to synaptic consolidation. As the TA pathway neurons receive cortical inputs and innervate the major hippocampal output CA1 neurons,⁶⁴ the entorhinal cortical delta activity may ultimately facilitate systems consolidation. Further studies are required to reveal the mechanism underlying delta-oscillation-induced memory consolidation.

Taken together, these findings demonstrate potent oscillatory activities of the entorhinal cortical neurons that rely on intrinsic membrane properties. The mechanism illustrated in this study uncovers a crucial entorhinal cortical hippocampal network that is critical for memory consolidation.

Limitations of the study

To investigate the effect of HCN channel blockade on memory consolidation during sleep, we infused the HCN channel antagonist ZD7288 during the light phase, when most sleep occurs in rodents. This blockade occurred after the encoding phase to isolate the effect of ZD7288 on consolidation, and ZD7288 did not significantly alter the power of theta oscillations, suggesting the specificity of the HCN channel blockade. However, in our study, we did not block HCN channels specifically during sleep, and the possibility that HCN channel blockade may affect memory consolidation that may occur during quiet waking in the light phase cannot be completely ruled out.

STAR★METHODS

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Jerry L. Yakel (yakel@niehs.nih.gov)

Material availability

All custom plasmids generated from this study will be available upon request.

Data and code availability

• All data will be available upon request.

• Custom analysis code is available at https://github.com/haam-lab/Haam_etal_TA_consolidation.

• Any additional information required to reanalyze the data reported in this work is available from the lead contact upon request.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Wfs1-cre/ERT2 (Wfs1-Tg3-CreERT2), GAD2-IRES-Cre, and wild-type C57B6 mice were initially obtained from Jackson Laboratory and Charles River and were bred in-house in the animal facility. Two to 8 months old male mice were used in this study. All animals were maintained in the light-dark cycle of 12h/12h. All animal handling and procedures were conducted in compliance with the National Institutes of Health Humane Care and Use of Animals.

METHOD DETAILS

Intracranial surgery for virus injection

Prior to stereotaxic surgery, anesthesia was induced with 4% isoflurane, and then ketamine/xylazine (100 mg/kg ketamine, 7 mg/kg xylazine) was injected intraperitoneally. For the TA pathway terminal recordings, AAVs carrying the GCaMP6f and tdTomato genes (AAV9-CaMIIα-GCaMP6f and AAV9-CaMIIα-tdTomato-WPRE) were injected into ECIII (AP −0.5; ML ± 3.45; DV, −4.44; the lambda as a reference) of wild-type C57B6 mice. For the labeling of GABAergic neurons in ECIII, AAVs carrying Cre-dependent GCaMP6f (AAV-hSyn-DIO-GCaMP6f) and tdTomato (AAV-hSyn-DIO-tdTomato) were injected into ECIII of GAD2-IRES-Cre mice. For the retrograde labeling of the TA pathway neurons, the retrograde AAV (retroAAV, hSyn-FAS-GCaMP6f) was injected into the CA1 area of the hippocampus (AP −2.27; ML, ±2.25; DV, −1.63; the bregma as a reference) of Wfs1-cre/ERT2 mice, where the terminals of TA pathway neurons are located. All viral solutions (500–700 nL/site) were injected at a rate of 100 nL/min using a Hamilton syringe (30 G) controlled by a syringe pump. After the virus infusion, the syringe remained for at least 5 min before being withdrawn from the injection site. In mice containing the ERT2-cre gene (i.e., Wfs1-cre/ERT2), Cre expression was induced by 7 days of tamoxifen injection. A 20 mg/mL tamoxifen solution was made by dissolving tamoxifen in corn oil (20 mg/mL) on a shaker in an incubator at 37°C overnight. Once tamoxifen was completely dissolved, the tamoxifen solution was kept at 4°C during the next 7 days of injections. To prevent any expression of GCaMP6f by Cre-positive neurons, tamoxifen injection started 1 day before the stereotaxic surgery and continued for the next 6 days including the day of surgery.

In vivo fiber photometry

Once mice were allowed to recover for at least 3 weeks after virus injection, optical probes (200 μm core, 230 μm cladding, NA 0.48) were implanted bilaterally, which were placed 500 mm dorsal to the region of interest. An optical probe was anchored by a stereotaxic arm to target a specific brain site and was connected to the spectrometer to measure the signals. For TA pathway terminal recordings, optical probes were implanted in the CA1 area (AP −2.35, ML −2.25, DV −1.5). For the recordings of retrogradely labeled TA pathway neurons, optical probes were implanted in ECIII (AP −0.5; ML ± 3.45 through 3.65; DV, −2.5 through −3.5; the lambda as a reference) of Wfs1-cre mice. For the comparison of oscillatory activities of different brain regions under anesthesia, photometry signals were recorded at least 30 s after an optical probe was positioned at a specific brain site (CA3 pyramidal, AP −1.65, ML ± 1.75, DV −1.25 or AP −2.25, ML ± 2.75, DV −1.5; CA1 pyramidal, AP −2.25, ML ± 2.25, DV −1.0; V1, AP −2.3, ML ± 2.25, DV −0.4 through −0.7). Once optical probes were positioned at the final target sites, the probes were anchored to the skull with dental cement. Animals were recovered on a heating pad until ambulatory. After at least 2 weeks of recovery, mice were used for in vivo recordings.

A spectrometer-based photometry system was used for all photometry recordings.²³ A 488 nm laser with a tip power output of 70–75 μW was used to excite GCaMP6f and tdTomato. Background subtraction was performed without ambient light in the room before the patch cable was connected to an optical implant on a mouse.

For anesthesia recordings, mice were placed in an induction chamber in which 1.5–2% isoflurane was supplied at a rate of 4 L/min. For sleep recordings, mice were placed in their home cage during the early phase of the light cycle. Video and photometry data were acquired simultaneously using a custom-made Arduino setup that sends 25Hz TTL signals to the recording devices.

Miniature endoscopy

After at least 4 weeks of recovery after the virus injection, miniature endoscopy lens implantation surgery was conducted. A craniotomy of 1.2 mm diameter was made onto the skull using a dental drill. Prior to lens implantation, a needle track was made using a 27G blunt needle, which was lowered at a rate of 10 μm/6 s up to a DV coordinate of −1 mm. Once the blunt needle was removed, a miniature endoscopy lens (1 mm diameter) was lowered to ECIII (AP −0.25; ML ± 3.65; DV, −2.6 through −2.8; the lambda as a reference) at a rate of 10 μm/6 s, during which fluorescent signals were monitored. Once the lens reached the final DV coordinate, the lens was anchored to the skull using dental cement. A baseplate cover was placed to protect the lens. After a minimum of 2-week recovery, mice were used for in vivo recordings.

Electroencephalogram

To record delta activity from the cortical region during non-REM sleep, 0.1-inch stainless screws soldered with a copper/tin lead were installed onto the skull above the frontal cortex (AP, +1.5; ML, ±1.7). A second EEG screw was also installed on the skull above the parietal cortex (AP, −3; ML, ±1.7) to record theta activity during REM and waking. A ground screw was installed at the following coordinate: AP, 3; ML, ±1.7, whereas a reference screw was installed at the following coordinates: AP, −5.75; ML, ±2. Once all screws were installed onto the skull, leads from the screws were soldered to the leads of the head mount. A pair of EMG wires were installed in the nuchal muscles. All EEG signals were acquired at 1000 Hz with a gain of 100 (Pinnacle, 8200-K1-SL 2 EEG/1 EMG Mouse System for Sleep).

Microinfusion

For the infusion of saline or ZD7288, mice were first implanted with guide cannulas bilaterally targeting ECIII. Before the implantation, small burr holes were made into the skull using a dental drill (Operative carbide burs #1/2 RA). The guide cannulas (P1 Technologies, 26 GA, 4 mm length) were slowly lowered to the coordinates targeting the site dorsal to ECIII (AP −0.65, ML+/−3.65, DV −2.8; lambda as a reference). Guide cannulas were anchored to the skull using cyanoacrylate and dental cement. After the implantation, dummy caps were placed onto the guide cannulas to prevent dust from getting into the cannulas. After a minimum of two weeks of recovery, mice were used for in vivo fiber photometry recordings or behavioral studies. The NNC 55–0396 dichloride stock solution (20 mM) was made using sterile saline and stored as frozen aliquots. On the day of infusion, 2.5 μL of the NNC 55–0396 stock solution was added to 46.6 μL of sterile saline to make the approximate concentration of 858 μM. The injection volume (0.5 μL) of this solution corresponds to the final dose of 0.25 μg. The ZD7288 stock solution (66 mM) was made using sterile saline and stored as frozen aliquots. On the day of infusion, the ZD7288 stock solution was 1:10 diluted in sterile saline (final concentration 6.6 mM). Before the infusion, mice were anesthetized using isoflurane, and the internal cannula (P1 Technologies, 33 GA, 5 mm length) was inserted into the guide cannula. Using syringe pumps, 500 nL of saline or ZD7288 solution was infused at a rate of 500 nL/min. After the infusion, the internal guide was left for 4 min, before being disconnected from the guide cannulas.

Behavioral testing

Before any behavioral experiments, mice were brought to the experimental room at least 30 min before the beginning of the experiments. On the training day, mice were placed in the fear conditioning apparatus (Med Associates, NIR-018SD or Lafayette Instrument 80014CGAT) and were presented with three blocks of tone, trace, and shock after a baseline period of 160 s. For trace conditioning, the tone (20 s) and shock (2 s) presentations were separated by 20 s. For delay conditioning, shock (2 s) was presented during the last 2 s of the 20 s-tone delivery. An intertrial interval was defined as the beginning of the trace period to the next tone presentation. For fear conditioning with microinfusion experiments, mice were infused with saline or ZD7288 (see the section Microinfusion) under isoflurane anesthesia immediately after the training session. For the XE991 experiment, XE991 (1 mg/kg in 0.1 cc saline/10g body weight) was injected intraperitoneally immediately after the training session. One day after the training, mice were subjected to the context test. The context test was conducted by placing the mice in the same context where the conditioning took place for 8 min but without tone or shock presentations. One day after the context test, mice were subjected to the cue test. The cue test was conducted by placing the mice in a different context with tone presentations without shocks. Two weeks after the initial conditioning and test, 2-week testing was conducted, which consisted of the context and cue test. Two weeks after the 2-week testing, retraining was conducted, during which the same conditioning paradigm as the initial training was used. The retraining was followed by the context and cue test sequentially with 1-day delays. The linear method was used to detect freezing behavior with a threshold of 19 and a minimum freezing duration of 30 frames for data generated from Med Associates fear conditioning system. Data from the Lafayette Instrument system were analyzed using custom MATLAB scripts to first remove randomly added video frames. The motion index was then calculated from the video data by comparing the pixel changes after applying a Gaussian filter, erosion, and dilation. Freezing was defined when the motion index is under the threshold of 1000 with a minimum duration of 0.75 s. Manual scoring of freezing was used to optimize and validate the freezing detection algorithm. Percent freezing was defined as time immobile/total session time × 100.

Post hoc analysis of calcium indicator expression and cannula locations

After the completion of in vivo and behavioral study, mice were deeply anesthetized with pentobarbital, and the brains were fixed via cardiac perfusion with 4% paraformaldehyde. Once brains were equilibration with 30% sucrose for at least three days, the brains were frozen with Tissue Freezing Media for cryosectioning. Using a cryostat, 50 μm-thick horizontal or coronal brain sections were made. The sections were placed on a glass slide and mounted with Fluoro Gel with DABCO. Sections were imaged with the Leica SP8 or DM6B microscope. Paxinos and Franklin’s The Mouse Brain Atlas was used as a reference to examine the locations of the cannulas and GCaMP6f expression.

Neural data analysis

All photometry data were analyzed using custom MATLAB scripts. For photometry recordings that contain both GCaMP6f and tdTomato, the linear spectral unmixing method was used to acquire GCaMP6f and tdTomato coefficients. Thereafter, the ratio-metric approach (i.e., the coefficient of GCaMP6f to that of tdTomato) was used to eliminate movement artifacts. All photometry data were presented as percent F/F_0. For the frequency domain analysis, the multi-taper method was used with a moving window width of 10 s and a step size of 1 s. Calcium signals have a slower decay time than electrophysiological signals. However, given the linear relationship between calcium signal and action potential frequency of a moderate range shown in the entorhinal neurons (1–30 Hz) and the relatively fast rise time of GCaMP6f (<0.1 s),^20,21 photometry calcium signals can reliably reflect the populational neuronal activities. As such, photometry signals can be used to estimate the average firing rate of a neuronal population and the degree of populational synchrony, analogous to Poisson distribution.

The Fano factor was calculated for each 10 s time window:

$$F_i=\frac{{\sigma }_i^2}{{\mu }_i}$$

where ${\sigma }_1^2,\dots ,{\sigma }_n^2$ are the variances of the signal in each 10 s-time window and μ_i,…, μ_n are the means of the signal for each 10 s-time window.

All miniature endoscopy data were analyzed using custom MATLAB and Python scripts. Individual neural activity was extracted from miniature endoscopy data using the CNMF-E approach,²⁷ which is available as the Python library CaImAn and the graphical user interface Mesmerize.^28,29 For the analysis of anesthesia recording sessions, AOs were identified by the presence of high power in 0–1 Hz during anesthesia (>0.5). For the analysis of sleep recording sessions, sleep and wake cells were identified based on activity levels in the sleep/wake epochs during the session. The Wilcoxon rank-sum test was used to compare activity levels that are positive during sleep and wake epochs, and the effect size was quantified using Cliff’s Delta. For a given cell, when the rank-sum test yielded a p value less than the Bonferroni adjusted p value of 0.05/n (where n was the number of neurons being compared for that session), then the neuron was identified as a wake or sleep cell depending on whether the effect size is positive or negative, respectively. When the p value did not meet the criteria, the cell was identified as no difference (n.d.) cell.

For all miniature endoscopy recordings of retrogradely labeled TA pathway neurons, a relatively low sampling frequency (6 Hz) was used due to the low expression of the fluorophore. Delta oscillations shown with miniature recordings were not due to theta frequency showing up in the delta frequency range due to the Nyquist frequency limit, as demonstrated by higher sampling frequency (20 Hz) recordings (Figure S10).

The multi-taper method was used for spectral analysis with a moving window of 10 s and 20 s for sleep and anesthesia data, respectively. The Hilbert method was used for phase synchrony analysis, The Pearson correlation coefficient was used to examine the correlation between pairs of individual neurons. Phase synchrony and correlation were calculated for 10 s or 20 s time bins for sleep and anesthesia data, respectively.

Video data analysis

All video data were analyzed using custom MATLAB scripts and Ethovision XT. Movement data, acquired using Ethovision XT, were used for sleep detection using a threshold-based or machine-learning approach. Video-based measurements of rodents’ mobility have been used extensively (e.g., freezing behavior fear conditioning). A threshold-based sleep detection was used for video data acquired simultaneously with miniature endoscopy recordings:

$$\theta =\text{b}+\sigma$$

where θ is the immobility threshold for the recording session, and b and σ are the baseline and the standard deviation of the signal, respectively. The baseline was calculated by the following:

$$\text{b}=\text{max}({\text{C}}_{\text{i}})$$

where C₁,…, C_k are the histogram counts with a bin width of 0.1.

One notable challenge with the threshold approach on our in vivo photometry recordings arises from the movement of the patch cable falsely signaling the mouse’s movement. This issue occurs when the optical patch cable is of a significant thickness and cannot be installed through a commutator, unlike the miniature endoscopy setup. This permits the patch cable to be placed occasionally in front of the camera, causing significant pixel changes by breathing, but not by mobility. In this case, threshold-based movement detection would cause higher errors (accuracy below 85%), therefore we implemented a machine learning approach to improve the accuracy of waking and sleep prediction.

Machine learning was conducted to classify sleep and wake status on 10 s-bin data of movement data. Seven features that represent the trend, periodicity, and irregularity were used: mean, standard deviation, median, base, and power of 0–1 Hz, 1–4 Hz, and 4–10 Hz. 10100 observations were used for training classifiers. Among 26 classifiers, the bagged ensemble method was chosen based on accuracy, sensitivity, and specificity. Randomly shuffled response data were used for the random model. Manual scoring of EEG, EMG, and video data was used as response data for machine learning. Manual scoring was based on the power spectra of EEG, EMG signals, and visual inspection of video data.

Statistical analysis

Statics were performed using MATLAB (Mathworks) and JASP (JASP Team). The normality of the samples was tested using the Shapiro-Wilk and Anderson-Darling tests. For the comparison of two samples, the Students’ t-test, Mann-Whitney test, or Wilcoxon signed-rank test were used. For the comparison of more than two samples, the Kruskal-Wallis test or ANOVA with post hoc tests was used. Statistical significance was defined when p < 0.05. All photometry recordings were included unless insufficient GCaMP6f expression was present (defined as GCaMP6f coefficient <150). All behavioral data except Cohort#2 (Cohort #1, 3, and 4) were included. All animals from Cohort#2 were excluded due to issues with microinfusion. The expression of the calcium indicator with retrograde AAV is considerably lower than that induced by other typical serotypes (e.g., AAV9 used in this study). Among the animals that were injected with retrograde AAV, animals that showed detectable levels of fluorescent signals within 4 months of virus injection were used for miniature endoscopy recordings (2 out of 11 injected animals). All animals that were injected with AAV9 (direct labeling of ECIII neurons) for miniature endoscopy recordings were included for data analysis (3 out of 3 injected animals).

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
AAV9-CaMIIα-GCaMP6f	NIEHS Viral Vector Core	N/A
AAV9-CaMIIα-tdTomato-WPRE	NIEHS Viral Vector Core	N/A
AAV9-hSyn-DIO-GCaMP6f	NIEHS Viral Vector Core	N/A
AAV9-hSyn-DIO-tdTomato	NIEHS Viral Vector Core	N/A
retroAAV-hSyn-FAS-GCaMP6f	NIEHS Viral Vector Core	N/A
Experimental models: Organisms/strains
Mouse, C57B6	Charles River	C57BL/6NCrl
Mouse, GAD2-IRES-Cre	JAX	RRID:IMSR_JAX:010802
Mouse, Wfs1-Tg3-CreERT2	JAX	RRID:IMSR_JAX:009103
Software and algorithms
MATLAB 2018, 2019, 2020, 2021, and 2022	Mathworks	RRID:SCR_001622
Ethovision XT14, 15, and 16	Noldus	RRID:SCR_000441
Inscopix nVista	Inscopix	RRID:SCR_017407
Video Freeze	Med-Associates	RRID:SCR_014574
CaImAn	https://caiman.readthedocs.io/	RRID:SCR_021152
Mesmerize	https://github.com/kushalkolar/MESmerize	N/A
Video-based sleep detection algorithm	https://github.com/haam-lab/Haam_etal_TA_consolidation	N/A
Photometry data processing algorithm	https://github.com/haam-lab/Haam_etal_TA_consolidation	N/A
Video-based freezing detection algorithm	https://github.com/haam-lab/Haam_etal_TA_consolidation	N/A

Highlights.

• TA pathway exhibits highly synchronous activities under anesthesia

• TA pathway neurons generate delta oscillations during sleep via HCN channels

• TA pathway consists of distinct subpopulations: sleep and wake cells

• HCN channel blockade during sleep impairs the consolidation of hippocampus-dependent memory

INCLUSION AND DIVERSITY

We support inclusive, diverse, and equitable conduct of research.