Theta-phase-specific modulation of dentate gyrus memory neurons
Abstract
The theta rhythm, a quasi-periodic 4–10 Hz oscillation, is observed during memory processing in the hippocampus, with different phases of theta hypothesized to separate independent streams of information related to the encoding and recall of memories. At the cellular level, the discovery of hippocampal memory cells (engram neurons), as well as the modulation of memory recall through optogenetic activation of these cells, has provided evidence that certain memories are stored, in part, in a sparse ensemble of neurons in the hippocampus. In previous research, however, engram reactivation has been carried out using open-loop stimulation at fixed frequencies; the relationship between engram neuron reactivation and ongoing network oscillations has not been taken into consideration. To address this concern, we implemented a closed-loop reactivation of engram neurons that enabled phase-specific stimulation relative to theta oscillations in the local field potential in CA1. Using this real-time approach, we tested the impact of activating dentate gyrus engram neurons during the peak (encoding phase) and trough (recall phase) of theta oscillations. Consistent with previously hypothesized functions of theta oscillations in memory function, we show that stimulating dentate gyrus engram neurons at the trough of theta is more effective in eliciting behavioral recall than either fixed-frequency stimulation or stimulation at the peak of theta. Moreover, phase-specific trough stimulation is accompanied by an increase in the coupling between gamma and theta oscillations in CA1 hippocampus. Our results provide a causal link between phase-specific activation of engram cells and the behavioral expression of memory.
Editor's evaluation
This study represents an important step toward unifying two strains of inquiry, one related to the functional role of hippocampal theta oscillations and one related to the behavioral impact of engram reactivation, and thus the findings have implications for our understanding of memory that will impact multiple subfields. In combination with additional context from the literature, the important findings are supported by solid evidence supporting the conclusion that memory recall operations occur preferentially at a specific phase of theta.
https://doi.org/10.7554/eLife.82697.sa0Introduction
The discovery of neurons whose activity correlates with memory activity, often referred to as engram cells, has offered evidence that certain memories are stored in a sparse set of neuronal ensembles across the hippocampus (Reijmers et al., 2007). These cells are active during the encoding of a memory and reactivate upon recall of that specific event (Denny et al., 2014; Tayler et al., 2013). Further, artificial reactivation of engram neurons induces recall-like behavior, thus establishing a causal role for these neurons in memory processing (Liu et al., 2012; Ramirez et al., 2013).
Previous studies have utilized activity-based neural tagging strategies, which link expression of a protein of interest (e.g. Channelrhodopsin) to the expression of immediate early genes (IEG) (most commonly Arc [Denny et al., 2014] and cFos [Liu et al., 2012]) to tag and modulate active populations during a specific event. The modulation of memories through activity-based tagging of engram neurons has often targeted cells in the hippocampus (Josselyn and Tonegawa, 2020; Tonegawa et al., 2015); a brain structure in the temporal lobe with a modular design that consists of three major sub-regions: the dentate gyrus (DG), CA3, and CA1, all with extensive interconnectivity (Andersen et al., 2009; Squire et al., 2004). Each of the sub-regions has distinct connectivity patterns that potentially provide unique stages in the processing of memories.
Recently, studies have probed this functional organization by driving engram neurons in different regions of the hippocampus, such as DG (Liu et al., 2012; Ramirez et al., 2013) and CA1 (Redondo et al., 2014; Ryan et al., 2015). In these studies, however, there have been discrepancies in the most effective stimulation frequency, both between regions and within regions (cCompare Ohkawa et al., 2015; Ryan et al., 2015). Further, studies have been limited to activating engram neurons using fixed-frequency stimulation, without taking into consideration the ongoing spontaneous network activity. In particular, hippocampal regions are dominated by a quasi-periodic 4–10 Hz network-wide theta oscillation in the field potential generated by temporally organized firing during memory processing (Buzsáki, 2002; Colgin, 2013).
The hippocampus is tasked with both encoding of new information and recalling past experiences. A prominent model, termed the Separate Phase of Encoding and Recall (SPEAR) model (Hasselmo et al., 2002), posits that encoding and retrieval are temporally interleaved at different phases of hippocampal theta oscillations. In the SPEAR model (Hasselmo et al., 2002; Hasselmo and Stern, 2014), the peak of the hippocampal theta oscillation (as measured in striatum pyramidale) is dominated by inputs from the entorhinal cortex, which carry sensory information that is potentially required for the encoding process. In contrast, the trough (negative peak) of theta oscillations occurs during strong CA3 activity, a region of the hippocampus known for pattern completion that is supported through strong recurrent connections (Leutgeb et al., 2007; Rolls, 2016; Senzai, 2019), and thus ideally suited for the retrieval of previously stored memories.
The SPEAR model has been supported by electrophysiological data in vivo (Hyman et al., 2003; Kerrén et al., 2018; Wang et al., 2020) and in vitro (Kwag and Paulsen, 2009). These studies have indicated that the peak of theta oscillations is associated with strong long-term potentiation, which can support the encoding process, while the trough of theta has strong long-term depression that supports the recall of previously stored memories (Douchamps et al., 2013; Manns et al., 2007). In behaving mice, memory performance can be altered through phase-specific inhibition of neurons in CA1, the output region of the hippocampus (Siegle and Wilson, 2014). Finally, a recent study in human subjects has shown a strong correlation between memory tasks and the phase of theta (Kragel et al., 2020). Despite these results, as well as the prior development of closed-loop optogenetics stimulation (Grosenick et al., 2015), the activation of engram neurons has been carried out at fixed frequencies, without taking into account the phase of theta oscillations.
Here, we tie the modulation of engram neurons to specific phases of theta oscillations. We hypothesize that trough stimulation is more effective than peak or fixed frequencies of stimulation for inducing artificial memory reactivation. To test this hypothesis, we implemented a real-time phase prediction algorithm and tested the behavioral and electrophysiological effects of theta-phase-specific stimulation of DG memory neurons in gating memory recall. We measured theta in CA1 to be consistent with past literature and the SPEAR model. We tagged and reactivated memory neurons in DG rather than CA3, as would be ideal in a test of the SPEAR model. We did this because reactivation of tagged CA3 neurons is not well studied and tends to generate seizure-like activity (unpublished data), presumably because of the higher probability of connection among CA3 pyramidal cells. In contrast, DG stimulation causes robust behavior without generating seizure-like activity. We compared the results of phase-specific stimulation with those using fixed-frequency stimulation at the previously established value of 20 Hz, as well as stimulation at 6 Hz, which provided a control representing the average stimulation frequency during phase-specific activation. Our results support the SPEAR model, with optogenetic stimulation of engram neurons in DG at the trough (recall) phase driving the most robust behavioral recall and the largest amount of coupling between the theta and gamma rhythms in CA1 region.
Results
Closed-loop photo stimulation at specific phases of theta oscillations
To date, engram work has used fixed-frequency optogenetic stimulation to reactivate tagged cells and drive behavior linked to memory recall. For example, most studies have used 20 Hz stimulation, which is well above the natural firing rates measured in the DG (1 Hz on average, with peak rates of 8 Hz) (GoodSmith et al., 2017; Senzai and Buzsáki, 2017). We hypothesized that a more physiologically realistic stimulus pattern, in terms of both firing rate and timing, would reactivate memories more effectivity. To test our hypothesis, we developed a protocol that phase-locked the stimulation time to the theta rhythm.
To deliver photo-stimulation at specific phases of the theta oscillation, we used a custom and a real-time phase detection-and-prediction algorithm. Our algorithm (Figure 1) reads the LFP from the CA1 region of the hippocampus and filters it using a finite impulse response filter between 4 and 10 Hz implemented in the Real-Time eXperimental Interface (RTXI) software (Lin et al., 2010; https://github.com/ndlBU/phase_specific_stim; Noueihed, 2022). The algorithm predicts the timing of the next desired extremum by averaging the duration of previous cycles and sends a TTL pulse to drive the laser at the predicted time of either the peak or the trough of the next theta cycle. The phase of theta oscillations is in reference to measures in striatum pyramidale in the hippocampus. If the location of the electrode, as determined by post hoc histological analysis (see Methods), was observed to be in another layer of the hippocampus, we corrected for this change (trough vs. peak) to ensure similar theta phase across animals.

Quantification of the real-time phase detection algorithm performance.
(A) LFP signal recorded from the hippocampus is amplified and processed in Real-Time eXperimental Interface (RTXI) (https://github.com/ndlBU/phase_specific_stim). The signal is first filtered using an FIR filter in the theta range (4–12 Hz) and then a real-time phase detection algorithm predicts the next extrema. At the predicted time, a TTL pulse is sent to the laser which delivers light through fiber optics to the dentate gyrus (DG) region of the hippocampus to activate tagged neurons. (B) Sample stimulation shows the superior performance of the predictive algorithm in comparison to the 6 Hz periodic stimulation. (C) (Ci) Normalized polar histogram shows the phases of stimulation in the cases of peak, trough, and periodic 20 Hz and 6 Hz stimulation. Dotted lines indicate the accepted peak (pink) and trough (blue) stimulation phase (within quarter cycle). Note that 20 Hz stimulation overlaps completely with 6 Hz stimulation because both are fixed frequencies. (Cii) Confusion matrices indicate that peak and trough stimulation are specific to the desired phase of the stimulation. Stimulations are considered true if they take place within –π/4 to π/4 of the desired phase. The false entry for the No Stim case represents the true negative rate (TNR) or specificity. The true entry for the Stim case represents the true positive rate (TPR) or sensitivity.
The average frequency during both peak- and trough-specific stimulation was 6 Hz. Hence, we compared the performance of our real-time algorithm and phase-specific stimulation with fixed-frequency stimulation at 6 Hz. We also performed experiments with 20 Hz stimulation, as this is a commonly used stimulation frequency in past studies. As shown in Figure 1Cii, the use of closed-loop stimulation resulted in significantly more phase-specific stimulation: 83%, 84% true negative rate (TNR, specificity) for peak and trough stimulation, respectively, as compared to 46% and 3% TNR with 6 Hz and 20 Hz stimulations, respectively. These results indicate that our algorithm has a low rate of stimulation outside of the desired phase, which is critical to testing our hypothesis. Moreover, sensitivity of the algorithm is moderately superior to the 6 Hz stimulation with a true positive rate (TPR) of 58% and 59% for peak and trough, respectively, as compared to 54% for 6 Hz. Since a 20 Hz stimulation rate is about two times higher frequency than theta oscillations, it results in a constant, non-specific stimulation (TNR and TPR of 3% and 97%, respectively). The specificity of the algorithm is critically important in ensuring we are not confounding the results with a high rate of stimulation at the opposite phase. In terms of sensitivity, a higher than 50% value from the algorithm is satisfactory as we do not expect that during natural memory processing the animal is encoding or recalling the memory during every theta cycle. Rather, when either of these processes happen, they are preferentially happening at either the peak or trough, as explained above.
Experimental setup for comparing different modes of stimulation
We developed an experimental setup to compare phase-specific and fixed-frequency reactivation of hippocampal engram neurons. Importantly, our setup was designed to test different modes of stimulation within the same animal. Further, our experimental design randomized the order of the stimulation in different animals as repeated engram reactivation can affect the original memory. The experiment took place in two different contexts: a neutral context A and a fearful context B. As detailed in the Methods section, the two different contexts were differentiated based on a variety of sensory stimuli (see Methods). Animals were first habituated to handling and exploration in the neutral context A. Habituation took place over 4 days, during which both fiber optics and the LFP electrodes were attached to acclimatize the mice to the setup. Light stimulation took place, but as no opsin was expressed yet the light resulted in no behavioral change (Figure 2Cii). As illustrated in Figure 2Ci, the animals were anxious on day 1 as shown by the higher amount of freezing (15%). However, they quickly became habituated to the setup as shown by a much-reduced level of freezing on the last 2 days of the habituation (5%).

Behavioral experiment design.
(A) Schematic of the behavioral experiment. Animals were habituated in context A for 4 days prior to the tagging of engram neurons in context B. During tagging, some animals underwent a foot shock (FS+) while others were placed in context B but without any stimulus (FS-). Following recall, animals were re-exposed to context A and underwent reactivation of the engram neurons with four different stimulation strategies on distinct days (trials). The experiment concluded with a final recall in context B. (B) FS+ (Bi) and FS- (Bii) animals had a similar number of dentate gyrus (DG) cells tagged as shown by the number of granule cells expressing EYFP-ChR2. FS+ animals received a foot shock in context A, while FS- animals did not. (C) Trends of mouse baseline freezing during habituation. (Ci) Habituation of animals over 4 days in context A resulted in decreased in freezing, indicating comfort with the setup. Black line shows the trend for the mean (n=26 animals). (Cii) Average of the percent time freezing over four epochs of trials on the last day of habituation (day 4) indicates 5% increase in percent time freezing due to fatigue later in the trial. This value serves as a baseline for subsequent analysis. ‘x’ indicates No Stim epochs. Shaded area indicates 95% confidence interval (n=26 animals). (D) Both FS+ and FS- animals showed minimal freezing in context B prior to the foot shocks (FS+ n = 17 animals, FS- n=9 animals; independent t-test with Bonferroni correction, p=0.7). However, FS+ animals showed significantly higher freezing post foot shock that persisted on the following day, indicating recall (FS+ n = 17, FS- n=9; independent t-test with Bonferroni correction, ****p<0.00001). (E) On the last day of habituation, both FS+ and FS- groups exhibited minimal baseline freezing (FS+ n = 17, FS- n=9; independent t-test with Bonferroni correction, p=0.6). However, post fear conditioning (FC), the FS+ group showed significantly higher baseline freezing (FS+ n = 17 animals, FS- n=9 animals; independent t-test with Bonferroni correction, ***p<0.0001). The elevated baseline freezing for the FS+ animals is sustained throughout all 4 days of the experiment. The black line shows the trend for the mean freezing. (For all figures, box shows the quartiles of the dataset, while whiskers show the rest of the distribution. Outliers are shown using diamonds.)
Each day, both during habituation and during the experiment, each animal had one trial consisting of four 3 min epochs in which the stimulus lights were turned off and on in the order of No Stim, Stim, No Stim, Stim. Animals were given 4 days of habituation prior to memory tagging (days 1–4), and had 4 days of the experiment after memory tagging (days 7–10). Animal freezing on the last day of habituation was used as a measure of baseline freezing change for the rest of the experiment. As shown in Figure 2Cii, freezing increased by about 5% during the trial. Following habituation, animals were taken off doxycycline 48 hr prior to fear conditioning to allow for the tagging of engram neurons. On the tagging day, animals freely explored context B for 5 min (Figure 2D, pre-FC). Over a second, 5 min interval, experimental animals (FS+) received four 1.2 mA foot shocks, while the control animals (FS-) were left to explore freely. Both experimental and control animals showed minimal freezing prior to the shocks, with only experimental animals exhibiting elevated freezing after the foot shocks (Figure 2D). Post fear conditioning, animals were put back on a doxycycline diet and returned to a new home cage. On the following day, animal recall of the fearful context was tested by re-introducing them to the fearful context B for 5 min. As shown in Figure 2D, only FS+ animals show an elevated level of freezing, indicating successful recall of the fearful memory.
Following tagging and recall of the fearful memory in context B, animals were re-exposed to the neutral context A in which they were originally habituated. Post fear conditioning, only experimental animals showed elevated baseline freezing (1.5% pre vs 10% post), indicating a potential generalization of the fearful context B memory to the neutral context A (Figure 2E). Despite the elevated freezing (10%), neutral context A freezing was lower than the freezing following fear conditioning (45%) or during recall (46%) of the fearful memory in context B (Figure 2D vs. 2E). However, the increased baseline freezing in neutral context A obscured the light-induced freezing and, therefore, required the artificial memory reactivation to generate higher freezing levels to be deemed effective. It is important to note that the elevated baseline freezing is similar across days, making it possible to pool data from different days. As a result of the elevated baseline freezing, an effective stimulation needed to be powerful enough to elicit a behavioral response (increase in percent time freezing) beyond the elevated baseline.
Trough stimulation leads to stronger and more robust recall
Next, we compared the effects of engram reactivation via different modes of stimulation using phase-specific and fixed-frequency stimulation. Because we did not detect any differences across measures taken at different days, we pooled the data across days. As shown in Figure 3Ai, only stimulation at the trough of theta could drive the expected increase in freezing during both stimulation epochs. Although 6 Hz stimulation was effective during the first stimulation epoch, this frequency failed to elicit significant freezing during the second epoch. Peak and 20 Hz stimulations showed a gradual increase in freezing that was similar to the habituation trial. Averaging the light-induced freezing across epochs indicated that only trough stimulation resulted in significantly higher freezing rates (Figure 3Aii; paired t-test, p<0.01).

Behavioral responses indicate that recall is largest when using stimulating at the trough of theta.
(A) (Ai) Average freezing per epoch for FS+ (solid line) and FS- (dashed line) animals during habituation (gray) and during the four modes of stimulation (20 Hz: green, 6 Hz: yellow, peak: pink, trough: blue). ‘x’ indicates No Stim epochs. Shaded region represents 95% confidence interval. (Aii) Average increase in freezing using no stimulation (epochs 1 and 3) and stimulation (epochs 2 and 4). Only trough stimulation reliably caused increased freezing that resulted from activation of engram neurons (n=17 animals, paired t-test with Bonferroni correction). (B) (Bi) Average light-induced freezing was calculated for both experimental (FS+, shaded boxes) and control (FS-, open boxes) animals by subtracting epochs 2 and 4 from epochs 1 and 3, respectively, and averaging the value. Only 6 Hz stimulation and trough stimulation showed light-induced freezing that differed significantly from the non-foot shocked group. Light-induced freezing of using peak and 20 Hz stimulation was not significantly different than the control group (n=17 animals, independent t-test with Bonferroni correction; 6 Hz: *p=0.02 < 0.05; 20 Hz: p=0.6; peak: p=0.07; trough: ***p=0.0002 < 0.0001). (Bii) Light-induced freezing on the last day of habituation prior to the experiment only differed significantly for trough stimulation (n=17 animals, independent t-test with Bonferroni correction; 20 Hz: p=0.8, 6 Hz: p=0.4, peak: p=0.8, trough: *p=0.02 < 0.05). (C) Paired comparison between trough and peak stimulation for animals that exhibited light-induced freezing indicated significantly higher levels of freezing induced by trough stimulation (n=13 animals, **p=0.007 < 0.001). (D) Significantly higher freezing was observed upon exposure to the fearful context B 4 days after artificial reactivation of engram neurons in context A (n=17 animals, paired t-test, **p=0.01).
To compare the effects of the stimulation between different stimulation patterns in the control and experimental animals, as well as between different stimulation patterns, we calculated the percent of light-induced freezing. For this measure, the change in freezing relative to the baseline prior to the stimulation was calculated by subtracting the percent time freezing in epochs 2 and 4 from epochs 1 and 3, respectively. The average of these two values is referred to as light-induced freezing, and represents the increased memory reactivation due to optogenetic stimulation of memory cells. As shown in Figure 3B, only 6 Hz and trough stimulations resulted in significantly higher values of light-induced freezing when compared with the control animals receiving the same stimulation. Further, paired comparisons of the amount of light-induced freezing during each stimulation indicated that only trough stimulation elicits light-induced freezing values significantly different than the habituation trials. As a result, the increase in freezing in the cases of peak stimulation, as well as 20 Hz and 6 Hz fixed-frequency stimulations, were like those expected from a general increase in the animal’s immobility in later trials (Figure 2C).
Direct comparisons of trough and peak stimulations within each animal showed significantly higher freezing values when using trough stimulation (Figure 3C). For this analysis, we focused only on the cases in which both the peak and trough caused light-induced freezing. The analysis indicated that activating engram neurons at the trough of theta was more effective at inducing artificial recall of a tagged memory. We also found that the experimental group showed significantly higher freezing post reactivation of engrams in the neutral context A, which we believe indicates a potential strengthening of the original tagged memory.
Electrophysiological hallmarks during different forms of stimulation indicate changes in gamma-theta coupling following trough stimulation
Having established that trough stimulation is the most effective stimulus in driving freezing behavior, we sought to identify physiological hallmarks of its efficacy. To start, we analyzed the LFP recordings from the CA1 region of the hippocampus. Although CA1 is downstream of DG, the optogenetically stimulated region, it presents an opportunity to uncover the resulting network dynamics from the stimulation. Representative traces show characteristic theta frequency oscillations. After filtering, the theta oscillations become more apparent (Figure 4A). Consistent with previous studies, theta oscillations were higher in frequency during locomotion (Figure 4B, see the ‘bumps’ in the blue lines in the right panels) when compared with measurements made during freezing (red lines). This trend held during all epochs of the experiment regardless of the presence or absence of the stimulation. Similarly, results were not different when using different modes of stimulation (data not shown). A more detailed analysis of theta power (Figure 1—figure supplement 2) demonstrates that theta power is not significantly different in freezing vs. non-freezing conditions. These straightforward measures of hippocampal activity match the literature and provide confidence that the recordings capture the network activity during the experiments, which sets the basis for more in-depth analysis.

LFP characteristics during locomotion and stimulation.
(A) Sample LFP recordings from CA1 during active exploration (blue) and freezing (red). (B) Spectrogram for the whole duration of an epoch (left), and the associated power spectral density graph. In spectrograms, freezing episodes are marked with a white line. As shown, during all three epochs (pre-stimulation: epoch 1, stimulation: epoch 2, and post-stimulation: epoch 3), power and the peak frequency of theta were lower during freezing episodes. This is indicated more clearly in the PSD graphs on the right as evidenced by the shifts in the peak of theta oscillations. Shaded areas represent 95% confidence intervals.
The 30–100 Hz gamma rhythm has been theorized to play a critical role in hippocampal memory processing (Lisman and Jensen, 2013). Electrophysiological studies have established that theta-gamma coherence is correlated with coordinated information transfer between different sub-regions of the hippocampus (Pernía-Andrade and Jonas, 2014). Moreover, cross-frequency coupling (CFC) has been demonstrated in a memory test experiment that established a correlation between the strength of this coupling and memory performance (Tort et al., 2009). These studies quantified the CFC using a metric termed the modulation index (MI) (Tort et al., 2010), which is calculated by measuring the distribution of the gamma amplitude within specific phases of theta. This is a measure of phase amplitude coupling (PAC) between theta and gamma oscillation. To test the role of gamma oscillations, we applied the MI metric to our recordings (from stratum pyramidale in CA1) and tracked the MI correlation to the efficacy of artificial memory modulation eliciting recall.
Our analysis indicated that the MI was highest between the phase of theta filtered at 4–8 Hz frequency and the amplitude of 55–85 Hz gamma, known as mid-gamma. After establishing the presence of this CFC, we sought to quantify its value during different epochs of the experiment, as well as during different modes of engram reactivation. Comparing the MI at baseline (pre-stimulation epoch) indicated no difference between the four stimulation patterns. However, during the stimulation period, only trough activation showed elevated values in MI, which then went back to control levels during the post-stimulation period (epoch 3, Figure 5B). Comparing pre-stimulation to stimulation epochs for each mouse confirmed that the MI was only significantly modulated in cases using trough stimulation, which was significantly higher than both pre- and post-stimulation. In other stimulation setups, we observed no differences in the MI between the three epochs (Figure 5C).

Theta-gamma cross-frequency coupling correlates with memory recall performance during trough stimulation.
(A) (Ai) Sample LFP recording indicating theta-nested gamma oscillations. (Aii) The modulation index (MI) was calculated for cross-frequency coupling between the mid-gamma (55–85 Hz) amplitude and the phase of theta for trough stimulation during epochs 1–3 corresponding to pre-stimulation, stimulation, and post-stimulation. Comodulograms only showed an increase in the MI during the stimulation epoch. (B) Boxplots show the MI during the three different epochs for the four stimulation modes (20 Hz: green, 6 Hz: yellow, peak: pink, trough: blue) (n=17 animals). (C) Direct comparison of the three epochs for the different stimulation modes indicates that only trough stimulation significantly increased the MI during the stimulation epoch (n=17 animals, paired t-test with Bonferroni correction). Shaded areas represent 95% confidence intervals. (D) Correlation of the stimulation efficacy (% light-induced freezing) and MI was only correlated significantly in the case of trough stimulation, and not during peak stimulation. Note, only mice that showed light-induced freezing were included in the analysis. Shaded areas represent 95% confidence intervals (peak: n=11 animals, trough: n=12 animals, independent t-test).
Consistent with the SPEAR model, our data support the hypothesis that the peak and trough of theta correspond to different modes of hippocampal function with regard to memory processing. In particular, we observed that with trough stimulation the behavioral response was stronger (higher light-induced freezing) and that the MI was higher (Figure 5D, right). This relationship was in the opposite direction, albeit not at a significant level, when using stimulation at the peak of theta (Figure 5D, left). Surprisingly, this result did not hold for slow gamma (Figure 5—figure supplement 1), as we would have expected from the literature related to natural recall (Colgin, 2015; Colgin, 2016; Fernández-Ruiz et al., 2017; Schomburg et al., 2014; Zhang et al., 2019). See the Discussion for more on this point.
Discussion
Using c-fos-dependent neuronal tagging of cells associated with fear conditioning, as well as theta-phase-specific photo-stimulation, we assessed the functional role of theta phase in memory processing. We show that activating DG memory neurons during the trough of the theta field potential is most effective at inducing recall of stored memories and yields the most robust behavioral outcome corresponding to successful artificial reactivation of the tagged memory. When artificial recall is elicited through phase-specific stimulation, the behavioral outcome is well correlated with an increase in phase-amplitude coupling between theta and gamma oscillations, which has been established as an electrophysiological correlate of memory performance (Kragel et al., 2020; Tort et al., 2009).
Previous studies have observed frequency dependence in eliciting behavioral responses across different regions of the hippocampus. For example, Ryan et al., 2015, activated CA1 memory neurons using 4 Hz optogenetic stimulation and observed consistent behavioral responses, while Ohkawa et al., 2015, were able to drive behavior with 20 Hz stimulation in the same region. In the case of DG, the majority of past studies have used 20 Hz stimulation to drive engram activation despite the rate being much higher than natural DG granule cell firing rates in vivo (GoodSmith et al., 2017; Senzai and Buzsáki, 2017).
Overall, our results support the SPEAR model (Hasselmo et al., 2002), as well as a general role for theta oscillations in organizing memory recall in the hippocampus. The effectiveness of trough stimulation in eliciting recall is also consistent with previous experiments in rodents (Douchamps et al., 2013; Manns et al., 2007; Siegle and Wilson, 2014) and human subjects (Kragel et al., 2020; Kerrén et al., 2018) performing memory tasks. However, given the artificial nature of our reactivation protocol, it is unlikely that we are regenerating quasi-natural, circuit-wide activity. For example, to avoid generating seizure-like activity, we stimulated in DG rather than CA3. It seems quite unlikely that our rather crude reactivation protocol replicates the intricate phase relationships of cellular activity, relative to theta, that are seen in careful measurement of cellar inputs and outputs in DG, CA3, and CA1 under natural conditions of recall (Mizuseki et al., 2009; Fernández-Ruiz et al., 2017). Additionally, because we are not measuring theta where we are reactivating, we do not know if our driven DG activity is properly phase-locked with the local theta rhythm in stratum granulosum, which is typically antiphase from theta in CA1 stratum pyramidale (Buzsáki, 2002). Nevertheless, our results align well with predictions of the SPEAR model, and it says something about the robustness of hippocampal functional organization that this artificial drive yields a positive result. Our results support the SPEAR model and move the field closer to more naturalistic manipulations in the brain. Direct measures of neuronal activity in CA1 and CA3 are required to further understand the downstream effects of in phase and out of phase engram activation in DG. Additional future work could investigate methods to induce more naturalistic recall optogenetically by incorporating previously measured temporal sequences of activity with high spatial resolution. Though exceedingly challenging, such experiments might potentially drive more robust behavioral responses and would represent a detailed test of the SPEAR model.
Our controls included open-loop stimulation at both 6 Hz, representing the average frequency of our closed-loop stimulation, and 20 Hz, representing a standard value for such studies. Although open-loop, 6 Hz stimulation was moderately effective in eliciting recall, recall only occurred when the baseline freezing rate was low. It should be noted that this mode of activation is close to the peak physiological firing rates of the DG granule and mossy cells (GoodSmith et al., 2017; Senzai and Buzsáki, 2017). Despite the success of 20 Hz stimulation in artificial recall of the tagged memory in DG (Liu et al., 2012; Ohkawa et al., 2015; Ramirez et al., 2013; Redondo et al., 2014), we did not observe significant light-induced freezing with this frequency of stimulation. A possible explanation for this discrepancy was the presence of an elevated baseline freezing rate post fear conditioning in the neutral context in our study (average 10% [Figure 2] versus less than 5% in previous studies [e.g. Cowansage et al., 2014]). As a result, the elevated baseline freezing could be masking the effect of the 20 Hz stimulation. Moreover, based on the distribution of enhanced and inhibited freezing with 20 Hz stimulation (Figure 3), we hypothesize that this frequency of stimulation could both activate or inhibit engram neurons. In support, work measuring the impact of mossy fiber stimulation on CA3 neuron firing rates (Lee et al., 2019) indicates that 20 Hz stimulation of DG mossy fibers can potentially lead to inhibition of CA3 neurons through feedforward inhibition. For this reason, the potential inhibition caused by 20 Hz stimulation could lead to a lack of stimulation efficacy. In contrast, Lee et al. also showed that 6 Hz stimulation can have a net positive effect on CA3 neuron activity, which is consistent with us observing elevated light-induced freezing during 6 Hz stimulation (Lee et al., 2019). Testing all four different forms of stimulation within a single subject to control for inter-subject variability, however, did not allow us to assess the specific effects of each type of stimulation on the original memory. Future experiments could conduct the same stimulation within one animal to assess the efficacy of certain forms of stimulation in eliciting artificial recall and its impact on the original memory and synaptic plasticity (Chen et al., 2019; Nabavi et al., 2014).
Theta-gamma coupling was found to be modulated only in the trough stimulation during the stimulation epoch relative to the pre- and post-stimulation baselines. In past literature, slow gamma (25–50 Hz) is associated with CA3 inputs to CA1 during recall, though the frequency ranges for each sub-band of gamma vary between publications (Colgin, 2015; Colgin, 2016; Fernández-Ruiz et al., 2017; Schomburg et al., 2014; Zhang et al., 2019). Surprisingly, we did not observe modulation in the slow gamma band. Instead, we saw significant modulation in the mid-gamma band of 55–85 Hz, typically associated with EC input during encoding (Colgin, 2015; Colgin, 2016; Fernández-Ruiz et al., 2017; Schomburg et al., 2014; Zhang et al., 2019). Consistent with prior literature, but not with the recall phase, we found that the mid-gamma modulation arrived at the descending phase of theta (Figure 5—figure supplement 2; Colgin et al., 2009; Fernández-Ruiz et al., 2017). We have two potential hypotheses for this discrepancy. First, the neuronal circuitry responsible for artificial and natural memory reactivation may be distinct, as the Tet-tag system has been shown to primarily label excitatory cells. This difference between natural and artificial memory reactivation could result in different LFP signatures. Second, it is possible that the induction of a fear memory also results in the encoding of the context. Evidence for this hypothesis comes from experiments demonstrating that engram stimulation in DG with opposite valence stimuli can re-associate the tagged cells with the new stimuli (Redondo et al., 2014).
We tried to limit variability of behavioral results by employing a number of exclusion criteria, as described in the Methods section. We tested the data to look for factors that could explain the remaining variability, and found that the MI was significantly correlated with freezing, but only for trough-phase stimulation (Figure 5D). This result suggests to us that the behavioral variability we see has underpinnings in neural processing.
Results presented here open the door for an exciting line of research with regard to a role of theta oscillations in the context of memory processing. Future studies using phase-specific activation of engram neurons will greatly benefit from combining stimulation with calcium imaging (Grienberger and Konnerth, 2012) or high-density electrode arrays (Jun et al., 2017), which could provide single neuron- and network-based mechanisms for hippocampal theta oscillation function during memory processing. Another potential research route for probing the role of theta oscillations in memory gating is through a comparison of encoding and recall engram neurons in the CA3 and EC regions, respectively. For example, we expect tagged CA3 neurons, which are associated with the recall of a fearful memory, to be more robustly reactivated when stimulating at the trough of theta. To our knowledge, however, no studies to date have demonstrated the successful artificial reactivation of memories in CA3. We also hypothesize that EC inputs to the hippocampus can be tagged during the encoding of an experience, with subsequent peak stimulation of the tagged neurons during a second experience disrupting the encoding of that event.
In conclusion, our systematic investigation of engram neuron activation using different modes of stimulation provides new insights regarding the impact of stimulation frequency and phase on engram reactivation, as well as the utility of using a closed-loop photo-stimulation approach. Work to follow should allow the community to refine these early experiments to reproduce behavioral and electrophysiological correlates of normal recall more closely.
Methods
Animals
All procedures were done in accordance with the National Institutes of Health Guide for Laboratory Animals and were approved by the Boston University Institutional Animal Care and Use and Biosafety Committees. We exclusively used adult C57BL/6 wild type male mice (aged 4–8 months). Exclusion of female mice from the study was based on the observation that female mice expressed elevated anxiety relative to male mice, which made assessing fear responses very difficult. Animals were acquired from The Jackson Laboratory.
Sampling and exclusion criteria
No statistical methods were used to determine the sample size; the number of subjects per group was determined based on previously published studies. We used 53 male mice in the current study, with 10 serving as control and 43 as experimental animals. Animals were randomly assigned to the experimental versus control group. Mice were included in the analysis based on pre-defined factors regarding viral expression, effects of the light stimulation, and the quality of LFP recording. If the experimental animals did not show at least 5% increase in their freezing as a result of engram reactivation on 2 out of 4 days of the experiment, they were excluded. Seventeen (17) animals showed lack of effects from the stimulation (Figure 1—figure supplement 1). The exclusion was further confirmed based on virus expression using post hoc manual inspection of brain slices. All mice that exhibited behavioral effects had high levels of expression. Further, since it was crucial for phase-locked stimulation to have an LFP signal with low noise, if the quality of LFP was deemed unsatisfactory to drive reliable phase-locked stimulation, the animal was removed from further analysis. Low-quality LFP affected 1 control animal and 10 experimental animals. The exclusion left 17 experimental animals and 9 control animals for the analysis. Additionally, if the LFP was post hoc located in DG rather than CA1, then the peak and trough data were switched due to the reversal of theta between the regions (Buzsáki, 2002). This change only affected 2 of the 26 mice.
Surgeries
To express the virus, as well as implant the fiber optics and the electrode, animals underwent stereotaxic surgery 3–4 weeks prior to the start of the behavioral experiments. Twenty-four hours prior to the surgery, animals were put on 40 mg/kg doxycycline diet. For the surgery, mice were anesthetized with isoflurane (1.8–2%) vaporized in room air. Bilateral holes were drilled above the dorsal DG at 2.2 mm anterioposterior (AP); ±1.3 mm mediolateral (ML) from Bregma. To express channel rhodopsin in engram neurons, animals were bilaterally injected with AAV9-TRE-ChR2-EYFP acquired from the Massachusetts Institute of Technology, at a depth of 1.8 mm dorsoventral (DV) from the surface. A total of 200 nl of the virus was injected using a 10 nl syringe (World Precision Instruments [WPI]) fitted with a 33-gauge needle (NF33BL; WPI), at a speed of 50 nl/min that was controlled via a microsyringe pump (UltraMicroPump 3–4; WPI). Post injection, one side was implanted with a fiber-optic cannula (200 μm core diameter, 0.39 numerical aperture; Doric Lenses). Following that, the other side was implanted with a costume single LFP electrode (diameter 125 μm, acquired from inVivo1) glued to a fiber optic. Cannulas were implanted at –1.6 mm DV above the injection site. The electrode was targeted to stratum pyramidale layer of CA1 region of the hippocampus. Location of the electrode was validated post hoc using Prussian Blue staining to identify the tip of the electrode and ensure correct phase of theta was measured. Craniotomy was secured with a layer of metabond followed by dental cement. Postop animals received an intraperitoneal injection of the analgesia Buprenorphin (0.2–0.5 mg/kg) which was continued for 48 hr post surgery every 8–12 hr.
Behavioral experiments
After recovery, prior to behavioral experiments, mice were handled daily to habituate them to the transportation and researchers. For the first few days, handling was accompanied with a small treat. Behavioral experiments started after animals were acclimatized to the experimenter, behavioral testing was conducted in a 30.5×24×21 cm3 conditioning chambers (Med Associates). During all the trials, both fiber optics and the LFP electrode were plugged in to normalize the effects of the distress between all trials.
Two contexts were designed for the experiment. The fearful context B contained the bare chamber with metal rods on the bottom, aluminum side walls and a 20 kHz, 40 dB noise source. In the neutral context A, walls and the floor were covered with striped papers along with an ambient white light.
Wooden cage bedding was present on the floor and walls were infused with an orange scent. Animals’ behavior was monitored using a near-infrared camera.
The experiment started with 4 days of habituation for the animals. During which animals were exposed to context A while both fiber optics and the LFP electrode were plugged in. The trial was designed to be the same as the experimental trial. On each day animals stayed in the chamber for four 3 min epochs alternating between no stimulation and stimulation. The habituation days got the animal acclimatized to the experimental environment, with data from the last day used as a baseline for the rest of the experiment.
Mice were taken off doxycycline 48 hr prior to the tagging experiment. To tag engram neurons, animals were introduced to context B, and after 5 min of exploration, the experimental group received four 1.2 mA electrical foot shocks over the next 5 min. Control group animals did not receive any foot shocks. Subjects were put back on the doxycycline immediately after the experiment. In following day, animals were put back in context B, to assess recall of the fearful behavior.
Testing the different stimulation setups was done in context A over 4 days. At the conclusion of the fourth day, mice were re-exposed to context B on day 5, after which mice were being sacrificed and perfused for histological analysis. This last step was performed to assess the long-term effects of artificial reactivations on the original memory.
Reactivation of engram neurons
To test the effectiveness of different stimulation setups, engram neurons for each animal were optogenetically activated in the context A over 4 days. The order of different stimulations was randomized to control for possible effects of repeated stimulation using an equal number of animals for each stimulation at each specific order. Engram reactivation on each day consisted of four 3 min epochs: no stimulation, stimulation, no stimulation, and last epoch of stimulation.
Electrophysiology and optogenetics stimulation
LFP signals were gathered via custom electrodes created by inVivo1. Signals were recorded using a head-stage connected to a Molecular Devices, Axon Instrument amplifier, and digitizer. Digitized signal was recorded at 1 kHz. The closed-loop algorithm implemented in RTXI (Lin et al., 2010), filtered the signal at 2–10 Hz and delivered a TTL pulse to drive a DPSS Single Longitudinal Mode 473 nm laser (optoEngine LLC) at a predicted time for stimulation (peak or trough or fixed frequency).
Behavioral scoring
In order to remove experimenter’s bias from the analysis, an open-source Python package named ezTrack (Pennington et al., 2019) was used to score the freezing during each trial. Parameters for the analysis of motion were manually tuned and kept consistent between trials of the same day in the same animal. For the mice to be considered freezing, it required the animal to remain still for at least 6 s. To ensure accuracy of the algorithm, random blinded trials were hand scored and compared with the automated analysis. Since there was no significant difference between the two, the automated analysis was kept for all trials.
Slice preparation and histology
After completion of experimental manipulations, animals were introduced to context B for the last recall. Ninety minutes after testing, animals were transcardially perfused with cold PBS and tissue fixed using 4% paraformaldehyde (PFA) in PBS with ferrocyanide dissolved in the solution (10%). Brains were extracted and after 24 hr storage in 4% PFA, 50–100 μm slices were prepared. For EYFP staining, slices were incubated at 4°C with PBS and 0.2% Triton and normal goat serum solution for 1 hr, followed by an incubation period with a primary anti GFP chicken antibody 1:1000 (Invitrogen, catalogue # A-10262). After 48 hr of incubation, slices were washed and then incubated with a secondary antibody 1:500, Alexa 488 goat-anti chicken (Invitrogen catalogue # A-11039) for 1 hr. Slices were washed and mounted on a slide with Vectashield (Vector Laboratories H-1000-10) containing a DAPI stain (stain for cell bodies). Olympus FV3000 was used for confocal imaging of slices and a z stack of 10 μm thick slices was acquired.
Statistical analysis
For each of the statistical analyses, both visual inspection of the distribution histogram and Shapiro-Wilk test of normality was performed. Following the determination of normal distribution, Levene’s test for equality of variance was performed to determine appropriate variables in the test. In all the cases, data were normal and paired or independent t-test was applied to test for significance. In cases of multiple comparison, the Bonferroni correction of multiple comparisons was used. All the statistical tests were done in Python, using custom scripts taking advantage of SciPy (Virtanen et al., 2020) package.
LFP analysis
LFP analysis was performed through custom Python scripts. Analysis took advantage of Scipy and a Python implementation (Branlard, 2022) of the Chronux toolbox (Bokil et al., 2010). For the spectral analysis, zero-mean signal was bandpass filtered from 1 to 100 Hz using a fourth-order Butterworth filter. A notch filter was applied to remove the 60 Hz noise (fourth-order Butterworth band-stop filter, with a center frequency of 60 Hz). Spectral content was estimated using a multi-taper method (9 tapers) with a 5 s sliding window, and 1 s overlap.
To identify the actual phase of theta, and determine sensitivity and specificity of the algorithm, post hoc analysis was performed. Data was filtered using fourth-order Butterworth filter (4–10 Hz bandpass) and a Hilbert transform was performed to determine the phase of signal. If the stimulation was within quarter cycle of the peak (0 degree) or trough (180 degree), it was considered on target for respective stimulations, otherwise it was considered off target. Sensitivity was calculated by dividing the number of stimulations by total number of in-phase extrema detected. Specificity was calculated by dividing number of out-of-phase extrema not stimulated by the total number of out-of-phase extrema.
PAC analysis
A metric adopted from a previous study (Tort et al., 2010), termed the MI, was applied to identify the coupling between gamma amplitude and theta phase. This metric is defined through comparing the amplitude distribution through a theta cycle with a uniform distribution. The phase from the Hilbert transform of the LFP filtered at theta range (4–10 Hz) was used as the phase signal. Amplitude of Hilbert transform of the LFP filtered at slow gamma frequency (35–55 Hz) and mid-gamma frequency (55–85 Hz) was used as the amplitude distribution. Data was binned in twenty 18 degree bins, with the amplitudes normalized by the average amplitude of the signal. This normalized distribution (P) was compared to a uniform distribution (U) using Kullback-Leibler (KL) divergence. MI was calculated by dividing the KL divergence by logarithm of the number of bins (N).
Comodulograms, like those in Figure 5, were created by calculating MI between theta and gamma by binning the phase frequencies in bands of 4 Hz, steps of 1 Hz and amplitude frequencies in bands of 20 Hz and steps of 5 Hz.
Data availability
Data collected for the purpose of this paper and the custom algorithms that were used in performing the analysis are available at https://doi.org/10.5061/dryad.k0p2ngfc0. The theta-phase detection algorithm is accessible at https://github.com/ndlBU/phase_specific_stim (copy archived at Noueihed, 2022). It can be run using the RTXI platform accessible through http://rtxi.org. Behavioral scoring was done using the ezTrack package (Penn, 2021).
References
-
BookThe Hippocampus BookOxford University Press.https://doi.org/10.1093/acprof:oso/9780195100273.001.0001
-
Chronux: a platform for analyzing neural signalsJournal of Neuroscience Methods 192:146–151.https://doi.org/10.1016/j.jneumeth.2010.06.020
-
Artificially enhancing and suppressing Hippocampus- mediated memoriesCurrent Biology 29:1885–1894.https://doi.org/10.1016/j.cub.2019.04.065
-
Mechanisms and functions of Theta rhythmsAnnual Review of Neuroscience 36:295–312.https://doi.org/10.1146/annurev-neuro-062012-170330
-
Rhythms of the hippocampal networkNature Reviews. Neuroscience 17:239–249.https://doi.org/10.1038/nrn.2016.21
-
Evidence for Encoding versus retrieval scheduling in the hippocampus by Theta phase and acetylcholineThe Journal of Neuroscience 33:8689–8704.https://doi.org/10.1523/JNEUROSCI.4483-12.2013
-
The timing of external input controls the sign of plasticity at local synapsesNature Neuroscience 12:1219–1221.https://doi.org/10.1038/nn.2388
-
ConferenceReal-time experiment interface for biological control applicationsAnnual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference. pp. 4160–4163.https://doi.org/10.1109/IEMBS.2010.5627397
-
Hippocampal Ca1 Spiking during Encoding and retrieval: relation to Theta phaseNeurobiology of Learning and Memory 87:9–20.https://doi.org/10.1016/j.nlm.2006.05.007
-
SoftwarePhase-specific stimulation, version swh:1:rev:36a0cc21dbfd742298ad78da0f82fa25d1745455Software Heritage.
-
Pattern separation, completion, and Categorisation in the hippocampus and neocortexNeurobiology of Learning and Memory 129:4–28.https://doi.org/10.1016/j.nlm.2015.07.008
-
Function of local circuits in the hippocampal dentate gyrus-Ca3 systemNeuroscience Research 140:43–52.https://doi.org/10.1016/j.neures.2018.11.003
-
The medial temporal lobeAnnual Review of Neuroscience 27:279–306.https://doi.org/10.1146/annurev.neuro.27.070203.144130
-
Measuring phase-amplitude coupling between neuronal Oscillations of different frequenciesJournal of Neurophysiology 104:1195–1210.https://doi.org/10.1152/jn.00106.2010
Decision letter
-
Laura L ColginSenior and Reviewing Editor; University of Texas at Austin, United States
-
Josh SiegleReviewer; Allen Institute, United States
-
Antonio Fernandez-RuizReviewer; Cornell University, United States
Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.
Decision letter after peer review:
Thank you for submitting your article "Theta phase specific modulation of hippocampal memory neurons" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by Laura Colgin as the Senior Editor. The following individuals involved in the review of your submission have agreed to reveal their identity: Josh Siegle (Reviewer #1); Antonio Fernandez-Ruiz (Reviewer #3).
The reviewers have discussed their reviews with one another, and the Senior Editor has drafted this to help you prepare a revised submission.
Essential revisions:
1) Clarify why the dentate gyrus was chosen for stimulation and discuss the implications of recording in CA1 while stimulating in the dentate gyrus.
2) The authors should expand γ analyses and better explain γ methods and results. See individual reviews below for details.
3) Attempt to account for the variability of effects across mice. Specific factors that may have contributed to different effect sizes, and different effects, across different mice are suggested below and should be explored.
4) Make the open-source platform more easily available for use by other labs.
5) Add key references related to separate encoding and retrieval phases of the theta cycle and theta-phase specific inputs to hippocampal regions. See individual reviews below for details.
6) Clarify data exclusion criteria.
7) Improve statistics. Corrections for multiple comparisons are needed in statistical analyses. Also, sample sizes and statistics are missing for some analyses. See individual reviews below for details.
8) The authors should better explain how they accounted for theta phase differences across different electrodes.
9) The authors should discuss how freezing behavior may affect the detection of the theta phase.
Reviewer #1 (Recommendations for the authors):
The number of mice in the experimental group drops from 43 to 17 after applying three exclusion criteria. Based on the description in the methods, it sounds like two of these criteria were not independent (lack of increase in freezing and low viral expression levels). It would be helpful to include the number of mice that were excluded based on only one factor, two factors, or all three factors.
The authors should cite two additional studies that provide correlational evidence for separate encoding and retrieval phases of theta:
Kerrén et al. (2018) Current Biology 28: 3383-3392.e6
Wang et al. (2020) Science 370: 247-250
The combined use of capital letters and roman numerals in the figure labels is unconventional and somewhat confusing. It would be clearer to just use consecutive letters for all panels.
Figure 1
The caption should state the criteria used to determine whether stimulation occurred at the peak or trough (stimulation within 1/4 cycle of the target). Although this is available in the methods section, it should be included here as well, since it's critical for interpreting these results. It would also be helpful to indicate the extent of this range in panel 1Ci.
The labels on Figure 1Cii are hard to interpret – it's not clear what "True" and "False" mean without carefully reading the text.
Figure 2
Please use consistent nomenclature throughout all panels. Panel A uses "Habituation Trials 1-4" and "Engram Reactivation Trials 1-4", which are later referred to as "Day 1-4" and "Day 7-10". In addition, the meaning of "FC", "FS", and "GC" should be defined in the caption, so the figure can stand on its own.
While Figure 3 shows all data points, Figure 2 only includes dots for outliers; ideally, all points should be shown in both figures.
Figure 3
Again, there is inconsistency in the labels. "FS+" and "FS-" are used in the caption, but "E" and "C" are used in the Figure legend (and are not defined).
In, panels Ai and Aii, the sub-panels should be vertically aligned within conditions.
The low-frequency open-loop stimulation condition should be labeled "6 Hz" instead of "06 Hz", to match the text and other figures.
Figure 5
Please add labels to the sub-panels in 5C and 5D-the colors don't appear when printed in black & white.
Reviewer #2 (Recommendations for the authors):
– Why stimulate in dentate instead of CA3 inputs to CA1? The rationale for dentate and its role in the SPEAR hypothesis should be explained in the introduction.
– It is unclear how the authors correct for theta phase differences based on the location of the electrode. The authors state if the recording location is deemed out of the striatum pyramidal then it was "corrected" but the correction method and accuracy are unclear.
– Could the mice see the laser light during stimulation? Did that affect their behavior?
– How many DG cells were labeled? How many were stimulated? Did that have an effect on freezing behavior? Is this study labeling and stimulating similar numbers of neurons and areas of DG as prior studies, especially those that did see a significant increase in freezing with 20Hz stimulation?
– The statistical approach needs revisions. Many statistical comparisons were made in freezing behavior without controlling for multiple comparisons. Multiple comparisons should be taken into account. In some cases the statistical tests used are unclear, e.g. Figure 3Bii, C, and D.
– The figures need clarification and lack key info. First, several figures/captions lack an adequate explanation of colors or acronyms. Some examples:
– Throughout color meaning should be indicated in the caption.
– Line plots throughout do not note errors being shown.
– Statistical details beyond p-values are required: n and the statistical tests used are also required.
– Figure 2Ci – caption is missing or mislabeled as C.
– Figure 2Cii: What are X and Stim on the x-axis? Needs to be explained.
– Figure 2E left graph – you cannot see the color of the smallest bars. They should be labeled with words/letters.
– Figure 3A – define E and C or use FS+ and FS- in the legend. Define x.
– Is each dot with the box plots an animal or a day or something else?
– Figure 3B – what do open vs filled box plots mean?
Second, some figure colors are unclear/confusing:
– Figure 1Ci – is there green (20Hz stim)? If so, I can't tell where.
– Figure 2 – the pink and purple are hard to distinguish. Looking at Figure 2E left graph I could not tell if that was Habituation or Engram Reactivation, eg pink or purple.
– Figure 3 – A schematic showing the stimulation paradigm and interleaved stimulation trials is warranted. I am confused about how long was given between different types of stimulation.
– The authors show that theta trough stimulation of DG cells increases medium γ coupling to theta while other stimulation parameters do not increase. Slow γ is thought to correspond to periods of stronger CA3 input to CA1, while medium γ is thought to correspond to periods of stronger EC input to CA1. The authors should explicitly examine these different gammas separately and their modulation by theta. What phase of theta is each γ coupled to after trough or peak stimulation? Does this shift from baseline coupling? If the authors find trough stimulation increased theta modulation of medium γ, in particular, they should explain how this fits into their hypothesis that this stimulation leads to stronger CA3-CA1 coupling when other works suggest this should result in more slow γ.
– The discussion and Figure 6 caption imply that peak stimulation is disrupting CA1 activity due to the arrival of the CA3 input at the same time as EC inputs. This could be described further to support the behavioral result in 5D. Is there any data to support this point?
– Is there any correlation with freezing behavior in the 20 or 6 Hz groups?
– What was the modulation index of animals that did not show light-induced freezing?
Reviewer #3 (Recommendations for the authors):
The authors tag and manipulate engram cells in the DG but they performed analysis on CA1 LFPs and used the CA1 theta phase as a reference signal. While this is not necessarily a problem per se, it needs to be explicitly addressed and its implication discussed. Likewise, through the text allusions to work on CA1 are used to interpret results from DG manipulations without a clear rationale.
– The abstract should state that engram cells are been tagged and reactivated in the DG. The title will also be more informative if instead of 'hippocampal memory neurons' it said 'dentate gyrus engram neurons'.
– The inputs that impinge DG granular cells during theta are different from those received by CA1 pyramidal cells. This needs to be explained, either in the Introduction or the Discussion. CA1 receives input from CA3 at the descending theta phase and from entorhinal layer 3 at the trough (e.g. Mizuseki et al., Neuron, 2009; Fernandez-Ruiz et al., Neuron, 2017). On the other hand, entorhinal input to DG granular cells originates in layer 2 and arrives at the trough of the CA1 pyramidal layer theta cycle (e.g. Mizuseki et al., Neuron, 2009; Fernandez-Ruiz et al., Science, 2021). It would be also useful to mention that the theta phase completely reverses from the CA1 pyramidal layer to the DG (e.g., Buzsaki, Neuron, 2002).
– If memory encoding and retrieval occur at the same theta phase in CA1 and DG is not yet clear. This needs to be acknowledged as most of the supporting references cited are for CA1. This can actually add additional value to the present paper and previous work (e.g. Siegle et al., eLife, 2014) was done in CA1.
– It needs to be explicitly discussed the potential relationships between stimulating DG cells, the behavioral effects observed, and the fact that LFP analysis was done in CA1. The current discussion on the topic lacks depth, and misses some details (e.g., different phases and layers of origin of entorhinal inputs to CA1 and DG). Excessive speculative statements (e.g., "Activating engrams neurons in DG at the trough of theta leads to in-phase activation of CA3 neurons") should be removed or supporting evidence provided.
Statistics and n's are missing for several analysis
– In figure legends, the n of both animals and sessions that went into each analysis should be clearly stated
– The metric plotted in each case should also be clearly stated. For example, what do the shaded error bars in Figure 2A represent, mean +/- SEM? What do boxplots represent, median +/- CI?
– Figure 4 is missing all statistics and n's (e.g., spectral power comparison across states). That whole figure should be better described. Also, panels A and B are missing voltage and colormap scales respectively.
Theta-γ analysis in Figure 5 has several issues that need to be corrected.
– It needs to be clearly stated from which layer the LFP signals came, as there are strong laminar differences in CA1 γ oscillations (e.g., Schomburg et al., Neuron, 2014; Lasztoczni and Klausberger, Neuron, 2014)
– The following sentence is not accurate: "Our analysis indicated that the MI was highest between the phase of theta filtered at 4-8 Hz frequency and the amplitude of 55-85 Hz γ, known as mid γ. This is similar to previous observations in CA1 (Jiang et al. 2020; Schomburg et al. 2014; Zhang et al. 2019) and consistent with CA3 inputs to CA1 driving recall of memories (Colgin 2015, 2016)". The papers cited, and many others (e.g., Colgin et al., Nature, 2009; Bieri et al., Neuron, 2014; Lasztoczni and Klausberger, Neuron, 2014; Fernandez-Ruiz et al., Neuron, 2017; Lopes dos Santos et al., Neuron, 2018) show that CA3 input to CA1 in dominant is a slower γ band (~30-50 Hz) while entorhinal input is of higher frequency (60-100 Hz), the so-called 'mid γ' sub-band.
– Related to the point above, in the comodulograms of Figure 5Aii I only see the mid-γ component (thus most likely the EC3 to CA1 input) but not the slow γ (the CA3 to CA1 input). The authors can refer to the papers mentioned above to see how at least 3 γ sub-bands are typically found with CFC analysis in CA1. This should be acknowledged in the manuscript. More importantly, the authors should try to separate slow and mid-γ sub-bands and then interpret their results in terms of CA3 and EC3 inputs to CA1. To further verify this separation, theta-phase γ-amplitude analysis can be conducted, as EC3 mid-γ input arrives at the CA1 theta peak and CA3 slow γ input at the descending phase.
[Editors' note: further revisions were suggested prior to acceptance, as described below.]
Thank you for resubmitting your work entitled "Theta-phase-specific modulation of dentate gyrus memory neurons" for further consideration by eLife. Your revised article has been evaluated by Laura Colgin (Senior Editor).
The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined in the individual reviews below:
Reviewer #1 (Recommendations for the authors):
In my initial review, I assumed (along with the authors) that the stronger theta-γ coupling observed during trough stimulation was associated with the enhanced flow of information between CA3 and CA1, as is expected during memory recall. However, the other two reviewers correctly pointed out that the modulation occurred at higher γ frequencies (55 Hz and above), which likely indicates stronger coupling between EC and CA1 (associated with stronger encoding). The authors now acknowledge the apparent contradiction here and have removed the schematic previously shown in Figure 6. Since this undermines a key argument of the original manuscript, they should discuss more possible explanations for their observations, rather than just attributing it to the "somewhat abnormal" effects of their optogenetic intervention. For example, there could actually be stronger coupling theta-slow γ coupling in a different portion or layer of hippocampus, but they've just failed to record it-they mention earlier in the discussion that the sites of recording and reactivation are not spatially aligned. In addition, it's highly plausible that the induction of a fear-related freezing state *does* lead to stronger encoding of the present environment, and hence stronger EC-CA1 interactions. The reactivation of the contextual memory may occur within the first few cycles of stimulation, after which the physiological measurements in the trough stimulation condition are most strongly influenced by differences in behavioral state. It would also be helpful to clarify what is meant by "future imaging experiments should help us understand why our γ modulation experiments differ from those seen in normal behavior."
In their response to reviewers, the authors state that "stringent inclusion criteria were used to ensure that mice had adequate viral expression levels." Although there is a note that the behavior-based exclusion criteria was "further confirmed" based on "post-hoc analysis of brain slice slides," there is no information about how this analysis was conducted. What was the threshold for adequate expression? Were there any mice that showed behavioral effects but had low expression levels?
The frequency range that corresponds to "mid γ" is not stated consistently across the manuscript. The authors should also mention that the specific boundaries (and name) of this frequency band are not universally agreed upon in the literature.
It's fine if the authors want to continue their approach of combining capital letters and roman numerals in the figure panel labels, but I still maintain that most readers would appreciate if they switched to the more commonly used convention of consecutive letters for all panels. Closely related panels can be associated with a single letter, as they already are in Figures 2E and 4B, for example.
In several figures, the "6 Hz" condition is still written as "06 Hz"
Figure 2 – The authors state they have changed days to trial numbers, but this change does not appear in the revised manuscript.
Figure 3 – The authors state they have changed the "E" and "C" labels, but this change does not appear in the revised manuscript.
Figure 5 – Sub-panel Aii actually shows the modulation index for a frequency range from 25-90 Hz. The axis labels for this panel should be changed to "γ frequency" and "theta frequency," since "frequency of amplitude" and "frequency of phase" are not valid terms.
https://doi.org/10.7554/eLife.82697.sa1Author response
Essential revisions:
1) Clarify why the dentate gyrus was chosen for stimulation and discuss the implications of recording in CA1 while stimulating in the dentate gyrus.
We chose to stimulate in DG rather than in CA3 for two, interrelated reasons. First, in most of the historical work on this topic, re-activation has been performed in DG. Repeating this approach allows us to compare our data with those from the bulk of the literature, including recent work from the Ramirez lab involving calcium imaging in CA1 (Zaki et al. 2022, Neuropsychopharmacology). Second, reactivation of memory-encoding cells in CA3 tends to led to seizures that confound results and are unpleasant for the mouse (Ramirez, personal communication), presumably because of the degree of mutual excitation in this structure. As discussed in detail in our response to Reviewer #3, this experimental design has important implications that we did not discuss adequately in the previous version of the manuscript. We have made changes throughout the revised manuscript to clarify these points and avoid overstating our results.
2) The authors should expand γ analyses and better explain γ methods and results. See individual reviews below for details.
Our response to this concern is given below.
3) Attempt to account for the variability of effects across mice. Specific factors that may have contributed to different effect sizes, and different effects, across different mice are suggested below and should be explored.
As described in the methods section, we took steps to attempt to control for animal-to-animal variability. For example, we excluded animals that did freeze in response to light drive of the channelrhodopsin. Additional factors such as baseline freezing both during habituation and during each trial were investigated on an individual basis, but lacked clear trends (data not shown). We did see that the modulation index was correlated with levels of light induced freezing during trough stimulation (Figure 5D), indicating that mice with more successful optogenetic manipulation of the natural brain rhythms responded with larger behavioral changes. This point is made in the revised discussion.
4) Make the open-source platform more easily available for use by other labs.
As noted in the revised manuscript, we have posted the software for real-time phase prediction on Github (https://github.com/ndlBU/phase_specific_stim).
5) Add key references related to separate encoding and retrieval phases of the theta cycle and theta-phase specific inputs to hippocampal regions. See individual reviews below for details.
The following two references were added to capture more of the relevant literature surrounding the theta rhythm and its role in memory.
Kerrén et al. (2018) Current Biology 28: 3383-3392.e6
Wang et al. (2020) Science 370: 247-250
6) Clarify data exclusion criteria.
We have included additional information in the methods section covering how many mice were excluded due to each of the exclusion criteria.
7) Improve statistics. Corrections for multiple comparisons are needed in statistical analyses. Also, sample sizes and statistics are missing for some analyses. See individual reviews below for details.
We have updated the figure captions to include the n values, statistical tests, and corrections applied for each set of experiments. Corrections for multiple comparisons were performed in all cases but were only mentioned in the methods section of the original manuscript. This omission has been rectified.
8) The authors should better explain how they accounted for theta phase differences across different electrodes.
In post-hoc anatomical analysis, we located the tips of the LFP electrodes. We found that the vast majority of electrodes (24/26) were in the correct place, with only 2/26 electrodes being placed too far ventrally, in DG. For these mice, the peak and trough phases were flipped because the phase of the theta oscillation relative to CA1 reverses in DG (Buzsáki 2002). This information has been added to the methods section.
9) The authors should discuss how freezing behavior may affect the detection of the theta phase.
The reviewers raise a good point: a substantial reduction theta power during freezing could interfere with our ability to predict theta phase. Fortunately, this is not the case. We went through all of our data and calculated theta power for both non-freezing and freezing animals. We found that theta power is not affected by freezing in this task (Appendix 1, Figure 4).
Reviewer #1 (Recommendations for the authors):
The number of mice in the experimental group drops from 43 to 17 after applying three exclusion criteria. Based on the description in the methods, it sounds like two of these criteria were not independent (lack of increase in freezing and low viral expression levels). It would be helpful to include the number of mice that were excluded based on only one factor, two factors, or all three factors.
This information has been included in the revised methods section.
The authors should cite two additional studies that provide correlational evidence for separate encoding and retrieval phases of theta:
Kerrén et al. (2018) Current Biology 28: 3383-3392.e6
Wang et al. (2020) Science 370: 247-250
These references have been added.
The combined use of capital letters and roman numerals in the figure labels is unconventional and somewhat confusing. It would be clearer to just use consecutive letters for all panels.
The figure labels were indicted in that manner to group similar experiments and analyses together for the reader. Each figure has sub-themes that are represented by the individual letters, though each of these sub-themes may require multiple presentations of data to convey the full meaning.
Figure 1
The caption should state the criteria used to determine whether stimulation occurred at the peak or trough (stimulation within 1/4 cycle of the target). Although this is available in the methods section, it should be included here as well, since it's critical for interpreting these results. It would also be helpful to indicate the extent of this range in panel 1Ci.
The success criteria for the stimulation (within ¼ cycle of the goal) has been added to the figure caption.
The labels on Figure 1Cii are hard to interpret – it's not clear what "True" and "False" mean without carefully reading the text.
We feel we must use these terms, which are standard in the field, but in the revised legend of Figure 1, we have explained how these entries of the confusion matrix related to sensitivity and specificity.
Figure 2
Please use consistent nomenclature throughout all panels. Panel A uses "Habituation Trials 1-4" and "Engram Reactivation Trials 1-4", which are later referred to as "Day 1-4" and "Day 7-10". In addition, the meaning of "FC", "FS", and "GC" should be defined in the caption, so the figure can stand on its own.
The “days” in Figure 2E have been replaced with engram reactivation trial numbers. Additionally, FS+, FS-, FC, and GC have been defined in the figure caption.
While Figure 3 shows all data points, Figure 2 only includes dots for outliers; ideally, all points should be shown in both figures.
Dots have been included for Figure 2
Figure 3
Again, there is inconsistency in the labels. "FS+" and "FS-" are used in the caption, but "E" and "C" are used in the Figure legend (and are not defined).
E and C have been replaced with FS+ and FS-, respectively.
In, panels Ai and Aii, the sub-panels should be vertically aligned within conditions.
This has been changed.
The low-frequency open-loop stimulation condition should be labeled "6 Hz" instead of "06 Hz", to match the text and other figures.
The title has been renamed to match.
Figure 5
Please add labels to the sub-panels in 5C and 5D-the colors don't appear when printed in black & white.
The labels have been added.
Reviewer #2 (Recommendations for the authors):
– Why stimulate in dentate instead of CA3 inputs to CA1? The rationale for dentate and its role in the SPEAR hypothesis should be explained in the introduction.
We chose to stimulate in DG rather than in CA3 for two, interrelated reasons. First, in most of the historical work on this topic, re-activation has been performed in DG. Repeating this approach allows us to compare our data with those from the bulk of the literature, including recent work from the Ramirez lab involving calcium imaging in CA1 (Zaki et al. 2022, Neuropsychopharmacology). Second, reactivation of memory-encoding cells in CA3 tends to led to seizures that confound results and are unpleasant for the mouse (Ramirez, personal communication), presumably because of the degree of mutual excitation in this structure. We have made this point more clearly in the revised manuscript.
– It is unclear how the authors correct for theta phase differences based on the location of the electrode. The authors state if the recording location is deemed out of the striatum pyramidal then it was "corrected" but the correction method and accuracy are unclear.
Because the phase of theta flips between CA1 and DG (Buzsáki 2002), if the recording electrode was too far ventral, then the Peak and Trough stimulations were reversed relative to the other mice. Thus, we flipped the phases for the 2/26 mice with LFP electrodes in DG. We have updated the methods to explain this correction.
– Could the mice see the laser light during stimulation? Did that affect their behavior?
Yes, the mice could see the light during stimulation. We habituated the animals to the light source prior to expression of the opsin and used control animals to ensure that the animals were not freezing in response to the light. We emphasize this point in the revised Methods section.
– How many DG cells were labeled? How many were stimulated? Did that have an effect on freezing behavior? Is this study labeling and stimulating similar numbers of neurons and areas of DG as prior studies, especially those that did see a significant increase in freezing with 20Hz stimulation?
Although we did not make a quantitative comparison, there were a similar number of DG cells labeled as in previous studies (Figure 2B). Viral expression and stimulation effects were controlled for with the stringent inclusion criteria.
– The statistical approach needs revisions. Many statistical comparisons were made in freezing behavior without controlling for multiple comparisons. Multiple comparisons should be taken into account. In some cases the statistical tests used are unclear, e.g. Figure 3Bii, C, and D.
In the original and revised manuscripts, Bonferroni correction was used in any case that required multiple comparisons, but we failed to mention this in the figure captions. Figure captions have been updated to emphasize this correction and the statistical tests have been added where missing.
– The figures need clarification and lack key info. First, several figures/captions lack an adequate explanation of colors or acronyms. Some examples:
These concerns have all been addressed and the updated manuscript is more clear.
– Throughout color meaning should be indicated in the caption.
The color meaning has been added to all relevant figure captions.
– Line plots throughout do not note errors being shown.
The errors are confidence intervals, and the figure captions have been updated to specify this.
– Statistical details beyond p-values are required: n and the statistical tests used are also required.
Statistics and n values were added to the figure captions.
– Figure 2Ci – caption is missing or mislabeled as C.
The figure caption has been added.
– Figure 2Cii: What are X and Stim on the x-axis? Needs to be explained.
The definition was added to the figure legend.
– Figure 2E left graph – you cannot see the color of the smallest bars. They should be labeled with words/letters.
The labels have been added.
– Figure 3A – define E and C or use FS+ and FS- in the legend. Define x.
The legend has been changed to use FS+ and FS-. Also, x has been defined in the figure caption.
– Is each dot with the box plots an animal or a day or something else?
Each dot is an animal, and figure captions have been updated to include this information.
– Figure 3B – what do open vs filled box plots mean?
The open box plots are FS-, while the filled box plots are FS+. This difference has been added to the figure caption and a legend has been included.
Second, some figure colors are unclear/confusing:
– Figure 1Ci – is there green (20Hz stim)? If so, I can't tell where.
There is green, but it overlaps entirely with the yellow. This information has been added to the figure caption.
– Figure 2 – the pink and purple are hard to distinguish. Looking at Figure 2E left graph I could not tell if that was Habituation or Engram Reactivation, eg pink or purple.
The pink and purple were used to be separate from the colors used later in the manuscript, and to be distinguishable by readers who may be color blind. We’ve added labels to make Figure 2E more clear.
– Figure 3 – A schematic showing the stimulation paradigm and interleaved stimulation trials is warranted. I am confused about how long was given between different types of stimulation.
The schematic in Figure 2 has been updated to be more clear and consistent with terminology, and we have explained the experimental paradigm more clearly. Now, trials are days of reactivation, while epochs are within a single day of stimulation. There were habituation trials, tagging, recall, and reactivation trials. Each mouse was stimulated with one of the frequencies on a single day, with the order of the stimulations across days balanced so that each stimulation type was equally represented across all four days.
– The authors show that theta trough stimulation of DG cells increases medium γ coupling to theta while other stimulation parameters do not increase. Slow γ is thought to correspond to periods of stronger CA3 input to CA1, while medium γ is thought to correspond to periods of stronger EC input to CA1. The authors should explicitly examine these different gammas separately and their modulation by theta. What phase of theta is each γ coupled to after trough or peak stimulation? Does this shift from baseline coupling? If the authors find trough stimulation increased theta modulation of medium γ, in particular, they should explain how this fits into their hypothesis that this stimulation leads to stronger CA3-CA1 coupling when other works suggest this should result in more slow γ.
As we note in the revised discussion, elicited increases in theta-γ coupling are by far strongest for medium γ, in contrast with results for natural recall. Light-stimulated effects on slow γ are not statistically significant (Appendix 1, Figure 2). We believe this anomalous effect is due to the artificial memory reactivation, where an entire group of neurons are activated simultaneously. Additionally, engram neurons expressing channelrhodopsin are typically excitatory cells, which further differentiates this manipulation from natural recall. In future work, when the downstream effects of engram reactivation are understood, a stronger explanation may emerge.
– The discussion and Figure 6 caption imply that peak stimulation is disrupting CA1 activity due to the arrival of the CA3 input at the same time as EC inputs. This could be described further to support the behavioral result in 5D. Is there any data to support this point?
The convergence of simultaneous inputs in CA1 is our working hypothesis of why peak stimulation does not elicit as strong of a behavioral effect as trough stimulation, though we have not investigated the physiological basis for this result. We hope that future work will elucidate the mechanisms underlying the behavior observed. As emphasized in the revised document, the discussed hypothesis is based on the previous literature.
– Is there any correlation with freezing behavior in the 20 or 6 Hz groups?
There was no correlation between freezing behavior and the modulation index in the 20 Hz or 6 Hz groups. Because the peak and trough groups are the only groups with physiological stimulation, we felt it was only necessary to show them in Figure 5D.
– What was the modulation index of animals that did not show light-induced freezing?
Because those animals did not show light induced freezing, they were excluded from all analysis. The lack of light induced freezing could stem from a range of reasons, with some so basic as having poor viral expression to where trying to categorize all of the cases would be tedious and potentially not very fruitful. For that reason, we elected to limit the scope of our analysis to mice showing behavioral effects.
Reviewer #3 (Recommendations for the authors):
The authors tag and manipulate engram cells in the DG but they performed analysis on CA1 LFPs and used the CA1 theta phase as a reference signal. While this is not necessarily a problem per se, it needs to be explicitly addressed and its implication discussed. Likewise, through the text allusions to work on CA1 are used to interpret results from DG manipulations without a clear rationale.
– The abstract should state that engram cells are been tagged and reactivated in the DG. The title will also be more informative if instead of 'hippocampal memory neurons' it said 'dentate gyrus engram neurons'.
We have made it clear in the revised Abstract and title that the tagged and reactivated cells were in DG. Due to shifts in the field over what constitutes an engram, that term was intentionally avoided.
– The inputs that impinge DG granular cells during theta are different from those received by CA1 pyramidal cells. This needs to be explained, either in the Introduction or the Discussion. CA1 receives input from CA3 at the descending theta phase and from entorhinal layer 3 at the trough (e.g. Mizuseki et al., Neuron, 2009; Fernandez-Ruiz et al., Neuron, 2017). On the other hand, entorhinal input to DG granular cells originates in layer 2 and arrives at the trough of the CA1 pyramidal layer theta cycle (e.g. Mizuseki et al., Neuron, 2009; Fernandez-Ruiz et al., Science, 2021). It would be also useful to mention that the theta phase completely reverses from the CA1 pyramidal layer to the DG (e.g., Buzsaki, Neuron, 2002).
– If memory encoding and retrieval occur at the same theta phase in CA1 and DG is not yet clear. This needs to be acknowledged as most of the supporting references cited are for CA1. This can actually add additional value to the present paper and previous work (e.g. Siegle et al., eLife, 2014) was done in CA1.
– It needs to be explicitly discussed the potential relationships between stimulating DG cells, the behavioral effects observed, and the fact that LFP analysis was done in CA1. The current discussion on the topic lacks depth, and misses some details (e.g., different phases and layers of origin of entorhinal inputs to CA1 and DG). Excessive speculative statements (e.g., "Activating engrams neurons in DG at the trough of theta leads to in-phase activation of CA3 neurons") should be removed or supporting evidence provided.
The reviewer raises several excellent points here. By not expressing ourselves clearly enough, we appeared to be overstating our results. In the revised manuscript, we have added the following clarifications.
We have revised the Abstract and the title to make it clear we tagged and reactivated cells in the DG. As explained in this version, we stimulated DG rather than CA3 because optogenetic stimulation of CA3 tends to generate seizures.
As we make more clear, the hypothesized “encoding” and “retrieval” phases of SPEAR model are defined relative to the theta rhythm in CA1 stratum pyramidale. Our goal was to drive CA3 inputs to CA1 using the practical but indirect method of stimulating tagged DG neurons. We did not demonstrate that we were driving CA3 inputs to CA1 at the trough, nor did we recreate natural recall-related activity in DG, CA3, and CA1. Our positive results add a new piece to the puzzle here, as we note in the revised Discussion.
As the reviewer notes, natural activity throughout the hippocampal formation is far more intricate and complex than in our experiments. We have modified the discussion, eliminating the last figure, which we think contributed to the perception that we were making overly bold claims. We have added suggested content to the Discussion about the intricacy of natural, phase-locked activity in the region.
Statistics and n's are missing for several analysis
– In figure legends, the n of both animals and sessions that went into each analysis should be clearly stated
n values and statistical tests have been added to all figure captions.
– The metric plotted in each case should also be clearly stated. For example, what do the shaded error bars in Figure 2A represent, mean +/- SEM? What do boxplots represent, median +/- CI?
The errors have been clarified in all of the figure captions.
– Figure 4 is missing all statistics and n's (e.g., spectral power comparison across states). That whole figure should be better described. Also, panels A and B are missing voltage and colormap scales respectively.
There are not statistics because the data are a representative sample. Scales bars have been added to panels A and B.
Theta-γ analysis in Figure 5 has several issues that need to be corrected.
– It needs to be clearly stated from which layer the LFP signals came, as there are strong laminar differences in CA1 γ oscillations (e.g., Schomburg et al., Neuron, 2014; Lasztoczni and Klausberger, Neuron, 2014)
The recording layer has been restated in the Results section for Figure 5, in addition to being specified in the Results section for Figure 1.
– The following sentence is not accurate: "Our analysis indicated that the MI was highest between the phase of theta filtered at 4-8 Hz frequency and the amplitude of 55-85 Hz γ, known as mid γ. This is similar to previous observations in CA1 (Jiang et al. 2020; Schomburg et al. 2014; Zhang et al. 2019) and consistent with CA3 inputs to CA1 driving recall of memories (Colgin 2015, 2016)". The papers cited, and many others (e.g., Colgin et al., Nature, 2009; Bieri et al., Neuron, 2014; Lasztoczni and Klausberger, Neuron, 2014; Fernandez-Ruiz et al., Neuron, 2017; Lopes dos Santos et al., Neuron, 2018) show that CA3 input to CA1 in dominant is a slower γ band (~30-50 Hz) while entorhinal input is of higher frequency (60-100 Hz), the so-called 'mid γ' sub-band.
This sentence was made in error, apologies. The Discussion section has been updated to include a more thorough review of how the theta-γ coupling results fit in with the prior literature.
– Related to the point above, in the comodulograms of Figure 5Aii I only see the mid-γ component (thus most likely the EC3 to CA1 input) but not the slow γ (the CA3 to CA1 input). The authors can refer to the papers mentioned above to see how at least 3 γ sub-bands are typically found with CFC analysis in CA1. This should be acknowledged in the manuscript. More importantly, the authors should try to separate slow and mid-γ sub-bands and then interpret their results in terms of CA3 and EC3 inputs to CA1. To further verify this separation, theta-phase γ-amplitude analysis can be conducted, as EC3 mid-γ input arrives at the CA1 theta peak and CA3 slow γ input at the descending phase.
This additional analysis has been conducted and included in the supplementary figures. While there is modulation in mid-γ, surprisingly there is minimal modulation in the slow-γ band. We believe this discrepancy is due to the non-physiological stimulation taking place, with only excitatory cells in DG being activated. We find it unsurprising that a non-physiological stimulation would result in shifted hallmarks of network activity downstream relative to non-perturbed hippocampal function. These ideas have been added to the Discussion section in a separate paragraph.
[Editors' note: further revisions were suggested prior to acceptance, as described below.]
The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined in the individual reviews below:
Reviewer #1 (Recommendations for the authors):
In my initial review, I assumed (along with the authors) that the stronger theta-γ coupling observed during trough stimulation was associated with the enhanced flow of information between CA3 and CA1, as is expected during memory recall. However, the other two reviewers correctly pointed out that the modulation occurred at higher γ frequencies (55 Hz and above), which likely indicates stronger coupling between EC and CA1 (associated with stronger encoding). The authors now acknowledge the apparent contradiction here and have removed the schematic previously shown in Figure 6. Since this undermines a key argument of the original manuscript, they should discuss more possible explanations for their observations, rather than just attributing it to the "somewhat abnormal" effects of their optogenetic intervention. For example, there could actually be stronger coupling theta-slow γ coupling in a different portion or layer of hippocampus, but they've just failed to record it-they mention earlier in the discussion that the sites of recording and reactivation are not spatially aligned. In addition, it's highly plausible that the induction of a fear-related freezing state *does* lead to stronger encoding of the present environment, and hence stronger EC-CA1 interactions. The reactivation of the contextual memory may occur within the first few cycles of stimulation, after which the physiological measurements in the trough stimulation condition are most strongly influenced by differences in behavioral state. It would also be helpful to clarify what is meant by "future imaging experiments should help us understand why our γ modulation experiments differ from those seen in normal behavior."
Thank you for the insight. We have updated the manuscript to reflect these ideas. The discrepancy between the mid-γ coupling during induced recall is a surprising result with many possible explanations. As you suggest, it is possible that the induction of a fear memory also results in the encoding of the context, as can be seen with memory re-association (Redondo et al., 2014). Another explanation is that the neuronal circuitry responsible for artificial and natural memory reactivation are distinct, thus resulting in different LFP signatures. In order to differentiate these possibilities, we believe calcium imaging provides more detailed insight into the neuronal mechanisms underlying memory reactivation. These points have been added, clarified, and expanded upon in the revised discussion.
In their response to reviewers, the authors state that "stringent inclusion criteria were used to ensure that mice had adequate viral expression levels." Although there is a note that the behavior-based exclusion criteria was "further confirmed" based on "post-hoc analysis of brain slice slides," there is no information about how this analysis was conducted. What was the threshold for adequate expression? Were there any mice that showed behavioral effects but had low expression levels?
Expression data were inspected manually, with each mouse checked for positive expression. Perhaps because of the stringent nature of the behavioral criteria, no mice exhibited positive behavior without high levels of opsin expression. We have updated the methods section to include this information.
The frequency range that corresponds to "mid γ" is not stated consistently across the manuscript. The authors should also mention that the specific boundaries (and name) of this frequency band are not universally agreed upon in the literature.
Fixed, and the variation in frequency bands has been mentioned in the discussion.
It's fine if the authors want to continue their approach of combining capital letters and roman numerals in the figure panel labels, but I still maintain that most readers would appreciate if they switched to the more commonly used convention of consecutive letters for all panels. Closely related panels can be associated with a single letter, as they already are in Figures 2E and 4B, for example.
Our apologies for missing this. We’ve fixed this problem in Figures 2-5.
In several figures, the "6 Hz" condition is still written as "06 Hz"
Figure 2 – The authors state they have changed days to trial numbers, but this change does not appear in the revised manuscript.
Fixed.
Figure 3 – The authors state they have changed the "E" and "C" labels, but this change does not appear in the revised manuscript.
Fixed.
Figure 5 – Sub-panel Aii actually shows the modulation index for a frequency range from 25-90 Hz. The axis labels for this panel should be changed to "γ frequency" and "theta frequency," since "frequency of amplitude" and "frequency of phase" are not valid terms.
Fixed. Thanks for catching this.
https://doi.org/10.7554/eLife.82697.sa2Article and author information
Author details
Funding
National Institute of Neurological Disorders and Stroke (R01 NS054281)
- John A White
National Institute of Biomedical Imaging and Bioengineering (R01 EB016407)
- John A White
National Institutes of Health (DP5 OD023106-01)
- Steve Ramirez
National Institutes of Health (Transformative R01)
- Steve Ramirez
Ludwig Family Foundation (Research Grant)
- Steve Ramirez
Brain and Behavior Research Foundation (Young Investigator Grant)
- Steve Ramirez
McKnight Foundation (Memory and Cognitive Disorders Award)
- Steve Ramirez
Pew Charitable Trusts (Pew Scholars Program in the Biomedical Science)
- Steve Ramirez
Air Force Office of Scientific Research (FA9550- 21-1-0310)
- Steve Ramirez
Boston University
- Bahar Rahsepar
- Jacob F Norman
- Jad Noueihed
- Steve Ramirez
- John A White
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
Acknowledgements
We acknowledge the use Boston University Biomedical Engineering Department core micro- and nano-imaging facilities. We thank Dr. Michael Hasselmo, Dr. Christopher Harvey, and Dr. David Boas for their feedback and guidance through the development of this research, and Mr. Daniel Carbonero for assistance with supplemental data analysis. This work was supported by NSF NRT: National Science Foundation Research Traineeship Program (NRT): Understanding the Brain (UtB): Neurophotonics DGE-1633516.
Ethics
This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All of the animals were handled according to approved institutional animal care and use committee (IACUC) protocols (PROTO201800599) of Boston University. The protocol was approved by the Committee on the Ethics of Animal Experiments of the University of Minnesota (Permit Number: 27-2956). All surgery was performed under sodium pentobarbital anesthesia, and every effort was made to minimize suffering.
Senior and Reviewing Editor
- Laura L Colgin, University of Texas at Austin, United States
Reviewers
- Josh Siegle, Allen Institute, United States
- Antonio Fernandez-Ruiz, Cornell University, United States
Version history
- Received: August 14, 2022
- Preprint posted: October 28, 2022 (view preprint)
- Accepted: July 3, 2023
- Accepted Manuscript published: July 4, 2023 (version 1)
- Version of Record published: July 21, 2023 (version 2)
Copyright
© 2023, Rahsepar, Norman et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Metrics
-
- 671
- Page views
-
- 140
- Downloads
-
- 1
- Citations
Article citation count generated by polling the highest count across the following sources: PubMed Central, Crossref, Scopus.
Download links
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Further reading
-
- Neuroscience
Consumption of food and water is tightly regulated by the nervous system to maintain internal nutrient homeostasis. Although generally considered independently, interactions between hunger and thirst drives are important to coordinate competing needs. In Drosophila, four neurons called the interoceptive subesophageal zone neurons (ISNs) respond to intrinsic hunger and thirst signals to oppositely regulate sucrose and water ingestion. Here, we investigate the neural circuit downstream of the ISNs to examine how ingestion is regulated based on internal needs. Utilizing the recently available fly brain connectome, we find that the ISNs synapse with a novel cell-type bilateral T-shaped neuron (BiT) that projects to neuroendocrine centers. In vivo neural manipulations revealed that BiT oppositely regulates sugar and water ingestion. Neuroendocrine cells downstream of ISNs include several peptide-releasing and peptide-sensing neurons, including insulin producing cells (IPCs), crustacean cardioactive peptide (CCAP) neurons, and CCHamide-2 receptor isoform RA (CCHa2R-RA) neurons. These neurons contribute differentially to ingestion of sugar and water, with IPCs and CCAP neurons oppositely regulating sugar and water ingestion, and CCHa2R-RA neurons modulating only water ingestion. Thus, the decision to consume sugar or water occurs via regulation of a broad peptidergic network that integrates internal signals of nutritional state to generate nutrient-specific ingestion.
-
- Neuroscience
Complex behaviors depend on the coordinated activity of neural ensembles in interconnected brain areas. The behavioral function of such coordination, often measured as co-fluctuations in neural activity across areas, is poorly understood. One hypothesis is that rapidly varying co-fluctuations may be a signature of moment-by-moment task-relevant influences of one area on another. We tested this possibility for error-corrective adaptation of birdsong, a form of motor learning which has been hypothesized to depend on the top-down influence of a higher-order area, LMAN (lateral magnocellular nucleus of the anterior nidopallium), in shaping moment-by-moment output from a primary motor area, RA (robust nucleus of the arcopallium). In paired recordings of LMAN and RA in singing birds, we discovered a neural signature of a top-down influence of LMAN on RA, quantified as an LMAN-leading co-fluctuation in activity between these areas. During learning, this co-fluctuation strengthened in a premotor temporal window linked to the specific movement, sequential context, and acoustic modification associated with learning. Moreover, transient perturbation of LMAN activity specifically within this premotor window caused rapid occlusion of pitch modifications, consistent with LMAN conveying a temporally localized motor-biasing signal. Combined, our results reveal a dynamic top-down influence of LMAN on RA that varies on the rapid timescale of individual movements and is flexibly linked to contexts associated with learning. This finding indicates that inter-area co-fluctuations can be a signature of dynamic top-down influences that support complex behavior and its adaptation.