PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of elifeeLifeRecent contentAbout eLifeFor authorsSign up for alerts
 
eLife. 2013; 2: e00647.
Published online 2013 June 25. doi:  10.7554/eLife.00647
PMCID: PMC3691575

Rigid firing sequences undermine spatial memory codes in a neurodegenerative mouse model

Howard Eichenbaum, Reviewing editor
Howard Eichenbaum, Boston University, United States;

Abstract

Hippocampal neurons encode spatial memories by firing at specific locations. As the animal traverses a spatial trajectory, individual locations along the trajectory activate these neurons in a unique firing sequence, which yields a memory code representing the trajectory. How this type of memory code is altered in dementia-producing neurodegenerative disorders is unknown. Here we show that in transgenic rTg4510 mice, a model of tauopathies including Alzheimer's disease, hippocampal neurons did not fire at specific locations, yet displayed robust firing sequences as animals run along familiar or novel trajectories. The sequences seen on the trajectories also appeared during free exploration of open spaces. The spatially dissociated firing sequences suggest that hippocampal neurons in the transgenic mice are not primarily driven by external space but by internally generated brain activities. We propose that tau pathology and/or neurodegeneration renders hippocampal circuits overwhelmed by internal information and therefore prevents them from encoding spatial memories.

DOI: http://dx.doi.org/10.7554/eLife.00647.001

Research organism: Mouse

eLife digest

Patients with Alzheimer's disease often forget where they have been or who they have just met. This happens because the neurons in those areas of the brain where memories are processed are dying. Indeed, by the time Alzheimer's disease has been diagnosed, many of the neurons in these regions have already died. The symptoms of Alzheimer's disease are then produced by the remaining neurons. However, the reasons why the remaining neurons cannot make new memories are unknown.

In normal mice the neurons in the hippocampus, a part of the brain that is important for memory, are called ‘place cells’ because they are turned on when the mouse is in a specific place. As a consequence, when the mice moves around, different neurons are turned on one by one, and this sequence of activation is believed to be a memory code that represents the places the animal has travelled.

Cheng and Ji have explored this phenomenon in mice that have been genetically engineered so that their neurons contain structures called ‘tau tangles’ that are thought to be involved in the death of neurons. More importantly, these transgenic mice suffer age-dependent neuron loss in a way that is similarly to people with Alzheimer's disease.

Cheng and Ji implanted tiny sensors into the hippocampus of these mice, and used these sensors to monitor the activity of the remaining hippocampal neurons as the mice moved around while searching for food. They found that the neurons were activated almost everywhere, which indicates that the hippocampal neurons in transgenic mice are no longer working as place cells. However, these neurons were still activated one by one in robust sequences. Moreover, the sequences generated by the transgenic mice were the same in many different surroundings, which suggests that these sequences are not memory codes of the animal's current surroundings. Cheng and Ji propose that the sequences reflect existing memories already stored in the brain, which would suggest that Alzheimer's patients cannot form new memories because the brain is preoccupied by old memories, and thus fails to store the new information that is coming in from the outside world.

DOI: http://dx.doi.org/10.7554/eLife.00647.002

Introduction

Alzheimer's disease (AD) is behaviorally characterized by cognitive declines such as memory loss (Carlesimo and Oscar-Berman, 1992; Salmon and Bondi, 2009), but pathologically defined by β-amyloid plaques and tau pathology including neurofibrillary tangles and tau-mediated neurodegeneration (Wenk, 2003; Ashe and Zahs, 2010). A current challenge is how to bridge between these two very different aspects of AD. Because memory deficits are necessarily produced by the memory-processing neural circuits in vivo (Dickerson and Eichenbaum, 2010), a key step toward this challenge is to understand the functional alterations in the memory circuits, including the hippocampus (HP), of the living brain with ongoing pathogenesis. Both the β-amyloid and tau pathologies are likely involved in the memory loss (Chapman et al., 1999; Yu et al., 2001; Cacucci et al., 2008; Ashe and Zahs, 2010; Palop and Mucke, 2010). In particular, tau pathology and subsequent neurodegeneration play a key role in mediating the memory symptoms seen at a mature stage of AD (Ashe and Zahs, 2010). Currently, it is unknown how the mnemonic functions of HP are altered by tau pathology in vivo.

One well-studied function of HP in vivo is the coding of spatial memories. It is believed that spatial memories in both rodents and humans are encoded by HP place cells that fire at one or a few specific locations of a given space (place fields) (O’Keefe and Dostrovsky, 1971; Wilson and McNaughton, 1993; Burgess and O'Keefe, 2003; Ekstrom et al., 2003). When the animal runs along a novel spatial trajectory, external sensory input at individual locations along the trajectory drives place cells to fire one after another in a unique sequence, which yields a memory code for the trajectory (Wilson and McNaughton, 1993; Harris et al., 2003; Dragoi and Buzsaki, 2006). After the novel experience, the sequence becomes internally established and can be reactivated during later memory recall and/or consolidation with or without partial external input (Buzsaki, 1989; Wilson and McNaughton, 1994; Skaggs and McNaughton, 1996; Nadasdy et al., 1999; Lee and Wilson, 2002; Foster and Wilson, 2006; Diba and Buzsaki, 2007; Ji and Wilson, 2007; Gelbard-Sagiv et al., 2008; Davidson et al., 2009; Karlsson and Frank, 2009; Carr et al., 2011). Therefore, in theory, on any given trajectory HP could either retrieve stored internal sequences, form new ones, or combine both, based on the external input that HP receives from the environment. It has been proposed that the interplay between internal and external inputs is important to memory processing in HP and abnormal interaction may lead to memory interference and/or intrusion (Nakazawa et al., 2002; McHugh et al., 2007; Colgin et al., 2008).

The availability of tauopathy animal models provides an opportunity to study how tau pathology alters HP mnemonic functions in vivo. To this end, we set out to examine the spatial memory code in the transgenic rTg4510 (Tau) mouse. This model was chosen because its well-characterized pathological and behavioral phenotypes mimic many key features of tau pathology in AD (Ramsden et al., 2005; Santacruz et al., 2005; Berger et al., 2007). In Tau mice, the overexpression of a mutated version (P301L) of the human tau gene restricted in the forebrain leads to age-dependent spatial memory deficits, tau neurofibrillary tangles, and neuronal loss. In this study, we recorded neurons in the CA1 region of HP using the tetrode recording technique (Wilson and McNaughton, 1993; Buzsaki, 2004; Ji and Wilson, 2007) from 7 to 9 month old Tau mice and their control littermates while they performed spatial navigation tasks. We found that HP neurons in Tau mice fail to fire at specific spatial locations and yet maintain robust firing sequences, suggesting that non-spatial, internal brain activities dominate the HP circuits and prevent them encoding external space.

Results

We recorded a total of 497 CA1 neurons from eight Tau and six wildtype (WT) control mice, all at 7–9 months old (Table 1). Confirming previous studies (Ramsden et al., 2005; Santacruz et al., 2005), there was massive neurodegeneration in Tau mice at this age and the thickness of the CA1 pyramidal layer in HP of Tau mice was on average only 48% of that of WT mice (Figure 1 and Table 1).

Table 1.
Details of the recorded mice
Figure 1.
Histological confirmation of recording sites and neurodegeneration in Tau mice.

CA1 neurons in Tau mice did not fire at specific locations on a familiar track

First, we asked whether CA1 place cell activities were altered in Tau mice when they ran along a familiar rectangular track (Figure 2A) two sessions a day, 15 min each session. In each session, the animals ran back and forth (two trajectories) and repeated 10–40 laps on each trajectory. We analyzed 87 putative pyramidal neurons from WT mice (WT neurons) and 93 from Tau mice (Tau neurons) that were active on at least one trajectory in at least one session. The median firing rate of Tau neurons during the track running sessions was not significantly different from that of WT neurons (median and [10% 90%] range rate: WT 1.21 [0.43 3.60] Hz, Tau 0.90 [0.41 2.75] Hz, p=0.08, ranksum-test, same below unless specified otherwise).

Figure 2.
Unstable firing locations of Tau neurons led to the loss of overall location-specificity on familiar trajectories.

As expected, WT neurons fired spikes at stable, specific locations in every lap that the animal traversed a trajectory. In contrast, spikes of Tau neurons were not confined to specific locations (Figure 2B). Close visual inspections revealed that Tau neurons’ spikes were still specific to one or a few locations in each individual lap, but the locations were changed from lap to lap (Figure 2B). We described a neuron’s firing profile on a trajectory by a rate curve (the two trajectories were analyzed separately for each neuron), which was the neuron’s firing rate as a function of the position along the trajectory. The rate curve was computed for every lap and then averaged across all laps. Whereas the averaged rate curves of typical WT neurons clearly showed one or a few peaks indicative of well-defined place fields, those of typical Tau neurons did not show such prominent peaks (Figure 2B, ‘bottom curves’). We quantified the location-specificity using spatial information (SI) (Skaggs et al., 1993). SI was computed for each neuron active on each trajectory at two levels, a lap SI to measure the location-specificity at individual lap level and a trajectory SI to measure the overall location-specificity across all laps based on its averaged rate curve. The median lap SI of WT neurons was only slightly (21%) greater than that of Tau neurons (Figure 2C; WT 2.03 [1.00 3.09] bits/spike, N = 263; Tau 1.68 [0.84 2.34] bits/spike, N = 262; p=7.7 × 10−11), but the median trajectory SI of WT neurons was much greater (180%) than that of Tau neurons (Figure 2D; WT: 1.29 [0.55 2.61] bits/spike; Tau: 0.46 [0.15 1.29] bits/spike; p=5.3 × 10−39). We also quantified the lap-to-lap location change by rate-stability, defined as the average correlation coefficient between any two laps’ rate curves. Tau neurons showed much lower rate-stability than WT neurons (Figure 2E; WT: 0.66 [0.30 0.86]; Tau: 0.075 [0.0042 0.34]; p=1.7 × 10−70). These results reveal that Tau neurons lost their overall location-specificity, mainly due to unstable firing locations.

CA1 neurons in Tau mice maintained robust firing sequences on the familiar track

Because WT neurons fired at stable locations, they fired one after another with a stable, position-locked sequence in every lap of a trajectory (Figure 3A). To our surprise, despite their unstable firing locations, Tau neurons still maintained stable firing sequences, that is, they fired with consistent orders across laps (Figure 3A).

Figure 3.
Tau neurons maintained stable firing sequences on familiar trajectories.

To quantitatively demonstrate this, we analyzed the high-order (≥4 neurons) firing sequences (Lee and Wilson, 2002; Ji and Wilson, 2007) of WT/Tau neurons on the familiar track. We first assigned letter identities to those neurons active on a trajectory. We arranged pairs of neurons according to their firing orders identified by their cross-correlations. We then derived a template sequence for the trajectory that agreed with all the ordered pairs. For example, for the five Tau neurons shown in Figure 3A, the cross-correlation between the second (assigned as neuron A) and the fourth (neuron H) had a positive peak at 0.8 s. We ordered them as AH. By ordering all pairs among the five neurons in this manner, we derived a template sequence FABHD for this trajectory. A total of 13 template sequences were derived from WT mice and 8 from Tau mice (Table 2). For a template sequence on a trajectory, we then detected the firing sequences in the spiking patterns of all laps that matched with the template. To evaluate the significance of the detection, we generated 1000 shuffled copies of the spike patterns by randomly swapping neuron identities and then detected the number of sequences in each copy. The number of detected sequences in the real spike patterns was expressed as a Z-score computed from the mean and standard deviation of the shuffling-generated counts. We found a significant number of detected sequences for 9 out of the 13 WT templates and 6 of the 8 Tau templates (p<0.05, Z-test; Table 2). The mean (±SEM) Z-scores for all WT and all Tau templates were similarly high (Figure 3B; WT: 2.5 ± 0.3; Tau: 2.3 ± 0.3; p=0.67, one tailed t-test, same below unless specified otherwise). Both WT and Tau template sequences were detected in almost every lap (Figure 3A). Although it appeared that Tau sequences occurred more frequently than WT sequences in a given lap (Figure 3A), this difference was not significant (Figure 3C; WT: 1.0 ± 0.1 times per lap; Tau: 1.4 ± 0.3 times per lap; p=0.10). These results show that Tau neurons, like WT neurons, formed robust firing sequences.

Table 2.
High-order sequence analysis results for individual template sequences derived on the familiar trajectories

However, there was a major difference between WT and Tau sequences. Whereas WT template sequences predominantly occurred at similar locations across laps, Tau sequences shifted their locations from lap to lap (Figure 3A). The mean location shift of the detected Tau template sequences was significantly greater than that of WT sequences (Figure 3D; WT: 12.7 ± 1.9 cm; Tau: 31.3 ± 6.2 cm; p=0.0035). Therefore, unlike WT sequences, Tau sequences were not anchored to specific spatial locations.

The firing sequences of CA1 neurons in Tau mice seen on the familiar track also appeared during free exploration of a familiar open box

We have shown that CA1 neurons maintained robust firing sequences during track running, even though they did not fire at specific locations. We next asked whether this was true during a very different behavior in a different familiar space. After completing the track running sessions, we kept recording the same CA1 neurons while the mice freely explored a familiar open box (Figure 4A) for one to two sessions, 15 min each session.

Figure 4.
Tau neurons fired with low location-specificity in a familiar open box.

We analyzed 74 WT and 68 Tau neurons that were active in at least one open box session. The median firing rate of these Tau neurons was lower than that of WT neurons (WT 1.3 [0.5 3.8] Hz, Tau 0.9 [0.5 2.7] Hz, p=0.0001), which would predict a greater location-specificity of Tau neurons. However, we observed the opposite. As illustrated by their rate maps (Figure 4B), firing rate color-plotted at each position of the two dimensional (2D) open box, firing activities of Tau neurons covered a much larger portion of the box (meaning less location-specific) than those of WT neurons. The location-specificity in the box was quantified by SI in the 2D open space (open SI) for each neuron active in each open box session. The median open SI of WT neurons was significantly greater than that of Tau neurons (Figure 4C; WT: 0.24 [0.04 0.76], N = 131; Tau: 0.11 [0.04 0.34], N = 119; p=1.9 × 10−7). The result indicates that Tau neurons fired with low location-specificity in the familiar open box.

We then examined whether there were organized firing sequences of multiple neurons during the open box sessions. Because of the random nature of the free exploration behavior, unlike the repeated trajectory running on the track, we did not expect repetitive high-order firing sequences to appear in the open box. However, when we arranged the neurons according to their firing orders in the template sequences derived during track running, it became apparent that Tau neuron sequences, but not WT sequences, during the track running also appeared frequently during the open box sessions (Figure 5A). We quantified this observation by searching through the spike patterns of open box sessions for sequences that matched with the trajectory template sequences. Here we refer ‘a search’ as to matching one open box session to one trajectory template. We found a significant number of sequences in 10 out of the 16 searches in Tau mice, whereas only in 2 of the 24 searches in WT mice (p<0.05, Z-test; Table 2). The mean Z-score of all searches in Tau mice was significantly higher than that in WT mice (Figure 5B; WT: −0.1 ± 0.3; Tau: 1.4 ± 0.4; p=0.001). Therefore, Tau sequences, but not WT sequences, seen on the track trajectories also appeared robustly during the open box sessions.

Figure 5.
Tau, but not WT, sequences seen on the familiar trajectories also appeared in the familiar open box.

Although the animals were allowed to move around freely in the open box, it is possible that some mice ran certain paths repeatedly and resulted in repeatedly occurring firing sequences. To examine this possibility, we plotted the animal’s paths between the start and end times of all detected Tau sequences in a search. These paths appeared to be randomly distributed in the box (Figure 5C), suggesting that the detected sequences were not linked to particular locations or a particular path. To quantitatively demonstrate this, we drew a vector from the start to the end location of each path. We then plotted all the vectors resulted from a search with their starting positions aligned at the origin (Figure 5D). If the paths were not randomly distributed, the vectors should show a preferred direction. However, the vectors of a typical search were randomly orientated (Figure 5D). For each search in Tau mice where a significant number of sequences were found, we computed a Rayleigh p value, which tested a null hypothesis that the vectors had no preferred direction. None of the p values for all the 10 searches was significant (Figure 5E, p≥0.07), indicating that the null hypothesis could not be rejected. Therefore, the detected Tau sequences in open box sessions were not due to repetitive running behavior. Together with the low location-specificity of Tau neurons in the open box sessions (Figure 4), we concluded that Tau sequences seen in the open box were not primarily driven by spatial locations. Even further, these Tau sequences seen in the open box were the same as those during the track running. Therefore, Tau sequences did not distinguish two completely different spaces during two very different behaviors. This is a striking result given the well-known phenomenon of ‘remapping’ of place cell activities between different environments (Muller and Kubie, 1987; Leutgeb et al., 2005; Colgin et al., 2008).

Robust firing sequences of CA1 neurons in Tau mice also appeared in novel environments

We have demonstrated that Tau neurons displayed robust firing sequences on a familiar track and in a familiar open box despite their apparent lack of spatial specificity. To understand whether this peculiar feature of Tau neurons requires spatial experience, we also recorded CA1 neurons while mice explored a novel open box for one to two sessions and then ran back and forth (two trajectories) on a novel track for one to two sessions in a novel room. The findings were similar as those seen in the familiar environments.

On the novel track (Figure 6A), 79 WT neurons and 92 tau neurons were active on at least one of its trajectories. Their median firing rates during the novel track running sessions were not significantly different (WT: 1.4 [0.4 2.7] Hz, Tau: 1.1 [0.4 3.3] Hz, p=0.71). Whereas the firing locations of WT neurons became stabilized quickly on the novel trajectories, those of Tau neurons stayed unstable from lap to lap (Figure 6B). As a result, the median trajectory SI of Tau neurons was much lower than that of WT neurons (WT 0.69 [0.23 1.5] bits/spike, Tau 0.28 [0.086 0.68] bits/spike, p=5.9 ×1 0−31), despite a similar median lap SI (WT 1.2 [0.58 1.8] bits/spike, Tau 1.2 [0.46 1.6] bits/spike, p=0.62). The rate-stability of Tau neurons was much lower than that of WT neurons (WT 0.47 [0.17 0.85], Tau 0.10 [0.03 0.28], p=1.6 × 10−51). In the novel open box (Figure 6C), 74 WT neurons and 61 Tau neurons were active in at least one session. The median firing rates of these WT and Tau neurons during the novel box sessions were not significantly different (WT: 1.1 [0.4 2.9] Hz, Tau: 1.0 [0.5 2.6] Hz, p=0.38). The rate maps of Tau neurons showed much broader firing areas than those of WT neurons (Figure 6D), suggesting a reduced location-specificity in Tau neurons. This was confirmed by the result that the median open SI of Tau neurons was significantly lower than that of WT neurons (WT: 0.28 [0.06 0.71] Hz, Tau: 0.08 [0.02 0.32] Hz, p=1.3 × 10−12). These results show that Tau neurons fired with little location-specificity either on the novel trajectories or in the novel open box.

Figure 6.
Tau neurons formed robust firing sequences on a novel track and in a novel open box, despite their low location-specificity.

Nevertheless, Tau neurons, like WT neurons, fired with robust sequences on the novel trajectories. But unlike WT neuron sequences, Tau sequences on the novel trajectories were not anchored to specific locations (Figure 6E) and same sequences also appeared in the novel open box sessions (Figure 6F). We derived 15 high-order template sequences on the novel trajectories from WT mice and 8 from Tau mice (Table 3). The mean Z-scores of WT and Tau template sequences on the novel trajectories were similarly high (Figure 6G, WT: 2.5 ± 0.2, Tau: 2.4 ± 0.4, p=0.91), but the mean location shift of Tau sequences among laps was greater than that of WT ones (Figure 6H; WT: 9.2 ± 0.9 cm, Tau: 18.5 ± 3.6 cm, p=0.0038). The mean location shift values of both WT and Tau sequences shown here were smaller than those on the familiar track, probably due to the shorter length of the novel track. Finally, Tau template sequences were detected in the novel open box sessions significantly more often than WT templates (Figure 6I, mean Z-scores: WT: 0.5 ± 0.2, N = 30 searches; Tau: 1.5 ± 0.3, N = 12; p=0.0029). Here we emphasize that the novel open box sessions preceded the novel track running. Therefore, Tau sequences seen on the novel trajectories were already preexistent in the prior open box sessions. These findings suggest that the formation of Tau sequences did not primarily depend on spatial experience in an environment.

Table 3.
High-order sequence analysis results for individual template sequences derived on the novel trajectories

External space contributed to the firing activities of CA1 neurons in Tau mice

Does the lack of location-specificity and experience-dependence mean that external space made no contribution to Tau neurons’ firing activities at all? To answer this, we first examined whether the firing locations of Tau neurons were completely random. Visual inspection of the lap-by-lap spike trains of Tau neurons revealed that the firing locations of Tau neurons were often consistent among a few laps on both the familiar (Figure 2B) and novel (Figure 6B) trajectories, suggesting that the locations were not completely random. Indeed, randomly sliding rate curves of individual laps further significantly reduced the median rate-stability of Tau neurons on both familiar (−0.0020 [−0.034 0.042], p=9.3 × 10−49 compared with actual rate-stability) and novel (−0.0020 [−0.032 0.049], p=1.8 × 10−60) trajectories, indicating that spatial locations still modulated Tau neurons’ firing activities.

Second, we examined whether Tau neuron sequences were trajectory-selective. On both the familiar and novel tracks, most animals ran two different trajectories (back and forth between two food wells) in each of the two running sessions. For each template sequence derived on a trajectory in a session, we counted the occurrence of the sequence (in Z-scores) not only on the trajectory itself (Self), but also on the same trajectory (ST) in the other session and on the other trajectory (OT) in the same session. On the familiar track, the mean Z-scores of Self and ST were similar for both WT (Self 2.5 ± 0.3, N = 13 templates; ST 2.3 ± 0.2, N = 13; p=0.53) and Tau (Self 2.3 ± 0.3, N = 8; ST 1.6 ± 0.5, N = 8; p=0.27) sequences, but there was a significant difference between Self and OT for both WT (OT 0.29 ± 0.3, N = 13; p=5.8 × 10−6) and Tau (OT 0.10 ± 0.2, N = 8; p=6.1 × 10−5) sequences (Figure 7A). The result shows that Tau sequences, like WT sequences, occurred consistently on same trajectories across different sessions and distinguished between different trajectories within same sessions. On the novel track, whereas the mean Z-scores of Self and ST remain comparable for both WT (Self 2.5 ± 0.2, N = 15; ST 1.7 ± 0.3, N = 15; p=0.06) and Tau (Self 2.4 ± 0.4, N = 8; ST 2.6 ± 0.6, N = 4; p=0.79) sequences, there was a significant difference between Self and OT only for WT (OT −0.26 ± 0.2, N = 15; p=3.0 × 10−9), but not for Tau (OT 1.9 ± 0.5, N = 8; p=0.51) sequences (Figure 7B), suggesting that Tau sequences, unlike WT sequences, did not distinguish the two trajectories on the novel track. This analysis shows that Tau sequences were originally not trajectory-selective on the novel track, but became so on the familiar track. Therefore, although Tau neuron’s firing activities were not location-specific and Tau sequences were not primarily driven by space, there was still a modulation of Tau neuron’s firing activities by spatial experience and spatial trajectory.

Figure 7.
Tau sequences were trajectory-selective on the familiar track, but not on the novel track.

There was no apparent spatial or temporal periodicity in the occurrence of Tau sequences

Since same Tau sequences sometimes occurred multiple times in a single lap (Figure 3A), we asked whether they tended to appear periodically in space or time. For all the sequences matched with a template in a session, we visualized the spatial/temporal gaps between neighboring sequences by a raster plot and computed the spatial/temporal auto-correlogram of the sequences using their locations/times to examine the periodicity (Figure 8). Neighboring WT sequences in track sessions displayed a consistent spatial gap and there were periodic peaks in their spatial auto-correlograms (Figure 8A, left). The spatial period was the same as the track length (~2 m for the familiar track, ~1 m for the novel track), which was expected, given the repetitively running of same track trajectories. The periodicity was much less obvious in the temporal domain (Figure 8A, right), because the amount of time that animals spent running and at the food sites varied from lap to lap. In contrast, Tau sequences during track sessions did not show any consistent spatial or temporal gaps in their raster plots and no obvious, periodic peaks were observed in either the spatial or temporal auto-correlograms (Figure 8B). The spatial periodicity at the track length disappeared apparently because Tau sequences were not anchored to track locations. The average auto-correlogram across all sessions confirmed this observation on both the familiar and novel tracks (Figure 8C,D). Using the same analysis, Tau sequences in open box sessions did not show apparent spatial or temporal periodicity either (Figure 8B,E). Therefore, there was no strong evidence for a periodic occurrence of Tau sequences.

Figure 8.
Tau sequences did not display obvious periodicity in space or time.

The robust firing sequences in Tau mice could not be explained by simple behavioral parameters

We have shown the existence of robust Tau sequences that were not anchored to specific locations and even did not distinguish different environments. Is it possible that these sequences were produced simply by some peculiar behaviors of Tau mice? Track running and free exploration of open boxes are simple behavioral tasks that both WT and Tau mice were able to perform. Figure 9A,B show the accumulative running paths and linearized individual laps of a WT and Tau mouse on the familiar rectangular track. Figure 9C shows the accumulative running paths of the same mice in the familiar open box. As seen in these raw plots, the overall behavior of the WT and Tau mouse was similar. We then quantified the behavior using the following parameters. The first, most straightforward parameter is the running speed during the running of a track trajectory or during an open box session. In addition, for each trajectory of a track, we measured the animal’s performance by the number of laps on the trajectory and quantified the quality of the running by computing the cumulative travel distance per lap, the number of stops (defined as speed <4 cm/s lasting ≥2 s) per lap, and the stopping duration per lap. These parameters were computed for 54 familiar (WT: 22, Tau: 32) and 46 novel (WT: 20, Tau: 26) trajectories and 27 familiar (WT: 11, Tau: 16) and 25 novel (WT: 12, Tau: 13) open box sessions. The results are shown in Figure 9D–H. On familiar trajectories, Tau mice completed less number of laps and ran with a slower speed than WT mice, but their running quality as measured by stopping and travel distance was similar. On novel trajectories, Tau mice behaved similarly as WT mice or even better on one measure (less number of stops per lap). In open boxes, Tau mice in general ran slightly faster than WT mice, but the difference was significant only in the novel open box. Since robust Tau sequences were found on both the familiar and novel tracks and these sequences appeared in both the familiar and novel open boxes, it is unlikely that the running behavior per se was responsible for the differences between WT and Tau sequences.

Figure 9.
Behavioral quantifications of WT and Tau mice.

Theta oscillations were qualitatively similar between WT and Tau mice

Theta (6–12 Hz) oscillation in the CA1 local field potentials (LFPs) is associated with normal place cell activities (Buzsaki, 2002) and is known to organize multiple place cells into firing sequences at theta time scale (O’Keefe and Recce, 1993; Skaggs et al., 1996; Dragoi and Buzsaki, 2006). We next asked whether theta oscillations were altered in Tau mice. As seen in Figure 10A, the raw and filtered LFP traces of a Tau mouse appeared similar to those of a WT mouse, except that the overall magnitude of the Tau mouse’s LFPs was reduced, which is expected because of the massive neurodegeneration. Power spectral density (PSD) analysis of the raw traces showed a clear peak within the theta band in the LFP PSDs of both mice, but with a reduced amplitude in the Tau mouse’s (Figure 10B, left). However, when the PSD values were normalized by the total LFP power, the peak of the Tau mouse’s PSD was comparable with that of the WT mouse (Figure 10B, right), indicating the presence of prominent theta oscillations in the LFPs of the Tau mouse.

Figure 10.
Theta oscillations were present in Tau mice.

We quantified the peak frequency and the absolute and normalized theta power for each LFP trace of a recording session. These parameters were compared among WT, Tau, and mice of two other control genotypes (+/−, −/+, see ‘Materials and methods’), which showed no obvious tau-mediated neurodegeneration in CA1 (Figure 1 and Table 1). There was a slight reduction in theta peak frequency in Tau mice (~8.5 Hz, compared with ~9.5 Hz in control mice) during familiar track and open box sessions. This difference became even less obvious in novel sessions (Figure 10C). The absolute theta power was apparently reduced in Tau mice from those of control animals (Figure 10D). However, the normalized theta power in Tau mice was significantly lower than that in control mice only during familiar track sessions, but not during open box sessions or novel track sessions (Figure 10E). We also restricted this analysis on LFPs within the start and end times of detected sequences in WT and Tau mice. The observation was similar (Figure 10F–H). Therefore, although LFPs in Tau mice were overall smaller, prominent theta oscillations were still present and qualitatively similar to those of WT mice, suggesting that abnormal theta oscillation is unlikely the cause of the difference between Tau and WT sequences.

There was a small gender difference in Tau mice

We also considered additional factors that may affect Tau neuron firing activities. First, we analyzed whether there was a gender difference in Tau mice. Rather than the sequence-related parameters, we used the rate-stability on track trajectories for this purpose because it allowed the comparison between male and female mice with sufficient samples (Figure 11A,B). Between those neurons recorded from male and those from female WT mice, there was no significant difference in the mean rate-stability either on familiar (male 0.70 [0.33 0.88], female 0.65 [0.29 0.86], p=0.21) or novel (male 0.54 [0.19 0.83], 0.46 [0.16 0.86], p=0.22) trajectories. Between those from male and female Tau mice, the mean rate-stability was similar on novel trajectories (male 0.10 [0.023 0.26], female 0.10 [0.037 0.31], p=0.51). But on the familiar trajectories, the mean rate-stability of the neurons from female Tau mice was significantly greater than that of male Tau mice (male 0.047 [−0.0035 0.14], female 0.13 [0.031 0.42], p=5.5 × 10−16). For both genders and on both the familiar and novel trajectories, the mean rate-stability of WT neurons was always much greater than that of Tau neurons (p≤1.3 × 10−14). These results indicate that the firing stability of CA1 neurons was severely compromised in Tau mice of both genders, but CA1 neurons in female Tau mice did better on familiar trajectories than those in male Tau mice.

Figure 11.
Gender and genotype differences.

Second, we analyzed an additional 163 neurons recorded from four mice with two other control genotypes (+/−, −/+). We found that they were much more stable than Tau neurons and largely similar to WT neurons. On familiar trajectories (Figure 11C), the median rate-stability values of both +/− and −/+ neurons were not different from that of WT ones (+/−: 0.62 [0.35 0.80]; −/+: 0.63 [0.39 0.85]; p≥0.11 compared with WT). Both were significantly greater than that of Tau neurons (p≤1.5 × 10−30). On novel trajectories (Figure 11D), the median rate-stability of −/+ neurons was not different from, but that of +/− was slightly smaller than, that of WT neurons (+/− 0.41 [0.090 0.69], −/+ 0.49 [0.21 0.76], p=0.016 between +/− and WT, p=0.90 between −/+ and WT). Both were significantly greater than the median rate-stability of Tau neurons (p≥1.4 × 10−21).

Discussion

We have examined the HP spatial memory code in a tauopathy mouse model, the transgenic rTg4510 mice. Our data show that CA1 neurons in the transgenic mice do not fire at specific locations, but still form robust firing sequences. These sequences are not anchored to precise locations of spatial trajectories and do not even distinguish between two very different spatial environments. The sequences seen on novel trajectories already exist in a prior session of free exploration in an open box. Therefore, CA1 firing sequences in the transgenic mice become so rigid that they no longer primarily respond to spatial environments or spatial experience, and therefore no longer capable of encoding spatial memories.

The lack of both spatial specificity and experience-dependence indicates that external spatial input in Tau mice is incapable of taking control of Tau neurons. Nevertheless, Tau neurons do not fall into disorganized firing patterns, but fire with robust sequences. If not external space, what could be the driven force that assembles Tau neurons into these sequences? It has been recently shown that internal brain activities, those related to memory recall (Gelbard-Sagiv et al., 2008; Pastalkova et al., 2008) and those encoding time (MacDonald et al., 2011; Eichenbaum, 2013), can produce CA1 firing sequences similar to those shown here. Therefore, here we propose that internal activities underlie the rigid firing sequences in the transgenic mice. On the other hand, although the firing activities of CA1 neurons in the transgenic mice were not location-specific, they are still modulated by external space and spatial experience (Figure 7). The existence of this external modulation together with the rigid firing sequences paint a picture of ‘cued false activation’. When the transgenic mice are placed in a space, instead of forming/retrieving the space's memory code, CA1 neurons are cued to activate those internally driven activity patterns irrelevant to the current space. According to this interpretation, the reason why these mice cannot form new spatial memories is because the HP network is dominated by internal brain activities.

This interpretation implies that there exists a direct competition between external and internal inputs in HP. Therefore, our data provide evidence that abnormal external-internal interaction may contribute to memory deficits in neurodegenerative diseases. Previous studies have shown that CA1 firing sequences can occur during immobility (Foster and Wilson, 2006; Diba and Buzsaki, 2007; Davidson et al., 2009; Karlsson and Frank, 2009; Carr et al., 2011) or when the animal is spatially restricted (Pastalkova et al., 2008; MacDonald et al., 2011). These sequences are believed to be internally generated because external spatial input stays stationary in these situations. In our experiments, the animals’ paths between the start and end times of Tau sequences spanned a distance on the tracks (Figures 3A and 6E) and in the open boxes (Figure 5C), clearly showing that the animals were moving when Tau sequences occurred. The sequences also occurred in accompany with prominent theta oscillations (Figure 10A,H). Therefore, the non-spatially driven firing sequences were found even when animals were actively moving through space. In this case, the external spatial input associated with the movement is overtaken by internal input in the control of CA1 neurons’ firing activities. This further extends the recent findings that pre-existing neuronal sequences may influence future memory codes (Gupta et al., 2010; Dragoi and Tonegawa, 2011).

What changes in the neural circuits of Tau mice could produce the abnormal completion between internal and external input? Internal activities could originate in CA3, an area where neurons can drive each other via its extensive recurrent connections (Lisman, 1999; Nakazawa et al., 2002). External sensory input reaches CA1 likely through the direct pathway via the entorhinal cortex. In Tau mice, there is evidence that tau pathology starts earlier in the cortex including the entorhinal cortex than in HP, and within HP, neuronal loss is more prominent in CA1 than in CA3 (Ramsden et al., 2005). It is likely that the external input to CA1 through the entorhinal cortex is weakened by the pathology, but the internal input from CA3 is relatively spared. In the entorhinal cortex, grid cells fire spikes in a spatially periodic fashion (Hafting et al., 2005). In our data, although Tau sequences sometimes occurred repeatedly in a running lap, there was no evidence for either spatial or temporal periodicity in Tau sequences (Figure 8). Also Tau cells’ firing activities were not tied to specific locations on the tracks (Figure 2B) and did not show obvious periodic patterns in the open boxes (Figure 4B). Therefore, Tau sequences are unlikely resulted from periodic firing of entorhinal grid cells. Other potentially relevant changes in Tau mice could include interneurons and theta oscillations, since they greatly shape the firing activities of HP place cells (O’Keefe and Recce, 1993; Skaggs et al., 1996; Buzsaki, 2002; Dragoi and Buzsaki, 2006). In Tau mice, the CamKII gene that induces tau pathology and neurodegeneration is not expressed in interneurons and thus should mainly affect pyramidal neurons in HP (Mayford et al., 1996; Sik et al., 1998; Ramsden et al., 2005). However, interneurons could still die or alter their properties as adaptive changes to the tau pathology and loss of pyramidal neurons. In our datasets, we only recorded 11 putative interneurons from WT mice and 12 from Tau mice. There was a trend that Tau interneurons fired with a lower rate than WT ones (WT: 24.2 [11.7 46.7] Hz, Tau: 14.8 [8.8 27.2] Hz, p=0.05). This rate change could be one of the adaptive changes in Tau mice to keep the overall activity of remaining HP pyramidal neurons at normal level. Regarding theta oscillations, although the medium septum, the structure important for theta generation in HP, could be affected by tau pathology, based on the known expression pattern of CamKII (Odeh et al., 2011), our data show that prominent theta oscillations are still present in the CA1 LFPs of Tau mice. In particular, the proportion of theta power in the total LFP power is in many cases similar between WT and Tau mice (Figure 10E,H). Therefore, theta oscillation unlikely plays a major role in the generation of abnormal Tau sequences. Based on these considerations, we propose that the internal input from CA3, by winning over the external input via the direct pathway from the entorhinal cortex, is the main driving force for producing the rigid Tau sequences. As such, Tau sequences mainly reflect the internal information such as old memories stored in the CA3 recurrent connections.

Our data reveal a subtle but significant difference between male and female Tau mice: The firing stability of CA1 neurons on familiar trajectories in female is greater than that in male mice. This is somewhat unexpected, given a previous report that female tau mice show higher level of abnormally phosphorylated tau and worse performance in the Morris water maze task (MWM) than male mice (Yue et al., 2011). Other previous studies on another tau-related mouse model have found more extensive β-amyloid, but not tau, pathology (Hirata-Fukae et al., 2008) and more severe impairments on MWM in females than in males (Clinton et al., 2007). One possible explanation for the discrepancy is the age of animals used in our study. The sex difference of Tau mice was previously examined at 5.5 months old (Yue et al., 2011). At much older ages (7–9 months) as in our study, tau pathology becomes much more extensive and possibly saturated and thus the sex difference may become less significant (Ramsden et al., 2005). Another possibility is that tau pathology may not only affect spatial memory representation, but also other factors important for MWM. For example, it has been proposed that behavioral testing produces more stress in females than males and thus greater behavioral deficits under pathological conditions (Clinton et al., 2007). In our experiments, the animals had been handled for at least 1 month (for tetrode adjustment and maze training) before recordings began. The mice in our study might have a reduced level of stress, which exposed the subtle advantage of females in the firing stability of CA1 cells. This advantage in females could be compensated by a stronger stress response to MWM, producing an overall greater impairment on task performance.

What pathological features in the transgenic mice are responsible for the observed rigid firing sequences? Our recordings were conducted on the transgenic mice at 7–9 month old, which is a late, mature stage of tau pathology that displays all typical pathological features including hyperphosphorylated tau, tau tangles, and neurodegeneration (Ramsden et al., 2005; Santacruz et al., 2005). Previous experiments show that stopping the production of the transgenic tau is sufficient to halt the memory deficits at the behavioral level (Ramsden et al., 2005; Santacruz et al., 2005), pointing to a role of the hyperphosphorylated tau protein itself. Another possibility is that neurodegeneration reduces the neuron number in HP and cortex and result in a limited capacity for memory storage. Future studies on tauopathy models at various pathological stages may uncover the exact features of tau pathology and/or neurodegeneration that cause the observed abnormalities. Nevertheless, the current study allows us to identify the functional alterations of CA1 neurons at a mature stage of tau pathology and lay the foundation for probing their pathological causes.

Currently, no animal models can fully recapitulate the pathological changes in human AD patients (Ashe and Zahs, 2010). The limitation in the animal model calls for caution in applying the finding to the human disease. However, tau pathology is a key feature of AD (Wenk, 2003; Ashe and Zahs, 2010) and plays a key role in its behavioral phenotypes (Ashe and Zahs, 2010). This model does mimic key features of human tauopathy including its age-dependent progression (Ramsden et al., 2005; Santacruz et al., 2005). The functional changes at the cellular level in HP identified in this animal model, such as the low location-specificity and rigid firing sequences, possibly also occur in the human brain with similar tau pathological features. Interestingly, our interpretation that internally generated activities obstruct the formation of new memories is consistent with two typical memory symptoms seen in AD patients: the inability to form new memories (Carlesimo and Oscar-Berman, 1992; Salmon and Bondi, 2009) and the intrusion of old memories (Butters et al., 1987; De Anna et al., 2008). This illustrates that studying the functional changes in memory circuits of animal models can generate novel insight into the memory symptoms of human neurodegenerative diseases including AD.

Materials and methods

Animals

The mouse colony was maintained by an activator and a responder mouse lines (Ramsden et al., 2005; Santacruz et al., 2005). The activator, bred in a 129S6 background strain, carried the tTA transgene under the CaMKIIa promoter. The responder, bred in the FVB/N strain, carried a transgene encoding the human four-repeat tau with the P301L mutation (hTauP301L). The F1 offspring of the two mouse lines that carries both the hTauP301L and tTA transgenes (hTau+/tTA+, Tau mice) has been shown to overexpress the human tau in the forebrain and display age-dependent progression of tau pathology and neurodegeneration in the hippocampus (HP) and the cortex (Ramsden et al., 2005; Santacruz et al., 2005). The animals used in this study were Tau mice, their wildtype (hTau−/tTA−, WT) littermates, and littermates with the other two control genotypes, hTau+/tTA− (+/−) and hTau−/tTA+ (−/+), all at 7–9 month old (Table 1). Both male and female animals were used.

Experimental procedure

A hyperdrive containing eight tetrodes was implanted during a surgery onto the skull of each mouse used in this study. Over the course of 2–4 weeks post surgery, tetrodes were slowly moved down to the CA1 pyramidal cell layer of HP. Starting about one week post surgery, the animal was food-deprived, with weight maintained above 85% of the ab libitum level, and trained in a familiar room to run back and forth (two trajectories) on a ~2 m long rectangular track (Figure 2A) for food reward (liquid condensed milk) and to freely explore in a 50 × 30 cm open box (Figure 4A). The animal was trained for at least 3 weeks and 15–30 min in each apparatus each day.

Recording in the familiar room started when multiple, stable single-units (neurons) had been obtained in CA1 and the animal achieved a performance of at least 10 laps per trajectory on the track. The daily recording procedure consisted of two sessions on the track, followed by one to two sessions in the open box. Each session lasted about 15 min and there was a 15-min break between any two sessions, during which the animal rested on an elevated dish (25 cm high, 18 cm in diameter) inside an enclosed box. The recording was repeated 3–10 days.

After completing the recordings in the familiar room, recording continued in a novel, different room while the animal freely explored a novel open box (30 × 50 cm, Figure 6C) and ran back and forth on a novel ~1 m long L-shaped track (Figure 6A). The animal had never been in the novel room or in the novel apparatuses before. The recording schedule was one to two open box sessions, followed by one to two sessions on the L track. Each session lasted about 15 min and there was a 15-min break between any two sessions.

After the recording, the animal was sacrificed using pentobarbital overdose (200 mg/kg, IP injection). Electrical current (30 µA, 6–10 s) was passed through the tetrodes to create small lesions at the recording sites. The animal’s brain was dissected out and fixed in 10% formalin for 24 hr. After its weight and size were measured, the brain was then sectioned for immunohistochemistry.

Surgery

The animal was fixed on a stereotaxic device with body temperature maintained at 36–38 °C and anesthetized with continuous flow of 0.5–2% inhalation anesthetic isoflurane. The flow was adjusted to keep the animal’s breathing rate at 40–80 per min. Ten anchor screws were mounted onto the skull. An exposure was made at the coordinates anteroposterior −2.0 mm, mediolateral 1.5 mm from Bregma. The hyperdrive cannula containing eight tetrodes was lowered to the exposure just above the brain. The drive was fixed in place by dental acrylic. Analgesic (ketoprofen, 5 mg/kg) was injected subcutaneously before the animal recovered from the anesthesia.

In vivo electrophysiological recording

Tetrode recording was made using a DigitaLynx acquisition system (Neuralynx, Bozeman, MT) and followed the published procedures (Ji and Wilson, 2007). For spike recording, voltage signals from four channels of a tetrode were digitally band-pass filtered between 600 Hz and 9 kHz. Spikes were detected with any of the four channels crossing a pre-set triggering threshold (50–70 μV) and were sampled at 32 kHz. Broad band (0.1 Hz–1 kHz) local field potentials (LFPs) were sampled at 2 kHz sampling rate. Two color diodes (red, green) were mounted over the animal's head to track its positions. Positions were sampled at 33 Hz with a resolution approximately 0.2 cm.

Cresyl violet staining

Coronal sections with 20–70 μm thickness were stained using 0.2% Cresyl violet and cover-slipped for storage. Tetrodes were identified (Figure 1), by matching the lesion sites with tetrode depths and their relative positions.

Data analysis

Data recorded from a total of 18 mice with four different genotypes were analyzed (Table 1). For each animal, one day’s data in the familiar room (the day with most neurons recorded) and one day’s data in the novel room (the first day of exposure) were included in the analysis. Neurons were manually sorted using xclust (Matthew A Wilson, MIT). In total, 333 CA1 neurons in the familiar room and 327 in the novel room were obtained. Among them, 246 in the familiar and 221 in the novel room were classified as putative pyramidal neurons active either on a trajectory of a track or in an open box session (mean firing rate: ≥0.5 Hz and <7 Hz). 11 in the familiar and 12 in the novel room were classified as putative interneurons (mean rate ≥ 7 Hz). Further analysis was performed only on the active, putative pyramidal neurons unless otherwise specified. Results are presented as median and [10% 90%] range values with significance determined by the non-parametric ranksum test, or as mean ± SEM values with significance determined by one tailed t-test, unless specified otherwise.

Rate curves, rate maps, and spatial information (SI)

A one-dimensional (1D) trajectory was linearized and divided into 2 cm spatial bins. Similarly, a 2D open box was divided into 2 × 2 cm grids. The rate curve or rate map of a neuron was obtained by computing the neuron’s firing rate at each bin or grid. The rate xi at the i-th bin or i-th grid was the number of spikes divided by the total time the animal spent at the bin/grid (occupancy time, ti). Then, SI was (Skaggs et al., 1993),

equation m1
[1]

where equation inf1 was the occupancy probability, equation inf2 was the mean firing rate, and N was the number of bins/grids. Trajectory SI was given by Equation [1] with xi and ti computed using the neuron’s spikes and the animal’s occupancy time during all laps of a trajectory. For lap SI, we first obtained SI for each single lap according to Equation [1] with the single lap’s rate curve and occupancy times. Lap SI was the mean of all laps’ SI values on a trajectory. SI on the 2D open box (open SI) was similarly given by Equation [1] with the open box session’s rate map and occupancy times.

Rate-stability

Given any two laps of a trajectory, suppose the neuron’s firing rates at the i-th bin are xi and yi. The correlation coefficient between these two laps’ spatial rate curves was defined as

equation m2
[2]

where equation inf3 and σx are the mean and standard deviation of [x1, x2, …, xN] and equation inf4 and σy are the mean and standard deviation of [y1, y2, …, yN], respectively. We computed C values for all combinations of laps and took their mean as the rate-stability.

High-order sequence template construction

Because it was impossible to build a template sequence from the neurons’ place fields on a trajectory as in previous studies (Lee and Wilson, 2002; Ji and Wilson, 2007), we relied on pair-wise cross-correlations (CCs). Given a pair of neurons, CC was computed for each lap on a trajectory. We divided the time period of the lap into 100-ms time bins. If the firing rates of the two neurons were [x1, x2, …, xN] and [y1, y2, …, yN], where N was the number of time bins, the correlation coefficient CT) at time lag ΔT was given by,

equation m3
[3]

The procedure for deriving a template sequence from CCs is the following. We assigned letter identifications to active neurons on a trajectory. For example, if eight neurons were active, we assigned them to A B C D E F G H. Then, we determined whether any pairs of neurons had a stable CC and if so, we used their CC peak times to order the neurons within the pairs (see Figure 3—figure supplement 1 for details). Next, we derived a sequence that agreed with all the ordered pairs. For example, if ordered pairs were found to be AB, FA, BH, BD, HD, we arrived at a sequence FABHD. However, it was not always as straightforward as this, because of the following complications. (1) Sometimes contradictory orders, for example AB, BC, CA occurred, due to the noisy nature of spike trains, especially when multiple neurons fired close in time. (2) There were often many undetermined orders, because not all pairs were found to have a stable CC. This resulted in too much freedom in template sequences and too many different sequences could agree with the ordered pairs to the same degree (as defined below).

Our strategy was to use the following algorithm to derive a template sequence from all ordered pairs of a trajectory:

  1. Determine the number of neurons (k) that participated in the ordered pairs. If k ≤ 10, generate all possible sequences of these neurons. Otherwise, randomly generate 500,000 sequences.
  2. Search through these initial sequences to identify the best sequences, which are those agree with the ordered pairs the most. The degree of agreement is determined by an ordered-pair score s: the percentage of all ordered pairs that match the orders in the sequence.
  3. Check if the best sequences meet the template criteria: First, only one best sequence is resulted or multiple equivalent best sequences co-exist only because some ordered pairs have peaks at 0 s. Second, they need to satisfy either (i) s > 80% and the all-pair score q > 90%, where q is the percentage of all possible pairs in the sequence that match with the ordered pairs, or (ii) the nearest-neighbor score b > 90%, where b is the percentage of all nearest neighbor pairs in the sequence that match with the ordered pairs.
  4. If the template criteria are met, a template sequence is obtained (if multiple equivalent sequences exist, randomly pick one) and the process is terminated. If not, eliminate from the best sequences and the ordered pairs the neuron that had stable relationship with the least number of other neurons. Use the remaining best sequences as initial sequences and repeat from step (2).
  5. If the number of remaining neurons is less than four, the process is terminated and no template is obtained.

Given that there were insufficient numbers of active neurons on some trajectories and that many neurons did not have stable CCs, not all trajectories resulted in a successful template sequence. The template sequences derived and their corresponding trajectories and animals are listed in Tables 2 and 3.

Template sequence matching

We modified our previously published method (Ji and Wilson, 2007) to search for firing sequences that matched with the template on a track trajectory in the spike patterns of all individual laps of the trajectory and in open box sessions (see details in Figure 3—figure supplement 2). Briefly, we transformed a multi-cell spike pattern into a long sequence, according to each neuron’s firing peak times. We then identified segments of the sequence that matched with the template by a rank-order correlation.

To illustrate the detected sequences, we arranged neurons according to their orders in the template and plotted their spike patterns in the spatial (vs positions on a trajectory, e.g., as in Figure 3A) or temporal (vs time, e.g., as in Figure 5A) domain. For illustration purposes only, peak times/positions of the neurons within each detected sequence were linear-regressed with their ranks in the template. The result was plotted as an angled line to highlight the sequence.

The absolute number of detected matches with a template sequence depends on the number of neurons in the template, the parameters that transform spike patterns to sequences, and how strict the rank-order correlation criterion is. Therefore, we expressed the number of detected sequences (n) as a Z-score relative to the chance-level, determined by the following shuffling procedure (Ji and Wilson, 2007). We randomly shuffled the neurons’ identities and generated 1000 copies of shuffled spike patterns. For each copy, we counted the number of sequences matched with the template. We computed the mean (m) and standard deviation (σ) of the counts for all the copies. The Z-score of n is Z = (nm)/σ. The number n was considered significant if Z > 1.645 (corresponding to p<0.05, Z-test). This Z-score normalization was a robust measurement of number of sequences, insensitive to the particular choice of a key parameter in the template generation process (Figure 3—figure supplement 4).

Sequence location shift

For each detected matching sequence, we mapped the animal’s locations on the trajectory between the start and end times of the sequence and computed the center between the start and end locations as the sequence’s center location. To quantify the location shift between any two laps, for each sequence in one lap we computed the distance between its center location and the center location of the nearest sequence in the other lap. If multiple sequences existed in any of the two laps, the distance was computed for every sequence and then averaged as the location shift between the two laps. The location shifts were computed as such for all combinations of laps and their mean value was determined as the location shift for the template sequence on its trajectory.

Sequence auto-correlation

For all sequences identified as matches with a template during a track running and open box session, we computed a temporal and a spatial auto-correlation to examine the periodicity of the sequence occurrence. For each sequence, we computed the time and the cumulative travel distance of the animal at the center of the sequence. We treated the times (distances) of all the sequences in a session as a ‘spike’ train and computed its binned rates [x1, x2, …, xN] with a temporal (spatial) bin size of 2 s (4 cm). The auto-correlation was computed using Equation [3] with two identical vectors (xi = yi, i = 1, 2, … N). The auto-correlogram (auto-correlations plotted against time/space lags) was only computed for those templates or open box sessions that yielded significant number of detected sequences (Z-score > 1.645, p<0.05, Z-test).

Power spectral analysis of LFPs

For each raw LFP trace during a track running and an open box session, we computed its power spectral density (PSD) at every 0.5 Hz by a multitaper method using Matlab. The absolute theta power was the integration of PSD values within the theta band (6–12 Hz). This frequency range was chosen because most PSDs in our data showed a peak centered at 8–10 Hz and spanning this range. The PSD values were normalized by the total power at frequencies >2 Hz. The power at ≤2 Hz was not included in the normalization because it was sensitive to the animal’s movement artifacts. The normalized theta power was the integration of the normalized PSD values within 6–12 Hz. The analysis was also performed on LFPs restricted within the start and end times of identified firing sequences.

Acknowledgements

We thank H-C Lu for helping set up the mouse colony, Y Dabaghian, C Kemere, H-C Lu, H Zheng and Ji lab members for helpful discussions, and R Davis, H Zoghbi and H Penagos for critical readings of the manuscript.

Funding Statement

The funder had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Funding Information

This paper was supported by the following grant:

  • National Institute of Neurological Disorders and Stroke R21NS075726 to Daoyun Ji.

Additional information

Competing interests

The authors declare that no competing interests exist.

Author contributions

JC, Acquisition of data, Analysis and interpretation of data.

DJ, Conception and design, Analysis and interpretation of data, Drafting or revising the article.

Ethics

Animal experimentation: All procedures and experiments described in this study followed the guidelines by the National Institute of Health and were approved by the Institutional Animal Care and Use Committee at Baylor College of Medicine (Animal protocol number: AN-5134).

References

  • Ashe KH, Zahs KR. 2010. Probing the biology of Alzheimer's disease in mice. Neuron 66:631–45. doi: 10.1016/j.neuron.2010.04.031. [PMC free article] [PubMed] [Cross Ref]
  • Berger Z, Roder H, Hanna A, Carlson A, Rangachari V, Yue M, et al. 2007. Accumulation of pathological tau species and memory loss in a conditional model of tauopathy. J Neurosci 27:3650–62. doi: 10.1523/JNEUROSCI.0587-07.2007. [PubMed] [Cross Ref]
  • Burgess N, O’Keefe J. 2003. Neural representations in human spatial memory. Trends Cogn Sci 7:517–9. doi: 10.1016/j.tics.2003.10.014. [PubMed] [Cross Ref]
  • Butters N, Granholm E, Salmon DP, Grant I, Wolfe J. 1987. Episodic and semantic memory: a comparison of amnesic and demented patients. J Clin Exp Neuropsychol 9:479–97. doi: 10.1080/01688638708410764. [PubMed] [Cross Ref]
  • Buzsaki G. 1989. Two-stage model of memory trace formation: a role for “noisy” brain states. Neuroscience 31:551–70. doi: 10.1016/0306-4522(89)90423-5. [PubMed] [Cross Ref]
  • Buzsaki G. 2002. Theta oscillations in the hippocampus. Neuron 33:325–40. doi: 10.1016/S0896-6273(02)00586-X. [PubMed] [Cross Ref]
  • Buzsaki G. 2004. Large-scale recording of neuronal ensembles. Nat Neurosci 7:446–51. doi: 10.1038/nn1233. [PubMed] [Cross Ref]
  • Cacucci F, Yi M, Wills TJ, Chapman P, O’Keefe J. 2008. Place cell firing correlates with memory deficits and amyloid plaque burden in Tg2576 Alzheimer mouse model. Proc Natl Acad Sci USA 105:7863–8. doi: 10.1073/pnas.0802908105. [PubMed] [Cross Ref]
  • Carlesimo GA, Oscar-Berman M. 1992. Memory deficits in Alzheimer's patients: a comprehensive review. Neuropsychol Rev 3:119–69. doi: 10.1007/BF01108841. [PubMed] [Cross Ref]
  • Carr MF, Jadhav SP, Frank LM. 2011. Hippocampal replay in the awake state: a potential substrate for memory consolidation and retrieval. Nat Neurosci 14:147–53. doi: 10.1038/nn.2732. [PMC free article] [PubMed] [Cross Ref]
  • Chapman PF, White GL, Jones MW, Cooper-Blacketer D, Marshall VJ, Irizarry M, et al. 1999. Impaired synaptic plasticity and learning in aged amyloid precursor protein transgenic mice. Nat Neurosci 2:271–6. doi: 10.1038/6374. [PubMed] [Cross Ref]
  • Clinton LK, Billings LM, Green KN, Caccamo A, Ngo J, Oddo S, et al. 2007. Age-dependent sexual dimorphism in cognition and stress response in the 3xTg-AD mice. Neurobiol Dis 28:76–82. doi: 10.1016/j.nbd.2007.06.013. [PMC free article] [PubMed] [Cross Ref]
  • Colgin LL, Moser EI, Moser MB. 2008. Understanding memory through hippocampal remapping. Trends Neurosci 31:469–77. doi: 10.1016/j.tins.2008.06.008. [PubMed] [Cross Ref]
  • Csicsvari J, Hirase H, Mamiya A, Buzsaki G. 2000. Ensemble patterns of hippocampal CA3-CA1 neurons during sharp wave-associated population events. Neuron 28:585–94. doi: 10.1016/S0896-6273(00)00135-5. [PubMed] [Cross Ref]
  • Davidson TJ, Kloosterman F, Wilson MA. 2009. Hippocampal replay of extended experience. Neuron 63:497–507. doi: 10.1016/j.neuron.2009.07.027. [PubMed] [Cross Ref]
  • De Anna F, Attali E, Freynet L, Foubert L, Laurent A, Dubois B, et al. 2008. Intrusions in story recall: when over-learned information interferes with episodic memory recall. Evidence from Alzheimer's disease. Cortex 44:305–11. doi: 10.1016/j.cortex.2006.08.001. [PubMed] [Cross Ref]
  • Diba K, Buzsaki G. 2007. Forward and reverse hippocampal place-cell sequences during ripples. Nat Neurosci 10:1241–2. doi: 10.1038/nn1961. [PMC free article] [PubMed] [Cross Ref]
  • Dickerson BC, Eichenbaum H. 2010. The episodic memory system: neurocircuitry and disorders. Neuropsychopharmacology 35:86–104. doi: 10.1038/npp.2009.126. [PMC free article] [PubMed] [Cross Ref]
  • Dragoi G, Buzsaki G. 2006. Temporal encoding of place sequences by hippocampal cell assemblies. Neuron 50:145–57. doi: 10.1016/j.neuron.2006.02.023. [PubMed] [Cross Ref]
  • Dragoi G, Tonegawa S. 2011. Preplay of future place cell sequences by hippocampal cellular assemblies. Nature 469:397–401. doi: 10.1038/nature09633. [PMC free article] [PubMed] [Cross Ref]
  • Eichenbaum H. 2013. Memory on time. Trends Cogn Sci 17:81–8. doi: 10.1016/j.tics.2012.12.007. [PMC free article] [PubMed] [Cross Ref]
  • Ekstrom AD, Kahana MJ, Caplan JB, Fields TA, Isham EA, Newman EL, et al. 2003. Cellular networks underlying human spatial navigation. Nature 425:184–8. doi: 10.1038/nature01964. [PubMed] [Cross Ref]
  • Foster DJ, Wilson MA. 2006. Reverse replay of behavioural sequences in hippocampal place cells during the awake state. Nature 440:680–3. doi: 10.1038/nature04587. [PubMed] [Cross Ref]
  • Franklin KBJ, Paxinos G. 2008. The mouse brain in stereotaxic coordinates, 3rd Ed Orlando, FL: Academic Press.
  • Gelbard-Sagiv H, Mukamel R, Harel M, Malach R, Fried I. 2008. Internally generated reactivation of single neurons in human hippocampus during free recall. Science 322:96–101. doi: 10.1126/science.1164685. [PMC free article] [PubMed] [Cross Ref]
  • Gupta AS, van der Meer MA, Touretzky DS, Redish AD. 2010. Hippocampal replay is not a simple function of experience. Neuron 65:695–705. doi: 10.1016/j.neuron.2010.01.034. [PubMed] [Cross Ref]
  • Hafting T, Fyhn M, Molden S, Moser MB, Moser EI. 2005. Microstructure of a spatial map in the entorhinal cortex. Nature 436:801–6. doi: 10.1038/nature03721. [PubMed] [Cross Ref]
  • Harris KD, Csicsvari J, Hirase H, Dragoi G, Buzsaki G. 2003. Organization of cell assemblies in the hippocampus. Nature 424:552–6. doi: 10.1038/nature01834. [PubMed] [Cross Ref]
  • Hirata-Fukae C, Li HF, Hoe HS, Gray AJ, Minami SS, Hamada K, et al. 2008. Females exhibit more extensive amyloid, but not tau, pathology in an Alzheimer transgenic model. Brain Res 1216:92–103. doi: 10.1016/j.brainres.2008.03.079. [PubMed] [Cross Ref]
  • Ji D, Wilson MA. 2007. Coordinated memory replay in the visual cortex and hippocampus during sleep. Nat Neurosci 10:100–7. doi: 10.1038/nn1825. [PubMed] [Cross Ref]
  • Karlsson MP, Frank LM. 2009. Awake replay of remote experiences in the hippocampus. Nat Neurosci 12:913–8. doi: 10.1038/nn.2344. [PMC free article] [PubMed] [Cross Ref]
  • Lee AK, Wilson MA. 2002. Memory of sequential experience in the hippocampus during slow wave sleep. Neuron 36:1183–94. doi: 10.1016/S0896-6273(02)01096-6. [PubMed] [Cross Ref]
  • Leutgeb S, Leutgeb JK, Barnes CA, Moser EI, McNaughton BL, Moser MB. 2005. Independent codes for spatial and episodic memory in hippocampal neuronal ensembles. Science 309:619–23. doi: 10.1126/science.1114037. [PubMed] [Cross Ref]
  • Lisman JE. 1999. Relating hippocampal circuitry to function: recall of memory sequences by reciprocal dentate-CA3 interactions. Neuron 22:233–42. doi: 10.1016/S0896-6273(00)81085-5. [PubMed] [Cross Ref]
  • MacDonald CJ, Lepage KQ, Eden UT, Eichenbaum H. 2011. Hippocampal “time cells” bridge the gap in memory for discontiguous events. Neuron 71:737–49. doi: 10.1016/j.neuron.2011.07.012. [PMC free article] [PubMed] [Cross Ref]
  • Mayford M, Bach ME, Huang YY, Wang L, Hawkins RD, Kandel ER. 1996. Control of memory formation through regulated expression of a CaMKII transgene. Science 274:1678–83. doi: 10.1126/science.274.5293.1678. [PubMed] [Cross Ref]
  • McHugh TJ, Jones MW, Quinn JJ, Balthasar N, Coppari R, Elmquist JK, et al. 2007. Dentate gyrus NMDA receptors mediate rapid pattern separation in the hippocampal network. Science 317:94–9. doi: 10.1126/science.1140263. [PubMed] [Cross Ref]
  • Muller RU, Kubie JL. 1987. The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. J Neurosci 7:1951–68. [PubMed]
  • Nadasdy Z, Hirase H, Czurko A, Csicsvari J, Buzsaki G. 1999. Replay and time compression of recurring spike sequences in the hippocampus. J Neurosci 19:9497–507. [PubMed]
  • Nakazawa K, Quirk MC, Chitwood RA, Watanabe M, Yeckel MF, Sun LD, et al. 2002. Requirement for hippocampal CA3 NMDA receptors in associative memory recall. Science 297:211–8. doi: 10.1126/science.1071795. [PMC free article] [PubMed] [Cross Ref]
  • Odeh F, Leergaard TB, Boy J, Schmidt T, Riess O, Bjaalie JG. 2011. Atlas of transgenic Tet-Off Ca2+/calmodulin-dependent protein kinase II and prion protein promoter activity in the mouse brain. NeuroImage 54:2603–11. doi: 10.1016/j.neuroimage.2010.11.032. [PubMed] [Cross Ref]
  • O’Keefe J, Dostrovsky J. 1971. The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Res 34:171–5. doi: 10.1016/0006-8993(71)90358-1. [PubMed] [Cross Ref]
  • O’Keefe J, Recce ML. 1993. Phase relationship between hippocampal place units and the EEG theta rhythm. Hippocampus 3:317–30. doi: 10.1002/hipo.450030307. [PubMed] [Cross Ref]
  • Palop JJ, Mucke L. 2010. Amyloid-beta-induced neuronal dysfunction in Alzheimer's disease: from synapses toward neural networks. Nat Neurosci 13:812–8. doi: 10.1038/nn.2583. [PMC free article] [PubMed] [Cross Ref]
  • Pastalkova E, Itskov V, Amarasingham A, Buzsaki G. 2008. Internally generated cell assembly sequences in the rat hippocampus. Science 321:1322–7. doi: 10.1126/science.1159775. [PMC free article] [PubMed] [Cross Ref]
  • Ramsden M, Kotilinek L, Forster C, Paulson J, McGowan E, SantaCruz K, et al. 2005. Age-dependent neurofibrillary tangle formation, neuron loss, and memory impairment in a mouse model of human tauopathy (P301L). J Neurosci 25:10637–47. doi: 10.1523/JNEUROSCI.3279-05.2005. [PubMed] [Cross Ref]
  • Salmon DP, Bondi MW. 2009. Neuropsychological assessment of dementia. Annu Rev Psychol 60:257–82. doi: 10.1146/annurev.psych.57.102904.190024. [PMC free article] [PubMed] [Cross Ref]
  • Santacruz K, Lewis J, Spires T, Paulson J, Kotilinek L, Ingelsson M, et al. 2005. Tau suppression in a neurodegenerative mouse model improves memory function. Science 309:476–81. doi: 10.1126/science.1113694. [PMC free article] [PubMed] [Cross Ref]
  • Sik A, Hajos N, Gulacsi A, Mody I, Freund TF. 1998. The absence of a major Ca2+ signaling pathway in GABAergic neurons of the hippocampus. Proc Natl Acad Sci USA 95:3245–50. doi: 10.1073/pnas.95.6.3245. [PubMed] [Cross Ref]
  • Skaggs WE, McNaughton BL. 1996. Replay of neuronal firing sequences in rat hippocampus during sleep following spatial experience. Science 271:1870–3. doi: 10.1126/science.271.5257.1870. [PubMed] [Cross Ref]
  • Skaggs WE, McNaughton BL, Gothard KM, Markus EJ. 1993. An information-theoretic approach to deciphering the hippocampal code. In: Hanson SJ, Cowan JD, Giles CJ, editors. , editors. Advances in neural information processing systems. Morgan Kaufmann; pp. 1030–7.
  • Skaggs WE, McNaughton BL, Wilson MA, Barnes CA. 1996. Theta phase precession in hippocampal neuronal populations and the compression of temporal sequences. Hippocampus 6:149–172. doi: 10.1002/(SICI)1098-1063(1996)6:2<>1.0.CO;2-Z. [PubMed] [Cross Ref]
  • Wenk GL. 2003. Neuropathologic changes in Alzheimer's disease. J Clin Psychiatry 64(suppl 9):7–10. [PubMed]
  • Wilson MA, McNaughton BL. 1993. Dynamics of the hippocampal ensemble code for space. Science 261:1055–8. doi: 10.1126/science.8351520. [PubMed] [Cross Ref]
  • Wilson MA, McNaughton BL. 1994. Reactivation of hippocampal ensemble memories during sleep. Science 265:676–9. doi: 10.1126/science.8036517. [PubMed] [Cross Ref]
  • Yu H, Saura CA, Choi SY, Sun LD, Yang X, Handler M, et al. 2001. APP processing and synaptic plasticity in presenilin-1 conditional knockout mice. Neuron 31:713–26. doi: 10.1016/S0896-6273(01)00417-2. [PubMed] [Cross Ref]
  • Yue M, Hanna A, Wilson J, Roder H, Janus C. 2011. Sex difference in pathology and memory decline in rTg4510 mouse model of tauopathy. Neurobiol Aging 32:590–603. doi: 10.1016/j.neurobiolaging.2009.04.006. [PubMed] [Cross Ref]
eLife. 2013; 2: e00647.
Published online 2013 June 25. doi:  10.7554/eLife.00647.022

Decision letter

Howard Eichenbaum, Reviewing editor
Howard Eichenbaum, Boston University, United States;

eLife posts the editorial decision letter and author response on a selection of the published articles (subject to the approval of the authors). An edited version of the letter sent to the authors after peer review is shown, indicating the substantive concerns or comments; minor concerns are not usually shown. Reviewers have the opportunity to discuss the decision before the letter is sent (see review process). Similarly, the author response typically shows only responses to the major concerns raised by the reviewers.

Thank you for sending your work entitled “Rigid firing sequences undermine spatial memory codes in a neurodegenerative mouse model” for consideration at eLife. Your article has been favorably evaluated by a Senior editor and 3 reviewers, one of whom is a member of our Board of Reviewing Editors.

The Reviewing editor and the other reviewers discussed their comments before we reached this decision, and the Reviewing editor has assembled the following comments to help you prepare a revised submission.

This is an outstanding and valuable study that provides new evidence on two major issues: A) the findings show that firing sequences of hippocampal principal cells are independent of the external stimuli (locations and other cues) to which they can become bound with experience; B) and the results highlight that, in an AD mouse model, the sequence structure is retained, but the ability to bind the sequences to behavior and location is diminished. The methods are largely straightforward recording and analysis methods for place cells. The analyses for sequence coding are creative and logical, and nicely controlled for a comparison to random sequences. So, the methods are appropriate, and the contrast between poor spatial coding and strong and rigid sequence coding is compelling.

1) The authors state in the Introduction: “on any given trajectory HP either retrieves stored internal sequences or forms new ones…” Are there data that support this statement? Are these two processes necessarily mutually exclusive, especially on the large scale of a “trajectory”? If not, this should be modified.

2) The data suggest that the rigid sequences in the mutants are not spatially anchored and thus repeat themselves independent of position. Do the current data allow the authors to estimate the spatial period of this repetition? If so, is there a fixed distance across environments and mice?

3) The last section of the Results deals with sex differences found in the mutants (males were worse than females). This is surprising given the 2011 Yao et al papers (Neurobio Aging) that found females had greater memory impairments and increased hyperphosphorylation of Tau compared to males. This discrepancy should be discussed.

4) Is there any evidence of what type of neurons are lost in the mutant’s CA1 (inhibitory vs pyramidal cells)? If so, how does this color your interpretation of the data?

5) The firing sequence templates used in the study are time templates, not spatial templates, detected via an algorithm based on cell-pair cross-correlation. This algorithm is actively searching for the temporally rigid sequences and we have two related concerns. First, the algorithm selects for animals and sessions that have a better representation of temporal sequences and detects 13 templates in 6 WT mice (2.1/animal) and 8 templates in 8 Tau mice (1/animal), twice more in WT. Given that detection of templates has a stage that eliminates neurons that are not reliably cross-correlated with other neurons, we wonder whether this method artificially enhances the detection of rigid sequences by excluding neurons and consequently sessions from the analysis. Second, the neurons of Tau mice seem to fire over more extended periods of time, while animals explore spatial environments which could give more opportunities for short sequences of 4–5 neurons to be detected, particularly given the case that multiple peaks from CA1 neurons are being considered in the template construction. The authors should address these concerns and show that their method does not artificially enhance the detection of sequences in Tau mice.

6) The authors should carefully evaluate/discuss which brain areas are affected in the Tau transgenic mice. Alzheimer's disease affects the cholinergic system and the entorhinal cortex, two areas crucial for theta generation and place field expression. Are entorhinal cortex and medial septum affected in these transgenics? The authors should quantify the theta oscillation in WT, Tau, and additional control animals. What does the EEG activity look like during the expression of the rigid sequences in the Tau mice? Related to this, how is the overall activity of CA1 interneurons in the Tau mice? The authors mention that the rigid sequences could be generated in the CA3. Is the CA3 area spared in the Tau animals?

7) Is there any regularity in the occurrence of the rigid sequences in time?

8) Aside from movement velocity, differences in the overall behavior of the Tau mice could potentially affect the results. Tau animals could be more confused and turn around more, which would preserve temporal sequences but would affect place cell activity. The authors should analyze animals’ trajectories on the linear tracks and show there are no differences between genotypes.

eLife. 2013; 2: e00647.
Published online 2013 June 25. doi:  10.7554/eLife.00647.023

Author response

1) The authors state in the Introduction: “on any given trajectory HP either retrieves stored internal sequences or forms new ones…” Are there data that support this statement? Are these two processes necessarily mutually exclusive, especially on the large scale of a “trajectory”? If not, this should be modified.

This sentence has been re-phrased to reflect that this is a theoretical hypothesis and the two types of sequences could be mixed.

2) The data suggest that the rigid sequences in the mutants are not spatially anchored and thus repeat themselves independent of position. Do the current data allow the authors to estimate the spatial period of this repetition? If so, is there a fixed distance across environments and mice?

We have added an analysis to examine the spatial and temporal periodicity of the detected sequences. The results (Figure 8) do not reveal any periodicity other than the expected periodic occurrence of WT sequences at a spatial period equal to the track length.

3) The last section of the Results deals with sex differences found in the mutants (males were worse than females). This is surprising given the 2011 Yao et al papers (Neurobio Aging) that found females had greater memory impairments and increased hyperphosphorylation of Tau compared to males. This discrepancy should be discussed.

We are grateful for the reference provided (Yue et al., Neurobiol Aging, 2011). We have added the reference and the corresponding paragraph in the Discussion section has been modified. This is a very important reference that provides direct evidence for a sex difference in the transgenic rTg4510 mice. We propose two possible reasons for the discrepancy between this reference and our results. (1) The mice in our study were much older than those in the reference. (2) Tau pathology may affect memory-unrelated factors, such as stress, in female rTg4510 mice more than those in males.

4) Is there any evidence of what type of neurons are lost in the mutant's CA1 (inhibitory vs pyramidal cells)? If so, how does this color your interpretation of the data?

There is evidence that the CamKII gene used to control the transgene expression is not expressed in interneurons. Therefore, although it has not been explicitly studied, the tau pathology and cell death processes are expected to take place only in pyramidal neurons. However, this does not necessarily mean that there would be no changes in interneurons. There could be still adaptive changes. We have computed the firing rates of the limited number of interneurons in our datasets to show that interneurons in Tau mice fired with a lower rate. This type of adaptive change could maintain the overall firing rate of remaining pyramidal neurons at a relatively normal level. The interneuron firing rate result and related discussion are included in the Discussion section, because this is a side result and it is appropriate for the topic at place.

5) The firing sequence templates used in the study are time templates, not spatial templates, detected via an algorithm based on cell-pair cross-correlation. This algorithm is actively searching for the temporally rigid sequences and we have two related concerns. First, the algorithm selects for animals and sessions that have a better representation of temporal sequences and detects 13 templates in 6 WT mice (2.1/animal) and 8 templates in 8 Tau mice (1/animal), twice more in WT. Given that detection of templates has a stage that eliminates neurons that are not reliably cross-correlated with other neurons, we wonder whether this method artificially enhances the detection of rigid sequences by excluding neurons and consequently sessions from the analysis. Second, the neurons of Tau mice seem to fire over more extended periods of time, while animals explore spatial environments which could give more opportunities for short sequences of 4–5 neurons to be detected, particularly given the case that multiple peaks from CA1 neurons are being considered in the template construction. The authors should address these concerns and show that their method does not artificially enhance the detection of sequences in Tau mice.

These are very important concerns. Since we do not know which sequences are “real”, it is hard to measure the “enhancement” of sequence detection. Therefore, we addressed the concerns by analyzing how our sequence detection results depend on (A) the threshold used in the cell-pair selection and (B) the temporal overlap of spikes of multiple cells (especially in Tau mice). The results of this analysis have been presented in Figure 3–figure supplement 4. As the threshold is reduced from a stringent to a low, more relaxed value for cell-pair selection, the number of templates is expectedly increased. However, the mean number of detected sequences (in Z-score) is hardly changed for a wide range of threshold values. Only when the threshold is very low (< 0.08) does the Z-score start to change significantly. Second, to probe the effect of temporal overlap of spiking activities, we binned each spike train in 1-s bins and eliminated spikes within randomly chosen bins. We then computed the number of detected sequence (in Z-score) with the down-sampled spike patterns. The down sampling is to systematically vary the level of spiking overlaps among neurons at a time scale of the sequences detected. We found that the resulted Z-score changes very slowly as more and more spikes are removed. Even with only 40% spikes remaining, the mean Z-scores of WT and Tau mice do not significantly change from the values computed using the original spike patterns. Therefore, although the number of templates and the number of detected sequences inevitably vary with the particular threshold chosen and the specific details in the firing patterns, the normalization (Z-scoring relative to the shuffle-generated number of sequences) provides a robust measure of the sequential structure in the multicell spiking patterns.

6) The authors should carefully evaluate/discuss which brain areas are affected in the Tau transgenic mice. Alzheimer's disease affects the cholinergic system and the entorhinal cortex, two areas crucial for theta generation and place field expression. Are entorhinal cortex and medial septum affected in these transgenics? The authors should quantify the theta oscillation in WT, Tau, and additional control animals. What does the EEG activity look like during the expression of the rigid sequences in the Tau mice? Related to this, how is the overall activity of CA1 interneurons in the Tau mice? The authors mention that the rigid sequences could be generated in the CA3. Is the CA3 area spared in the Tau animals?

(A) We have now added to the Discussion that entorhinal cortex is affected by the pathology more than CA3. The medial septum has not been specifically examined, but should be affected based on the expression pattern of CamKII-tTA (Odeh, et al., Neuroimage, 2011). [Our own histology data could not confirm this without the staining of identified septal cholinergic neurons using antibodies specific to phosphorylated tau.] (B) In the Discussion, we have added that interneurons in Tau mice have lower firing rates than those in WT mice. See the response to point #4. (C) We have now quantified theta oscillations in WT and Tau mice (Figure 10). We found that prominent theta is still present in the local field potentials (LFPs) of Tau mice. Although the absolute LFP power is lower in Tau mice, as expected by the cell loss, the proportion of theta power is comparable between WT and Tau mice. Restricting the analysis to LFPs within rigid sequences returned a similar result. (D) Based on the result of EEG analysis and the structural differences in the pathology, we propose that CA3 plays a major role in driving the rigid sequences in Tau mice.

7) Is there any regularity in the occurrence of the rigid sequences in time?

We have examined the temporal periodicity of Tau sequences and found no sign of periodic occurrence of Tau sequence in time (Figure 8). See also the response to point #2.

8) Aside from movement velocity, differences in the overall behavior of the Tau mice could potentially affect the results. Tau animals could be more confused and turn around more, which would preserve temporal sequences but would affect place cell activity. The authors should analyze animals’ trajectories on the linear tracks and show there are no differences between genotypes.

Besides number of laps completed per session and running speed, we have added several additional parameters to quantify the animals’ trajectories: the total travel distance per lap (which captures the turning around behavior), the number of stops per lap, and the mean duration of stopping per lap. The results, as well as example raw running trajectories, are plotted in Figure 9. The total travel distance and stopping duration were found to be similar between WT and Tau mice.


Articles from eLife are provided here courtesy of eLife Sciences Publications, Ltd