|Home | About | Journals | Submit | Contact Us | Français|
The integrity of the hippocampus is critical for both spatial navigation and episodic memory but how its neuronal firing patterns underlie those functions is not well understood. In particular the modality by which hippocampal place cells contribute to spatial memory is debated. We found that administration of the cannabinoid receptor agonist CP55940 induced a profound and reversible behavioral deficit in the hippocampus-dependent delayed spatial alternation task. On the one hand, despite severe memory impairment, the location-dependent firing of CA1 hippocampal place cells remained largely intact. On the other hand, both spike-timing coordination between place cells at the theta timescale and theta phase precession of spikes were reversibly reduced. These results raise the possibility that cannabinoids impair memory primarily by altering short-term temporal dynamics of hippocampal neurons. We hypothesize that precise temporal coordination of hippocampal neurons is necessary for guiding behavior in spatial memory tasks.
The hippocampus supports both spatial navigation and episodic memory but how these functions relate to the firing patterns of hippocampal neurons is not well understood (Eichenbaum et al., 1999; Buzsáki, 2005; O’Keefe and Burgess, 2005; Mizumori, 2006). On the one hand, overwhelming evidence supports the early observation (O’Keefe and Dostrovsky, 1971) that hippocampal pyramidal cells fire in particular locations of the environment (Muller, 1996) and movement trajectories can be predicted from the population activity of these ‘place cells’ (Wilson and McNaughton, 1993). Such unit recording studies, combined with lesion experiments, are consistent with the cognitive map theory in which the collective activity of hippocampal place cells reflects an internal representation of the environment, which can be used by the rest of the brain to assist spatial navigation (O’Keefe and Nadel, 1978). On the other hand, recent studies have challenged the view that hippocampal neurons carry only spatial information by reporting that firing patterns of hippocampal neurons can be modulated by non-spatial events (Markus et al., 1995; Hampson et al., 1999; Wood et al., 1999, 2000; Frank et al., 2000; Ferbinteanu and Shapiro, 2003; Pastalkova et al., 2008). These findings have been proposed to provide coding mechanisms underlying the hippocampal contribution to episodic memory (Eichenbaum, 2004; Squire et al., 2004; Hasselmo, 2005; Shapiro et al., 2006; Mizumori, 2006). In addition to discharge rate-based coding, the hippocampus exhibits a clear example of temporal coding characterized in the reliable relationship between the position of the rat and the phase of place cell spikes relative to hippocampal theta oscillations (O’Keefe and Recce, 1993). From the apparent independence of rate and phases of place cells spikes (Hirase et al., 1999; Huxter et al., 2003) it has been proposed that the navigation and memory functions of the hippocampus were supported by theta spike-timing and firing rate mechanisms, respectively (O’Keefe et al., 2005).
The postulated existence of different coding mechanisms underlying hippocampal functions raises the question of their respective behavioral relevance. To examine the relationship between memory, spatial representation and firing patterns of hippocampal neurons, we analyzed place cell activity recorded from the CA1 region in rats engaged in a hippocampus-dependent spatial memory task (delayed alternation in a modified T- Maze), and compared them to firing patterns obtained during a behavioral impairment induced by systemic injection of the synthetic cannabinoid CP55940. This pharmacological tool was chosen because cannabinoids impair hippocampus-dependent memories in humans (Miller and Branconnier, 1983; Curran et al., 2002; Ilan et al., 2004) and rats (Lichtman and Martin, 1996; Lichtman et al., 1995; Hampson and Deadwyler, 1998, 2000) and interfere with the synchrony of hippocampal network activity (Constoe et al., 1975; Hájos et al., 2000; Robbe et al., 2006, Hajós et al., 2008). We found that while CP55940 impaired the rat’s performance in the spatial memory task, the location-dependent firing of hippocampal neurons was largely preserved. In contrast, temporal aspects of firing patterns at the theta timescale were altered under the drug’s influence, proportionally with the magnitude of the behavioral alteration. These results raise the possibility that impairment of cell assembly coordination underlies the observed memory deficit.
All protocols were approved by the Institutional Animal Care and Use Committee of Rutgers University. Full methods are available online as Supplementary Detailed Methods
Five Long Evans rats (male, 300–400 g) were water-deprived and trained to perform a hippocampus-dependent delayed T-maze alternation task (Fig. 1) (Ainge et al., 2007; Pastalkova et al., 2008). Once animals performed the task very well (>85%) for several consecutive sessions, access to water was provided for a full day before surgery.
Following recovery from surgical implantation of tetrode array or silicon probe, re-establishment of accurate/stable performance of the alternation task and recording of CA1 ripple local field potential (LFP) oscillations and large amplitude units in the home cage, recordings were performed in the maze before and after a systemic injection of a synthetic cannabinoid (CP55940, potent cannabinoid receptor agonist, 0.1 mg/kg body weight in a vehicle composed of 5% ethanol, 5% Cremophor® (a derivative of castor oil and ethylene oxide, Sigma) and 90% Saline). Vehicle injection had no effect either on behavior performance or on neuronal activity (Supplementary Fig. 10)
To compute firing maps, the space visited by the animal in the maze was divided into a grid of 150×150 pixels (1 pixel = 1 sq.cm) and in each pixel the number of spikes fired was divided by the amount of time an animal spent in that pixel. Multiple successive sessions recorded in a given condition (Control, Drug or Recovery) were concatenated to generate a single firing map per condition. The resulting firing maps were smooth with a Gaussian kernel of horizontal and vertical standard deviation of 2.4 pixels. Place fields were automatically detected (see Detailed Methods).
The spatial information content was calculated according to the following equation: where i is the pixel number, Pi is the probability of occupancy of pixel i, Ri is the mean firing rate in pixel i, and R is the overall mean firing rate (Skaggs et al., 1993)
For each place cell, the pixel-by-pixel correlation coefficient between Control and Drug firing rate maps was calculated. Only pixels in which the rats spent at least 10 msec in both Control and Drug sessions were used to generate the correlation coefficients. This correlation coefficient was compared to the pixel-by-pixel correlation between firing maps generated from the first and second halves of the Control recording. A similar procedure was performed between Recovery and Drug firing maps.
Instantaneous theta phase was derived from the Hilbert transform of the filtered (4–12 Hz) LFP. To estimate the strength of phase precession of place fields, the phases of the spikes were systematically rotated by steps of 1 degree. For each of the 360 rotations, the correlation between spikes phase and position was quantified using a Spearman rank correlation. The correlation between phase and position for the rotation that gave the regression line with the most negative coefficient was taken as indicator of phase precession.
To calculate the theta timescale lag between the spikes of two overlapping place fields, cross-correlograms were computed and filtered (1–20 Hz). The theta timescale lag was determined as the time of the cross-correlograms peak with highest amplitude between −100 msec and 100 msec.
Local field potential (LFP) and multiple single units (Supplementary Table; Supplementary Fig. 1) were recorded from the CA1 pyramidal layer of 5 rats trained to perform a hippocampus-dependent, delayed spatial alternation task (Detailed Methods, Ainge et al., 2007; Pastalkova et al., 2008)). In this task, a rat must remember its previous choice in order to turn into the arm opposite to the one he visited in the preceding trial (Fig. 1A). Return to the previously visited arm was not rewarded and was regarded as error. Following an error trial, rats kept alternating from the erroneous choice. Behavioral/recording sessions lasted for 15–20 minutes. Experiments began by recording 1 to 3 control sessions. Following the last control session, rats were injected with CB1 agonist CP55940 (0.1mg/kg body weight, I.P.), a dose which affected firing patterns and performance of the animal but did not interfere with the ability of the animal to run in the maze (Robbe et al., 2006). Starting half an hour after the injection, 1 to 7 sessions were recorded while the animals were under the effect of the drug. Additionally, one or two recovery sessions were recorded starting 6 hours after the injection or on the following day (Supplementary Table). In the Control condition, rats made very few errors, performing on average at 88% correct (Fig. 1B, C left panel). After cannabinoid injection, choice errors increased significantly and behavioral performance was, on average, close to chance level (53%, Fig. 1B, C left panel, p=0.0001, two-sided signed rank test). As expected from the well described locomotor effects of cannabinoids (Monory et al., 2007), reduced choice accuracy under the drug’s influence was associated with decreased running speed (Fig. 1D) and fewer trials performed per session (Fig. 1B, C right panel, p= 0.0002). The effect of the cannabinoid injection was reversible both in terms of choice accuracy and running speed (Fig. 1B–D, p=0.9 for accuracy and p=0.8 for the number of trials per 5 minutes; Control vs. Recovery). Altogether these behavioral observations confirm that under the influence of cannabinoids rats fail to perform a spatial memory task. Additionally this cognitive impairment is associated with slower running speed that will have to be taken in account when analyzing firing patterns of hippocampal neurons.
In the hippocampus, the location dependency of pyramidal cell firing is an outstanding and robust neuronal correlate of behavior. However, how these place cells contribute to the memory function of the hippocampus is debated. Different mechanisms have been proposed, including spatial and rate remapping (Colgin et al., 2008). We therefore started by quantifying and comparing basic features of place fields recorded before and after the cannabinoid injection. Of the 14 experiments performed in 5 rats, one was discarded from further analysis because the running pattern of the animal was strongly affected by the drug (Supplementary Fig. 2). In the remaining 13 experiments, 58 putative pyramidal cells were unambiguously recorded through at least two consecutive behavioral conditions (see Supplementary Table for the number of cells and place fields analyzed per experiment, and Detailed Methods for the criteria used to detect place fields). Despite strong impairment in behavioral accuracy, spatial remapping of place cells was not observed in Drug condition, as quantified by the spatial correlation measure of place cells (see Methods, Fig. 2A, B, p=0.23, two-sided signed rank test; Supplementary Fig. 4 shows the firing maps of all the place cells reliably tracked across experimental conditions). The spatial information content of the place cells’ firing was not significantly affected either (Fig. 2B, p=0.38). Similarity of position-coding by the place cell was further illustrated by the similar population vectors (Gothard et al., 1996) across Control, Drug and Recovery conditions (Supplementary Fig. 5). We next compared other features of place fields. We observed a small but significant decrease of place field area in Drug condition (8%, Fig. 2D, p<0.0001) and a small decrease in the ratio between spikes fired in the field(s) and the total number of spikes fired by the cell (Fig. 2E, p=0.01). More noticeably, we found a strong decrease of both in-field peak firing rates and in-field average firing rates (Fig. 2F, G, p<0.0001). Because locomotion speed is known to affect firing rates of pyramidal neurons (McNaughton et al., 1983; Czurkó et al., 1999; Huxter et al., 2003) and the drug effectively reduced the in-field running speed of the rat (Fig. 2H, p<0.0001), we attempted to quantify the contribution of speed decrease on the observed changes in firing rate. We performed an analysis of covariance between in-field average running speed and both infield peak and average firing rates. These comparisons revealed that the cannabinoid effect on both peak and average firing rates was essentially explained by the decrease in running speed (Fig. 2I–J, p=0.14 and p=0.12, analysis of covariance).
Units that could be reliably tracked across conditions represented only a third of the total number of well-isolated cells (Supplementary Table), whereas the majority of the recorded cells were sufficiently stable within but not across sessions. We used the entire dataset to perform independent population analysis of the place cell features examined above. Place cells analyzed independently in Control, Drug and Recovery conditions confirmed the observations made on the subset of cells monitored throughout conditions (Supplementary Fig. 6). We also noticed that following cannabinoid injection, the number of putative pyramidal cells clustered per electrodes significatively increased (p=0.017, Supplementary Figure 3), suggesting that a fraction of previously silent cells began to emit spikes (Thompson and Best, 1989). Because this effect did not reverse (p=0.8, Supplementary Figure 3) and was not accompanied by an increase in the number of place cells (p=0.25, Supplementary Figure 3) it is difficult to relate it to the observed behavioral impairment. This result is reminiscent of the observation in head-restrained rats, in which the number of coactive neurons increased following cannabinoid injection and is in support of the hypothesis that neurons tend to ‘escape’ population control following cannabinoid injections (Robbe et al., 2006).
The previous set of analyses revealed that despite the strong memory impairment caused by the cannabinoid, place cells firing pattern dynamics at behavioral time-scale were largely preserved, especially if the effect of running speed on firing rate was taken in account. In behaving rodents, the firing rate of hippocampal neurons is strongly modulated by the 5 to 10 Hz theta rhythm (Buzsáki, 2002), and theta-based mechanisms have been proposed to support episodic memory (Hasselmo, 2005; Jensen and Lisman, 2005). In head-restrained animals, cannabinoids altered the firing pattern and coordination of hippocampal neurons at short (+/− 150 msec) timescale (Robbe et al., 2006). Therefore, we next investigated whether theta timescale dynamics of hippocampal place cells were affected by the cannabinoid administration in the present behavioral setting.
First, autocorrelograms and power spectrograms were computed from the spike trains and LFP recorded inside each place field. Both methods showed a strong drug-induced decrease in the power and frequency of unit oscillation (Fig. 3A–D, unit power, p=0.0002, two-sided signed rank test; unit frequency p<0.0001) and LFP (Supplementary Fig. 7A, B) in the theta band. In principle, both these changes could be attributed to reduced running speed (Geisler et al., 2007; Jeewajee et al., 2008; Montgomery et al., 2009). An analysis of covariance showed that most of the drug-induced decrease of unit theta power was explained by speed (Fig. 3E, p=0.23). In contrast, the drug-induced decrease in oscillatory frequency of unit activity remained strongly significant even once the running speed effect was taken into account (Fig. 3F, p<0.0001). A similar, speed-independent decrease in theta LFP frequency was also observed (Supplementary Fig. 7C). A further regression analysis between unit and LFP theta frequency revealed that the drug had a distinct effect on those two variables (Supplementary Fig. 7D, the slopes of the linear regression in Control and Drug conditions were statistically different, p<0.0001, analysis of covariance). These results suggest that the cannabinoid, independently of changes in running speed, reduced differentially the oscillatory (theta) frequency of place cells and LFP.
When rats perform in a maze, the theta oscillation frequency of place cells is faster than that of the ongoing LFP (theta) oscillation, resulting in a gradual shift of their spikes relative to the simultaneously recorded phase of theta (phase precession, O’Keefe et al., 1993). Therefore, a direct measure of the drug’s influence on the spike-LFP theta temporal relationship can be obtained by comparing the strength of the phase precession (see Methods), a measurement that has the advantage of being independent of running speed (O’Keefe et al., 1993; Skaggs et al., 1996; Huxter et al., 2003; Geisler et al., 2007; Diba and Buzsáki, 2008). To reduce the possibility that changes in overt behavior of the animal biased the phase precession analysis, spikes fired while animals were rearing and running at a speed less than 5 cm/sec were discarded from the analysis. Both the correlation coefficient and the slope of the linear regression between spikes phase and position of the animal were decreased under the influence of cannabinoid (Fig. 4A–C, correlation coefficient p<0.0001; slope, p=0.01, two-sided signed rank test). To determine if this result was related to phase estimation errors, due to decreased LFP theta power, we examined, for each place field recorded across experimental conditions, the correlation between the change in phase precession coefficient and the change in theta LFP power but did not find a significant effect (Spearman’s rank correlation coefficient, r=−0.11, p=0.38).
To further investigate how theta timescale temporal coding was altered by the cannabinoid, we took advantage of the observation that within the theta cycle, the time lag between the spikes of overlapping place cells is correlated with the spatial distance between the respective place fields. Thus, this theta-scale time lag provides information about the past, present and future positions of the animal (Skaggs et al., 1996; Dragoi and Buzsáki, 2006). The distance vs. theta-scale time-lag correlation is referred as sequence compression index, and has been shown to be independent of the running speed (Geisler et al., 2007; Diba et al., 2008). We compared theta-scale sequence compression index in Control, Drug and Recovery conditions in place cell pairs with overlapping spatial representations (see Methods). This analysis was performed using only the spikes fired inside the overlapping place fields. Similarly to the phase precession analysis, spikes fired while animals reared and ran at a speed less than 5 cm/sec were discarded. The correlation between theta timescale time lag and place fields distance was strongly and reversibly reduced after cannabinoid injection (Fig. 5A–C, correlation coefficient between field distance and theta time lag was 0.83, 0.41 and 0.83 for respectively Control, Drug and Recovery, Control vs. Drug Z-score = 3.06; Drug vs. Recovery Z-score = 2.42 and Control vs. Recovery Z score = 0.50, permutation test described in Detailed Methods; see also Supplementary Figure 9). To understand the origin of this increased temporal jitter in the Drug condition, we calculated the power spectrogram of the unit pair cross-correlograms and averaged them to obtain a group measure for each condition. A strong and reversible decrease in both power and frequency of the cross-correlograms was caused by the drug (Fig. 5D, Control vs. Drug and Recovery vs. Drug, Z-score>2 for all frequencies between 8 and 12 Hz, permutation test). In a separate analysis, we constrained our analysis to a subset cell pairs, whose cross-correlograms were theta modulated (see Detailed Methods). In this subset of cell pairs the correlation remained significantly smaller in the Drug condition (Supplementary Fig. 8, Control vs. Drug, Z-score =4.96, Recovery vs. Drug, Z-score =3.52, Control vs. Recovery Z-score= 0.32, permutation test). Altogether this analysis showed that theta timescale coordination between spikes of neuron pairs is reversibly reduced by cannabinoid administration. Finally, in two experiments, (rat #5, Supplementary Table), sufficiently large numbers of overlapping pairs of place fields were recorded in many successive sessions. In these datasets, a strong correlation was found between the percent of correct trials and theta-scale sequence compression index (Figure 5D, r=0.82, p<0.0001), further supporting the hypothesis that alteration in theta timescale spike coordination is linked to the cognitive impairment.
We found that cannabinoid administration resulted in a profound behavioral deficit in a hippocampus-dependent task. Despite the severe memory impairment, the location-dependent firing of place cells was largely spared, whereas both spike-timing coordination between place cells at the theta timescale and theta phase precession of spikes were reversibly reduced. These findings imply that spatial memory depends primarily on the temporal coordination of neurons and can be dissociated from the stability of the spatial map.
According to the cognitive map theory (O’Keefe et al., 1978), position-dependent activity of hippocampal neurons represents an internal map, which enables flexible spatial navigation and assists in solving hippocampus-dependent spatial memory tasks (O’Keefe and Speakman, 1987; Lenck-Santini et al., 2001, 2002; Cacucci et al., 2008), such as the delayed spontaneous alteration task, used in the present experiment. From this perspective, an unexpected finding of our experiment is that administration of the cannabinoid receptor agonist CP55940 left the spatial map representation largely intact, while inducing a severe memory deficit. No spatial ‘remapping’ of place cell activity was observed under the drug’s influence either by visual inspection of the firing maps or by using the quantitative pixel-by-pixel correlation method. The spatial resolution of the place cells did not deteriorate as shown by the stable information content of individual place cells, and the similar population vectors across conditions. In addition to these preserved spatial parameters, we also found a reversible decrease of both average and peak firing rates of place fields following cannabinoid administration. However, it is unlikely that changes in firing rate of place cells were responsible for the memory deterioration (Colgin et al., 2008) because the rate decrease was accounted for by the decreased running speed within the individual place fields (McNaughton et al., 1983; Czurkó et al., 1999; Huxter et al., 2003). In addition, it has been shown previously that reduced firing rates during slower runs in intact rats do not affect the quality of place fields. In summary, the profound drug-induced memory deficit was not correlated with detectable changes in the spatial map and the underlying behavioral-scale temporal dynamic
Our findings complement a previous experiment, which addressed the relationship between spatial memory and hippocampal map representation. Jeffery and al. (2003) trained rats on a tone-cued spatial navigation task in a black box and tested them later in a white box in the same room. Although the change from black to white induced remapping of most place cells, navigational performance remained largely above chance level (Jeffery et al., 2003). These findings, combined with ours, are at variance with the widely held view that navigation is guided by the map, maintained by the hippocampal place cells (O’Keefe et al., 1987; Lenck-Santini et al., 2001, 2002; Cacucci et al., 2008). In support of the hippocampal map-dependent navigation, one can argue that an extra-hippocampal mechanism maintained memory performance in both the navigation task of Jeffrey et al., and ours. However, this interpretation is not likely since hippocampal lesion essentially abolished performance in both tasks (Jeffery et al., 2003; Ainge et al., 2007; Pastalkova et al., 2008). Furthermore, several observations point to the hippocampus as an important target of systemic cannabinoid injection and the mediator of the cognitive impairment. First, intra-hippocampal injection of the same cannabinoid used in our study reproduced the spatial deficits induced by systemic cannabinoid injection (Lichtman et al., 1995). Second, the effects of systemic cannabinoid injection in memory tasks resemble the effects of hippocampal lesions (Hampson et al., 1998). Third, recreational (systemic) use of cannabis in humans deteriorates hippocampus-dependent episodic memory (Miller et al., 1983; Curran et al., 2002; Ilan et al., 2004). In conclusion, the mechanisms responsible for the severe memory deficit seen in our experiments are likely to involve the hippocampus but may be different than those supported by the position-specific firing of place cells.
In addition to the firing rate-based computation, temporal aspects of place cell activity have been proposed to provide coding mechanisms for hippocampal functions (Skaggs et al., 1996; Hasselmo 2005; Jensen and Lisman, 2005; O’Keefe et al., 2005). When the rat crosses the place field of a neuron, spikes are fired at progressively earlier phases of the theta LFP oscillations (O’Keefe et al., 1993). Furthermore, place field sequences and distance representation of neurons on the track are represented as temporal (phase) offsets within single theta cycles in a compressed manner (Skaggs et al., 1996; Dragoi et al., 2006). Both of these temporal codes are independent of running speed (O’Keefe et al., 1993; Huxter et al., 2003; Geisler et al., 2007; Diba et al., 2008). Recently, a dual coding mechanism has been proposed, suggesting that the theta phase of spikes would encode spatial position of the rat, whereas the firing rate would be available for encoding episodic memories (Huxter et al., 2003; O’Keefe et al., 2005). At variance with this model, our study shows that representation of position by discharge rates was not affected following cannabinoid injection and memory impairment but both spike phase precession and distance-time compression were altered. In a previous head-restrained study, synchrony of hippocampal neurons at the short timescale (+/−150 ms) was severely impaired by cannabinoids, while the firing rates of pyramidal cells and interneurons were only mildly affected (Robbe et al., 2006). The current results obtained in behaving rats further support that synchrony is the main network target mechanism of cannabinoids (Constoe et al.,1975; Buonamici et al., 1982; Hájos et al., 2000; Bernard et al., 2005; Robbe et al., 2006; Hajós et al., 2008). Additionally our findings reinforce the conclusion of a recent study showing that phase precession, theta-scale distance compression and performance in the water maze task were impaired in an animal model of temporal lobe epilepsy (Lenck-Santini and Holmes, 2008). Here, since alteration of short-timescale events coincided with the behavioral impairment, we hypothesize that proper temporal coordination of spikes across cell assemblies is critical for normal spatial memory performance. Theta timescale compression of the representation of past, present and future locations may be needed for bringing these neurons in the temporal window of plasticity (Skaggs et al., 1996; Dragoi et al., 2006). In addition, or alternatively, synchrony may be critical to effectively transfer the content of hippocampal cell assemblies to neocortical targets and/or properly link cell assembly sequences in time (Pastalkova et al., 2008). Yet another possibility is that theta timescale coordination of neuronal firing is necessary to convey the hippocampal spatial representation to other structures, which assist in guiding behavioral performance. Although these mechanisms have yet to be elaborated, our findings imply that theta timescale coordination of neuronal activity is critical for memory performance, whereas properly anchored spatial representation is not sufficient.
We thank A. Amarasingham, K. Diba, C. Geisler, K Mizuseki, E Pastalkova and A. Sirota, for valuable discussions, comments on the manuscript and support with data analysis. Supported by National Institutes of Health (NS034994; MH54671), National Science Foundation (SBE 0542013), the J.D. McDonnell Foundation (GB), the Human Frontier Science Program and the Ramon-y-Cajal Young Investigator Program (DR).