Peyrache et al. Replay of rule-learning related neural patterns



Download 146.31 Kb.
Page2/2
Date conversion16.05.2018
Size146.31 Kb.
1   2

DISCUSSION

This work shows new links between learning and the dynamics of replay in the mPFC and hippocampus with 3 main results. First, mPFC replay occurs in transient episodes, apparently corresponding to the activation of distinct cell assemblies. Second, prefrontal replay, while occurring throughout the following sleep episode, was most likely coincided with hippocampal sharp wave events, and therefore with increased hippocampal replay and hippocampal/neocortical interactions. Third, and most importantly, mPFC replay during hippocampal sharp waves principally concerned neural activity patterns emerging only after the acquisition of a new rule.

The results demonstrated the relation between memory replay, the task phase where the activity patterns originated, and task performance level in a dynamic setting, wherein the rat were obliged to continually adapt to new rules. Interestingly, when the rat behaved according to the newly acquired rule, the patterns that contributed the most to memory replay in mPFC during hippocampal SPWRs were those appearing when the rat commits to choose one target arm. During the trial, the contribution to hippocampal-related replay in the mPFC climbed steadily on the departure arm (where activity may predict the choice42, to peak at the arms intersection. Furthermore, preferentially replayed patterns arose just when rats began a series of correct trials at criterion with respect to the new rule. Prefrontal cortical activity reflecting a newly learned associations has been found to emerge only after that the associations have been acquired34. The time course that we observe for replay is similar: co-activations preferentially replayed in SPWRs arise after the rat starts to perform at criterion. These effects were likely not a consequence of different neural activity statistics at the choice point, as they remain specific for POST SPWR patterns and are not found for patterns extracted from PRE or from inter-SPWR intervals.

The hippocampal involvement in the formation of these novel cell assemblies is likely to be critical: strikingly, times in the task linked to preferential replay were also marked by increased coherence between the hippocampus and mPFC LFP in the theta range (Benchenane et al. Soc. Neurosci. Abs 690.15, 2008). Thus, upon learning, the choice point is the site of an increase hippocampal-prefrontal coherence. Cell assemblies activated at those times are prominently replayed in the mPFC, during SPWRs, that is, when the hippocampal-neocortical interaction is at its peak19-22. Coactivations may reflect prefrontal (or prefrontal-hippocampal) cell assemblies representing new information. The nature of this new information could be two-fold: on one hand, it could represent an emerging representation of the newly learned rule; on the other hand, it might reflect processes that take place after learning has taken place, for example, a representation that is activated by consistent stream of reward. The reward signal may “tag” the representations making them more likely to be replayed. In support of this hypothesis, we analyzed those periods in which rats were searching for the correct rule by trial and error. We showed that when the rat,makes spontaneous strategy shifts to strategies different from the rewarded one, patterns from before or after the shift are indifferently replayed in sleep (Supplementary Fig. 13). Thus, the consequent changes in the general sensory experience, do not cause the resulting neural patterns to be replayed more or less strongly. Also, this seems to rule out interpretation of our result in terms of repetitive experience resulting in stronger replay18. Because preferential replay only occurs during SPWRs, we speculate that regulation of hippocampal-prefrontal interactions could result from dopaminergic, reward-related signals43. The replay of reward tagged patterns, in concert with the hippocampal replay during SPWRs14,44 would mark the initial period of systems consolidation.



The high temporal resolution of our coactivation measures revealed another remarkable feature of prefrontal replay, its detailed time course. Replay is largely accounted for by brief events, with durations on the order of 100 ms or less (e.g. Fig. 2). In each event (similar to ‘network spikes'; see also Supplementary Discussion), a substantial number of cells is co-activated. The observed highly irregular time course of reactivation strength suggests that replay is not simply due to changes in probability of co-firing for cell pairs (a straightforward consequence of a change in the efficacy of the connection between the two cells18), but is more likely generated by global network effects, induced for example by the excitatory feedback connections within the mPFC.
Transient replay is concentrated during SWS (Supplementary Fig. 5). This differs from studies of hippocampal reactivation which showed non-zero restful, non-sleep periods14. This may be an intrinsic difference between the hippocampus and the mPFC, or it could be due to a different sensitivity of our analysis methods. Within SWS, replay occurs principally in proximity to hippocampal SPWRs (Fig. 3-5). Thus, besides playing a special role for the hippocampal replay of newly formed memories14, SPWRs also coincide with an increase of overall reactivation in hippocampal output structures. Moreover, sharp-waves are related to increases in cortical activity and transitions to UP states20,21. It is therefore possible that the correlation between cortical replay and SPWRs observed here may be due to a hippocampal influence on mPFC. At present, though, it is difficult to speculate whether reactivation events originate in the hippocampus or the neocortex: our data show that prefrontal replay peaks about 40 ms after SPWR occurrences, a latency which is compatible with observations of prefrontal responses to hippocampal stimulation39, and thus with hippocampus triggering this extra-hippocampal replay. On the other hand, it is known that the neocortical slow oscillations45 influence the membrane potential of hippocampal cells23 and the probability of emission of SPWR22; such an influence would allow the neocortex to contribute to selecting the information reactivated in the hippocampus. As a third alternative, the concurrent prefrontal cortical replay and hippocampal SPWRs may be the result of particular neocortical network states that simultaneously regulate when hippocampal reactivation takes place and facilitate the transfer of reactivated patterns between brain regions20. In a more radical version of this hypothesis, the representation of a memory item could integrally involve both hippocampus and neocortex already at the time of initial encoding, and thereafter this would reactivate as a whole and manifest itself simultaneously in both structures. In fact, analyzing data from these same experiments we show that signs of hippocampal/neocortical interplay were already present when the replayed representation was likely formed, at the level of theta-band coherence.

In conclusion, the acquisition of new rules involves the hippocampo-cortical network; during the ensuing sleep, the PFC activity patterns during hippocampal SPWRs reflect the neural patterns that occurred during the task phase, particularly when a rule has been learned, just at the time when hippocampo-cortical coherence is enhanced. Whether the predominating causal influence in this dialogue is cortical or hippocampal, or rather this corresponds to an emergent system-wide representation of information, our results show a possible mechanism by which task-relevant learned information can be expressed and reactivated in the prefrontal cortex. This would be contingent to SPWRs for newly formed memories, but more uncoupled to the hippocampus for more distant memory traces.



METHODS

Animals Four Long-Evans (pigmented) male rats (René Janvier, Le Genest-St-Isle, France) weighing 250-300g at arrival, were handled and pre-trained as described in the Supplementary Information. All experiments were in accord with institutional (CNRS Comité Opérationnel pour l’Ethique dans les Sciences de la Vie), international (NIH guidelines) standards and legal regulations (Certificat no. 7186, French Ministère de l’Agriculture et de la Pêche) regarding the use and care of animals.

After pre-training, rats were anesthetized with intramuscular Xylazine (Rompun 0.1 ml), and intra-peritoneal pentobarbital (35 mg/kg). A drive containing 7 tetrodes (six recording, plus one reference) was implanted on the skull above the right medial prefrontal cortex (AP 3.5-5 mm, ML 0.5-1.5 mm). Each tetrode was contained in a 30 ga hypodermic tube, with the tubes arranged in two parallel, adjacent rows. Tetrodes were twisted bundles of 13 m diameter polymide-coated nichrome wire (Kanthal, Palm Coast, FL); the drive allowed independent adjustment of tetrode depth. After dura retraction, the rows of cannulae were implanted parallel to the sagittal sinus, so that they targeted, respectively, the superficial and deep layers of the medial bank of the cortex. A separate micro-drive containing three tetrodes was targeted to the ventral hippocampus (AP -5.0 mm ML 5.0 mm). The tetrodes were electrically connected in a single-electrode configuration (all channels shorted together) and used for local field potential (LFP) recordings. For all LFP recordings, a screw implanted on the occipital bone above the cerebellum was used as a reference. The hippocampal tetrodes were lowered to the CA1 pyramidal layer; the depth was adjusted with the help of LFP signs (flat sharp waves, maximum amplitude of ripple oscillations) After surgery, rats were allowed to recover for at least 2 weeks, while the tetrodes were gradually lowered to reach the Prelimbic area (PL; main drive), and the CA1 pyramidal layer (hippocampal micro-drive). During the experiment, they were gradually lowered to probe, in different sessions, different dorso-ventral levels of the prelimbic cortex.



Behavioral task Four rats performed an attentional set shift task on a Y-maze (see Supplementary information for a description of the apparatus). Such extradimensional set shift tasks were shown to require the function of the medial prefrontal cortex (mPFC) in rats36. This parallels the involvement of the human prefrontal cortices in the Wisconsin Card Sorting Task, which inspired the present experimental design. Each recording session consisted of a 20-30 minutes sleep or rest epoch (PRE epoch) in which the rat remained undisturbed in a padded flowerpot placed on the central platform of the maze, followed by an AWAKE epoch, in which the rat performed the behavioral task described below for 20-40 minutes, and by a second sleep or rest epoch (POST epoch; same situation as in PRE) of 20-30 minutes.

The first recording sessions corresponded to the first time rats encountered the behavioral task. Rats started each trial in the same arm (the departure arm). One of the two other (choice) arms was illuminated at random (pseudo-random schedule: runs of more than 4 consecutive trials with the same illuminated arm were avoided, as were repeated bouts of imposed alternation between the two arms). After that, the central platform was lowered, allowing the rat to access the choice arms.

Only one of the choice arms was rewarded, according to one of 4 contingency rules. Two contingency rules were spatially guided (always go to the right arm, or to the left arm), the other two were cue guided (go to the illuminated arm, or to the dark arm). The rule employed at any given moment in time was not signaled to the rat in any way, so that the animal had to learn the rule by trial and error. Once the rat reached a criterion of 10 consecutive correct trials, or only one error out of 12 trials, the rule was changed with no further cue warning to the rat. Rule changes were extra-dimensional, that is, from a spatially-guided rule to a cue-guided rule, and vice versa. Because of an operator mistake, an intradimensional shift was operated once. Typically, the trials necessary for acquisition of one rule spanned more than one session, which implies that rule shifts occurred only in a fraction of the sessions.

All four rats learned the Right and Light rules (at least 10 consecutive correct trials), whereas only two learned the Left task, one learned the Dark task.



Electrophysiology Tetrode recordings were obtained from 6 tetrodes implanted in the Prelimbic and Infralimbic areas of the rat neocortex, hippocampal local field potentials were obtained from two further tetrodes implanted in the CA1 pyramidal layer. The tetrodes were gradually lowered in the cortical tissue during the course of the experiment. Because they were not moved after each recording session, it is possible that the same cells were recorded more than once, however, usually in different behavioral situations (rewarded strategy, performance level, etc.).

For electrophysiological recordings, signals from all electrodes were fed into a unit-gain headstage pre-amplifier (HS-54; Neuralynx, Bozeman, MT) and then, through a tether cable, to programmable amplifiers (Lynx-8, Neuralynx, Bozeman, MT). All signals were there amplified 2000x, Signals for single unit recordings were bandpass filtered between 600 and 6000 Hz, signals for LFP recordings were filtered between 0.1 and 475 Hz (cortex) and 1 and 475 Hz (hippocampus). Data were digitized and stored on hard disk by a Power 1401 (CED, Cambridge, UK) acquisition system, controlled by the Spike 2 software (also by CED). Single unit data were sampled at 25 kHz, and a 1.3 ms sample was timestamped and stored for all the channel in a tetrode whenever any of the 4 channels exceeded a threshold. Local field potentials were sampled and stored at a 2 kHz sampling rate.


Learning: The first trial of a block of at least three consecutive correct ones and from which the performance until the end of the session was above 80% was defined as the learning point (i.e. the rule was acquired).

.

Histology At the end of the experiments, a small electrolytic lesion was made by passing a small cathodal DC current (20 μA, 10 s) through each recording tetrodes to mark the location of its tip. The rats were then deeply anesthetized with pentobarbital. Intracardial perfusion with saline was followed by 10 % formalin saline. Histological sections were stained with cresyl violet. The electrode tracks were reconstructed, verifying that the recording sites were in the Prelimbic cortex, or in exceptional cases, in the dorsal end of the infralimbic cortex.



Slow Wave Sleep Detection Slow Wave Sleep (SWS) was automatically detected. Power spectrograms of cortical and hippocampal LFPs were computed for each sleep sessions with bins of 1 seconds. Power in the cortical delta band (1-4 Hz), hippocampal theta (5-10 Hz), cortical spindles (10-20 Hz) and speed of head motion were clustered with a K-means algorithm. Clusters corresponding to high values of delta and spindle powers, and to a low degree of head movement, were considered as corresponding to SWS. Successive SWS clusters occurring within intervals of less than 1 minute were merged and finally, resulting time intervals of SWS smaller than 10s were dropped.

Data Preprocessing. For single unit activity discrimination, the first three principal components of the energy-normalized waveforms were computed from spike waveforms for the 4 tetrodes, generating a 12-dimensional vector describing each spike. Those vectors were the input of the KlustaKwik46 clustering program. The resulting classification was manually refined using the Klusters47 interface.

For ripple detection, the hippocampal LFP from the CA1 pyramidal layer was bandpass filtered in the ripples frequency range (100-300 Hz) to detect SPWRs. Then, the RMS signal was convolved with a 20 ms gaussian window. Only intervals of time for which the resulting envelope was 2 SD above the raw filtered signal were kept. Ripples events were time stamped with the times of the deepest LFP troughs in these intervals if the latter were at or below 5 SD from the RMS signal baseline. Finally, discrimination of SPWRs were considered reliable when at least 40 ripples were detected. In most sessions, at least 200 SPWRs were detected.



For spindles detection, cortical LFP signal were filtered between 10 and 20 Hz. The RMS was low-pass filtered with a gaussian window of 100 ms. Then time intervals 1 SD above the filtered signal, which were closer than 100 ms apart were merged. Of the resulting intervals, only those at least 500 ms long were kept. Spindles troughs are the minima of the filtered signal during those intervals. For delta waves detection, cortical LFP signal was filtered between 0.1 and 4 Hz. Delta waves peaks were the minima below 2 SD of the filtered signal.

Computational methods. Reactivation strengths describe the instantaneous replay at each time, during rest sessions, of neural co-activation patterns that were characteristic of the AWAKE epoch. Briefly, the spike trains from all cells recorded in the AWAKE epoch were binned (in 100 ms bins) and z-transformed; then, principal component analysis (See SOI; Supplementary Fig. 3) was used to extract the characteristic activity patterns. PRE and POST epoch spike trains were similarly binned and transformed, and reactivation strengths for each component at any given time during the rest epoch were computed from the projection of the binned activity vector on each principal component, which measures the instantaneous reactivation of that activity pattern. By summing over all components and all times, one obtains a measure of similarity between the correlation matrices in the AWAKE period and in the PRE and POST periods analogous to previously published reactivation measures12,14. High rank (which we will refer to as signal) principal components (associated to large eigenvalues), capturing fundamental processes , are defined as those associated to an eigenvalue greater than a threshold value. As a threshold, we took the maximum theoretical value (λmax) for eigenvalues of principal components for random uncorrelated spike trains48. (see SOI, Fig. 1C). The normalized eigenvalues, or encoding strengths Φ = λ/λmax, quantify the weight of each component during the AWAKE epoch in a uniform way across sessions.

Conversely, in order to establish which moments during the task expressed the patterns that account for most of the sleep replay, principal components can be computed by the same procedure from the activity in the POST SWS epoch (PRE SWS as a control), and reactivation strength can be computed as defined above from the AWAKE activity.

An in-depth description and a schematic (Supplementary Fig. 3) of the mathematical method can be found in the Supplementary information

Note: Supplementary information is available on the Nature Neuroscience website.


Acknowledgements We thank Dr. P. Tierney for valuable discussions and help with the surgical procedures, Prof. J.-M. Deniau, Drs. Y. Gioanni, A.-M. Thierry, M. Zugaro, M. Cencini and A. Aubry for interesting discussions. S. Doutremer for histology, Prof. D. Hopkins and N. Quenech’du for the anatomical reconstructions, V. Douchamps for help with the experiments, Drs. K. Gothard, K. Hoffman and A. Treves for critical readings of an earlier version of the manuscript and Prof. A Berthoz for support throughout the project. Supported by Fondation Fyssen (FPB), Fondation pour la Recherche Medicale (AP), EC contracts FP6-IST 027819 (ICEA), FP6-IST-027140 (BACS), and FP6-IST-027017 (NeuroProbes).
Author Contributions SIW, FPB and MK designed the experiment, FPB, MK, and AP performed the experiment, AP, FPB and KB designed the analysis techniques, AP analyzed the data, FPB, AP and SIW wrote the paper.
Author information Reprints and permission information is available at http://www.nature.com/reprints. Correspondence and request for materials should be addressed to SIW (sidney.wiener@college-de-france.fr).
1. Malenka, R. C. & Nicoll, R. A. Long-term potentiation--a decade of progress? Science 285, 1870-1874 (1999).

2. Marr, D. Simple memory: a theory for archicortex. Philos Trans R Soc Lond B Biol Sci 262, 23-81 (1971).

3. McClelland, J. L., McNaughton, B. L. & O'Reilly, R. C. Why there are complementary learning systems in the hippocampus and neocortex: insights from the successes and failures of connectionist models of learning and memory. Psychological Review 102, 419-457 (1995).

4. Dudai, Y. The neurobiology of consolidations, or, how stable is the engram? Annu Rev Psychol 55, 51-86 (2004).

5. Squire, L. R. Memory and the hippocampus: a synthesis from findings with rats, monkeys, and humans [published erratum appears in Psychol Rev 1992 Jul;99(3):582]. Psychological Review 99, 195-231 (1992).

6. Moscovitch, M., Nadel, L., Winocur, G., Gilboa, A. & Rosenbaum, R. S. The cognitive neuroscience of remote episodic, semantic and spatial memory. Curr Opin Neurobiol 16, 179-190 (2006).

7. Winocur, G., Moscovitch, M. & Sekeres, M. Memory consolidation or transformation: context manipulation and hippocampal representations of memory. Nat Neurosci 10, 555-557 (2007).

8. Tse, D., Langston, R. F., Kakeyama, M., Bethus, I., et al. Schemas and memory consolidation. Science 316, 76-82 (2007).

9. Marshall, L., Helgadóttir, H., Mölle, M. & Born, J. Boosting slow oscillations during sleep potentiates memory. Nature 444, 610-613 (2006).

10. Walker, M. P. & Stickgold, R. Sleep, memory, and plasticity. Annu Rev Psychol 57, 139-166 (2006).

11. Wagner, U., Gais, S., Haider, H., Verleger, R. & Born, J. Sleep inspires insight. Nature 427, 352-355 (2004).

12. Wilson, M. A. & McNaughton, B. L. Reactivation of hippocampal ensemble memories during sleep. Science 265, 676-679 (1994).

13. Nadasdy, Z., Hirase, H., Czurko, A., Csicsvari, J. & Buzsaki, G. Replay and time compression of recurring spike sequences in the hippocampus. J Neurosci 19, 9497-9507 (1999).

14. Kudrimoti, H. S., Barnes, C. A. & McNaughton, B. L. Reactivation of hippocampal cell assemblies: effects of behavioral state, experience, and EEG dynamics. Journal of Neuroscience 19, 4090-4101 (1999).

15. Hoffman, K. L. & McNaughton, B. L. Coordinated reactivation of distributed memory traces in primate neocortex. Science 297, 2070-2073 (2002).

16. Ji, D. & Wilson, M. A. Coordinated memory replay in the visual cortex and hippocampus during sleep. Nat Neurosci 10, 100-107 (2007).

17. Euston, D. R., Tatsuno, M. & McNaughton, B. L. Fast-forward playback of recent memory sequences in prefrontal cortex during sleep. Science 318, 1147-1150 (2007).

18. O'Neill, J., Senior, T. J., Allen, K., Huxter, J. R. & Csicsvari, J. Reactivation of experience-dependent cell assembly patterns in the hippocampus. Nat Neurosci 11, 209-215 (2008).

19. Siapas, A. G. & Wilson, M. A. Coordinated interactions between hippocampal ripples and cortical spindles during slow-wave sleep. Neuron 21, 1123-1128 (1998).

20. Sirota, A., Csicsvari, J., Buhl, D. & Buzsaki, G. Communication between neocortex and hippocampus during sleep in rodents. Proc Natl Acad Sci U S A 100, 2065-2069 (2003).

21. Battaglia, F. P., Sutherland, G. R. & McNaughton, B. L. Hippocampal sharp wave bursts coincide with neocortical "up-state" transitions. Learn Mem 11, 697-704 (2004).

22. Isomura, Y., Sirota, A., Ozen, S., Montgomery, S., et al. Integration and segregation of activity in entorhinal-hippocampal subregions by neocortical slow oscillations. Neuron 52, 871-882 (2006).

23. Hahn, T. T., Sakmann, B. & Mehta, M. R. Phase-locking of hippocampal interneurons' membrane potential to neocortical up-down states. Nat Neurosci 9, 1359-1361 (2006).

24. Buzsaki, G. Two-stage model of memory trace formation: a role for "noisy" brain states. Neuroscience 31, 551-570 (1989).

25. Buzsaki, G. Hippocampal sharp waves: their origin and significance. Brain Research 398, 242-252 (1986).

26. Steriade, M., Nuñez, A. & Amzica, F. A novel slow (< 1 Hz) oscillation of neocortical neurons in vivo: depolarizing and hyperpolarizing components. J Neurosci 13, 3252-3265 (1993).

27. Frankland, P. W. & Bontempi, B. The organization of recent and remote memories. Nat Rev Neurosci 6, 119-130 (2005).

28. Gais, S., Albouy, G., Boly, M., Dang-Vu, T. T., et al. Sleep transforms the cerebral trace of declarative memories. Proc Natl Acad Sci U S A 104, 18778-18783 (2007).

29. Jay, T. M. & Witter, M. P. Distribution of hippocampal CA1 and subicular efferents in the prefrontal cortex of the rat studied by means of anterograde transport of Phaseolus vulgaris-leucoagglutinin. J Comp Neurol 313, 574-586 (1991).

30. Jay, T. M., Burette, F. & Laroche, S. NMDA receptor-dependent long-term potentiation in the hippocampal afferent fibre system to the prefrontal cortex in the rat. Eur J Neurosci 7, 247-250 (1995).

31. Takashima, A., Petersson, K. M., Rutters, F., Tendolkar, I., et al. Declarative memory consolidation in humans: a prospective functional magnetic resonance imaging study. Proc Natl Acad Sci U S A 103, 756-761 (2006).

32. Floresco, S. B., Braaksma, D. N. & Phillips, A. G. Thalamic-cortical-striatal circuitry subserves working memory during delayed responding on a radial arm maze. Journal of Neuroscience 19, 11061-11071 (1999).

33. Jones, M. W. & Wilson, M. A. Theta rhythms coordinate hippocampal-prefrontal interactions in a spatial memory task. PLoS Biol 3, e402 (2005).

34. Pasupathy, A. & Miller, E. K. Different time courses of learning-related activity in the prefrontal cortex and striatum. Nature 433, 873-876 (2005).

35. Baeg, E. H., Kim, Y. B., Kim, J., Ghim, J. W., et al. Learning-induced enduring changes in functional connectivity among prefrontal cortical neurons. J Neurosci 27, 909-918 (2007).

36. Birrell, J. M. & Brown, V. J. Medial frontal cortex mediates perceptual attentional set shifting in the rat. J Neurosci 20, 4320-4324 (2000).

37. Beggs, J. M. & Plenz, D. Neuronal avalanches are diverse and precise activity patterns that are stable for many hours in cortical slice cultures. J Neurosci 24, 5216-5229 (2004).

38. Mongillo, G., Barak, O. & Tsodyks, M. Synaptic theory of working memory. Science 319, (2008).

39. Degenetais, E., Thierry, A. M., Glowinski, J. & Gioanni, Y. Synaptic influence of hippocampus on pyramidal cells of the rat prefrontal cortex: an in vivo intracellular recording study. Cereb Cortex 13, 782-792 (2003).

40. Amzica, F. & Steriade, M. Electrophysiological correlates of sleep delta waves. Electroencephalogr Clin Neurophysiol 107, 69-83 (1998).

41. Steriade, M. Neuronal substrates of spike-wave seizures and hypsarrhythmia in corticothalamic systems. Adv Neurol 97, 149-154 (2006).

42. Baeg, E. H., Kim, Y. B., Huh, K., Mook-Jung, I., et al. Dynamics of population code for working memory in the prefrontal cortex. Neuron 40, 177-188 (2003).

43. Seamans, J. K., Floresco, S. B. & Phillips, A. G. D1 receptor modulation of hippocampal-prefrontal cortical circuits integrating spatial memory with executive functions in the rat. J Neurosci 18, 1613-1621 (1998).

44. Lee, I. & Kesner, R. P. Differential contribution of NMDA receptors in hippocampal subregions to spatial working memory. Nat Neurosci 5, 162-168 (2002).

45. Steriade, M. Grouping of brain rhythms in corticothalamic systems. Neuroscience 137, 1087-1106 (2006).

46. Harris, K. D., Henze, D. A., Csicsvari, J., Hirase, H. & Buzsáki, G. Accuracy of tetrode spike separation as determined by simultaneous intracellular and extracellular measurements. J Neurophysiol 84, 401-414 (2000).

47. Hazan, L., Zugaro, M. & Buzsáki, G. Klusters, NeuroScope, NDManager: a free software suite for neurophysiological data processing and visualization. J Neurosci Methods 155, 207-216 (2006).

48. Sengupta, A. M. & Mitra, P. P. Distributions of singular values for some random matrices. Phys Rev E Stat Phys Plasmas Fluids Relat Interdiscip Topics 60, 3389-3392 (1999).

49. Eytan, D. & Marom, S. Dynamics and effective topology underlying synchronization in networks of cortical neurons. J Neurosci 26, 8465-8476 (2006).

50. Cossart, R., Aronov, D. & Yuste, R. Attractor dynamics of network UP states in the neocortex. Nature 423, 283-288 (2003).

FIGURE LEGENDS

.
Figure 1



Signal components and their behavioral correlates

A: Signal components and behavioral correlates of their component mPFC cells: Peri-event time histograms of all cells from a recording session, aligned with trial initiation (vertical dashed line). Cells are sorted by their weight in the first signal component (scale above), showing that cells with similar weights tended to have similar behavioral correlates and cells with opposite signed large weights had complementary behavioral correlates. B: AWAKE epoch correlation matrix eigenvalues from the same session as Figs. 2A and 3A. Dotted line: the signal threshold, defined as the theoretical upper bound for eigenvalues in the random spike trains case (see SOI). Filled squares: eigenvalues associated with the six (suprathreshold) signal components. Hollow squares: non-signal components. C Trial-by-trial time scores of the first three principal components over the activity during the awake epoch, plotted as a function of the linearized position of the rat on the maze. The right panels summarize the rat behavior: a beige background color indicated trials in which the rat reliably chose the right arm, the green background indicated trials in which the rat, to a large extent alternated between the two arms (with no other discriminable strategy). The black dots in the first column (“Arm”) denote trials in which the rat chose the left arm. The black dots in the second column (“Correct”) indicate rewarded trials. (Here the Light rule was imposed and neither strategy was successful.) Note that PC scores are displayed instead of reactivation strength (see below) because in this case the sign is important, since large positive PC scores denote activation of those cells with positive PC weights (see fig. 2A) and large negative PC scores denote activation of those cells with negative PC weights.
Figure 2

Time course of memory replay

A: Reactivation strength (black/white traces; right axis) of the signal component shown in Fig. 2A during sleep PRE (left) and POST (right) epochs superimposed on the mPFC LFP spectrogram (left axis). White traces: SWS periods; black trace: non SWS. The spectrogram shows periods of elevated slow/delta and spindle oscillations often coinciding with SWS. Reactivation strengths show high peaks during POST epoch SWS, usually concomitant with periods of strong oscillations. B: Bar plot of the epoch-wide average reactivation strengths, grouped by their encoding strength (normalized eigenvalue), Φ = λ/λmax, for the PRE and POST epochs (data from 63 sessions and 1692 recorded cells; error bar: S.E.M.). C: Average difference in reactivation strength between PRE and POST, for PCs grouped by their encoding strengths (error bars: SD). D: Incidence of reactivation strengths during rest periods (left, PRE; right, POST) for the same session as Figs. 2A and 3A. Black filled zones: during SWS; gray trace: non SWS. The POST SWS histogram has a heavy tail, reflecting strong transient reactivation events. E: Average difference between POST and PRE of skewness of reactivation strength incidence histogram (3C) for components grouped by their encoding strength, showing that POST epoch histograms were generally more skewed than the PRE. Error bars: SEM. F: Cumulative contribution to the difference between total reactivation strength in POST and PRE for signal components (black; grouped by encoding strength) and for non-signal components (gray). Red diamonds: 99th percentile of reactivation strength distribution (in POST). About half of the reactivation was accounted for by reactivation strengths over the 99th percentile.
Figure 3

Memory replay, SPWRs and cell activity in a rest session

A: Reactivation strength for one signal component. B: Occurrences of SPWRs. C: Averaged firing rate of the cells that contributed the highest weights to the signal component. Note that peaks correspond to events in A and B. D: Average firing rate for all cells recorded in this session.
Figure 4

Example of reactivation strength peaks coinciding with hippocampal SPWR. A: Reactivation strength (white/gray traces; right axis) of the signal component superimposed on the mPFC LFP spectrogram (left axis). This is expanded from the zone in Figure 3A (POST) delimited by the dashed line. Black dashed line: normalized population firing rate. B: The bandpass filtered hippocampal LFP (100-300 Hz) shows ripple events (red asterisks); signal is normalised by its SD. C: bandpass filtered (0-5 Hz) PFC LFP. Delta waves are denoted by green asterisks. D: Raster plot of spike trains from the PFC cells sorted by principal component weight magnitude (as in Fig. 1A). E: Expansion of the 300 ms surrounding the peak indicated by an arrow in A. Red rasters represent spikes occurring in the bin of peak reactivation strength. The example also show two delta waves in the cortical LFP, with preceding and following increases in population activity (or UP states41). Note that, in the absence of SPWRs only a smaller reactivation peak was obtained (see Supplementary Fig. 7 for statistical analysis of reactivation/slow oscillations interactions).
Figure 5

A: Average reactivation strength for non-signal (Φ < 1) and signal components (grouped by encoding strength) during PRE and POST SWS for SPWR bins (left) and non-SPWR bins (right); error bars: S.E.M. B: Event Triggered Average (ETA) of reactivation strength centered on hippocampal SPWR for all analyzed signal components during PRE SWS (blue) and POST SWS (red) and for non-signal components (respectively black and gray), showing an increase around SPWRs during POST for signal components only; error bars: S.E.M; Grey bars indicate significantly higher reactivation strengths (p < 0.001, t-test) for signal components during POST SWS with respect to baseline (defined as the average reactivation strength from -4 s to -2 s and from 2 s to 4 s from delta wave peak). C: Expanded view of the central portion of C for POST SWS (time bins of 20 ms) with increased reactivation peak 40 ms after SPWRs. D: ETA of spiking probability density of multi-unit activity relative to SPWR occurrences and averaged over all recording sessions. grey: PRE SWS; black: POST SWS. No difference was observed between PRE and POST. E Autocorrelogram for POST SWS reactivation strengths (grey) and SPWR occurrences (black), showing similar decay time constants.
Figure 6. Reactivation strength relative to mPFC LFP events.

A: Cortical delta wave peaks. Top: ETA of signal components’ reactivation strength centered on cortical delta wave peaks, for PRE (grey) and POST SWS (black). Error bars: S.E.M.; grey bars indicate significantly higher reactivation strengths (p < 0.001, t-test) as defined in Fig. 5Bf. Reactivation had a significant peak preceding the delta wave. Middle: Cross-correlogram of SPWR occurrences relative to delta peaks. SPWRs tended to occur more frequently just before delta peaks, similarly to reactivation. Bottom: Spiking probability density of multi-unit activity relative to delta waves and averaged over all recording sessions. grey: PRE SWS; black: POST SWS. Prefrontal cells exhibited a strong decrease in firing at the time of the delta peak, preceded and followed by activity increases. B: Same plot as in A but centered on DOWN to UP transition (as defined by population average firing rate). Results are comparable with A except for spiking probability which only shows a dramatic deflection during the DOWN state . C: Same plots as in A but centered on spindle troughs. Top: Reactivation strength was significantly higher for over 1 s before spindles. Middle:, Hippocampal SPWR ETA centered on spindles showed a similar profile Bottom: Spike activity increased at the time of spindles. D: Comparison between time relationship of reactivation strength and SPWR occurrence relative to the three cortical events presented in A-C showing a high correlation in each cases.



Figure 7

Replayed activity during SPWRs is correlated to rule acquisition



A: The activation strength of principal components computed from POST SPWR activity is displayed in color peri-event rasters, for two examples of sessions with rule learning (in different rats). The maze was linearized and divided into 25 equal bins. Rewarded trials are marked with white dots at right. The arrows indicate the trial where learning is considered to be achieved. B: Incidences of PCs computed from PRE or POST SPWRs or inter-SPWR epochs during sessions with learning (n = 10, 4 rats). (black bar is after learning occurred in the session, grey bars are before). Significantly more POST SPWR activity patterns were positively correlated with activity after learning than with activity before learning (p < 0.05, two-ways ANOVA followed by t-test). No such difference was observed for PRE SPWR patterns or for inter-SPWR patterns, or when a comparison was made between the first and second half of non-learning sessions or when the rat switched between two erroneous, unrewarded strategies. Error bars: 95% confidence interval defined for multinomial distributions (see Methods). C: Average spatial distribution of all fifteen principal components computed from POST SPWR which were significantly higher on average after learning (black trace) than before (grey trace). The grey bar (above) marks significantly different bins (p < 0.05, t-test, n=15). Dashed lines displays SEM. D: Factorial analysis scores of data from C exhibits two profiles with peaks at the goal arms and the central choice point respectively (see also supplementary information).

SUPPLEMENTARY DISCUSSION

Memory replay and network spikes.

The transient co-activations which make up most of replay resemble network spikes, transient activations of groups of cells, which have been described in mathematical models of recurrent neural networks, in which dynamics is governed by synaptic facilitation and depression38. These events are an intrinsic feature of network dynamics in several neural systems, e.g. organotypic cultures49 and in neocortical slices50. From a theoretical viewpoint, network spikes may be triggered by small fluctuations in activity, activating a subset of cells encoded in the synaptic matrix, then terminating because of synaptic depression. Network spike firing may also, through synaptic facilitation, increase the likelihood of a successive activation of the same group of cells38 in a self-sustained, repetitive process, favoring synaptic plasticity and a more permanent encoding of that cell assembly in the connectivity matrix. Thus, by themselves, network spikes do not denote learning or information processing. However, they provide a mechanism for the expression of learning in neural activities: patterns of activity may be “tagged”, for example by synaptic facilitation, or long-term synaptic plasticity, and be more likely to re-emerge later under the form of a network spike. Moreover, several groups of cells likely to give rise to network spikes may co-exist, and activate at different moments during sleep. Conceivably, the sleep activity of the mPFC could be largely composed of network spikes, each involving a different cell group. Of these groups, our technique can detect but a few among those that are related to the immediately preceding experience. Another interesting point is that different reactivating cell groups are unlikely to be active at the same time, as shown by the cross-correlograms for reactivation strengths relative to different PCs (Supplementary Figure 10). This suggests that some sort of pattern separation mechanism (possibly through feedback inhibition), takes place specifically during sleep. This result does not contradict studies16,17 demonstrating the replay of neuronal ensemble activation in sequential order, spanning up to several hundreds of ms: our technique will search for the patterns of co-activation accounting for the largest fraction of the activity variance in the AWAKE epoch. With respect to sequences, two situations may arise. First the sequence is shorter than the bin size we used (100 ms), then it would be completely contained in one of our co-activation patterns. For longer sequences, or in any case sequences that are cut in two at the bin boundaries, nothing ensures that all parts of the sequence will be detected as co-activation patterns by our analysis, so that the sequential character of replay can be analyzed. In general this will not be the case, because here PCs are computed by AWAKE analysis without reference to precise, repetitive behavioral templates, as it is the case for analyses of sequential activation of ensembles. Thus, the present method and sequence-based methods highlight complementary aspects of sleep replay.

1   2


The database is protected by copyright ©dentisty.org 2016
send message

    Main page