02 - 1 - Wilson 1994 - Reactivation of hippocampal ensemble memories during sleep.pdf

Science IWaaas Reactivation of Hippocampal Ensemble Memories During Sleep Author(s): Matthew A. Wilson and Bruce L. McNaughton Source: Science , Jul. 29, 1994, New Series, Vol. 265, No. 5172 (Jul. 29, 1994), pp. 676-679 Published by: American Association for the Advancement of Science Stable URL: ht...

Science IWaaas Reactivation of Hippocampal Ensemble Memories During Sleep Author(s): Matthew A. Wilson and Bruce L. McNaughton Source: Science , Jul. 29, 1994, New Series, Vol. 265, No. 5172 (Jul. 29, 1994), pp. 676-679 Published by: American Association for the Advancement of Science Stable URL: https://www.jstor.org/stable/2884280 JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at https://about.jstor.org/terms American Association for the Advancement of Science is collaborating with JSTOR to digitize, preserve and extend access to Science JSTOR This content downloaded from 145.127.58.139 on Mon, 02 Oct 2023 07:16:32 +00:00 All use subject to https://about.jstor.org/terms (Fig. IB), which indicates that an intact RIC may be necessary for FKBP-12 binding. We also tested FKBP-12A5 (a clone of FKBP-12 lacking the first five amino acids) for interaction with a panel of type I receptor molecules (Fig. ID), all of which showed weakened interaction with FKBP-12A5. Thus, an intact NH2-terminus of FKBP-12 is required for optimal interaction. Co-immunoprecipitation was used to con¬ firm the specific interaction between FKBP-12 and R4C in vitro. We used fusion proteins of LexA-R4C or LexA-RlC-JM (the juxtamembrane domain of the RIC) (Fig. IC, lane 2) made in yeast and lysates of mink lung cells, which are a rich source of FKBP-12 as con¬ firmed by protein immunoblotting (19). FKBP-12 was coprecipitated with LexA-R4C but not with LexA-RlC-JM (Fig. 2). Two immunosuppressant drugs, FK506 and rapamycin, bind to the same site on FKBP-12 (22). We therefore investigated whether the type I receptors and the drugs share the same binding site on FKBP-12. If the binding sites are the same or partially overlap, drug binding to FKBP-12 would com¬ pete with the cytoplasmic domain of the type I receptors. This would result in a decrease of the P-galactosidase activity in the yeast trans¬ formants. We added FK506 to a liquid culture of the yeast transformants at two concentra¬ tions that did not affect yeast growth. FK506 effectively competed with R4C for FKBP-12 binding at a concentration of 1 p,M (Fig. 3), but not at 100 nM. When cyclosporin was used at various concentrations in the same assay, no competition was observed. To fur¬ ther confirm that the type I receptors and the drugs share the same binding site on FKBP12, we tested in the two-hybrid system a mutant FKBP-12 (D37G) (18), which was shown to be defective in binding to FK506 and rapamycin. It failed to interact with the type I receptors (19). Our results suggest that FKBP-12 inter¬ acts with type I receptors in a specific manner. A K230R mutation on R4, which abolishes R4 signaling activity (12), de¬ creased binding of R4C to FKBP-12, as did a D37G mutation on FKBP-12, which sug¬ gests that the interaction may be function¬ ally important. In mammalian cells, ligand binding may promote the type I receptors to bind FKBP-12; alternatively, it may stabi¬ lize a preexisting type I receptor-FKBP-12 complex. Although the function of FKBP12 in the type I receptor-mediated signal¬ ing pathway needs to be clarified in TGFP-responsive cell lines, our data indicate that the binding sites on FKBP-12 for R4C and FK506 may be shared or overlap, which suggests that the type I receptor may be a natural ligand for FKBP-12. Because the competition of FK506 for R4C to bind FKBP-12 can be easily monitored, this yeast system provides a potential screen for other 676 candidate immunosuppressant drugs. If these interactions are confirmed to occur in mammalian cells, rapamycin and FK506 may prove useful in analyzing the down¬ stream actions of TGF-p family ligands. REFERENCES AND NOTES 1. J. L. Wrana eta/., Ce//71, 1003 (1992). 2. H. Y. Lin et al., ibid. 68, 1 (1992); L. S. Mathews and W. W. Vale, ibid. 65, 973 (1991); L. L. Georgi etal., ibid. 61, 635 (1990). 3. H. Y. Lin and H. F. Lodish, Trends Cell Biol. 3, 14 (1993). 4. W. W. He etal., Dev. Dynam. 196, 133 (1993). 5. L. Attisano etal., Cell75, 671 (1993). 6. R. Ebner et al., Science 262, 900 (1993). 7. P. Frazen etal., Cell 75, 681 (1993). 8. K. Matsuzaki ef al., J. Biol. Chem. 268, 12719 (1993). 9. P. ten Dijke et al., Science 264, 101 (1994). 10. K. Tsuchida, L. S. Matthews, W. W. Vale, Proc. Natl. Acad. Sci. U.S.A. 90, 11242 (1993). 11. Y. Yamazaki, T. Saito, T. Nohno, DNA Sequence 3,' 297 (1993). 12. C.-H. Bassing etal., Science263, 87 (1994). 13. R.-H. Chen etal., ibid. 260, 1335 (1993). 14. J. Gyuris etal., Cell 75, 791 (1993). 15. S. Fields and O.-K. Song, Nature 340, 245 (1989). 16. A. S. Zervos etal., Cell 72, 223 (1993). 17. R1C (residues 147 to 509) was amplified with the polymerase chain reaction (PCR) with Eco RI and Bam HI attached at the 5’ and 3' ends, respec¬ tively, and subsequently inserted into the Eco RI and Bam HI sites of the multicloning region of PEG202 (14), with the LexA DNA binding domain fused 5’ to R1C. 18. Abbreviations for the amino acid residues are: D, Asp; G, Gly; K, Lys; P, Pro; R, Arg; and S, Ser. Arg, Mutations are indicated as follows: Lys230 19. T. Wang, P. K. Donahoe, A. S. Zervos, unpub¬ lished results. 20. The entire R4C (residues 147 to 501) was ampli¬ fied by PCR and then subcloned into the Eco RI and Xho I sites of PEG202. 21 . The constructs were generated by PCR and sub¬ cloned as described for LexA-R1C (17), except for the LexA-aRIIC construct. DNA sequencing and protein immunoblotting confirmed the ex¬ pression of the fusion proteins. 22. G. D. VanDuyne, R. F. Standaert, S. L. Schreiber, J. Clardy, J. Am. Chem. Soc. 113, 7433 (1991). 23. C. Wittenberg and S. I. Reed, Cell 54, 1061 (1988). 24. The neonatal rat heart cDNA library was from A. Green (Beth Israel Hospital, Boston). An R4 K230R pCMV6 plasmid was provided by X.-F. Wang. The LexA-aRIIC construct was from R. Perlman and R. A. Weinberg (Whitehead Insti¬ tute). We thank T. Haqq for technical assistance; R. Brent for encouragement, support, and valu¬ able suggestions; T. E. Starzl and W. W. He for helpful discussions; A. Green for the rat neonatal cDNA library; and X.-F. Wang, R. Perlman, R. A. Weinberg, and B. Bierer for materials critical for this study. We particularly thank S. Schreiber for supplying FK506 and antibody to FKBP-12; B. Gladstone for the FKBP-12 mutant; R. Finley, P. Baciu, M. Kozlowski, A. Vincent, P. Busto, H. Y. Lin, and J. Schnitzer for technical advice; and R. Brent, P. Goetink, S. Schreiber, J. Avruch, L. Perkins, D. MacLaughlin, and J. Teixeira for com¬ ments on the manuscript. Supported by a Nation¬ al Institute of Child Health and Human Develop¬ ment (NICHD) Reproductive Sciences Training Grant (P-32 HD07396) (T.W.) and by grant CA1 7393 from the National Cancer Institute and by P30 Reproductive Sciences grant (P-30 HD28138) from NICHD (P.K.D.). A.S.Z. was fund¬ ed under an agreement between Massachusetts General Hospital and Shiseido. 31 March 1994; accepted 24 June 1994 K230R. Reactivation of Hippocampal Ensemble Memories During Sleep Matthew A. Wilson* and Bruce L. McNaughton Simultaneous recordings were made from large ensembles of hippocampal “place cells” in three rats during spatial behavioral tasks and in slow-wave sleep preceding and following these behaviors. Cells that fired together when the animal occupied particular locations in the environment exhibited an increased tendency to fire together during subsequent sleep, in comparison to sleep episodes preceding the behavioral tasks. Cells that were inactive during behavior, or that were active but had non-overlapping spatial firing, did not show this increase. This effect, which declined gradually during each post-behavior sleep ses¬ sion, may result from synaptic modification during waking experience. Information acquired during active behavior is thus re-expressed in hippocampal circuits during sleep, as postulated by some theories of memory consolidation. The selective strengthening of interac¬ tions among small sets of neurons engaged in the encoding of specific external events has been a foundation of modern theories of neural information storage. Yet, despite indirect evidence that changes in synaptic Division of Neural Systems, Memory, and Aging and Department of Psychology, University of Arizona, Tuc¬ son, AZ 85724, USA. •Present address: Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, MA 02139, USA. SCIENCE VOL. 265 29 JULY 1994 This content downloaded from 145.127.58.139 on Mon, 02 Oct 2023 07:16:32 +00:00 All use subject to https://about.jstor.org/terms efficacy may be the basis of learning within the hippocampus (1, 2), there has been no demonstration of changes in functional interaction among hippocampal cells spe¬ cific to the representation of a given expe¬ rience (3). Recent experiments with par¬ allel recording methods, however, have revealed rapid changes in neuronal ensem¬ ble codes for space within the hippocam¬ pus during exposure to a novel environ¬ ment (4). Also, Pavlides and Winson (5) previously demonstrated an increase in Reports firing rate, during sleep, of those hippo¬ campal cells that had been active during the prior waking period. Although this finding is suggestive of mnemonic opera¬ tions, without knowledge of the distribut¬ ed structure of the activity across the Flg. 1. Spatial firing characteristics of popu¬ lations of simultaneous¬ ly recorded hippocam¬ pal pyramidal cells from three rats (rat 1: 49 cells; rat 2: 74 cells; rat 3: 69 cells). Each panel represents the average firing rate of a particular cell as a function of lo¬ cation in the recording apparatus. The colored areas represent visited regions in the appara¬ tus, with red indicating high firing rate (—10 Hz) and blue indicating no firing. (Several of the si¬ lent cells from each rat are not shown, due to space constraints.) neuronal population it cannot be deter¬ mined whether this activity is organized into coherent representations of the pre¬ ceding experience (memories). Here we address the question of whether information acquired during periods of ac¬ muuummm mummumn umuummm mmnmunm tive behavior results in the modification of neuronal circuits within the brain and whether the reactivation of such memory traces can be detected in the hippocampus during sleep. In principle, this question can be studied by the examination of statistical pair-wise interactions between neuronal spike trains. However, because the expect¬ ed number of cells involved in encoding any particular event is small (4, 6), the probability of observing and quantifying such interactions in small samples of neu¬ rons is remote. To overcome this problem, we have studied the activity correlations among large sets of simultaneously recorded cells (4), because the number of possible interactions that can be observed increases as the square of the sample size. manaimi lEiinmm Fig. 2. Representative cross-corre¬ lation histograms from cell pairs with spatially overlapped (lower left) and non-overlapped (lower right) firing during the RUN phase (data from rat 2; color code as in Fig. 1). Periodicity during the RUN phase reflects theta rhythm modu¬ lation (9). This modulation is ab¬ sent during the PRE and POST sleep periods. Comparison of the central peaks (100 ms) of the PRE and POST correlation histograms reveals an increase in coactivity for the cell pair (left column), which was coactive during running (cen¬ ter left histogram) because of spa¬ tially overlapped firing. This in¬ crease does not appear to cells that are active during running but are not coactive (right column). The y-axis gives correlation in terms of the number of spike pairs that occurred at the specified time lag. The bin size is 10 ms. Non-overlapping Overlapping fields SCIENCE • VOL. 265 • 29 JULY 1994 This content downloaded from 145.127.58.139 on Mon, 02 Oct 2023 07:16:32 +00:00 All use subject to https://about.jstor.org/terms Fig. 3. Mean cross-correlations (±100 ms) for cell pairs during PRE, RUN, and POST phases (Fig. 2). Only cells with significant spatial firing on the apparatus are included (10). Cell pairs were grouped on the basis of the amount of overlap between their spatial firing distribu¬ tions. Cells that were correlated (OVR, solid bars) during the RUN phase exhibited a signif¬ icant (analysis of variance, P < 0.05) increase in correlation during the post-run (POST) sleep relative to pre-run (PRE) sleep (♦) or to cells that were non-overlapped (NON, hollow bars), and hence uncorrelated during running (plus signs). Correlations between cells that were active during sleep but inactive during RUN did not change between POST and PRE and were not different from the average PRE values of the cells active during RUN. The numbers of cell pairs contributing to each histogram were as follows: (A) Rat 1, NON 437 and OVR 60; (B) Rat 2, NON 719 and OVR 95; (C) Rat 3, NON 941 and OVR 147. 677 Rats were surgically implanted with mi¬ crodrive arrays and then trained on one of two food-reinforced, spatial behavioral tasks (7). With the use of multielectrode record¬ ing techniques (4), the activity of 50 to 100 single cells in area CAI of the hippocampus was monitored simultaneously during sleep periods before and after brief behavioral episodes (Fig. 1). Spike-train cross-correla¬ tions were computed for all pairs of principal cells during the pre-behavioral sleep (PRE) , running (RUN), and post-behavioral sleep (POST) periods. For each cell pair, the short latency (±100 ms) cross-corre¬ lation of spike trains was computed over each period (8); each period was approxi¬ mately 20 min in duration (9). The cor¬ relation pairs were sorted by the amount of spatially overlapped activity between cells in the pair (10), as determined by the spatial firing characteristics (“place fields”) of these cells during the running phase (RUN) (Fig. 2). During the RUN phase, as expected, cells with overlapping place fields exhibited highly positively cor¬ related activity, whereas cell pairs with nonoverlapping fields did not exhibit such activity (Fig. 3). On the hypothesis that hippocampal activity during sleep reflects the reactivation of population events rep¬ resenting the experiences of the prior wak¬ ing behavior, cell pairs that were coactive during the RUN phase (because of their place field overlap) were examined during the POST phase and compared with cell pairs that were active but not coactive during behavior (non-overlapping place fields) (Figs. 2 and 3). Cells that were coactive during behavior showed a signif¬ icant increase in correlation from their PRE level compared to pairs in which both cells displayed spatial firing but not in the same region of space (fl). This selective reactivation of correlated states declined during the POST phase with a time con¬ stant of approximately 12 min (12). The increased correlation is consistent with an underlying associative synaptic mod¬ ification at some stage in the system, although not necessarily in the hippocampus. Indeed, there is little direct connectivity among pyra¬ midal cells within CAI, and, therefore, the emergence of experience-specific correlated states in this region is likely to arise from changes in commorj inputs, such as those from area CA3 (which has extensive intrinsic connectivity) or entorhinal cortex. Changes in neocortical or subcortical pathways might underlie the increased correlations. However, on the basis of current understanding of the physiological dynamics of the hippocampus during slow- Fig. 4. Diagram of the effective connectivity matrix of a network of 42 cells (selected at random) from rat 2. Individual cells are represented as dots around the perimeter of a circle. Lines indicate a positive correlation between the pair, with color reflecting the magnitude of the correlation [red, high (0.2); blue, low (0.002)]. The lower three panels show all positive correlations for each of the PRE, RUN, and POST conditions and illustrate the dramatic contrast in overall correlation structure between slow-wave sleep and active behavior. The upper three panels show a subset of the same data having positive correlations above 0.05. Bold lines indicate cell pairs that were correlated during RUN and also correlated during either PRE or POST. These panels reveal that most of the highly correlated pairs that appear during the run phase also appear in the POST phase but are typically absent from the PRE phase. 678 SCIENCE • VOL. 265 • 29 JULY 1994 This content downloaded from 145.127.58.139 on Mon, 02 Oct 2023 07:16:32 +00:00 All use subject to https://about.jstor.org/terms wave sleep, it is possible to construct a plausible hypothesis for a hippocampal origin of the phenomenon. During behav¬ ioral inactivity and slow-wave sleep, the hippocampus displays a significant change in its electrical activity compared either to active waking behavior or to paradoxical rapid eye movement (REM) sleep. During slow-wave sleep, cells discharge in intermit¬ tent, synchronized bursts (ripples) (9), asso¬ ciated with electroencephalogram (EEG) sharp-wave (SPW) activity. It has been proposed that information transfer to neocor¬ tex occurs during sleep in general (5, 6) and particularly during the synchronized bursts (13). To investigate this hypothesis, corre¬ lation effects were examined both during ripples and during the intervals between them. Correlations were significantly larger during ripples than in the intervening inter¬ vals (14). This difference is consistent with previous theoretical models of autoassociative memory (6), in which recurrent exci¬ tation is used to reproduce a previously stored activity pattern, even put contains very little of when the in¬ the relevant information. In such networks, even ran¬ dom background noise can occasionally lead to the emergence of correlated states, manifested as population bursts (ripples), reflecting stored patterns. The self-termi¬ nating nature of these bursts (9) would then reset the network to it’s “ground state” and allow the process to start anew. Consistent with this notion, recent physiological stud¬ ies (15) have shown that SPWs and ripples are initiated in CA3, and that the output layers of entorhinal cortex, but not the input layers, exhibit neuronal activity cor¬ related with CAI SPWs. This relation sug¬ gests that the induced correlations during SPWs arise from modifications within the hippocampus itself and are propagated to the output layers of entorhinal cortex. The foregoing observations demon¬ strate that, during sleep, the activity of hippocampal cells engaged in place-specif¬ ic activity during prior waking (5) retains the structure of distributed representations of the visited locations (Fig. 4). Accord¬ ing to the proposed hypothesis, initial storage of event memory occurs through rapid synaptic modification, primarily within the hippocampus. During subse¬ quent slow-wave sleep, synaptic modifica¬ tion within the hippocampus itself is sup¬ pressed (16) and the neuronal states en¬ coded within the hippocampus are “played back” as part of a consolidation process by which hippocampal information is gradu¬ ally transferred to, the neocortex (6, 17). REFERENCES AND NOTES 1. T. V. P. Bliss and A. R. Gardner-Medwin, J. Physiol. 232, 357 (1973); C. A. Barnes and B. L. McNaughton, Behav. Neurosci. 99, 1040 (1985); Reports T. J. Teyler and P. Discenna, Annu. Rev. Neurosci. 10, 131 (1987); R. G. M. Morris, S. Davis, S. P. Butcher, Philos. Trans. R. Soc. London Ser. B 329, 187 (1990); B. L. McNaughton and R. G. M. Morris, Trends Neurosci. 10, 408 (1987); P, W. Landfield and S. A. Deadwyler, Eds., Long-Term Potentiation: From Biophysics to Behavior. (A, R. Liss, New York, 1987). 2. B. L. McNaughton, R. M. Douglas, G. V. Goddard, Brain Res. 157, 277 (1978). 3. Coactivity (simultaneous or near-simultaneous activityj-dependent synaptic change was proposed as a basis for memory by D. 0. Hebb [The Organization of Behavior (Wiley, New York, 1949)]. Since then, Hebbian synaptic plasticity has been identified in the mammalian central nervous system (2), and conditioning studies indicate that responsiveness of single cells can be altered selectively by repeated stimulus pair¬ ings [C. D. Woody and J. Engel Jr., J. Neurophys¬ iol. 35, 230 (1972)]. Such changes can be depen¬ dent on behavioral state [E. Ahissar etal., Science 257, 1412 (1992)]. This dependence suggests that persistent, rapidly induced changes in func¬ tional connectivity between hippocampal cells might be induced through coactivity of these cells during behavior. 4. M. A. Wilson and B. L. McNaughton, Science 261, 1055 (1993). 5. C. Pavlides and J. Winson, J. Neurosci. 9, 2907 (1989). 6. D. Marr, Philos. Trans. R. Soc. London Ser. 8262, 23 (1971); D. J. Amit and A. Treves, Proc. Natl. Acad. Sci. U.S.A. 86, 7871 (1989); B. L. Mc¬ Naughton and L. Nadel, in Neuroscience and Connectionist Theory, M. A. Gluck and D. Rummelhart, Eds. (Lawrence Earlbaum Associates, Hillsdale, NJ, 1990); A. Treves and E. T. Rolls, Network 2, 371 (1991); W. G. Gibson and J. Robinson, Neural Networks 5, 645 (1992). 7. In one task, the rat searched steadily within an enclosed box (62 by 62 cm; rats 1 and 3) for a randomly scattered food reward. In the second (spatial working-memory) task, an elongated, X-shaped, four-arm track (167 by 25 cm; rat 2) was used with two adjacent arms designated as start and the opposite two arms designated as goals. The correct goal arm was a function of the randomly selected start arm. Animals were male Fisher 344 rats, approximately 300 g and 9 months of age. All surgical procedures were carried out according to NIH guidelines. 8. Both principal cells (excitatory pyramidal cells) and inhibitory interneurons were recorded during these sessions. Interneurons, identified on the basis of wave shape, firing rate, and spike interval characteristics, were not included in the study. Cross-correlations were normalized by spike counts [D. H. Perkel, G. L. Gerstein, G. P. Moore, Biophys. J. 7, 419 (1967); G. L. Gerstein and D. H. Perkel, Science 164, 828 (1969)]. 9. Hippocampal EEGs from eight of the unit record¬ ing sites were continuously monitored. Sleep phases were predominantly characterized by in¬ termittent SPW and ripple activity [J. B. Ranck Jr., Exp. Neurol. 41, 461 (1973); J. O’Keefe and L. Nadel. The Hippocampus as a Cognitive Map (Oxford Univ. Press, Oxford, 1978), pp. 150-153; G. Buzsaki, Brain Res. 398, 242 (1986)] with population bursts of variable duration (100- to 300-Hz band EEG, mean duration 74 m, mean inter-burst interval 1.7 s) with little or no REM sleep. The behavioral phase was dominated by theta activity (6- to 9-Hz modulation of EEG and unit discharge), which has been correlated with locomotion [C. H. Vanderwolf, Electroencephalogr. Clin. Neurophysiol. 26, 407 (1969)]. 10. Hippocampal neurons exhibit robust selectivity for spatial location. The preferred location for a given cell is called its “place field" (J. O'Keefe and J. Dostrovsky, Brain Res. 34, 171 (1971)]. For each cell pair in which both members exhibited significant spatially related firing in the apparatus, the distance between the locations of their peak firing was used as a measure of overlap (Fig. 1). The criterion for overlap was a distance between peaks of <16 cm (approximately the diameter of an average place field). To eliminate any possibility of spurious over¬ lap due to incomplete isolation of single units on a given probe, cell pairs taken from the same probe were eliminated from the analysis. The mean firing rates of cells in coactive cell pairs were the same as those for non-coactive pairs. Thus, the increased correlations were not due to firing rates per se. 11. Figure 2 reveals a tendency for pairs that were negatively correlated (not overlapping) during be¬ havior (see animals 1 and 2 in Fig. 3) to result in reduced correlation during the POST phase. This effect was small and statistically significant only for rat 1. 12. The time constant tor decay of correlation was estimated by dividing the initial 10 min of the POST sleep phase into two 5-min periods and computing the mean correlation for overlapping cell pairs in each period. A third point obtained from the mean correlation -of the PRE phase was assumed to be the asymptotic base line for the decay. The mean time constant was determined by the fitting of a single exponential to these three points. Rat 1 latency to sleep onset was 20 min, sleep duration was 18 min, and the time constant estimate was 15 min. Rat 2 latency was 10 min, duration was 20 min, and the time constant was 9 min. Rat 3 latency was 8 min, duration was 10 min, and the time constant was 13 min. 13. B. L. McNaughton, in The Neurobiology of the 14. 15. 16. 17. 18. Hippocampus, W. Seifert, Ed. (Academic Press, London, 1983), pp. 609-610; G. Buzsaki, Neuro¬ science 31, 551 (1989). For rat 1, cell overlap correlation during ripples was 0.017 ± 0.003 SEM versus 0.005 ± 0.003 (P < 0.01) for non-ripples. For rat 2, correlation was 0.069 ± 0.008 for ripples and 0.011 ± 0.004 (P < 0.01) for non-ripples. For rat 3, appropriate EEG information was not available. J. J. Chrobak and G. Buzsaki, J. Neurosci., in press. B. J. Leonard, B. L. McNaughton, C. A. Barnes, Brain Res. 425, 174 (1987). J. J. Kim and M. S. Fanselow, Science 256, 675 (1992); J. L. McClelland, B. L. McNaughton, R. O'Reilly, Soc. Neurosci. Abstr. 18, 1216 (1992); J. L. McGaugh and M. J. Hertz, Memory Consolida¬ tion (Albion, San Francisco, 1972); L. R. Squire, N. J. Cohen, L. Nadel, in Memory Consolidation, H. Weingartner and E. Parker, Eds. (Earlbaum Press, Hillsdale, NJ, 1984), p. 185. We thank C. A. Barnes, L. Nadel, and J. L. McClelland for helpful comments on the manu¬ script. Supported by grant MH46823 from the National Institute of Mental Health (B.L.M.), the Office of Naval Research (B.L.M.), NSF grant 901449 (M.A.W.), and the McDonnell-Pew Cogni¬ tive Neuroscience Program. 14 February 1994; accepted 9 May 1994 Dependence on REM Sleep of Overnight Improvement of a Perceptual Skill Avi Kami,* David Tanne, Barton S. Rubenstein, Jean J. M. Askenasy, Dov Sagi Several paradigms of perceptual learning suggest that practice can trigger long-term, expe¬ rience-dependent changes in the adult visual system of humans. As shown here, performance of a basic visual discrimination task improved after a normal night’s sleep. Selective disruption of rapid eye movement (REM) sleepresulted in no performance gainduring a comparable sleep interval, although non-REM slow-wave sleep disruption did not affect improvement. On the other hand, deprivation of REM sleep had no detrimental effects on the performance of a similar, but previously learned, task. These results indicate that a process of human memory consol¬ idation, active during sleep, is strongly dependent on REM sleep. — Perceptual learning the improvement of perceptual skills through practice is a type of human learning that may serve as a paradigm for the acquisition and retention of procedural knowledge, “habits,” or “how to” memories (1). Recent results suggest that when observers practice a simple tex¬ ture discrimination task the large and con¬ sistent improvements that occur over the course of several consecutive daily sessions — A. Kami, Department of Neurobiology, Brain Re¬ search, Weizmann Institute of Science, Rehovot 76100, Israel, and Department of Neurology and Os¬ car Sager Sleep Research Lab, Chaim Sheba Medical Center, Tel Hashomer 52621, Israel. D. Tanne and J. J. M. Askenasy, Department of Neurology and Oscar Sager Sleep Research Lab, Chaim Sheba Medical Center, Tel Hashomer 52621, Israel. B. S. Rubenstein and D. Sagi, Department of Neuro¬ biology, Brain Research, Weizmann Institute of Sci¬ ence, Rehovot 76100, Israel. *To whom correspondence should be addressed at Laboratory of Neuropsychology, National Institute of Mental Health, Bethesda, MD 20892, USA. SCIENCE • VOL. 265 • 29 JULY 1994 This content downloaded from 145.127.58.139 on Mon, 02 Oct 2023 07:16:32 +00:00 All use subject to https://about.jstor.org/terms are subserved by discrete changes depen¬ dent on retinal input and within an early stage in the stream of visual processing (2). Psychophysical data implicate neuronal mechanisms of figure-ground segmentation at a stage in the processing pathway as early as the primary visual cortex in mediating (by becoming more efficient and faster) the learning of this basic visual skill (2, 3). These results, as well as results from several other perceptual learning paradigms (4), suggest that different levels of visual pro¬ cessing may, under specific retinal input and task-defined conditions, undergo long¬ term, experience-dependent changes (func¬ tional plasticity) (5) . Recently, we and others have found that an improvement in perceptual performance occurs neither during nor immediately after practice but rather 8 to 10 hours after a training session has ended, suggesting a slow, latent process of learning (6). As the im¬ proved visual skills were not forgotten even 679