R EPORTS

temperature of 51°C, and 40 cycles on a PTC-200 Rc_ubiG1, 5Ј-CTTGTCACTGTCTAATTCTTC-3Ј; Rc_u- region of the DnaE gene in R. felis (9). There is no thermocycler (MJ Research). A positive control was biG2, 5Ј-AAATATTAGACGTCGGTTGC-3Ј; Rc_ubiG3, significant similarity between this RS3-like sequence then performed under the same PCR condition with 5Ј-CAGATTGACGCACAAATTAC-3Ј; and Rc_ubiG4, and our newly described RPE sequence. genomic DNA as template. All primer pairs gener- 5Ј-AATAAACCCTATTCGCCTTG-3Ј. 24. Reviewed in F. B. Perler, Nucleic Acids Res. 27, 346 ated the expected amplicons. RT-PCR were then 20. Amino acid sequence identities for the repeat-de- (1999). performed on the RNA extract with a different rived peptides (and the rest of the proteins) are 43% 25. We examined the following structures whose ho- thermocycler. All primer pairs generated amplicons (85%) for KdtA, 31% (83%) for RP545, and 82% mologs are known to contain inteins: 1TGO, 1AB4, of the expected sizes and sequences. (95%) for PolA. 1ECL, 2REC, 1RLR, and 1SKY. 18. For the location and the designation of selected 21. R. G. Lloyd and G. J. Sharples, Nucleic Acids Res. 18, primers, see Web fig. 4 (11). 26. V. Monchois, personal communication. 6503 (1990). 19. Primer sequences are as follows: Rc_gltX1, 5Ј-TCTT- 27. J. J. Muller, Mutat. Res. 1, 2 (1964). TGGAATCTTGGCTGTAG-3Ј; Rc_gltX2, 5Ј-GATTAG- 22. S. Bachellier, J.-M. Cle´ment, M. Hofnung, Res. Micro- 28. H. Tao, C. Bausch, C. Richmond, F. R. Blattner, T. TGAGCAGGAAATA-3Ј; Rc_gltX3, 5Ј-GGATTGACGT- biol. 150, 627 (1999). Conway, J. Bacteriol. 181, 6425 (1999). Ј Ј 23. As revealed by a partial, low-significance match be- ACAAATTAC-3 ; Rc_gltX4, 5 -CAAAGACTGTAGAG- 29. We thank C. Abergel, D. Gautheret, and J. Weissen- Ј Ј tween RSA and 23 amino acid residues at the 3Ј end ATATTGG-3 ; Rc_hemC1, 5 -TACAGATAGCTTCCAA- bach for their help at various stages of this work. CATC-3Ј; Rc_hemC2, 5Ј-CAAACCAATTTTATGCTC of a Salmonella typhimurium hypothetical ORF GG-3Ј; Rc_hemC3, 5Ј-CGGATTGACGTACAAATTAC- (SwissProt accession number P16656). A RS3-like Ј Ј Ј 3 ; Rc_hemC4, 5 -GTATCTAGATGCTAATTGCC-3 ; inverted repeat is also present in the NH2-terminal 7 April 2000; accepted 24 August 2000

as based on procedural learning (which is Replaying the Game: preserved in ), the amnesic group demonstrated impoverished learning in Hypnagogic Images in Normals comparison with normal novices, suggest- ing that Tetris proficiency may also depend on declarative systems. This inter- and Amnesics pretation is supported by the finding that Robert Stickgold,1* April Malia,1 Denise Maguire,1 left and right hippocampal glucose utiliza- David Roddenberry,1 Margaret O’Connor2 tion during Tetris play decreases during training in proportion to improvement in Participants playing the computer game Tetris reported intrusive, stereotypical, performance, suggesting a shift away from visual images of the game at onset. Three amnesic patients with extensive hippocampal dependence as learning pla- bilateral medial temporal lobe damage produced similar hypnagogic reports teaus (22). despite being unable to playing the game, suggesting that such imagery Three of the five patients produced a may arise without important contribution from the declarative memory sys- total of eight reports of Tetris imagery tem. In addition, control participants reported images from previously played across three nights. These reports account- versions of the game, demonstrating that remote can influence the ed for 7.4% of all hypnagogic reports col- images from recent waking experience. lected, similar to that seen in normal par- ticipants (7.2%; Fig. 2A) (23). In contrast, People who engage in novel physical or men- (9–12). Studies of the role of sleep in there was only a single report of a thought tal activities for extended periods of time learning and memory have generally ig- about Tetris without clearly associated im- often experience a hallucinatory replay of the nored the hypnagogic period, focusing in- agery (Ͻ1%; Fig. 2B). As with the normal activities as they fall asleep, a phenomenon stead on possible roles of REM sleep and participants, the visual imagery reported noted in both literary and scientific sources deep, slow wave sleep (13–16). was highly stereotyped (24). (1–3). Although the origin of these images in We asked 27 participants to play 7 hours Normal controls without prior experi- waking experience appears clear, the source of the computer game Tetris (17) over 3 days, ence playing Tetris had first game scores within the brain and the function of such 2 hours at the first exposure and 1 hour each averaging 786, significantly higher than imagery are uncertain. If these images are subsequent morning and evening (18). Three that of the amnesics (25). They showed mediated by declarative memory systems, groups of participants were studied: 12 nov- considerable improvement in performance amnesic patients with bilateral hippocampal ices with no prior Tetris experience, 10 ex- over the 3 days and six sessions (Fig. 1) lesions should not experience them. Because perts with considerable Tetris experience, (26). The appearance of hypnagogic Tetris amnesics are capable of reporting ex- and 5 amnesics with extensive, bilateral me- images in novices was very similar to that periences (4, 5), the question can be mean- dial temporal lobe damage (19). Participants seen with amnesics. Nine of 12 Tetris nov- ingfully addressed. were repetitively prompted for mentation re- ices produced a total of 19 reports of Tetris Most studies of dreaming involve re- ports during the first hour of attempted sleep imagery over two nights of recording, a ports of mental activity after forced awak- (20). report rate very similar to that of the am- enings from rapid eye movement (REM) Even though the five amnesic patients nesics (Fig. 2A). The imagery described sleep (6, 7) or after spontaneous awaken- were of average intelligence, they were un- showed the same highly stereotyped con- ings in the middle of the night or morning able to learn and retain new episodic infor- tent as seen with the amnesics (27). Nov- (8). Hypnagogic mentation represents an- mation regardless of modality of presentation ices produced 12 reports of thoughts of other source of sleep-state mental imagery or the nature of the material. The perfor- Tetris without imagery (Fig. 2B). Such re- mance of the amnesic patients showed only ports, clearly linked to declarative memo- 1Laboratory of Neurophysiology, Department of Psy- minimal, albeit significant, improvement ries, were virtually absent in the amnesics. chiatry, Harvard Medical School, 74 Fenwood Road, over 7 hours of play (Fig. 1). The mean score Ninety percent of the reports with images 2 Boston, MA 02115, USA. Memory Disorders Research increased from 537 points on participants’ from these novices (17 out of 19) were ob- Center, Boston University Medical School, Division of Behavioral Neurology, Beth Israel Deaconess Medical first game to a first session average of 651. tained on the second night. In contrast, two- Center, Boston, MA 02118 USA. Over the next five sessions, the scores in- thirds of the reports of thoughts alone (8 out *To whom correspondence should be addressed. E- creased by 36% to 884 points per game of 12) occurred on the first night (Fig. 2). mail: [email protected] (21). Although game playing is often seen Although this difference in distributions is

350 13 OCTOBER 2000 VOL 290 SCIENCE www.sciencemag.org R EPORTS highly significant (28), further studies will be Identifying the sources of memories third night. These images, both conceivably needed to confirm the finding. Nevertheless, used in dream construction is normally con- sleep onset REM reports, were more complex the distribution of images is not what would tingent on the investigator’s subjective in- and bizarre than the others in our sample be expected if the images were the product of terpretation of obscure symbolic transfor- (34). either demand characteristics or a simple and mations of the original memory or on the The questions of how material is selected automatic replay of recent, repetitive, sensory dreamer’s own subjective evaluation. In the for and then incorporated into and stimuli. case of hypnagogic reports of Tetris imag- what function this entire process serves re- Normal participants with extensive ex- es, no such subjective interpretations are main largely unresolved. But the data pre- perience playing Tetris, but who in general needed. Only reports that explicitly de- sented here, looking at sleep onset hypnago- had no recent practice (29), produced high scribed images or thoughts related to Tetris gic imagery, provide some insight into these scores from the start and showed no signif- were counted. Indeed, what was most strik- issues. icant improvement over the six sessions ing about the data was the strong similarity Out of 27 participants, 17 (63%) report- (Fig. 1) (30). Nevertheless, they also pro- in reports from different individuals. In all ed stereotypical hypnagogic images of Te- duced hypnagogic Tetris imagery. Over the cases, reported images included seeing tris during the first nights of play, imagery three nights, 5 of the 10 experts produced Tetris pieces, usually falling in front of clearly induced by the earlier Tetris play. 11 reports of hypnagogic Tetris imagery one’s eyes and sometimes rotating or fitting Unlike the images of REM sleep dreams, (4.7% of 235 reports), as well as 13 reports into empty spaces at the bottom of the these images were relatively accurate rep- of thoughts without images (5.5% of 235 screen. On rare occasions, participants also resentations of the actual visual imagery reports). As with the other groups, the im- reported seeing completed lines disappear- perceived during play of the game, albeit agery reported was stereotyped (31). In ing. Among the 38 reports of imagery, there abstracted, with removal of nonessential addition to these stereotyped images, two were no reports of seeing the picture sur- game elements. This highly consistent im- of the five experts who reported Tetris rounding the play window, the scoreboard, agery across participants and groups sug- imagery described images from earlier ver- or the keyboard or of typing on the key- gests that image construction in all groups sions of the game, which they had not board, and there were only two reports of extracted and abstracted the most salient played in the last 1 or 5 years, respectively seeing the computer itself. In none of these aspects of the experience by a similar pro- (32). In these cases, the image construction reports did the imagery appear bizarre. cess. This, along with the report rate from process followed associative links and in- None appeared in inappropriate settings or amnesic participants, argues strongly corporated older memories in preference to associated with inappropriate thoughts. against demand characteristics playing a images from recent play. In addition to collecting hypnagogic re- major role in the production of reports of Normal participants (novices and ex- ports at sleep onset, participants were also Tetris imagery (35). perts combined) showed significantly dif- instructed to report any thoughts, images, or The apparent 24-hour delay between the ferent patterns for reports of images and feelings about Tetris that occurred at other start of play and the subsequent appearance thoughts across the sleep onset period (Fig. times. Seven of 12 novices and 4 of 10 of game images in the reports of novices 3). After an initial 2.5-fold increase from experts reported a total of 44 images of Tetris suggests that the rules for selection of im- presleep to 15 s into sleep (4.5% compared during waking. As with hypnagogic images, ages for incorporation at sleep onset in- with 11.3%), the frequency of Tetris imag- the vast majority of daytime images occurred volves a complex algorithm. Similar lags es then dropped in half (to 5.7%) after just only after a 24-hour delay, with only four have also been reported between waking 2 min. In contrast, thoughts about Tetris reports (9%) occurring on the first day of play experiences and their apparent incorpora- dropped in half between the presleep period (33). Thus, the production of this imagery is tion into REM dreams (36–38). The time and 15 s into sleep (7.8 to 3.9%) and then not limited to the sleep state but can occur in course of appearance of Tetris images remained constant across the next 3 min waking as well. Two spontaneous reports of across the sleep onset period is also strik- (4.2% at 3 min). hypnagogic Tetris imagery were reported by ing. The rapid decay in frequency of Tetris novices at sleep onsets other than when com- images across the first 2 to 3 min of sleep puter-initiated awakenings were performed. occurred in the face of a steady or even These included one report of sleep onset im- rising rate of report of images in general agery from an afternoon nap on the first day over this time interval (9, 10). Thus, the of the protocol and one of sleep onset after a occurrence of specific Tetris images seems spontaneous 2:30 a.m. awakening on the uniquely linked to sleep onset itself.

Fig. 2. Report frequen- cies. For each partici- pant group, the cumu- lative number of re- ports containing (A) vi- sual images of Tetris or (B) thoughts of Tetris without images is giv- en, normalized to the total number of reports collected across three Fig. 1. Game performance. Mean scores, aver- nights. Reports from aged across participants, for each play session. novices were collected Participants receive about 20 points for each on only the first two piece put into play. Session ”0“ is first game nights, and graph score of first session. Solid circles, experts; open points are normalized for three nights. The upward curvature of the curves in (A) indicates that reported circles, novices; hatched circles, amnesics. Error frequency of images increased across the nights, whereas the downward curvature in (B) indicates that bars ϭ SEM. the frequency of thoughts alone decreased across nights (53).

www.sciencemag.org SCIENCE VOL 290 13 OCTOBER 2000 351 R EPORTS Fig. 3. Images and thoughts of Upon my way to sleep . . . , Tetris across the sleep onset pe- And I could tell riod. (A) Percentage of reports What form my dreaming was about to take. collected before sleep onset Magnified apples appear and disappear, Stem end and blossom end, (wake) and after sleep onset And every fleck of russet showing dear. (sleep) that contained either im- My instep arch not only keeps the ache, ages of Tetris (solid bars) or It keeps the pressure of a ladder-round. thought of Tetris without images I feel the ladder sway as the boughs bend.’’ (open bars). Although thoughts 2. D. O. Hebb, Psychol. Rev. 75, 466 (1968). were most common in the 3. D. L. Schacter, Psychol. Rev. 83, 452 (1976). presleep period, images were 4. R. Greenberg, C. Pearlman, R. Mayer, E. Hartmann, most common after sleep onset; Arch. Gen. Psychiatry 18, 203 (1968). the difference was significant 5. H. P. Cathala, F. Laffont, S. Esnault, C. Sereni, M. (chi-square test: ␹2 ϭ 4.6, df ϭ Siksou, Neurophysiol. Clin. 18, 173 (1988). ϭ 6. J. Antrobus, Psychophysiology 20, 562 (1983). 1, P 0.033). (B) Percentage of 7. D. Foulkes, Dreaming: A Cognitive-Psychological reports collected after sleep on- Analysis (Erlbaum, Hillsdale, NY, 1985). set that contained either images 8. G. W. Domhoff, Finding Meaning in Dreams: A Quan- of Tetris (solid circles) or titative Approach (Plenum, New York, 1996). thoughts of Tetris without imag- 9. D. Foulkes and G. Vogel, J. Abnorm. Psychol. 70, 231 es (open circles). Although the (1965). frequency of thoughts remained constant over the 3 min of sleep onset (P Ͼ 0.4), the frequency 10. J. Rowley, R. Stickgold, J. A. Hobson, Conscious. Cogn. of images decreased with increasing periods of sleep from 11.3% at 15 s to 5.7% at 2 min (r ϭ 0.86, 7, 67 (1998). df ϭ 3, P ϭ 0.06). 11. A. Mavromatis, : The Unique State of Between Wakefulness and Sleep (Rout- ledge & Kegan Paul, London, 1987). Because reported images originated that will require further investigation. 12. D. L. Schacter, Psychol. Bull. 83, 452 (1976). from recent experiences, one might assume This lack of hippocampal involvement 13. R. Greenberg and C. A. Pearlman, Perspect. Biol. Med. that hypnagogic image construction is me- in image construction could explain many 17, 513 (1974). 14. C. Smith, Behav. Brain Res. 69, 137 (1995). diated by the hippocampal declarative of the formal properties of dreams (46). 15. E. Hennevin, B. Hars, C. Maho, V. Bloch, Behav. Brain memory system. Our data argue against this Without the anchor of temporal and spatial Res. 69, 125 (1995). assumption. If image construction simply associations found in hippocampal declar- 16. R. Stickgold, Trends Cogn. Sci. 2, 484 (1998). 17. The Macintosh version of Tetris (copyright 1988, reflected reactivation of recent memories, ative memories, much of the bizarreness of Sphere, Inc.) was obtained from Spectrum HoloByte the images should have been most common dreams, including their discontinuities, in- (Alameda, CA). For a more complete description of on the first night. Instead, only 10% of the congruities, and uncertainties (47), would the game, see Science Online (51). images reported by novices occurred on appear almost inevitable. This would be 18. Sessions began between 7 and 11 a.m. and between 7 and 11 p.m. the first night, whereas 90% occurred on especially true during REM sleep, when 19. Novices were 18 to 25 years old, had less than 10 the second night. Two of five experts reported there is evidence not only that hippocampal hours of lifetime video game experience, and had imagery incorporating abstracted details from output is blocked (48), but also that weak no prior experience with Tetris. Experts were also 18 to 25 years old. They averaged over 200 hours Tetris experiences that took place 1 to 5 years associations in cortical memory systems of lifetime experience with Tetris (50 to 500 earlier. These reports underscore the influence are preferentially accessed (49), thus the hours), but only one participant reported having of remote memories (that may be less depen- hypnagogic report from an afternoon nap of played in the 3 weeks before testing. Amnesic patients were 21 to 62 years old [42.8 Ϯ 14.9 (SD)] dent on hippocampal systems) on image con- Tetris pieces falling onto a garden path and were all apparently novices. All had extensive struction. Most telling is the presence of these (34). medial temporal lobe damage, including the hip- same images in the hypnagogic reports from We have very little information con- pocampal formation. Three suffered lesions sec- amnesics. All of the patients were densely am- cerning the function of these images or of ondary to anoxia, one to Herpes simplex enceph- alitis, and one to a cerebrovascular accident. Pa- nesic and could not produce declarative mem- the physiological processes underlying tients had 12 to 20 years of education (15.2 Ϯ 3.6 ories of recent events. Indeed, none of them them. The probability of having Tetris im- years). They had normal IQs ( Wechsler Adult In- ϭ Ϯ remembered the Tetris game or the experiment- ages for both the novice and expert groups telligence Scale III, FSIQ 94 11) but severely impaired memory functions ( Wechsler Memory er from one session to the next (39). These was inversely related to initial perfor- Scale III—General Memory Index: 48.6 Ϯ 3.2). participants produced hypnagogic images of re- mance, suggesting a relation to the learning Additional neuropsychological test results can be cent events even though they did not have process itself (50). Experts incorporated found on Science Online (51). All participants pro- vided internal review board–approved informed declarative knowledge of them. The idea that related memories from earlier play, memo- consent before participation in the study. Partici- the declarative memory system is not active at ries that were presumably activated by the pants in the three groups provided similar numbers sleep onset is further supported by the results of recent Tetris play. One possible function of hypnagogic reports. Novices averaged 7.6 re- ports per night, experts 7.8 per night, and amne- Wyatt et al. (40, 41), who demonstrated that for the activation of these memories would siacs 7.2 per night. In addition, nonamnesic partic- explicit recall of words and word pairs was be to alter their strengths, structures, or ipants were instructed to carry the recorder with impaired when words were presented within 3 associations in ways that are adaptive for them during the day and report any occurrences of ”thoughts, images, or feelings“ related to Tetris min of sleep onset. the organism. With such predictable asso- that they experienced while awake. Reports from The relative preservation of hypnagogic ciations being made and presumably nonamnesics were dictated into Pearlcorder S950 images in amnesia suggests that they do not strengthened at sleep onset, one can only hand-held microcassette recorders that automati- depend on medial temporal structures in- surmise that during REM sleep more un- cally stamped the time and date onto the record- ing. Those from amnesics were recorded by the volved in declarative memory (42). Instead, predictable, potentially valuable, but fre- experimenter, who in all cases prompted the pa- it is likely that these perceptual representa- quently useless, associations are tested and, tient for the report, often as a supplement to the tions are mediated by pro- when appropriate, similarly strengthened. computer-generated prompt. Novices were probed during the first two nights of play, as were 3 of the cesses, as seen in other studies with amne- 10 experts. On the basis of the data from these sic patients (43–45). It should be noted that References and Notes participants, the final seven experts and all amne- the amnesics were substantially older than 1. R. Frost, North of Boston (Holt, New York, 1915). siacs were probed for three nights. Frost described the sensation in his 1914 poem “After 20. Experimental awakening was done with the Night- the normal novices and that the contibution Apple-Picking”: cap sleep monitoring system as described by Row- of age remains an alternative explanation “ . . . I was well ley et al. (10). Reports were collected from the

352 13 OCTOBER 2000 VOL 290 SCIENCE www.sciencemag.org R EPORTS

presleep period, when participants were attempt- 27. Thus, one participant reported, ”I’m just watching 36. M. Jouvet, Ann. Practicien 29, 27 (1979). ing to fall asleep, as well as after intervals of 15 to tiles fall down“ (DW: day 2), whereas another 37. R. A. Powell, T. A. Nielsen, J. S. Cheung, T. M. Cer- 180 s of objective sleep, during the first hour after reported, ”I’m rotating it and I’m fitting the blocks venka, Percept. Mot. Skills 81, 95 (1995). participants retired for the night. Multiple forced in like I’m supposed to“ (AM: day 2). A third 38. T. A. Nielsen and R. A. Powell, Psychiatr. J. Univ. Ott. awakenings were made each night, in pseudoran- reported ”just seeing Tetris shapes floating around 14, 561 (1989). dom order, balanced across participants and nights, in my head like they would in the game, falling 39. On one occasion, after playing her evening session after 15, 45, 75, 120, and 180 s of eyelid quies- down, sort of putting them together in my mind“ of Tetris, participant PLS showered and put on her (JEG: day 2). It should be emphasized that none of cence (10), and additional reports were requested pajamas in preparation for bed. Upon returning to these reports described episodic memories. For after 3, 6, and 9 min of wake before the initial sleep her bedroom, she was clearly surprised to find the example, no one reported being in their room, onset and after 3 min of wake after any forced experimenter sitting by her bed, and asked him, awakenings, when such wake periods occurred. sitting at their desk, or using their computer. As ”Who are you, and what are you doing in my There were a maximum of 10 report requests per with the amnesics, the images occurred without bedroom?“ Nevertheless, she subsequently pro- night and no more than 5 from periods of wake. reference to time or place and appeared, rather, to duced hypnagogic Tetris imagery. The complete The five periods of sleep (15 to 180 s) produced be extracted and abstracted from the actual dissociation of from hypnagogic between 48 and 55 reports each; there were 154 experience. 2 imagery was seen most clearly in the case of DF, wake state reports. Frequencies of requests and 28. Chi-square test: ␹ ϭ 10.6, df ϭ 1, P ϭ 0.001 for 2 who explicitly reported being unaware of the reports of Tetris imagery or thoughts did not differ reports; ␹ ϭ 8.2, df ϭ 1, P ϭ 0.004 for participants. source of these images. significantly between novices and experts. Partici- 29. Of the 10 participants, only one had played within pants were instructed to dictate reports of any the last 3 weeks, and two had not played in the 40. J. K. Wyatt, R. R. Bootzin, J. Anthony, S. Bazant, Sleep ”thoughts, images, or feelings“ that they were previous year. 17, 502 (1994). experiencing before being prompted to report or, if 30. Repeated measures ANOVA across the first game 41. J. K. Wyatt, R. R. Bootzin, J. J. B. Allan, J. L. Anthony, appropriate, to state that there were none. Partic- and six sessions: df ϭ 6, F ϭ 1.24, P ϭ 0.30. Sleep 20, 512 (1997). ipants were normally awakened by digitized voice 31. Reported imagery included ”a Tetris screen and the 42. L. R. Squire, Psychol. Rev. 99, 191 (1992). recordings when the computer identified an appro- pieces fitting together in various ways“ (ANJ: day 2), 43. D. L. Schacter and E. Tulving, Memory Systems 1994 priate wake or sleep interval and instructed to and ”[I’m] seeing in my mind how the game pieces (MIT Press, Cambridge, MA, 1994). dictate their reports into microcassette recorders. kind of float down and fit into the other pieces, and 44. M. Verfaellie and M. M. Keane, in Varieties of Uncon- In some cases, amnesic participants were awak- am also rotating them“ ( TPR: day 2). scious Processing: New Findings and New Compari- ened by the experimenter. 32. One expert reported, ”It’s weird . . . because I don’t sons, B. Degelder, E. De Haan, C. Heywood, Eds., in 21. Because of missed sessions, individual patients see it the way we play it on this computer. I see it like press. played four to six sessions. Repeated measures anal- I used to play it at home where its in different colors 45. M. Verfaellie and M. O’Connor, Semin. Neurol.,in ysis of variance (ANOVA) showed significant im- and stuff like that and the boxes look a little differ- press. provement over four sessions (n ϭ 5, F ϭ 6.65, P ϭ ent.“ She also reported hearing the music that ac- 46. J. A. Hobson, The Dreaming Brain (Basic Books, New 0.007). All patients completed at least 18 games companied the game on her original Nintendo ver- York, 1988). during the first session and showed no significant sion, whereas the version used in the experiment was 47. ࿜࿜࿜࿜ , S. A. Hoffman, R. Helfand, D. Kostner, Hum. improvement across these 18 games (individuals: r Ͻ silent. Neurobiol. 6, 157 (1987). 0.11, P Ͼ 0.3 for each; group: r ϭ 0.01, P ϭ 0.98). By 33. Binomial distribution analysis, P Ͻ 0.0001. 48. G. Buzsa´ki, Cerebr. Cortex 6, 81 (1996). contrast, age-matched normal novices (51 Ϯ 8.7 34. In one instance, the participant reported dozing off years) all showed significant improvement (individu- in the afternoon while ”thinking about a project I 49. R. Stickgold, L. Scott, C. Rittenhouse, J. A. Hobson, J. als: r Ͼ 0.5, P Յ 0.03 for each; group: r ϭ 0.73, P ϭ have for work that involves designing a garden Cogn. Neurosci. 11, 182 (1999). 0.0006) and improved an average of 107% across the space indoors and as I was thinking about it in my 50. The 3 novices (out of 12) who produced no reports of one session. The 36% increase across six sessions mind little Tetris pieces kept falling down into the Tetris images performed twice as well as the remain- seen in patients is comparable to the 29% increase garden spaces, which they wouldn’t really [do] ing 9 during the initial 2-hour training session (mean seen over six sessions in a single amnesic patient by because [they are] geometrical and garden spaces scores 2984 compared with 1516; unpaired t test: ϭ ϭ ϭ Winter (52). are nice and flowing.“ df 10, t 3.69, P 0.004). Similarly, the number 22. R. J. Haier et al., Brain Res. 570, 134 (1992). 35. One concern in this study was whether demand of Tetris images reported by experts showed a neg- 23. Normals, consisting of 12 novices and 10 experts, characteristics would bias participants to report ative correlation with initial performance that ap- produced a total of 30 reports of Tetris imagery in Tetris images when they did not, in fact, occur. proached significance (r ϭϪ0.54, df ϭ 8, P ϭ 0.11). 418 reports (see text). Several aspects of the actual reports argue against Amnesics showed a similar negative correlation, but 24. In all cases, reported images included Tetris-like this. The reports from amnesics obviously were not it did not approach significance (r ϭϪ0.45, df ϭ 3, shapes, usually falling in front of the participants’ fabricated to please us, because participants could P Ͼ 0.2). Thus, Tetris imagery was negatively corre- eyes and sometimes rotating or fitting together. In not even remember that they had played Tetris. lated with initial performance. no case did the participant report seeing or touch- Second, the extreme similarity between reports 51. Supplementary material is available at www. ing the keyboard or actually playing the game and would not be expected. Thus, we received no sciencemag.org/feature/data/1050057.shl. in no case was the imagery seen in inappropriate reports of participants ”playing“ Tetris, which we 52. W. Winter, Implicit and explicit cognitive functioning contexts. Thus, one amnesic reported, ”I was just would have expected from invented reports. In in hippocampal amnesia (UMI Microform 9605682, trying to figure out those shapes and how to get addition, the preponderance of images from nov- Ann Arbor, 1995). them aligned“ (JM: day 2), whereas another report- ices on the second night and the higher rate of 53. Total numbers of hypnagogic reports and of par- ed ”thinking about little squares coming down on a images reported after sleep onset compared with ticipants reporting images and thoughts were as screen and trying to put them in place“ (PLS: day the presleep period are counterintuitive to the follows: novices (open circles): 183 reports from 12 3). The third amnesic reported seeing ”images that expectations of demand characteristics. A more participants, 9 of whom (75%) provided 19 reports are turned on their side. I don’t know what they are subtle concern is that more general geometric of images and 6 of whom (50%) gave 12 reports of from, I wish I could remember, but they are like images were being interpreted as Tetris images. To thoughts without images; experts (filled circles): blocks“ (DF: day 1). Despite references to ”those look for nonspecific effects, we searched the text 235 reports from 10 participants, 5 of whom (50%) shapes“ and ”a screen,“ participants gave no indi- of 796 hypnagogic reports from an earlier study gave 11 reports of images and 5 of whom (50%) cation of recalling the origin of these images. (10) for the words square[s], rectangle[s], star[s], gave 13 reports of thoughts without images; and Rather they appear to be images of events for diamond[s], block[s], geometr[y,ic], sparkl[e,ing], amnesics (hatched circles): 108 reports from 5 which they had no conscious recollection. kaleidoscop[e,ic], shape[s], pattern[s], tile[s], participants, 3 of whom (60%) provide 8 reports of ϭ ϭ ϭ 25. Two-tailed t test, df 14, t 3.1, P 0.008. For fall[ing], rotat[e,ed,ing], turn[ed], mov[e,ing], images and 1 of whom (20%) reported a single this and all other analyses of novice scores, one fit[ting,ted], Tetris, game[s], screen[s], and video[s]. thought without images. participant, whose initial game score was 24 SD Only four reports (0.5%) contained these words in 54. This research was supported by a grant from the above the mean of the other 11 novices, was relevant forms (e.g., references to Harvard MacArthur Foundation’s Mind-Body Network, by excluded. ”Square“, ”Star“ Trek, ”falling“ asleep, and being National Institute of Mental Health grants MH- 26. Repeated measures ANOVA across the first game out of ”shape,“ were excluded). The words rectan- ϭ ϭ Ͻ 13,923 and MH-48,832, and by National Institute and six sessions: df 6, F 17.1, P 0.0001. gle, diamond, sparkling, kaleidoscope, tile, geomet- of Neurological Disorders and Stroke program Post-hoc contrasts showed significant improve- ric, rotate, and Tetris did not appear in any context. project grant NS26985. We thank an anonymous ments after the first game (P Ͻ 0.0001) and first Taken together, these findings suggest that there is Ͻ ϭ ϭ reviewer for suggesting the analysis of reports by (P 0.0001), second (P 0.005), and third (P no important contamination of our data from re- sleep time. 0.01) sessions. There was a trend toward additional sponses to demand characteristics, although the improvement after the fourth session as well (P ϭ possibility of some minimal contribution cannot be 0.11). eliminated. 2 March 2000; accepted 1 September 2000

www.sciencemag.org SCIENCE VOL 290 13 OCTOBER 2000 353