Interocular transfer of perceptual skills after sleep

Abstract

Abstract

Abstract:

Abstract  Several studies suggest that sleep improves perceptual skills in the visual texture discrimination task (TDT). Here we report that besides consolidation, sleep also generalizes the learned perceptual abilities to the untrained eye. Healthy volunteers (n = 32) were trained on the TDT, in which they had to discriminate between horizontal and vertical target textures briefly presented in the periphery of the visual field (left upper quadrant). After a 10-hr interval filled with either sleep or wakefulness, they were retested first on the trained eye in the trained quadrant and then on the untrained eye and quadrant. In line with prior findings, visual discrimination was globally higher after sleep than after wakefulness, as compared to performance levels at the end of training. Furthermore, discrimination performance was significantly improved only in the sleep condition for the untrained eye in the same quadrant, but also showed a trend to generalize to the untrained eye and untrained quadrant. Our results suggest that sleep-dependent perceptual skills continue developing at a later visual-process stage than the V1 area, where learning is not monocular anymore.

Introduction

Visual perceptual skills may continue to improve over time and sleep after the end of a training session, suggesting an off-line consolidation process (see Peigneux & Smith, 2010, for a review). Using the paradigmatic texture discrimination task (TDT) devised by Karni and Sagi (1991), it has been shown that novel perceptual learning evolves across two successive stages, with an initial fast, within-session improvement of performance followed by a slower off-line, sleep-dependent consolidation process eventually leading to enhanced performance at delayed testing without intermediate practice (Karni & Sagi, 1993). In the TDT, subjects are trained to discriminate more and more rapidly horizontal from vertical patterns briefly displayed in a specific quadrant in the periphery of the visual field. Performance is computed as the minimal stimulus presentation time (or stimulus onset asynchrony, SOA) required to keep performance at a satisfyingly accurate level. In the initial training session, the formation of task-specific routines (Poggio, Fahle, & Edelman, 1992; Ullman, 1984) allows fast learning to take place in the first minutes to hours of training (Fahle, 1994; Fahle & Edelman, 1993). Fast perceptual learning is restricted to the trained eye, location (i.e., quadrant) and orientation, suggesting that experience-dependent changes are taking place at an early stage in the visual cortex (Karni et al., 1995), where monocularity and retinotopic organization of the visual inputs are still retained and different orientations are handled separately (Hubel, 1982; Zeki, 1978). 

Fast learning apparently needs to be consolidated in specific off-line conditions. Indeed, performance deteriorates rather than improving or stabilizing when participants are re-exposed several times to the same TDT within a same day (Censor, Karni, & Sagi, 2006; Gais, Plihal, Wagner, & Born, 2000; Mednick, Arman, & Boynton, 2005; Mednick et al., 2002; Mednick, Nakayama, & Stickgold, 2003) or are administered a single but extended, intensive training session (Censor et al., 2006; Censor & Sagi, 2008; Ofen, Moran, & Sagi, 2004, 2007). This deterioration effect presents a different specificity profile than at learning. Indeed, deterioration is retinotopically specific to the trained visual quadrant (Mednick et al., 2002; Mednick et al., 2003) and to the target orientation, but not to the trained eye (Mednick et al., 2005; Ofen et al., 2007) or to the background orientation (Mednick et al., 2005). This pattern of effects also indicates that performance decline in the TDT is not merely due to on-task fatigue. Furthermore, neither circadian influences, subjects' motivation, nor difficulty levels (e.g., long versus short interstimuli intervals) have a significant impact on perceptual decrements in the TDT (Mednick et al., 2002). Retinotopically specific deterioration might be due to the saturation of local networks in the early visual cortex through repeated testing (Mednick et al., 2002) and/or connectivity saturation (Censor & Sagi, 2008) resulting from undifferentiated strengthening of synapses contributing to noise as well as synapses contributing to the signal. Accordingly, an association has been found using functional MRI between deterioration in visual perceptual skills and decreased activity in V1 after a wake interval, suggesting that the deterioration is due to bottom-up fatigue of V1 neurons rather than decreased top-down attentional modulation (Mednick, Drummond, Arman, & Boynton, 2008). Conversely, performance stops deteriorating, and may even improve, when subjects are allowed to sleep after the training session, suggesting a restorative and slow-acting consolidating role of sleep (Karni, Tanne, Rubenstein, Askenasy, & Sagi, 1994; Walker et al., 2003). 

Sleep-dependent consolidation of performance in the TDT is a robust, often replicated effect (Gais et al., 2000; Karni et al., 1995; Karni & Sagi, 1991; Mednick et al., 2002; Mednick et al., 2003; Plihal & Born, 1997; Walker et al., 2003). Sleep-dependent consolidation mostly depends on the first night posttraining, even if additional performance improvement can be observed up to 4 days without practice after a single training episode (Stickgold, James, & Hobson, 2000). Also, short posttraining naps prevent performance deterioration (Mednick et al., 2002), whereas longer naps are associated with off-line gains in performance (Mednick et al., 2003), suggesting that sleep resets visual contrast thresholds to a lower baseline. Besides global sleep-dependent improvement in perceptual skills, consolidation effects appear to be determined by the time spent in specific sleep stages. However, findings are more contradictory in this respect. Karni et al. (1994) found that selective disruption of rapid eye movement (REM) sleep but not of slow wave sleep (SWS) prevented overnight gains in performance, whereas Aeschbach, Cutler, and Ronda (2008) found that SWS disruption also limits sleep-dependent improvement in the TDT. Gais et al. (2000) evinced performance improvement after 3 hr of early sleep (SWS dominant) but not after late sleep (REM sleep dominant), although to a lesser extent than after a full night of sleep. Additionally, Stickgold, Whidbee, Schirmer, Patel, and Hobson (2000b) found a correlation between overnight gains in performance and the combined amount of time spent in SWS during the first quarter of the night and in REM sleep during the last quarter of the night. It was further shown that a 60-min posttraining nap prevents deterioration of TDT performance, whereas a 90-min nap containing both SWS and REM sleep improves it (Mednick et al., 2003), suggesting a double-step consolidation process in which SWS would support performance stabilization and REM sleep would favor its improvement. 

As mentioned previously, sleep-dependent improvement in the TDT is monocular (Karni & Sagi, 1991; Schwartz, Maquet, & Frith, 2002; but see Schoups & Orban, 1996), specific to the trained quadrant (Karni & Sagi, 1991; Mednick et al., 2002; Schwartz et al., 2002) and to the orientation of the background elements but not to the orientation of the target elements (Karni & Sagi, 1991). At the functional neuroanatomical level, sleep-dependent perceptual learning appears to involve changes in local connections within the primary visual cortex, thus at the earliest stages of cortical processing (Karni & Sagi, 1991; Schwartz et al., 2002; Walker, Stickgold, Jolesz, & Yoo, 2005). However, Censor and Sagi (2008) showed a transfer of resistance to performance deterioration to the untrained locations of the visual field following short practice periods, casting doubts on the specificity of visual training and suggesting the potential involvement of higher order brain areas. 

With this perspective, the present experiment aimed at investigating the hypothesis of a sleep-dependent generalization of visual discrimination abilities to the untrained eye and/or the untrained visual quadrant in the TDT. Participants were tested either in a Wake (n = 16) condition with morning learning and evening testing or in a Sleep (n = 16) condition with evening learning and morning testing. In each condition, participants received monocular training on the left or right eye (n = 8 per subgroup). At each trial, participants had to report two simultaneously briefly presented targets, namely the identity of a centrally displayed (foveal vision) rotated letter (T or L) and the vertical or horizontal orientation of three aligned diagonal bars, laterally displayed (peripheral vision) in a background of 19 × 19 horizontal bars (Figure 1). Stimulus onset asynchrony (SOA) or stimulus presentation time was progressively decreased across 11 successive blocks of 51 trials each (360 ms, 260 ms, then 220 ms to 60 ms in 20-ms-steps). The shortest SOA allowing 80% correct responses for both letter and orientation targets was computed as an index of performance. At testing, TDT blocks were administered again under four possible monocular conditions in a fixed succession order: the same visual quadrant (SVQ) at the trained eye (TE) then at the untrained eye (UE), and then the other visual quadrant (OVQ) at TE then at UE. Performance evolution was computed as the δ (difference) between optimal SOA at learning versus each test condition (i.e., a positive SOA δ means improvement, whereas a zero or negative δ means no improvement or deterioration). 

Figure 1

 

Timeline of a single trial presentation in the visual texture discrimination task (TDT).

Results

Participants in the Sleep (S) and Wake (W) conditions did not differ according to age [S vs. W: mean = 23.94 ± standard deviation of 1.57 years vs. 25.12 ± 2.25; t(30) = −1.73, p = 0.09], educational level [15.94 ± 0.93 years of education vs. 16.37 ± 0.88; t(30) = −1.36, p = 0.18], sleep quality over the last month according to the Pittsburgh Sleep Quality Index (PSQI; Buysse, Reynolds, Monk, Berman, & Kupfer, 1989) total score (4.00 ± 0.95 vs. 3.31 ± 1.54; Mann–Whitney U = 96.5, p = 0.24), or self-reported sleep duration [7.65 ± 0.51 vs. 7.29 ± 0.80; t(30) = −1.51, p = 0.14] and latency [14.31 ± 9.68 vs. 18.19 ± 12.27; t(30) = 0.99, p = 0.33] during the night preceding the training session. 

A repeated-measures ANOVA performed on computerized psychomotor vigilance scores (psychomotor vigilance task [PVT]; Dinges & Powell, 1985) with between-factor Group (Wake vs. Sleep group) and within-factor Session (Training vs. Testing) did not reveal any main effect of Group, F(1, 30) = 2.18, p = 0.15, or Session, F(1, 30) = 0.18, p = 0.67, nor any interaction, F(1, 30) = 0.001, p = 0.97, indicating similar levels of vigilance before the learning and recall phases in the Wake and Sleep conditions. 

Texture discrimination task

Learning session

A two-way ANOVA with between-factors Group (Wake vs. Sleep group) and Trained Eye (Right vs. Left) failed to disclose any significant main effects or interaction (all Fs < 1 and ps > 0.34) on SOA scores, suggesting that the two groups had similar performance at baseline. 

Delayed testing

Learning was estimated computing the δ (difference) SOA between the SOA at the Learning session and the SOA at delayed testing in the four different conditions (TE-SVQ, UE-SVQ, TE-OVQ, UE-OVQ). A repeated-measures ANOVA performed on δ SOAs with within-factors Tested Quadrant and Tested Eye (same as or different from Learning) and between-factors Group (Wake vs. Sleep group) and Trained Eye (Right vs. Left) disclosed a main effect of Group, F(1, 28) = 5.1, p = 0.032, with higher performance gains in the Sleep than in the Wake condition (δ SOA = 24.38 ± 7.33 ms vs. 0.94 ± 7.33). The triple Quadrant × Eye × Group interaction was not significant, F(1, 28) = 2.22, p = 0.62. All other main effects and interactions were not significant (all ps > 0.5). 

To explore further the learning effects in each testing condition (TE-SVQ, UE-SVQ, TE-OVQ, UE-OVQ) separately in the Wake and Sleep groups (Figure 2), independent one-sample t tests were computed against the null hypothesis. In the Sleep group, this analysis disclosed positive, above-chance-level δ SOAs in the same visual quadrant for both the trained [TE-SVQ; t(15) = 5.12, p < 0.001; Cohen's d = 1.30] and untrained [UE-SVQ; t(15) = 4.75, p < 0.001; Cohen's d = 1.19] eyes. In the OVQ condition, the test against the null hypothesis did not reach significance for the trained eye [TE-OVQ; t(15) = 1.62, p = 0.125; Cohen's d = 0.41] but was marginally significant for the untrained eye [UE-OVQ; t(15) = 2.035, p = 0.059; Cohen's d = 0.51]. We also observed higher within-group variability in the TE-OVQ condition. At variance in the Wake group, δ SOAs were slightly negative but never above chance levels. The t tests against the null hypothesis were all markedly nonsignificant: TE-SVQ: t(15) = −0.246, p = 0.818, Cohen's d = 0.06; UE-SVQ: t(15) = 1.065, p = 0.304, Cohen's d = 0.27; TE-OVQ: t(15) = 0.504, p = 0.621, Cohen's d = 0.13; UE-OVQ: t(15) = −0.65, p = 0.525, Cohen's d = 0.16. 

Figure 2

 

Stimulus onset asynchrony improvement from Learning (δ SOA) in the Sleep (a) and Wake (b) groups in the trained eye on the same visual quadrant (TE-SVQ), untrained eye on the same visual quadrant (UE-SVQ), trained eye on other visual quadrant (TE-OVQ), and untrained eye on other visual quadrant (UE-OVQ). Asterisks indicate a significant difference (t test against the null hypothesis): **p < 0.01. Error bars are standard deviations.

Finally, at a more qualitative level we computed the percentage of participants in each group exhibiting at least a 20-ms gain from Learning condition to the Testing condition (Table 1). Frequency counts showed that across 16 subjects within each group, 66% of the participants in the Sleep group exhibited increased performance in the TE-OVQ and UE-OVQ conditions, whereas only 31% did so in the Wake group. 

Table 1

 

Percentage of participants showing delayed improvement in SOA performance ≥ 20 ms from Learning to Testing conditions in the Sleep and Wake groups (n = 16 each). Notes: TE-SVQ = trained eye, same visual quadrant; UE-SVQ = untrained eye, same visual quadrant; TE-OVQ = trained eye, other visual quadrant; UE-OVQ = untrained eye, other visual quadrant.

	TE-SVQ	UE-SVQ	TE-OVQ	UE-OVQ
Sleep	93.8	87.5	65.6	65.6
Wake	37.5	25.0	31.2	31.2

Discussion

In the present study, we tested the hypothesis of a sleep-dependent generalization of visual discrimination abilities to the untrained eye and/or the untrained visual quadrant in the TDT. In line with previous findings (Censor et al., 2006; Gais et al., 2000; Matarazzo, Franko, Maquet, & Vogels, 2008; Mednick, Drummond, Boynton, Awh, & Serences, 2008; Stickgold, James, et al., 2000; Stickgold, Whidbee et al., 2000b; Walker et al., 2005), we found an overnight, sleep-dependent improvement of perceptual skills. The average performance gain was about 24 ms after a night of sleep, whereas it was negligible (< 1 ms) after a daytime wake interval, whether tested on the same or different eye or visual quadrant. The lack of posttraining improvement in all conditions including same eye and same visual quadrant in the Wake group is inconsistent with Karni and Sagi's original (1993) report indicating increased performance after 8 hr posttraining spent awake. Nonetheless, it fits with others studies suggesting that improvement of visual perceptual skills in the TDT does not occur with the simple passage of time but needs sleep to take place (Mednick et al., 2002; Mednick et al., 2003). 

Regarding our initial hypothesis driven by Censor and Sagi's (2008) finding of a transfer of resistance to performance deterioration to the untrained locations of the visual field following short practice periods, our study demonstrates that sleep may actually support the transfer of learned perceptual skills to the untrained eye. Consequently, visual perceptual learning in the TDT seems not to be entirely monocular, at variance with the conclusions by Karni and Sagi (1991) and Schwartz et al. (2002) that learning is specific to the trained eye and takes place at early processing stages in the visual system, where monocularity and retinotopic organization of the visual stimulus are still retained. It must be noticed, however, that due to their experimental design, only cerebral activity—not performance—was compared between conditions in the fMRI study of Schwartz et al. (2002), who found higher activity in the corresponding retinotopic area of the visual cortex for targets presented at the trained eye compared to the untrained eye within the same quadrant. Accordingly, although the experiment by Walker et al. (2005) was not designed to disclose interocular transfer of visual perceptual skills after sleep, fMRI data revealed task activation patterns beyond primary visual cortex in ventral and dorsal regions. This latter finding supports the hypothesis that sleep may favor neuronal changes at later cortical stages in the visual process, where information of the two eyes merges. It should be noticed also that in the study by Karni, Weisberg, Lalonde, and Ungerleider (1996), eight subjects presented an overnight interocular transfer and were excluded from their analyses. In 1993, the authors observed that fast (within-session) learning is binocular whereas slow (between-session) learning is monocular (Karni & Sagi, 1993). Hence, beside the conclusion by Karni and Sagi (1993) that interocular transfer may occur solely within session during the fast-learning phase, reflecting the setting of task-specific routines, our results suggest that interocular transfer may also take place and be promoted off-line during the slow posttraining phase when the retention interval is filled with sleep. Methodological differences between our experiment and Karni and Sagi's (1993) study may explain this discrepancy between their and our own results. For Karni and Sagi, stimuli were more briefly presented (10 ms vs. 16 ms in our study), increasing the difficulty of the task and potentially reducing the interocular-transfer effect. Moreover, they tested nine participants only, while our sample size included 16 subjects in the Sleep group and 16 subjects in the Wake group. Thus, interindividual variability may have masked the interocular-transfer effect in Karni and Sagi's study. Further studies should investigate this issue. 

Regarding the retinotopic character of visual perceptual learning in the TDT, our results are globally in agreement with those of Karni and Sagi (1991) and Schwartz et al. (2002) in suggesting that overnight performance improvement is sleep dependent and specific to the trained quadrant. Intriguingly, however, our data also suggest a possible generalization of the learning process to the other visual quadrant in the Sleep group only, as shown Figure 2. Indeed, whereas SOA δ is at chance level in all conditions in the Wake group, it is positive at a statistical trend above chance level (although admittedly not surviving corrections for multiple comparisons) in the UE-OVQ condition. Performance in the TE-OVQ condition presents larger variability in the Sleep group, making it difficult to interpret. However, frequency counts disclosed that across 16 subjects within each group, 10 participants in the Sleep group exhibited increased performance by at least 20 ms in the two OVQ conditions, whereas only five did so in the Wake group (Table 1). Taken together, the overnight interocular transfer and this potential trend for a retinal-location generalization suggest that perceptual learning might take place beyond V1, allowing wider generalization of skills improvement. At present, these results must be interpreted with caution and call for further studies that may confirm a retinal-location generalization in perceptual skills. 

Other studies have evidenced performance deterioration over repeated practice when participants are kept awake in the intermediary period (Censor et al., 2006; Gais et al., 2000; Mednick et al., 2005; Mednick et al., 2002; Mednick et al., 2003). In the present study, we found only slightly decreased performance in the Wake group, although nonsignificant performance deterioration was observed in the same condition compared to Learning (TE-SVQ), supporting the hypothesis of a progressive synaptic saturation that is retinotopically specific but not specific to trained eye (Mednick et al., 2005; Ofen et al., 2007). A nonsignificant deterioration was also observed in the untrained eye in the other visual quadrant (UE-OVQ). Since the UE-OVQ condition takes place after the three other conditions, the transfer of performance deterioration to the other visual quadrant may be explained by an overload of the visual learning system due to the practice of the three other conditions. Further studies are needed to investigate these issues. 

Finally, it may be noticed that the comparison between a daytime wakefulness and a nocturnal sleep interval does not entirely rule out a potential circadian confound. Controlling for time-of-day effects would require testing an additional group in which participants are trained in the evening then retested in the morning after a sleep deprivation night. Although this would control for the circadian confound, it would have the drawback of having one population tested in a state of sleep deprivation that is likely detrimental to the discrimination performance. The other solution would be to have participants sleep during the day after learning in the morning, which is difficult to achieve without a prior night of sleep deprivation, again detrimental to performance, but at the Learning phase. Also, it is worth mentioning that that our data show that performance at Learning was similar between groups—that is, whatever the evening or morning moment of learning or the trained eye—suggesting no significant circadian impact. This result is also in agreement with previously published data from Mednick et al. (2002), where participants were trained at different times of the day and no circadian influence was reported on perceptual decrements in the TDT. Finally, should a circadian confound have biased our data, it should have similarly affected the performance in all eye/quadrant conditions, which is not the case here. Altogether, it suggests that the reported results cannot be ascribed to a circadian confound. 

To sum up, our results confirm prior findings that visual perceptual skills benefit from sleep for both the trained and untrained eyes in the same visual quadrant. Additionally, our results suggest for the first time that sleep may also favor generalization of visual discrimination abilities to untrained visual quadrants, besides retinotopic specificity in V1. 

Materials and methods

Participants

Thirty-five healthy students gave written informed consent to participate in this study, which was approved by the local ethics committee and followed the tenets of the Declaration of Helsinki. Three subjects were excluded from statistical analyses because of poor performance in the training session: They did not obtain 80% correct responses (both letter and orientation) in the three first blocks (SOA = 460 ms, 360 ms, and 260 ms). The 32 remaining students (21 women) were selected according to the following criteria: normal or corrected-to-normal vision, no history of neurological disorders, no sleep disturbances, no mood disorders, and intermediate or neutral chronotype (Horne & Ostberg, 1976). Subjects were required to keep regular sleep patterns during the week before and throughout the experiment and to refrain from drinking alcohol and stimulant drinks (e.g., caffeine, tea, cola) before each testing session. They were asked to fill in sleep logs to control for regularity of sleep habits. 

Materials

Participants performed a texture discrimination task adapted from Karni and Sagi (1991). They were seated in front of a 15-in. PC screen located at 50 cm from the eyes. The target display size was 10.2° of visual angle. For each trial, subjects had to report two targets: the rotated letter T or L located in the center of the display and the orientation of three aligned diagonal bars (vertical or horizontal orientation) in a background of 19 × 19 horizontal bars (Figure 1). These diagonal bars were presented in the peripheral visual field at a distance of 2.29° to 3.43° of visual angle from the center of the screen. Identification of the central letter was aimed at enforcing foveal fixation in the screen's center, hence processing of orientations in the peripheral visual field. Each trial started with a fixation cross located in the center of the monocular visual field (the other eye was occulted using a tissue patch). After 250 ms, the target screen was presented for 16 ms, followed by a black screen for a variable duration and a mask for 100 ms to erase the retinal image of the target display. Finally, a black screen was presented again for 3000 ms, during which subjects had to report the central letter and the target orientation by pressing the corresponding keys on the keyboard. The time between the target screen and the mask (stimulus-to-task-onset asynchrony, SOA) was variable and progressively reduced from 460 to 60 ms (460, 360, 260, 220, 180, 160, 140, 120, 100, 80, 60) across 11 51-trial blocks. Thus, discrimination difficulty progressively increased through reduction of the SOA. Performance measure was the shortest SOA at which 80% correct orientation and letter answers were obtained. Improvement on the task was measured as the δ (difference) in shortest SOAs between the Learning session and Testing conditions. 

Procedure

The 32 participants were randomly divided into two groups: Wake and Sleep. In the Wake group (n = 16), subjects learned the TDT at 9:00 a.m. and retesting took place at 7:00 p.m. after a 10-hr daytime interval. In the Sleep group (n = 16), they were trained on the TDT at 8:00 p.m. and retested at 8:00 a.m. after a 12-hr interval containing an episode of night sleep. In each group (Wake and Sleep), half of the participants were trained on the left eye (n = 8 per group) and the other half on the right eye. To ensure monocularity, a tissue patch covered the untrained eye. 

Training sessions consisted of 11 blocks of 51 consecutive trials each. During the training (Learning) session, participants were tested on the TDT with one eye occluded (trained eye left or right). The target was shown in the left quadrant of the visual field for all subjects. Test sessions began with an individually determined SOA defined as the training SOA + 40 ms. Each eye was tested for each quadrant (left and right), resulting in four testing conditions, always presented in the same order: trained eye, same visual quadrant (TE-SVQ); untrained eye, same visual quadrant (UE-SVQ); trained eye, other visual quadrant (TE-OVQ); and untrained eye, other visual quadrant (UE-OVQ). 

Additionally, the psychomotor vigilance task (PVT) (Dinges & Powell, 1985) was administered before training and testing sessions to estimate objective vigilance levels. 

Acknowledgments

GD and RS are respectively a research fellow and a postdoctoral fellow at the Belgian Fonds National de la Recherche Scientifique (FNRS). The authors declare no competing financial interests. The authors thank Yasmine Schroeder for her help in data acquisition. 

Commercial relationships: none. 

Corresponding author: Philippe Peigneux. 

Email: [email protected]. 

Address: Neuropsychology and Functional Neuroimaging Research Unit, Centre de Recherches en Cognition et Neurosciences, ULB Neurosciences Institute, Université Libre de Bruxelles, Brussels, Belgium.