Lack of frequency-tagged magnetic responses suggests statistical regularities remain undetected during NREM sleep

Hypnopedia, or the capacity to learn during sleep, is debatable. De novo acquisition of reflex stimulus-response associations was shown possible both in man and animal. Whether sleep allows more sophisticated forms of learning remains unclear. We recorded during diurnal Non-Rapid Eye Movement (NREM) sleep auditory magnetoencephalographic (MEG) frequency-tagged responses mirroring ongoing statistical learning. While in NREM sleep, participants were exposed at non-awakenings thresholds to fast auditory streams of pure tones, either randomly organized or structured in such a way that the stream statistically segmented in sets of 3 elements (tritones). During NREM sleep, only tone-related frequency-tagged MEG responses were observed, evidencing successful perception of individual tones. No participant showed tritone-related frequency-tagged responses, suggesting lack of segmentation. In the ensuing wake period however, all participants exhibited robust tritone-related responses during exposure to statistical (but not random) streams. Our data suggest that associations embedded in statistical regularities remain undetected during NREM sleep, although implicitly learned during subsequent wakefulness. These results suggest intrinsic limitations in de novo learning during NREM sleep that might confine the NREM sleeping brain’s learning capabilities to simple, elementary associations. It remains to be ascertained whether it similarly applies to REM sleep.

S2a. NREM / Exposure group. Individual topographies for tone-and tritone-related frequency-tagged responses in STAT an RDM conditions in the Sleep nap opportunity and subsequent Wake sessions. Note that participants S19 and S22 were exposed to 5 minutes of STAT, but less than 5 minutes of RDM streams during the sleep opportunity (see Supplementary Material Figure S1 and Table T1), and were thus excluded from statistical analyses (contrast STAT vs. RDM) for this period. SNR amplitude colour codes are not homogenized between conditions. S06 S07 S08 S09 S12 S13 S15 S23 S24 S19

Supplementary Material Figure S2b
S2b. No NREM / No Exposure group. Individual topographies for tone-and tritonerelated frequency-tagged responses in STAT an RDM conditions in the Wake (2nd) session. SNR amplitude colour codes are not homogenized between conditions. S02 S04 S05 S10 S14 S16 S26 S18

NREM / No Exposure participants
Potential between-group differences (NREM / Exposure vs.No NREM /No Exposure) in the temporal dynamics of tone-and tritone-related responses were assessed using linear mixed models (LMM). For the second five minutes STAT and RDM streams, the most appropriate model included the categorical MINUTES, STREAM and GROUP predictors but no interaction term (i.e. SNR ~ MINUTES + STREAM + GROUP). Tritone-related SNR tended to decrease during minutes 2-3 (estimate of the difference=-0.18 +/-0.11, t(133.8)=-1.7, p=0.09) and 3-4 (estimate of the difference =-0.23 +/-0.12, t(57.4)=-1.8, p=0.06), as compared to the minutes 1-2, but not during minutes 4-5 (estimate of the difference =-0.044 +/-0.14, t(24.97)=-0.3, p=0.7). Importantly for our purpose, tritone-related SNR did not differ between STAT and RDM streams (estimate of the difference =0.12 +/-0.10, t(21.7)=1.2, p=0.2) and exhibited a trend to be higher in the Prior Sleep/Exposure than in the Prior Wake/No Exposure group (estimate of the difference =0.21 +/-0.12, t(23.5)=1.7, p=0.09). Figure S6 S6. Individually fitted random slopes and intercepts (SNR) for the first 5-min STAT (left) and RDM streams (right). Figure S7, and time courses of tone-related SNR were examined separately for the first and the second 5-minutes STREAMs. Based on the AIC, random intercepts and random slopes for MINUTES and STREAM were included in the model.

Similarly, the time course of the tone-related responses averaged within sensors of interest (left and right temporal sensors) is illustrated in Supplementary Material
For the first five minutes streams, the most appropriate models for the fixed effect included the three predictors and interaction terms between continuous MINUTES and STREAM predictors for both linear and quadratic MINUTES terms (i.e. SNR ~ STREAM*MINUTES and SNR ~ STREAM*MINUTES 2 , AIC=765.4). The tone-related SNR responses linearly decreased across minutes for RDM streams (estimate of the minute slope = -0.6 +/-0.2, t(37.82)= -3.1, p<0.01, intercept estimate=8.0 +/-0.9) but not for STAT streams (estimate of the minute slope= -0.00 +/-0.2, t(37.82)=0.0, p=0.99; intercept estimate =5.4 +/-0.9). Intercepts of the MINUTES regression line between STAT and RDM streams were significantly different (estimate of the difference in intercept = 2.6 +/-0.9, t(37.12)=2.7, p=0.01). The interaction between MINUTES and STREAM was also significant (estimate of the slope difference between RDM and STAT = 0.6 +/-0.2, t(105)= -3.0, p=0.003). Hence, at the beginning of auditory streams exposure, tone-related SNR for STAT streams was lower than for RDM streams, but this difference was no longer significant at the end of the streams.
In addition, tone-related SNR in RDM streams linearly decreased as a function of time. Figure S7 S7. Temporal dynamics of tone-related SNR for the first (left) and second (right) 5minutes exposure during the Wake (2nd)  For the second 5-min streams, the most appropriate model for the fixed effect included predictors STREAM and continuous MINUTES (i.e., SNR ~ STREAM + MINUTES). Tonerelated SNR decreased across minutes both for STAT (slope estimate =-0.23 +/-0.11, t(21)=2.4, p=0.03, intercept estimate = 6.0 +/-0.7) and RDM (slopes estimate=-0.28 +/-0.11, t(21)=-2.4, p=0.03, intercept estimate = 6.7 +/-0.8) streams. There was no difference in intercept between streams (estimate of the difference=0.75 +/-0.44, t(21)=1.7, p=0.1). In other words, tone-response did not differ between STAT and RDM streams, and linearly decreased over time. Figure S8 S8. Transition probabilities matrices in statistical streams. The STAT set is used for exposure during the Nap opportunity and the subsequent Wake sessions, the TEST set is additionally used in the behavioural discrimination (2AFC) test (see 2AFC section below). Black, grey and white boxes respectively indicate transitional probabilities of 0, 33 and 100%. STAT and TEST sets share no positive transitional probabilities. None of the tritones from STAT and TEST sets started or ended with the same tones. Figure S9 S9. Neighbours definition template based on the triangulation method for Neuromag