Supplementary Figure 1 Hippocampal recordings. a. (top) Post-operative MRI (left, depicting a depth electrode implanted along the longitudinal hippocampal axis) and co-registered preoperative MRI (right) of a sample participant. (bottom) Power spectra (NREM/WAKE) for all 10 electrode contacts of the same participant, showing a maximum in the spindle range (12-16 Hz, dashed vertical lines) in a posterior hippocampus contact which was then carried forward to the group analysis (red arrow). (inset) NREM/WAKE power spectrum for scalp electrode Cz. Y axes are identical across panels, where a ratio of 1 demarks the same spectral power during NREM and WAKE. b. Hippocampal contacts included in the analysis for all participants. Each panel shows sagittal and coronal slices of the post-operative (top) and the co-registered pre-operative (bottom) MRI for a participant. Crosshairs highlight the contact showing the largest spindle amplitudes during NREM sleep relative to WAKE of all hippocampal contacts (as shown in a for a sample participant). MRIs were unavailable for 3 of the 12 participants. Note the different implantation scheme in one participant (right bottom).
Supplementary Figure 2 Event-locked analysis of scalp EEG (Cz) PAC. Left: SO-spindle PAC. (a) Grand average unfiltered EEG trace across participants (mean s.e.m.), aligned to the maximum of the SO trough (time 0). (b) Average of spindle-trough-locked TFR (% change from pre-event baseline). Y-axis starts at 5 Hz to circumvent the dominance of power in the SO range. (c) Statistically significant change from pre-event baseline (P <.05, corrected). Inset shows unit circle of preferred phases of the SO-spindle modulation for each participant, which illustrates the preferred clustering of spindle power towards the SO peak (328, white line). Yellow circles represent participants whose Rayleigh test for non-uniformity was significant at P <.05. Right: Spindle-ripple PAC. While the EEG trace in (a) reveals the typical waxing and waning pattern of sleep spindles, no reliable ripple power modulation was observed in the spindle-peak-locked TFR (b) (see also Supplementary Table 3b; presumably due to attenuated signal-to-noise ratio of scalp EEG recordings for higher frequencies). Note that the spindle s mean potential is above zero, reflecting the grouping of spindles in the SO peak (.25 s to +.25 s, t(11) = 3.45, P <.01). This is further illustrated in the inset, which shows the grand average EEG trace bandpass filtered from 0.5 1.25 Hz (SO range) and from 12 16 Hz (spindle range), respectively.
Supplementary Figure 3 Peri-event time histograms (PETH). (top) PETH of SO occurrences time-locked to SO down-states for Cz (left) and HC (right), highlighting multiple peaks at the distance of one SO cycle (multi-event oscillations). (bottom) PETH of spindle occurrences time-locked to spindle centers, revealing discernable peaks repeating at the SO frequency.
Supplementary Figure 4 Example hippocampal EEG traces illustrating multi-event SOs along with nested spindles. Top: Unfiltered raw EEG traces, highlighting data segments identified as a SO (green) and spindle (blue) by our automated algorithm. Bottom: Same data segments after applying the event-specific bandpass filters (SO: 0.5 1.25 Hz, spindle: 12 16 Hz).
Supplementary Figure 5 Temporal relation of spindles with respect to ripples. Bar graphs show the peri-event time histogram (PETH) of spindle center occurrences relative to ripple centers. Spindle centers show the tendency to occur after ripple centers: Within a search interval from -1 s to +1 s around ripple centers, 15.2% (s.d. = 5.0) of all ripples were followed by spindle centers and 12.6% (s.d. = 4.6) of all ripples were preceded by spindle centers. Note however that this difference in occurrence probabilities was not reliable across participants (t(11) = 1.44, P =.177). Importantly, the tendency of spindle centers to occur after ripple centers does not indicate that ripples trigger spindles, as the spindle onset occurs at least 250 ms before the spindle center (given our spindle detection duration criterion of 500 ms). The inset shows the onsets of spindles with respect to the onset of ripples, revealing that 17.1% (s.d. = 5.4) of all ripple onsets were preceded by a spindle onset from -1 to 0 s, whereas only 9.9% (s.d. = 3.2) of all ripple onsets were followed by a spindle onset from 0 to +1 s (t(11) = 4.08, P =.002). This indicates that ripples reliably occur after a spindle has already started. Note that the same statistical result was obtained when reducing the time window to.5 s to +.5 s around ripple onset.
Supplementary Figure 6 Spindle-ripple coupling as a function of associated SOs. a. Peri-event time histogram (PETH) showing the probability of SO occurrence (SO down-state) relative to spindle centers (t = 0 s, dashed vertical line). Dark gray area shows time interval used to determine whether a given spindle is located in the up-state of an SO (i.e., a SO down-state is temporally preceding a spindle at 1 to 0.2 s). Note that while there is a clear peak of SO occurrences in the temporal vicinity of spindles (with the asymmetry pointing to a predominant SO -> spindle sequence), many spindles occurred independently of SOs. The subsequent analyses separated spindle-ripple coupling into spindles immediately preceded by SO downstates (left column, average number of events across participants = 124, range = 30 315) and spindles not immediately preceded by SO down-states (right column, average number of events across participants = 697, range = 154 1467). b. Event-locked analysis. (i) (top) Grand average unfiltered ieeg trace across participants (mean s.e.m.), aligned to the maximum of the spindle trough. Insets show the bandpass filtered trace from 0.5 1.25 Hz (SO range) and from 12 16 Hz (spindle range) to better illustrate the relation of the spindle to the SO down-state. Note the pronounced SO emerging in spindle-locked raw traces on the left. (middle) Average of spindletrough-locked TFR (% change from pre-event baseline). (bottom) Statistically significant change from pre-event baseline (P <.05, corrected, with the initial cluster threshold also set to P <.05 to accommodate smaller trial numbers). Inset shows unit circle of preferred phases of the spindle-ripple modulation for each participant. Both V tests (testing for preferred phase at 180 ) are significant: spindles immediately preceded by SO down-states: V = 4.36, P =.037, spindles not immediately preceded by SO down-states: V = 5.75, P =.010. (ii) Direct comparison of spindle-trough-locked TFRs (immediately preceded by SO down-states vs. not immediately preceded by SO down-states), revealing only a relative power decrease at SO down-states, but no differential spindle-ripple modulation. (iii) Example data from one participant (same as in main Fig. 2-4), showing the nesting of ripples in spindle troughs for both event types. c. Comodulogram analysis. PAC was calculated for data segments around spindles ( 2.5 s to +2.5 s relative to the maximum spindle trough). Maps show cluster-corrected comparisons against phase-scrambled surrogate data (all P <.05). Direct comparison of MI averaged from 12 16 Hz phase-providing frequency and 80 100 Hz amplitude-providing frequency revealed no difference between spindles immediately preceded by SO down-states vs. spindles not immediately preceded by SO down-states (using surrogate corrected MI or non-surrogate-corrected MI, both t(11) < 0.71, P >.49).
Supplementary Figure 7 Example hippocampal EEG traces containing a SO-spindle-ripple interaction. Top: Unfiltered raw EEG traces, highlighting data segments identified as a SO (green), spindle (blue) and ripple (red) by our automated algorithm. Bottom: Same data segments after applying the event-specific bandpass filters (SO: 0.5 1.25 Hz, spindle: 12 16 Hz, ripple: 80 100 Hz). Insets zoom in on the ripple event, highlighting its oscillatory nature and its nesting in spindle troughs.
Supplementary Figure 8 Cross-regional CFC between Cz and hippocampus. Peri-event time histograms (PETH) depicting the occurrence probability of hippocampal SOs time-locked to Cz SOs (a) and of hippocampal spindles time-locked to Cz spindles (b).
Supplementary Figure 9 Hippocampal phase-amplitude coupling (PAC) during NREM sleep. Clusters showing a significant Modulation Index (MI) when comparing NREM PAC with WAKE PAC (left) and with trial-shuffled NREM surrogate data (right). Maps are thresholded to show t value clusters whose P < 0.05 and whose size exceeds 25 contiguous frequency pairs.
Supplementary Figure 10 Event-locked analysis of hippocampal PAC after re-referencing hippocampal data against the most anterior (11 participants) or lateral (1 participant) contact on the same depth electrode. (a) Contact used for re-referencing in a sample participant (red line). (b) Grand average unfiltered EEG trace across participants (mean s.e.m.) and (c) average of event-locked TFR (% change from pre-event baseline), locked to SO peak (left) and spindle trough (right). Spindle power (12 16 Hz) shows a significant increase vs. baseline from.25 to.75 s after SO peak (t(11) = 2.69, P =.021) and ripple power (80 100 Hz) shows a significant increase vs. baseline from.25 to +.25 s around maximal spindle trough (t(11) = 4.00, P =.002). Spindle power (12 16 Hz) also showed a significant increase in the re-referenced data when aligning the TFR to the ripple center (as done in main text Fig. 4a, t(11) = 4.81, P <.001,.25 to +.25 s around ripple center).
Supplemental Tables Table S1: Mean proportions of time spent in different sleep stages out of total time asleep. Sleep efficiency denotes time spent in Stages 1-4 or REM between sleep onset and last wake up. mean s.d. Stage1 28% 21% Stage2 41% 15% Stage3 12% 9% Stage4 3% 4% REM 16% 9% Sleep time 8.42h 2.00h Sleep efficiency 90% 10% Table S2a: Length of artifact free NREM data segments ( dataminutes ), total number of events and event densities (n events/dataminutes) for SOs, spindles and ripples detected in the hippocampus. SOs spindles ripples participant dataminutes n events density n events density n events density p01 122 421 3.4 674 5.5 75 0.6 p02 104 319 3.1 378 3.6 280 2.7 p03 215 1015 4.7 1487 6.9 194 0.9 p04 169 530 3.1 572 3.4 207 1.2 p05 162 450 2.8 523 3.2 107 0.7 p06 120 382 3.2 517 4.3 83 0.7 p07 187 743 4.0 1340 7.2 162 0.9 p08 160 886 5.6 1725 10.8 348 2.2 p09 147 479 3.3 683 4.6 112 0.8 p10 145 579 4.0 953 6.6 83 0.6 p11 50 186 3.7 184 3.7 104 2.1 p12 157 544 3.5 816 5.2 233 1.5 mean 145 545 3.7 821 5.4 166 1.2 s.d. 42 236 0.8 470 2.2 88 0.7 HC
Table S2b: Length of artifact free NREM data segments ( dataminutes ), total number of events and event densities (n events/dataminutes) for SOs and spindles detected at Cz. SOs spindles participant dataminutes n events density n events density p01 180 1045 5.8 1494 8.3 p02 169 748 4.4 1215 7.2 p03 277 1504 5.4 2407 8.7 p04 231 1096 4.7 1466 6.3 p05 244 748 3.1 2299 9.4 p06 179 1045 5.9 1074 6.0 p07 257 1422 5.5 1506 5.9 p08 165 784 4.7 1391 8.4 p09 227 1123 4.9 1903 8.4 p10 148 673 4.6 1182 8.0 p11 86 676 7.8 513 5.9 p12 189 782 4.2 1313 7.0 mean 196 971 5.1 1480 7.5 s.d. 53 283 1.2 523 1.2 Cz Table S3a. EEG SO metrics and results for SO-spindle coupling using various event detection criteria. EEG metrics include the number of detected events (avg n) as well as the across-participant average of the peak, trough and peak-to-peak amplitude after bandpass filtering the EEG in the SO frequency ranges. PAC results include (i) the change in spindle power (12 16 Hz, relative to the 2.5 s to 1.5 s pre-event baseline) during the SO up-state (.25 s to.75 s relative to maximal down-state) along with results of a two-tailed t-test against 0, (ii) P values of a V-test against the hypothesized mean direction of 180 (i.e., modulation by the SO up-state) and (iii) number of significant participant-specific Rayleigh tests against the H 0 of uniform distribution of preferred phase angles across all detected events (in the overall group of 12 participants). Percentiles denote the p2p amplitude cut-off, such that only the top 25% SOs were included for the 75 th percentile criterion. Mean + 2SD denotes the SOs whose p2p amplitude mean + 2 standard deviations of all SO candidates. Multi-event SOs denotes SOs that were preceded and/or followed by another SO down-state within 1.3 s (~.75 Hz). 75th perc., cons.art.det. denotes SOs whose p2p amplitude 75th percentile (i.e., the top 25%), as used in the main text, but after raising the thresholds for automated artifact detection (resulting in fewer rejected data segments).
SO EEG PAC Hippocampus avg n peak trough p2p spindle power (.25.75 s, 12 16 Hz) V-test (180 ) P value individually significant (Rayleigh test) % change t p 75 th percentile 545 134.5-130.4 264.9 45.6 3.90.003.001 9 75 th percentile (both halfwaves 400 ms) 75 th percentile (for both half-waves separately) 527 130.2-128.0 258.2 44.0 3.75.003.001 9 258 150.9-151.1 302.1 51.5 3.60.004.005 9 75 th - 95 th percentile 482 115.9-109.4 225.3 39.3 3.80.003.001 11 80 th percentile 482 137.6-134.1 271.7 48.5 4.07.002.001 9 85 th percentile 362 148.7-144.5 293.2 52.3 3.86.003.000 9 90 th percentile 241 161.5-161.3 322.8 56.8 3.62.004.003 9 mean + 2SD 93 194.8-202.9 397.7 57.7 3.80.003.047 7 multi-event SOs 133 136.7-135.8 272.5 21.7 3.03.015.041 8 75 th perc., cons.art.det. 781 146.7-140.4 287.0 49.5 4.40.001.001 10 SO EEG PAC Cz avg n peak trough p2p spindle power (.25.75 s, 12 16 Hz) V-test (0 ) P value individually significant (Rayleigh test) % change t p 75 th percentile 971 31.1-31.2 62.2 27.3 5.73.000.004 10 75 th percentile (both halfwaves 400 ms) 75 th percentile (for both half-waves separately) 872 31.3-31.5 62.8 28.2 5.82.000.004 10 487 36.8-36.9 73.7 35.8 4.83.001.003 11 75 th - 95 th percentile 804 26.4-26.4 52.8 23.5 5.97.000.020 8 80 th percentile 803 33.1-33.3 66.4 29.1 5.54.000.004 12 85 th percentile 602 36.2-36.6 72.8 34.3 6.33.000.001 11 90 th percentile 402 40.5-41.1 81.7 40.0 5.15.000.002 10 mean + 2SD 202 47.5-48.2 95.8 45.9 4.70.001.009 9 multi-event SOs 317 32.3-32.8 65.1 21.5 4.99.000.001 10 75 th perc., cons.art.det. 1226 33.5-33.8 67.3 28.9 5.41.000.008 12
Table S3b. EEG spindle metrics and results for spindle-ripple coupling using various event detection criteria. EEG metrics include the number of detected events (avg n) as well as the across-participant average of the peak, trough and peak-to-peak amplitude after bandpass filtering the EEG in the spindle range. PAC results include (i) the change in ripple power (80 100 Hz, relative to the 2.5 s to 1.5 s preevent baseline) from.25 s to +.25 s relative to the maximal spindle trough (HC) or peak (Cz) along with results of a two-tailed t-test against 0, (ii) P values of a V-test against the hypothesized mean direction of 180 (HC, i.e., modulation by the spindle trough) or 0 (Cz, i.e., modulation by the spindle peak) and (iii) number of significant participant-specific Rayleigh tests against the H 0 of uniform distribution of preferred phase angles across all detected events (in the overall group of 12 participants). Percentiles denote the RMS amplitude cut-off, such that only the top 25% spindles were included for the 75 th percentile criterion. Mean + 2SD denotes the spindles whose RMS amplitude mean + 2 standard deviations of all spindle candidates. 75th perc., cons.art.det. denotes spindles whose RMS amplitude 75th percentile (i.e., the top 25%), as used in the main text, but after raising the thresholds for automated artifact detection (resulting in fewer rejected data segments). Hippocampus avg n peak spindle EEG trough p2p ripple power (.25 s, 80 100 Hz) % change t p PAC V-test (180 ) P value individually significant (Rayleigh test) 75 th percentile 821 59.8-59.8 119.6 17.9 4.59.001.010 11 80 th percentile 664 63.6-63.6 127.1 17.9 5.89.000.010 12 85 th percentile 496 68.2-68.2 136.5 19.9 6.76.000.007 11 90 th percentile 317 74.5-74.5 149.1 20.3 6.89.000.006 11 mean + 2SD 150 83.9-84.0 167.8 21.0 3.63.004.002 10 75 th perc., cons.art.det. 1170 63.6-63.5 127.1 19.8 4.41.001.011 12 Cz avg n peak spindle EEG trough p2p ripple power (.25 s, 80 100 Hz) % change t p PAC V-test (0 ) P value individually significant (Rayleigh test) 75 th percentile 1480 12.5-12.5 25.0 0.3 0.59.567.829 2 80 th percentile 1182 13.2-13.2 26.4-0.4 0.58.573.724 3 85 th percentile 877 14.1-14.1 28.2 0.0 0.03.975.906 1 90 th percentile 558 15.3-15.3 30.7-0.3 0.26.802.826 0 mean + 2SD 257 17.2-17.2 34.5 1.1 0.93.371.893 0 75 th perc., cons.art.det. 1805 12.7-12.7 25.4 1.0 1.98.074.650 2
Table S3c. Results for spindle-ripple coupling in ripple-locked TFR using various event detection criteria. PAC results include the change in spindle power (12 16 Hz, relative to the 1.5 s to 1.0 s pre-event baseline) from.25 s to +.25 s relative to the ripple peak along with results of a two-tailed t-test against 0. Percentiles denote the RMS amplitude cut-off, such that only the top 1% ripples were included for the 99 th percentile criterion. Mean + 3SD denotes the ripples whose RMS amplitude mean + 3 standard deviations of all ripple candidates. 99th perc., cons.art.det. denotes ripples whose RMS amplitude 99th percentile (i.e., the top 1%), as used in the main text, but after raising the thresholds for automated artifact detection (resulting in fewer rejected data segments). PAC Hippocampus avg n spindle power (.25 s, 12 16 Hz) % change 99 th percentile 166 65.5 4.49.001 98 th percentile 354 43.0 4.45.001 mean + 3SD 203 60.0 4.37.001 99 th perc., cons.art.det. 238 67.5 4.53.001 t p Table S4. Participant s drug regimen at the time of recordings participant Anticonvulsant Antidepressant p01 Clobazam, Valproat - p02 Clobazam, Valproat - p03 Lamotrigin, Levetiracetam - p04 Lacosamide, Levetiracetam - p05 Lamotrigin, Levetiracetam - p06 Lamotrigin, Oxcarbazepin - p07 Lamotrigin, Oxcarbazepin Citalopram p08 Lamotrigin - p09 Levetiracetam - p10 Lamotrigin Sertralin p11 Levetiracetam, Oxcarbazepin - p12 Oxcarbazepin -
Table S5. Documented effects of anticonvulsant and antidepressant drugs on sleep architecture/oscillations. increase decrease. N1-3 sleep stages according to Iber, C., Ancoli- Israel, S., Chesson, A. & Quan, S.F. The AASM manual for the scoring of sleep and associated events: rules, terminology and technical specifications. (2007). 1... D. Neckelmann, B. Bjorvatn, A. Bjørkum, R. Ursin, Behavioural brain research 79, 183 (1996). 2... A. L. van Bemmel, R. H. van den Hoofdakker, D. G. Beersma, A. L. Bouhuys, Psychopharmacology 113, 225 (1993). 3... S. Wilson et al., European neuropsychopharmacology 14, 367 (2004). 4... S. V. Jain, T. A. Glauser, Epilepsia 55, 26 (2014). 5... J. D. Hudson et al., Seizure 25, 155 (2014). 6... R. D. Jindal et al., Journal of clinical psychopharmacology 23, 540 (2003). Drug Effects on sleep structure Citalopram REM ; REM alpha power ( 1-3 ) Clobazam N2, N1, SWS ( 4 ) Lacosamide none ( 5 ) Lamotrigin REM, N2, SWS ( 4 ) Levetiracetam N2, REM ( 4 ) Oxcarbazepin unknown Sertralin delta power during first sleep cycle ( 6 ) Valproat N1, REM ( 4 )