Increased neuronal signatures of targeted memory reactivation during slow-wave up states

It is assumed that slow oscillatory up-states represent crucial time windows for memory reactivation and consolidation during sleep. We tested this assumption by utilizing closed-loop targeted memory reactivation: Participants were re-exposed to prior learned foreign vocabulary during up- and down-states of slow oscillations. While presenting memory cues during slow oscillatory up-states improved recall performance, down-state cueing did not result in a clear behavioral benefit. Still, no robust behavioral benefit of up- as compared to down-state cueing was observable. At the electrophysiological level however, successful memory reactivation during up-states was associated with a characteristic power increase in the theta and sleep spindle band. No oscillatory changes were observable for down-state cues. Our findings provide experimental support for the assumption that slow oscillatory up-states may represent privileged time windows for memory reactivation, while the interplay of slow oscillations, theta and sleep spindle activity promotes successful memory consolidation during sleep.


Supplementary
Data are means ± SEM; Numbers indicate absolute or relative values of correctly recalled words that where either presented during SOs up-and down-states or remained uncued. For cued recall testing, the number of correctly recalled words during the learning phase before and the retrieval phase after the retention interval are indicated. Change (% Change) refers to the absolute (relative) difference in performance between the learning and retrieval phases.  Teil  part  deur  Tür  door  dief  Dieb  thief  dijk  Teich  pond  doek  Tuch  cloth  dorp  Dorf  village  duim  Daumen  thumb  eed  Eid  oath  fles  Flasche  bottle  fout  Fehler  fault  geur  Geruch  odor  gif  Gift  poison  hak  Absatz  heel  hei  Heide  heath  hiel  Ferse  heel  hout  Holz  wood  hulp  Hilfe  help  hut  Hütte  hut  inkt  Tinte  ink  jas  Jacke  jacket  kast  Schrank  closet  kerk  Kirche  church  kok  Koch  cook  kras  Kratzer  scratch  kruk  Krücke  crutch  kus  Kuss  kiss  kust  Küste  coast  kwal  Qualle  jellyfish  lens  Linse  lens  lijf  Leib  body  lijm  Kleber  glue  lip  Lippe  lip  loof  Laub  foliage  melk  Milch  milk  mes  Messer  knife  mond  Mund  mouth  mug  Mücke  mosquito  muts  Mütze  cap  muur  Mauer  wall  neef  Neffe  nephew  neus  Nase  nose  nier  Niere  kidney   oog  Auge  eye  pad  Pfad  path  piek  Gipfel  peak  pijn  Schmerzen  pain  pijp  Pfeife  pipe  pols  Puls  pulse  pont  Fähre  ferry  prik  Spritze  syringe  rek  Regal  rack  rib  Rippe  rib  rijst  Reis  rice  rit  Fahrt  drive  Wordlist used for memory task. Dutch-German word pairs used during the memory task.

Supplementary Figure 1: Slow Wave Detection Algorithm State Diagram
Online Detection Finite-State Machine Diagram Implementation of the slow-wave detection algorithm as a finite state machine. The algorithm starts at the black dot and traverses through the states while it is running.

Supplementary Figure 2: ERPs for Remembered and Non-Remembered cues
Comparison of ERPs for remembered and non-remembered words. ERPs for up-(blue) and down-state (red) remembered (solid line) and non-remembered (dashed line) words are shown. There is no significant difference between remembered and non-remembered word cues.

Supplementary Figure 3: Phase Distribution across the Scalp
Topographical distribution of phase. Signal phase at stimulus release for up-state cues (left) and down-state cues (right). While the algorithm detects slow-waves at the Fz electrode only, the phase distribution across the scalp is uniform at the time of cue onset. Up-state phase is around -20° and down-state phase around 120°.

Supplementary Figure 4: Phase Accuracy for each Subject
Phase accuracy for each subject at trial level. Up-state cues are shown in blue. Down-state cues are shown in red. Trial level phase accuracy for each individual subject shows a clear distinction between up-and down-state cues for all subjects.
Oscillatory analysis of up-versus down-state cues. Panel a) illustrates the power contrast between words presented during SOs up-and down-states. Black bars (significant cluster in frequency band analysis) with white lines below and above the time-frequency plot indicate the number of significantly differing electrodes for the theta and spindle band respectively. The full height of the bar corresponds to 100% (31) electrodes. Panel b) topographical distribution of the areas marked with a dashed box in a) for the spindle (top row) and theta (bottom row) band, pre-stimulus (left column) and post-stimulus (right column). Significant electrodes are shown as filled black dots. Panel c) shows the same data averaged across time, frequency and significant channels within the respective cluster. The power is scaled between -1 and 1 for both panels a) and b). Time-frequency data is shown for the Fz electrode.

Sleep stage specific EEG results
For memory cues played during the up-state in sleep stage N2 (n = 72.31 ± 11.72) we to not-remembered words. Averaged over time, channels and frequency band, within these clusters this difference was significant in the theta band (t15 = 2.00, P = 0.032; see b right column, bottom row) and in the spindle band (t15 = 2.73, P = 0.008; see b, right column, top row). For words presented during down-states c) no significant difference emerged between remembered and forgotten words, neither in the sleep spindle nor the theta band. Consequently, averaged activity in those clusters observed in the analysis of SO up-states did not reveal any significant differences for down-state cues, neither in the theta (t15 = -0.35, P = 0.635) nor the spindle band (t15 = -1.50, P =0.923). The difference between the two contrasts of up and down e) showed enhanced power in the spindle band, but not in the theta band. Averaged activity in the spindle cluster (t15 = 2.34, P = 0.017; see f right column, top row) showed a significant difference, while theta activity (averaged over the duration of the SO up-state theta cluster) showed no statistical difference (t15 = 1.44, P = 0.109; see f right column, bottom row). Mean ± s.e.m. are indicated. **: P < 0.01; *: P < 0.05. Averaged over time, channels and frequency band, within these clusters this difference was significant in the theta band (t15 = 2.43, P = 0.014; see b right column, bottom row) but not in the spindle band (t15 = 1.28, P = 0.109; see b, right column, top row). For words presented during down-states c) no significant difference emerged between remembered and forgotten words, neither in the sleep spindle nor the theta band. Consequently, averaged activity in those clusters observed in the analysis of SO up-states did not reveal any significant differences for down-state cues, neither in the theta (t15 = 0.05, P = 0.480) nor the spindle band (t15 = 0.07, P =0.472). The difference between the two contrasts of up and down e) again showed no enhanced power in the spindle band, nor in the theta band. Averaged activity in the spindle cluster (t15 = 0.33, P = 0.374; see f right column, top row) showed no significant difference, but averaged theta activity showed statistical difference (t15 = 1.84, P = 0.043; see f right column, bottom row). . b) right column shows the averaged relative difference between N2 and SWS averaged over time frequency and channels. We find a significant difference between N2 and SWS for both theta (t15 = 7.04, P < 0.001; b) bottom right) and spindles (t15 = 2.76, P = 0.007; b) top right). Mean ± s.e.m. are indicated. **: P < 0.01; ***: P < 0.001. Figure 10: ERP for 0.5 -1 Hz Bandpass filtered Signal 0.5 -1 Hz bandpass filtered ERP. To validate whether the targeting algorithm (lowpass filtered at 1.5 Hz) captures the correct states of slow wave (0.5 -1.0 Hz) the data was bandpass filtered between 0.5 -1 Hz. Comparable to the main results ( Figure 1) the ERPanalyses revealed that up-state cues were located at the down-to-up transition of the cortical slow wave (beginning of slow oscillatory up-state), and that down-state cues were played at the up-to-down transition (beginning of slow wave down-state), confirming that the slow wave detection algorithm does indeed detect the proposedly critical states of the slow wave. Data is shown for electrode Fz.