Scream's roughness grants privileged access to the brain during sleep

Abstract

During sleep, recognizing threatening signals is crucial to determine when to wake up and when to continue vital sleep functions. Screaming is perhaps the most efficient way for communicating danger at a distance or in conditions of limited visibility. Screams are characterized by rapid modulations of sound pressure in the so-called roughness range (i.e., 30-150 Hz) which are particularly powerful in capturing attention. However, whether these rough sounds are also processed in a privileged manner during sleep is unknown. We tested this hypothesis by presenting human participants with low-intensity vocalizations, including rough screams and neutral, low-roughness vocalizations during wakefulness and during a full night of sleep. We found that screams evoked cortical responses with higher theta phase-consistency as compared to neutral vocalizations during both wakefulness and NREM sleep. In addition, screams boosted sleep spindle power, suggesting elevated stimulus salience. These findings demonstrate that, even at low sound intensity (e.g., from a distant source), vocalizations' roughness conveys stimulus relevance and enhances exogenous processing in both the waking and sleeping states. Preserved differential neural responses based on stimulus salience may ensure adaptive reactions in a state where the brain is mostly disconnected from external inputs.

Keywords: Auditory perception; EEG; Emotions; Sleep.

Fig. 1

Stimulation protocol and acoustic stimuli characterization. (A) Example hypnogram of one participant depicting the presentation of stimuli as a function of sleep stages over one night. Stimuli were first played in a brief session of wakefulness before sleep, during the three first hours of sleep, during the two last hours of sleep and after sleep, during another session of wakefulness. (B) Examples of two auditory stimuli waveforms in each condition: neutral (dark gray) and screamed (orange) vocalizations. Overlaid black line visually depicts that neutral voices are smoother than rough screams. (C) Eight screams and eight neutral vocalizations (from Arnal et al. 2015) were presented to participants in a pseudo-randomized fashion. This subset of vocalizations was chosen to cover a wide range of fundamental frequencies (f0, paired across neutral and screamed vocalizations) and roughness levels. As quantified in Arnal et al. 2015, the scatter plot reveals that screams and neutral vocalization mainly differ in their roughness levels. Crosses represent the mean (center) and standard error of the mean (tails) of pitch (x-axis) and roughness (y-axis) for screams (orange) and neutral vocalizations (dark gray).

Fig. 2

EEG responses to vocalizations during wakefulness. (A) Screams (orange) and neutral vocalizations (gray) evoke typical N1, P2 and N2 ERP components on Fz (top panel), and more prominent responses on Cz (middle panel, and related topographies below). P2 magnitude appears larger for screams than neutral vocalizations although this difference does not resist correction for multiple comparisons. (B, C) The power and ITPC (respectively B and C) time-frequency map of the average responses across all electrodes indicates that vocalizations evoke typical waves in the delta-theta (1–8 Hz) range. To analyze consistently wakefulness and sleep data, we excluded the delta (1–4 Hz) frequency range that could potentially be influenced by sleep oscillations (i.e., slow waves). The dashed black rectangle indicates the theta (4–8 Hz) frequency range investigated in further analyses. (D) Power response was averaged in the theta frequency range for screams (orange trace) and for neutral vocalizations (gray trace) but no measurable difference of power between them was observed. (E) ITPC response was also averaged in the theta range and shows that auditory phase-locked responses are larger for screams (orange trace) than for neutral vocalizations (gray trace). Shaded surfaces in plots indicate SEM. Significant statistical differences (corrected using cluster-based permutations) are plotted as thick horizontal lines (orange for screams against 0; gray for neutral vocalizations against 0 and black for the difference screams vs. neutral vocalizations).

Fig. 3

EEG responses to vocalizations during NREM sleep. (A) Screams (in orange) and neutral vocalizations (in gray) evoke slow-waves, with maximal amplitude on Fz electrode (top panel) and less prominent on Cz (middle panel) although initial auditory-evoked potentials can be visually observed on Cz before 0.6 s. Topographies (at the bottom) of the P200, N550 and P900 are typical of evoked slow-waves during sleep, with maximal amplitude on frontal electrodes. (B) Power time-frequency map averaging responses across all electrodes reveals enhanced power in the delta, theta and sigma frequency ranges. (C) Power responses averaged in the sigma (top panel), theta (middle panel) and delta (bottom panel) frequency ranges reveal an increase in all ranges investigated for both screams and neutral vocalizations except in the sigma range where only screams induce a significant increase. Note how the power in all bands qualitatively appears to be spread out over time relative to the awake state. (see Fig. 2D) (D) ITPC frequency map indicates that, as during wakefulness, vocalizations evoke phase-locked responses in the delta-theta (1–8 Hz) range. (E) Averaged ITPC response in the theta (4–8 Hz) range are larger for screams (orange trace) than for neutral vocalizations (gray trace), which are merely detectable. Plain lines indicate the mean response across participants and shaded surfaces the SEM in plots of D and E. Significant statistical differences (corrected using cluster-based permutations) are plotted as thick horizontal lines (orange for screams against 0; gray for neutral vocalizations against 0 and black for the difference screams vs. neutral vocalizations).

Fig. 4

Roughness (but not pitch) enhances evoked responses during sleep. (A, B) Regression of evoked responses (at each time point and frequency) by stimulus roughness (A) or pitch (B). Black dotted rectangles indicate time-frequency windows of interest in sleep-related frequency ranges, namely in the sigma (12–15 Hz) and delta (1–4 Hz) ranges, for subsequent statistical analyses. (C) Time-course of sigma, theta and delta power regressions by roughness (brown) and pitch (green). While the pitch of vocalizations does not seem to influence brain responses during sleep, roughness induces a significant delta- followed by a sigma-power increase during auditory stimulation. Shaded surfaces indicate SEM.