Sound elicits stereotyped facial movements that provide a sensitive index of hearing abilities in mice

Abstract

Sound elicits rapid movements of muscles in the face, ears, and eyes that protect the body from injury and trigger brain-wide internal state changes. Here, we performed quantitative facial videography from mice resting atop a piezoelectric force plate and observed that broadband sounds elicited rapid and stereotyped facial twitches. Facial motion energy (FME) adjacent to the whisker array was 30 dB more sensitive than the acoustic startle reflex and offered greater inter-trial and inter-animal reliability than sound-evoked pupil dilations or movement of other facial and body regions. FME tracked the low-frequency envelope of broadband sounds, providing a means to study behavioral discrimination of complex auditory stimuli, such as speech phonemes in noise. Approximately 25% of layer 5-6 units in the auditory cortex (ACtx) exhibited firing rate changes during facial movements. However, FME facilitation during ACtx photoinhibition indicated that sound-evoked facial movements were mediated by a midbrain pathway and modulated by descending corticofugal input. FME and auditory brainstem response (ABR) thresholds were closely aligned after noise-induced sensorineural hearing loss, yet FME growth slopes were disproportionately steep at spared frequencies, reflecting a central plasticity that matched commensurate changes in ABR wave 4. Sound-evoked facial movements were also hypersensitive in Ptchd1 knockout mice, highlighting the use of FME for identifying sensory hyper-reactivity phenotypes after adult-onset hyperacusis and inherited deficiencies in autism risk genes. These findings present a sensitive and integrative measure of hearing while also highlighting that even low-intensity broadband sounds can elicit a complex mixture of auditory, motor, and reafferent somatosensory neural activity.

Keywords: acoustic startle reflex; auditory brainstem response; auditory cortex; autism; behavior; corticofugal; hearing loss; inferior colliculus; pupil.

Figure 1 –. Facial movements provide a sensitive and specific behavioral index of hearing in mice

(A) Schematic depicts mouse, camera, speaker, and summed output of three piezoelectric transducers attached to the force plate. The multi-peaked startle waveform is copied from the first 115dB SPL trial shown in D. (B) Video frames from a single trial depict changes in pinna, jaw, nose, pupil, and eyelid positions determined by DeepLabCut. (C) Motion energy calculated from a region of interest positioned caudal to the vibrissa array. (D) Movement amplitudes for 7 consecutive trials in a representative mouse. Y-axis scales presented at right. (E) Mean peristimulus force plate and facial movement responses from 8 mice. (F) Mean ± SEM movement amplitudes. Vertical axes on the left and right refers to the videographic movement and startle reflex, respectively. (G) Left: As per A, except that high-resolution video recordings (150 frames/s) are made of the hind paw. Right: Blue circle indicates DeepLabCut tracking of a point on the hind paw in four frames relative to the onset of a noise burst. (H) Mean peristimulus force plate and hind paw video responses from 8 mice. Force plate and hind paw pseudocolor scales match the startle reflex and nose videography plots presented above in E. (I) Startle thresholds for each mouse (circle) and sample mean. Arrow indicates one outlying value outside of the plotted range. There was no significant difference in startle reflex threshold measured via the piezoelectric force plate or hind paw videography (paired t-test, p = 0.17). (J) Left, mean FME amplitude in response to a high-contrast drifting visual grating compared to a 60 dB SPL white noise burst (N = 8). Right, moderate intensity noise bursts elicited significantly greater FME than visual gratings (paired t-test, p =3 × 10⁻⁵). (K) Minimum sound intensity that elicited movement presented for each mouse (circle) and sample mean (horizontal bar). Threshold varied significantly across movement types (one-way repeated measures ANOVA, F = 13.45, p = 2 × 10⁻⁸) with post-hoc comparisons finding significant differences between FME and jaw (p = 0.01) and startle reflex (p = 0.0002). For all figures, black and gray horizontal bars denote significant (p < 0.05) and non-significant differences, respectively, with FME after Holm-Bonferroni corrections for multiple comparisons. (L) Sound-evoked movement latencies presented for each mouse (circle) and sample mean (horizontal bar). Response latency varied significantly across movement types (one-way repeated measures ANOVA, F = 550.67, p = 8 × 10⁻³⁴) with post-hoc comparisons finding significant differences between FME and nose (p = 0.01), pupil dilation (p = 2 × 10⁻⁷), eyelid (p = 0.01), and startle reflex (p = 0.0003). Gray arrows for pupil denote that values were outside of the y-axis range (mean latency = 627 ms). (M) Trial-to-trial variability measured with the coefficient of variation presented for each mouse (circle) and sample mean (horizontal bar). Trial-to-trial variability was significantly different between movement types (one-way repeated measures ANOVA, F = 23.18, p = 6 × 10⁻¹¹) with post-hoc comparisons finding that FME was significantly less variable than all other movement types (p < 0.0004 for all comparisons). (N) Inter-subject variability measured with the coefficient of variation across subjects for each trial (circle) and sample mean (horizontal bar). Inter-subject variability was significantly different between movement types (one-way repeated measures ANOVA, F = 59.12, p = 1 × 10⁻³⁴) with post-hoc comparisons finding that FME was significantly less variable than all other movement types (p < 0.005 for all comparisons). See also Figure S1.

Figure 2 –. Facial movements synchronize to slow changes in the sound pressure envelope.

(A) Silent gaps of varying durations were introduced in a constant background of 50 dB SPL white noise. Mean FME before, during, and after the silent gap. (B) FME significantly increased with gap duration (one-way repeated measures ANOVA, F = 9.43, p = 3 × 10⁻¹¹). Individual mice and sample mean (N = 8) are plotted as thin gray lines and thick black line, respectively. (C) Gap detection thresholds for each mouse. (D) Spectrogram depicts downward frequency modulated sweeps presented at 2.5 Hz with a 50% duty cycle (white) at 70 dB SPL. Mean ± SEM FME amplitude for an example mouse (red) shows a facial twitch elicited by each of the six consecutive FM sweeps. (E) Fourier analysis of FME responses from 8 mice to the FM sweep sequence presented at 2.5 Hz yields a peak at the presentation rate (dashed vertical line). Individual mice and sample mean are plotted as thin gray lines and thick red line, respectively. (F) Facial synchronization was calculated as the power at the stimulus presentation rate relative to the noise floor. Synchronization significantly decreases across higher FM sweep presentation rates (one-way repeated measures ANOVA, F = 52.73, p = 6 × 10⁻¹⁸).

Figure 3 –. Decoding phonemes in background noise via facial movements.

(A) Spectrograms of two English speech tokens digitally resynthesized to span the mouse hearing range without distorting the spectrotemporal envelope of the source signal. (B) Spectrogram plots six presentations of the phoneme, Gee, presented at 1 Hz (grayscale, right vertical axis). Mean ± SEM FME from a representative mouse elicited by each speech token (red, left vertical axis). (C) Mean ± SEM FME for Gee and Ha presented at 70 dB SPL without background noise and five levels of increasingly intense background noise (N = 8). (D) FME synchronization to the speech token presentation rate decreased significantly, but equivalently, for both phonemes across increasingly levels of background noise (2-way repeated measures ANOVA, main effect for noise level [F = 12.36, p = 2 × 10⁻⁸], main effect for phoneme [F = 0.78, p = 0.39]). (E) Single trial speech token classification accuracy with actual and shuffled (shuff) assignment of stimulus identity. Chance classification = 50%. Classification accuracy was significantly greater for actual than shuffled stimulus label assignments for all background noise levels (paired t-tests, p < 0.002 for all), except at the highest noise level (50 dB SPL, p = 0.07). Thin lines and thick horizontal bars denote individual mice and sample means, respectively. See also Figure S2.

Figure 4 –. Noise-induced SNHL causes a combination of high-frequency threshold shift and excess low-frequency gain in the ABR

(A) Two hours of exposure to 16–32 kHz octave-band noise (OBN) at 103 dB SPL causes sensorineural damage in the high-frequency base of the cochlea, as determined with pinna-vertex ABR measurements. (B) Tone-evoked ABR measurements reveal significant threshold elevation for test frequencies above 11.3 kHz measured before vs two weeks after noise exposure (2-way repeated measures ANOVA, N=9, main effect for group, F = 146.95, p = 9 × 10⁻⁹; group × frequency interaction, F = 39.73, p = 9 × 10⁻¹⁶]). Asterisks denote post-hoc pairwise comparisons (p < 0.05 for all). (C) Mean peristimulus FME responses from 8 mice for pure tone and broadband noise stimuli. (D) Mean ± SEM FME amplitudes over sound levels. Broadband noise evokes significantly larger facial movements than pure tones [2-way repeated measures ANOVA, N = 8, main effect for stimulus type, F = 54.95, p = 4 × 10⁻⁶; group x frequency interaction, F = 10.87, p = 3 × 10⁻⁵]. (E) OBN centered at 8 and 32 kHz elicits a robust multi-peaked ABR before acoustic trauma. Two weeks after noise-induced high-frequency SNHL, the ABR response to 32 kHz noise band is virtually absent at sound levels up to 80 dB SPL, whereas responses to the 8 kHz noise band appear unaffected or slightly larger than baseline measurements. Arrows indicate the appearance of ABR waves (w) 1–5. (F) OBN ABR thresholds after noise exposure (two-way repeated measures ANOVA, N = 8; main effect for frequency [F = 38.26, p = 2 × 10⁻⁵]; main effect for timepoint [F = 46.59, p = 8 × 10⁻⁶], frequency x timepoint [F = 38.259, p = 1 × 10⁻⁴]). (G) Schematic illustrating the neural generators of ABR waves 1–5 and the expected transition from slight attenuation in the 8 kHz OBN level x amplitude input-output function for early waves to excess central gain measured in later waves. NA indicates that the neural generators of the ABR do not include central auditory structures above the midbrain. (H) Mean ± SEM 8 kHz OBN-evoked normalized wave amplitude growth functions plotted relative to threshold. Inset: 8 kHz OBN-evoked normalized wave amplitudes were averaged within a 30–45 dB range above threshold. Solid black line represents no change, gray lines represent individual subjects, thick gray and dashed black lines represent baseline and 2 weeks post-exposure, respectively. Asterisks denote p < 0.05.

Figure 5 –. Changes in sound-evoked facial movement after noise-induced SNHL parallel modifications in late ABR waves

(A) Mean ± SEM facial movements evoked by broadband noise, octave-band noise (OBN), and pure tones at 70 dB SPL. OBN and pure tone responses are averaged across 8 and 32 kHz. (B) OBN responses are elicited at low sound levels and grow monotonically with sound level (One-way repeated measures ANOVA [F=189.9, p = 6 × 10⁻⁶⁶, N = 20]). (C) Sound-evoked facial movements elicited by OBN centered at 8 and 32 kHz are stable over a 17 day (D) measurement period spanning sham noise exposure (2-way repeated measures ANOVA, N = 8, main effect for Day, [F = 1.2, p = 0.33]; main effect for Frequency, [F = 40.1, p = 4 × 10⁻⁴]). (D) Top: Thresholds for sound-evoked facial movements with the 8 kHz OBN are unchanged over time after noise exposure and do not differ between SNHL and sham groups (2-way repeated measures ANOVA, N = 8/10 sham/SNHL, main effect for Day [F = 0.14, p = 0.93]; main effect for Group [F = 0.44, p = 0.52]). Bottom: Thresholds for sound-evoked facial movements with the 32 kHz OBN are elevated after SNHL but not sham noise exposure (2-way repeated measures ANOVA, N = 18, main effect for Day, [F = 32.87, p = 1 × 10⁻¹¹]; main effect for Group, [F = 39, p = 1 × 10⁻⁵]). (E) Schematic illustrating hypothetical changes in sound-evoked facial movement amplitudes over the same 17-day period before and after a SNHL-inducing noise exposure (vertical gray line). The cartoon model assumes central gain is progressively enhanced at successive stages of the central pathway, promoting faster and more complete recovery of the high-frequency response in the damaged region of the cochlea and hyper-responsiveness to the low-frequency noise band. (F) Mean peristimulus FME responses from 8 mice over a 17-day period spanning noise-induced hearing loss for an 8 and 32 kHz OBN (top and bottom, respectively). (G) Actual data after SNHL are compared against the cartoon model shown in E by plotting the fold change in facial movement amplitudes relative to baseline (mean of D-1 and D-2) for sound levels at threshold or 10 dB above threshold (i.e., 0 dB and 10 dB sensation level, SL). Data from individual mice (N = 10) and group mean are shown as thin gray and thick dashed lines, respectively. Asterisks denote that the change in sound-evoked facial movements are either significantly elevated relative to baseline (top) or significantly suppressed relative to baseline (bottom), as assessed with one-sample t-tests relative to a population mean of 1.0 (p < 0.02 for all significant time points). (H) Changes in ABR amplitude and sound-evoked facial movements elicited by the 8 kHz noise band after noise-induced high-frequency SNHL are calculated for a subset of mice with data from both measurement types (N = 7). Dashed line presents the linear fit of the data. Facial movement changes reflect the average of 0 and 10dB SL sound levels. Increased 8 kHz-evoked facial movements are not correlated with changes in ABR wave 1 (Pearson r = -0.05, p = 0.92) or ABR wave 2 (r = 0.02, p = 0.96), but are significantly correlated with changes in ABR wave 4 (r = 0.92, p = 0.003). (I) The pattern of increased responses at 8 kHz and decreased responses at 32 kHz observed two weeks after SNHL is not observed following sham exposure (Two-way mixed model ANOVA, main effect for Group [F = 0.65, p = 0.42]; main effect for Frequency [F = 22.45, p = 3 × 10⁻⁴]; Group x Frequency interaction [F = 15.65, p = 0.001]). Thin and thick lines represent data from individual mice and group means from the SNHL group (N = 10) and sham exposure group (N = 8), respectively.

Figure 6 –. Suppressing auditory cortex activity facilitates sound-evoked facial movements

(A) Extracellular recordings were made from all layers of the primary auditory cortex (A1) with a 64-channel linear probe during contralateral sound presentation and facial videography. Electrophysiological responses are filtered offline to separate spiking activity (white trace) and the current source density (CSD). White arrow in CSD trace identifies the early current sink in layer (L) 4 elicited by a 70 dB SPL 50 ms white noise burst that is used to assign units to layers. (B) FME is increased following sound presentation (orange line) but facial twitches also occur spontaneously (dashed box). Action potentials (purple) from a single regular spiking (RS) unit are evoked by the combination of sound and movement but also during spontaneous facial movements. (C) Neurograms present the z-scored firing rates before and after bouts of spontaneous facial movements from 438 single RS units grouped into superficial (L2/3 and L4) and deep (L5 and L6) layers of the cortical column. Units are sorted by their mean activity. The line plot presents the mean FME over the same period. (D) Pie charts represent whether and how single unit firing rates were modulated by spontaneous facial movements. (E) Mean ± SEM absolute value of firing rate changes during spontaneous facial movements along the cortical column. Spike rate modulation is significantly elevated with increasing depth in the cortical column (one-way ANOVA, n = 438, main effect of depth [F = 2.09, p = 0.006]). (F) Left: Schematic illustrates that optogenetic suppression of auditory cortex (ACtx) spiking could eliminate sound-evoked facial movements if it were an obligatory sensorimotor relay. Right: Alternatively, optogenetic suppression of ACtx spiking could amplify or attenuate sound-evoked facial movements if it modulated a subcortical sensorimotor relay. (G) Mean ± SEM changes in spike rate in response to 20Hz activation of parvalbumin-expressing (PV) GABAergic interneurons in fast-spiking putative PV units (n = 99), superficial RS units (n = 33), and deep layer RS units (n = 373) from N = 4 mice. (H) Experimental paradigm to test the two hypothetical scenarios described in F. Bilateral activation of PV interneurons were interleaved with sound-only trials. (I) Sound-evoked FME on interleaved laser on and laser off trials. Top: Mean ± SEM FME. Thick black and blue lines depict the relative timing of sound and laser timing, respectively. Bottom: Mean FME over a range of broadband noise levels (N = 10). (J) Scatterplot presents mean sound-evoked FME during laser on/photoinhibition versus laser off trials. Each symbol represents the trial-averaged mean from a single animal color coded according to the level of the sound relative to threshold. Data points above the line of unity (dashed diagonal) are enhanced during ACtx suppression. Inset: mean FME within 10 dB SPL above threshold is plotted for each mouse during laser off and on trials. Asterisk denotes significant difference (paired t-test, p = 0.005). See also Figure S3.

Figure 7 –. Hyper-responsive sound-evoked facial movements in mice with an autism risk gene mutation.

(A) ABR thresholds are not significantly different in mice with Ptchd1 deletion (KO) and wildtype littermate controls (two-way repeated measures ANOVA, main effect for frequency [F = 131, p < 0.001], main effect for genotype [F = 0.31, p = 0.58], frequency x genotype interaction [F = 0.74, p = 0.60]). (B) Bi: Placement of pupil markers and ROI for FME calculation. Bii: Representative traces of pupil diameter changes in a wildtype (WT) and KO mouse. Solid and dashed arrows denote the timing of noise bursts in the corresponding recording session. (C) Baseline pupil diameter in KO mice is significantly larger than WT controls (two-sample t-test, [t = −4.64, p < 0.001]). Bars and errors bars represent mean ± SEM. Each data point represents an individual mouse (N = 14/11, WT/KO). (D) Top: Representative FME traces in a wildtype (WT) and KO mouse. Solid and dashed arrows denote the timing of noise bursts in the corresponding recording session. Bottom: Mean ± SEM FME for noise bursts of increasing intensity in WT (left) and KO (right) groups (N = 15/11, WT/KO). (E) Sound-evoked facial movement grow significantly more steeply across sound level in KO mice compared to WT controls (two-way mixed design ANOVA, main effect for sound level [F = 43.46, p = 0.001], main effect for genotype [F = 2.29, p = 0.14], level x genotype interaction [F = 2.67, p = 0.02]).