Pupillometry, a simple and non-invasive technique, is proposed as a method to determine hearing-in-noise thresholds in normal hearing animals and animal models of various auditory pathologies.
Noise exposure is a leading cause of sensorineural hearing loss. Animal models of noise-induced hearing loss have generated mechanistic insight into the underlying anatomical and physiological pathologies of hearing loss. However, relating behavioral deficits observed in humans with hearing loss to behavioral deficits in animal models remains challenging. Here, pupillometry is proposed as a method that will enable the direct comparison of animal and human behavioral data. The method is based on a modified oddball paradigm – habituating the subject to the repeated presentation of a stimulus and intermittently presenting a deviant stimulus that varies in some parametric fashion from the repeated stimulus. The fundamental premise is that if the change between the repeated and deviant stimulus is detected by the subject, it will trigger a pupil dilation response that is larger than that elicited by the repeated stimulus. This approach is demonstrated using a vocalization categorization task in guinea pigs, an animal model widely used in auditory research, including in hearing loss studies. By presenting vocalizations from one vocalization category as standard stimuli and a second category as oddball stimuli embedded in noise at various signal-to-noise ratios, it is demonstrated that the magnitude of pupil dilation in response to the oddball category varies monotonically with the signal-to-noise ratio. Growth curve analyses can then be used to characterize the time course and statistical significance of these pupil dilation responses. In this protocol, detailed procedures for acclimating guinea pigs to the setup, conducting pupillometry, and evaluating/analyzing data are described. Although this technique is demonstrated in normal-hearing guinea pigs in this protocol, the method may be used to assess the sensory effects of various forms of hearing loss within each subject. These effects may then be correlated with concurrent electrophysiological measures and post-hoc anatomical observations.
Pupil diameter (PD) can be affected by a wide number of factors and the measurement of PD that changes over time is known as pupillometry. PD is controlled by the iris sphincter muscle (involved in constriction) and the iris dilator muscle (involved in dilation). The constriction muscle is innervated by the parasympathetic system and involves cholinergic projections, whereas the iris dilator is innervated by the sympathetic system involving noradrenergic and cholinergic projections1,2,3. The best-known stimulus to induce PD changes is luminance-constriction and dilation responses of the pupil can be produced by variations in ambient light intensity2. PD also changes as a function of focal distance2. It has been known for decades, however, that PD also shows non-luminance-related fluctuations4,5,6,7. For example, changes in internal mental states can elicit transient PD changes. The pupil dilates in response to emotionally charged stimuli or increases with arousal4,5,8,9. Pupil dilation could also be related to other cognitive mechanisms, such as increased mental effort or attention10,11,12,13. Because of this relationship between pupil size variations and mental states, PD changes have been explored as a marker of clinical disorders such as schizophrenia14,15, anxiety16,17,18, Parkinson's disease19,20, and Alzheimer's disease21, among others. In animals, PD changes track internal behavioral states and are correlated with neuronal activity levels in cortical areas22,23,24,25. Pupil diameter has also been shown to be a reliable indicator of the sleep state in mice26. These PD changes related to arousal and the internal state typically occur on long time scales of the order of several tens of seconds.
In the domain of hearing research, in normal hearing as well as in hearing impaired subjects, listening effort and auditory perception have been assessed using pupillometry. These studies typically involve trained research subjects27,28,29,30 that perform various kinds of detection or recognition tasks. Because of the aforementioned relationship between arousal and PD, increased task engagement and listening effort have been shown to be correlated with increased pupil dilation responses30,31,32,33,34,35. Thus, pupillometry has been used to demonstrate that increased listening effort is expended to recognize spectrally degraded speech in normal-hearing listeners29,36. In hearing impaired listeners, such as humans with age-related hearing loss27,30,37,38,39,40,41 and cochlear implant users42,43, pupil responses also increased with decreasing speech intelligibility; however, hearing impaired listeners showed greater pupil dilation in easier listening conditions compared to normal hearing subjects27,30,37,38,39,40,41,42,43. But experiments that require the listener to perform a recognition task are not always possible – for example, in infants, or in some animal models. Thus, non-luminance related pupil responses evoked by acoustic stimuli could be a viable alternative method to assess auditory detection in these cases44,45. Earlier studies demonstrated a transient and stimulus-linked pupil dilation as part of the orienting reflex46. Later studies have demonstrated the use of stimulus-linked pupil dilations to derive frequency sensitivity curves in owls47,48. Recently, these methods have been adapted to assess sensitivity of the pupil dilation response in human infants48. Pupillometry has been shown to be a reliable and non-invasive approach to estimate auditory detection and discrimination thresholds in passively listening guinea pigs (GPs) by using a wide range of simple (tones) and complex (GP vocalizations) stimuli49. These stimulus-related PD changes typically occur at faster time scales of the order of several seconds and are linked to stimulus timing. Here, pupillometry of stimulus-related PD changes is proposed as a method to study behavioral impacts of various kinds of hearing impairment in animal models. In particular, pupillometry protocols for use in GPs, a well-established animal model of various types of auditory pathologies50,51,52,53,54,55,56 (also see reference57 for an exhaustive review) is described.
Although this technique is demonstrated in normal-hearing GPs, these methods can be easily adapted to other animal models and animal models of various auditory pathologies. Importantly, pupillometry can be combined with other non-invasive measurements such as EEG, as well as with invasive electrophysiological recordings in order to study the mechanisms underlying possible sound detection and perception deficits. Finally, this approach can also be used to establish broad similarities between human and animal models.
For all experimental procedures, obtain approval from the Institutional Animal Care and Use Committee (IACUC) and adhere to NIH Guidelines for the care and use of laboratory animals. In the United States of America, GPs are additionally subject to United States Department of Agriculture (USDA) regulations. All the procedures in this protocol were approved by the University of Pittsburgh IACUC and adhered to NIH Guidelines for the care and use of laboratory animals. For this experiment, three male wildtype, pigmented GPs between 4 and 10 months of age, with ~600-1,000 g weight were used.
1. Surgical procedure
2. Animal acclimation to the experimental setup
NOTE: Experiments typically take place in a sound-attenuated chamber or booth (see Table of Materials). The time required to familiarize an animal to the setup varies from subject to subject. Typical acclimation times are noted below. A well-acclimated animal will tolerate head-fixation with minimal body motion, and result in better pupil diameter measurements.
3. Calibration of pupil camera
NOTE: The camera used for pupillometry outputs a video via USB to the pupillometry software suite. From this video, the pupil diameter is extracted using an ellipse fit and user-adjustable threshold value by the pupillometry software suite (see Table of Materials). The software then interfaces with a digital-to-analog card. The card outputs an analog voltage value that is proportional to the pupil diameter. Calibration is needed to convert this voltage value back to pupil diameter in units of length.
4. Pupillometry data acquisition
5. Call-in-noise detection and categorical discrimination using a modified oddball paradigm
NOTE: The stimuli for pupillometry experiments consisted of GP vocalizations that were recorded in an animal colony58. The vocalization samples can be found in the following repository: https://github.com/vatsunlab/CaviaVOX. In particular, wheek and whine calls were used to elicit the pupil responses shown in the representative results. From each category, choose vocalizations whose lengths are approximately equal. To account for differences in the recording amplitude and temporal envelopes of the vocalizations, normalize the vocalizations by their root mean square (r.m.s.) amplitudes, if needed.
6. Analysis and statistics
NOTE: All the analyses were performed using custom code written in MATLAB (available at https://github.com/vatsunlab/GP_Pupil). Two main analysis methods are described, which address the reliability and the time course of pupil responses, respectively. The choice of one or both the methods will be dictated by experimental design.
Pupillometry was performed in three male pigmented GPs, weighing ~600-1,000 g over the course of the experiments. As described in this protocol, to estimate call-in-noise categorization thresholds, an oddball paradigm was used for stimulus presentation. In the oddball paradigm, calls belonging to one category (whines) embedded in white noise at a given SNR were employed as standard stimuli (Figure 2A), and calls from another category (wheeks) embedded in white noise at the same SNR (Figure 2A) as deviant stimuli. Standard and deviant stimuli were randomly chosen, with resampling, from eight exemplars of each category. In each experimental session, stimuli were presented with high temporal regularity (Figure 2B), with at least 20 presentations of standard stimuli between deviant stimuli. Data were acquired corresponding to a particular SNR level in each experimental session. Across sessions, a broad range of clean and noisy SNRs were sampled (-24, -18, -12, -6, -3, 0, 3, 6, 12, 40 dB SNR).
The PD changes to the standard stimuli did not differ significantly from the baseline (blue line in Figure 3A). The deviant stimuli evoked robust and significantly larger PD changes than those elicited by the standard stimuli (gray lines in Figure 3A), reflecting call category discrimination. The response magnitude and the percentage of trials with statistically significant pupil responses were highest at the cleanest SNR and decreased gradually with decreasing SNR (Figure 3A,B). Using GCA, pupil responses to deviant stimuli were found to be statistically significant at SNRs above -18 dB (Figure 3C), which was taken to be the call-in-noise categorization threshold (green line in Figure 3A). The percentage of significant trials at each tested SNR level was well-fit by a psychometric function (Figure 3D). The SNR level necessary to reach the half-maximum of the psychometric curve was about -20 dB SNR (Figure 3D). Anecdotally, for this case, the reliability-based and time course-based metrics yielded similar values of call-in-noise categorization thresholds.
Figure 1: Pupillometry setup, and stimulus-evoked and motion-related PD changes. (A) The pupillometry setup with video frame images of sound evoked pupil dilation (top). The baseline PD is shown by dashed green circles. (B) An exemplar PD trace (top) and exemplar motion trace (bottom) from a single experimental session. Vertical black lines correspond to onset time deviant stimulus presentations. Red ticks correspond to automatically detected motion events. Gray horizontal dashed line corresponds to 5 SD threshold. (C) The PD changes (ΔPD) evoked by deviant stimulus (top) and related to motion events (bottom) from one experimental session. Stimulus onset is shown by vertical black line; the detection of motion event is shown by vertical red line. Note that pupil dilation onset precedes the onset of motion. Please click here to view a larger version of this figure.
Figure 2: Call spectrograms and call-in-noise categorization paradigm structure. (A) Representative spectrograms of a guinea pig whine and wheek, in clean conditions and at 0- and -18-dB SNR, respectively. Noisy calls were obtained by adding white noise. (B) Structure of the oddball paradigm used to estimate call-in-noise categorization thresholds. Whine calls were randomly chosen from eight exemplars and used as standard stimuli. Wheek calls were randomly chosen from eight exemplars and used as deviants. In each experimental session, the noise was added at a different SNR level (-24, -18, -12, -6, -3, 0, 3, 6, 12 dB SNR). The calls are 1 s long and the time between stimuli is 3 s. Please click here to view a larger version of this figure.
Figure 3: Pupillometry estimates of call-in-noise detections and categorization thresholds. (A) Average pupil responses from three animals. Mean pupil responses to standard whine stimuli are represented by blue line, and shading corresponds to ±1 standard error of mean (s.e.m.). Gray lines and shading correspond to mean and ±1 s.e.m. of pupil responses evoked by deviant wheek stimuli. Gray shading intensity corresponds to SNR. Green line and shading correspond to average pupil trace at threshold SNR (about -18 dB SNR). Red vertical line corresponds to stimulus onset; orange vertical line corresponds to air puff onset; teal dashed lines correspond to GCA window (PD changes rising phase). (B) GCA fit to the rising phase of PD changes. Dots are mean pupil diameter in 100 ms time bins, whiskers correspond to ±1 s.e.m. Solid lines correspond to mixed-effects model fits. Line colors as in A. (C) GCA weight estimates. Weights of the intercept is in blue, slope is in red, and acceleration is in purple. Whiskers correspond to ±1 s.e.m. Asterisks show statistically significant regression weights (linear hypothesis test on linear regression model coefficients). (D) Psychometric function fit to the percent of trials with significant PD changes elicited by the deviant stimulus as a function of SNR. Whiskers correspond to ±1 s.e.m. Note that 50% of the maximum is reached at about -20 dB SNR (green dashed line). Please click here to view a larger version of this figure.
This protocol demonstrates the use of pupillometry as a non-invasive and reliable method to estimate auditory thresholds in passively listening animals. Following the protocol described here, call-in-noise categorization thresholds in normal hearing GPs were estimated. Thresholds estimated using pupillometry were found to be consistent with those obtained using operant training62. Compared to operant training, however, the pupillometry protocol was relatively straightforward and quick to set up and acquire data. Each data acquisition session (per SNR level) lasted about 12 min, which resulted in 1-2 h of experimental sessions (across SNR levels) per animal per day49. Data acquisition could be completed in about 7-10 days (depending on the number of SNR levels used). Although the oddball paradigm was used for call-in-noise categorization threshold estimation in this manuscript, this pupillometry protocol can be adapted to easier versions of oddball paradigms, where just one call exemplar is used, or to other stimulus paradigms using a wide range of complex or simple stimuli49.
The method is not without disadvantages. First, the current protocol requires the implant of a head post to fix the head during these experiments. Head post implant surgery and recovery would add a minimum of 2 weeks to the timeline of the experimental protocol. It is possible that this step can be avoided by using other methods of non-invasively immobilizing awake animals during experiments-for example, by using custom 3D-printed helmets63 or deformable thermoplastics64. Further experiments are necessary to explore these solutions. Second, animals could rapidly habituate to deviant stimuli as well, resulting in decreasing pupil dilation responses over the course of an experimental session. This effect could be minimized by restricting experimental sessions to short durations (~12 min) and presenting only a limited number (8) of deviant stimuli. Furthermore, an air puff delivered after the deviant stimuli can ensure that animals remain engaged with the auditory stimuli. Third, because of this rapid habituation, several days may be required to complete data acquisition. By only testing SNR values that densely sample the steepest parts of the psychometric curve, the total number of experimental days can be minimized. Fourth, animals may not keep still during experiments, or blink excessively or close their eyes during experiments. These factors are a function of species and acclimation and show a high degree of individual variability. GPs are naturally docile, and by acclimating them well to the experimental setup, motion and blink artifacts can be minimized. Spontaneous blinks and saccades are typically quite rare in guinea pigs49, but this could be a function of the species as well. Finally, as mentioned earlier, pupil dynamics in humans have been associated with a number of neuropsychiatric disorders. While the experimental animals used here are assumed to be neurotypical, this caveat must be kept in mind while interpreting results.
While one hardware implementation of pupillometry is described here (using a commercially available eye tracker and neural data acquisition system), the equipment required is expensive and not economical to scale-up. However, other custom solutions based on the same underlying principle of infrared-based eye tracking that are more cost-effective, are available. For example, one study used custom components and custom video processing algorithms to extract pupil diameter from the recorded video22,25. Recently developed deep-learning algorithms are also able to extract pupil diameter from videographic data65,66. These solutions could more than halve the cost of pupillometry rigs. The trade-off here is between expense and time-while commercial solutions are more expensive, they are turn-key solutions that can be used out of the box. On the other hand, custom solutions are cost-effective and scalable, but require expertise to set-up, and the time needed to develop custom analysis pipelines.
Although the protocol detailed here was performed in normal hearing GPs, pupillometry could be relatively easy to use in other animal models of hearing impairment with appropriate changes to stimulus type and parameters. This would allow for characterizing the effects of hearing loss across a range of stimulus types and species, which could potentially yield novel observations. Since pupillometry is a non-invasive technique that has also been used extensively in humans, by using the same stimuli used for animal subjects, pupillometry can be used to compare the effects of various auditory pathologies across species. For example, a recent meta-analysis in humans showed that speech-in-noise perception deficits arising from moderate noise exposure were best observed when complex and temporally varying stimuli were used67. The estimation of call-in-noise categorization thresholds by pupillometry demonstrated here could be used as one such task using complex stimuli to evaluate the effects of noise exposure in GPs. The assessment of hearing at a behavioral level using these methods would complement electrophysiological and anatomical methods and could be part of the standard toolkit for evaluating various known hearing disorders.
In conclusion, the following points are critical for successful acquisition of pupillometric data. First, to ensure high data yield, it is critical to familiarize the animals well to the experimental setup. A lack of patience in this step could degrade the quality of data that is eventually obtained or necessitate the repetition of several sessions to make up for the lost sessions. Second, to avoid luminance-related PD changes, it is important to perform experiments in constant illumination conditions, maintaining these conditions between sessions and subjects as much as possible. Third, to minimize the number of experimental sessions needed, it is important to perform pilot experiments to identify critical parameter ranges for dense sampling. Fourth, to minimize habituation of the animals to the stimuli, it is important to perform experiments in short sessions containing only a few presentations of deviant stimuli. An air puff may be additionally used to maintain high engagement with the auditory stimuli.
The authors have nothing to disclose.
This work was supported by the NIH (R01DC017141), the Pennsylvania Lions Hearing Research Foundation, and funds from the Departments of Otolaryngology and Neurobiology, University of Pittsburgh.
Analog output board | Measurement Computing Corporation, Norton, MA | PCI-DDA02/12 | |
Anechoic foam | Sonex One, Pinta Acoustic, Minneapolis, MN | ||
Condenser microphone | Behringer, Willich, Germany | C-2 | |
Free-field microphone | Bruel & Kjaer, Denmark) | Type 4940 | |
Matlab | Mathworks, Inc., Natick, MA | 2018a version | |
Monocular remote camera and illuminator system | Arrington Research, Scottsdale, AZ | MCU902 | Infrared LED array + camera with infrared filter |
Multifunction I/O Device | National Instruments, Austin, TX | PCI-6229 | |
Neural interface processor | Ripple Neuro, Salt Lake City, UT | SCOUT | |
Piezoelectric motion sensor | SparkFun Electronics, Niwot, CO | SEN-10293 | |
Pinch valve | Cole-Palmer Instrument Co., Vernon Hills, IL | EW98302-02 | |
Programmable attenuator | Tucker-Davis Technologies, Alachua, FL | PA5 | |
Silicon Tubing | Cole-Parmer | ~3 mm | |
Sound attenuating chamber | IAC Acoustics | ||
Speaker full-range driver | Tang Band Speaker, Taipei, Taiwan | W4-1879 | |
Stereo Amplifier | Tucker-Davis Technologies, Alachua, FL | SA1 | |
Tabletop – CleanTop Optical | TMC vibration control / Ametek, Peabody, MA | ||
Viewpoint software | ViewPoint, Arrington Research, Scottsdale, AZ |