Abstract
Deviance detection describes an increase of neural response strength caused by a stimulus with a low probability of occurrence. This ubiquitous phenomenon has been reported for humans and multiple other species, from subthalamic areas to the auditory cortex. Cortical deviance detection has been well characterized by a range of studies using a variety of different stimuli, from artificial to natural, with and without a behavioral relevance. This allowed the identification of a broad variety of regularity deviations that are detected by the cortex. Moreover, subcortical deviance detection has been studied with simple stimuli that are not meaningful to the subject. Here, we aim to bridge this gap by using noninvasively recorded auditory brainstem responses (ABRs) to investigate deviance detection at population level in the lower stations of the auditory system of a highly vocal species: the bat Carollia perspicillata (of either sex). Our present approach uses behaviorally relevant vocalization stimuli that are similar to the animals’ natural soundscape. We show that deviance detection in ABRs is significantly stronger for echolocation pulses than for social communication calls or artificial sounds, indicating that subthalamic deviance detection depends on the behavioral meaning of a stimulus. Additionally, complex physical sound features like frequency- and amplitude modulation affected the strength of deviance detection in the ABR. In summary, our results suggest that the brain can detect different types of deviants already in the brainstem, showing that subthalamic brain structures exhibit more advanced forms of deviance detection than previously known.
Significance Statement
Bats, like all mammals, rely on the identification of regulations and deviations in their acoustic environment. This phenomenon, called deviance detection, has been studied intensively in the past and keeps gaining attention in the field of electrophysiology. Over time, an impressive complexity of deviance detection could be shown, in both animal and human studies. However, complex forms of auditory deviance detection were so far only demonstrated for high-level brain structures. In this study, we show that complex deviance detection beyond simple frequency changes of auditory stimuli is already present in the lowest stations of the auditory pathway, the brainstem. These potentially feedback mediated effects could contribute significantly to the saving of resources very early in the processing of acoustic sounds.
Introduction
Like all echolocating bats, Carollia perspicillata navigates in the dark by emitting stereotypical acoustic pulses and listening to the echoes reflected off objects in its environment. In addition, this bat species has a large repertoire of social communication calls (Knörnschild et al., 2014; González-Palomares et al., 2021), which is a consequence of its social lifestyle, with groups of >100 individuals sharing the same roost (Cloutier and Thomas, 1992). This broad variety of social communication calls makes it an excellent animal model to study vocal communication. Echolocation pulses and social communication calls differ from each other in their carrier frequencies and durations, with echolocation pulses being higher in frequency and shorter in time (see Fig. 1a for an example echolocation pulse and social communication call). Both vocalization types represent fundamentally different behaviors (navigation and social communication) and can alternate in rapid succession for freely behaving bats. This raises a question that has puzzled neuroethologists for years: How does the bat brain process echolocation and social sounds in a fast and energy-efficient way, when they occur in the same acoustic stream? A theoretical model that explains how the brain efficiently deals with the tremendous amount of input it receives is the predictive coding framework and, in relation to this, the ability of deviance detection (Parras et al., 2017; Carbajal and Malmierca, 2018). According to the predictive coding theory, the brain is constantly creating predictions about the incoming stimuli (Friston, 2005). When the system encounters an unexpected signal, expectations are updated which is represented by a prediction error component in the neural response. This makes the identification of regularities and deviants in the incoming stream of signals (i.e., deviance detection) crucial for the predictive coding framework. The present study investigates deviance detection to naturally occurring sounds—echolocation pulses and social communication calls—in the bat species C. perspicillata. We focused on studying deviance detection in subthalamic neural populations of the auditory pathway by combining a naturalistic oddball stimulation paradigm (Fig. 1b) with noninvasively recorded auditory brainstem responses (ABRs). Two experiments were performed: In Experiment 1, an echolocation pulse and a social communication call (Fig. 1a) were presented in an oddball paradigm (Fig. 1b). Additionally, by using two control paradigms [the “Many-Standards” (MS) and the 50% control; Fig. 1c], we aimed to shed light on the underlying neural mechanisms responsible for deviance detection, namely, deviant enhancement and repetition suppression of the standard response. In Experiment 2, the effect of different acoustic parameters (e.g., carrier frequency and temporal structure) and the behavioral meaning of the auditory input on subthalamic deviance detection was evaluated by performing a cross-comparison of the responses to different stimuli. The stimuli considered ranged from natural vocalizations on one end, to artificially generated vocalization-mimics, noise bursts that resemble the vocalizations in their frequency range and duration but not in their temporal structure, on the other end (Fig. 1d).
Used stimuli and stimulation protocols. a, Oscillograms (top) and spectrograms (bottom) of an echolocation pulse (green frame) and a social communication call (orange frame) of C. perspicillata that were used as stimuli in this study. The communication signal is a syllable of a so-called distress call. b, Schematic representation of the oddball paradigm; blue, standard; red, deviant. c, Schematic representation of the two control sequences used; Many-Standards control (top) and 50% control (bottom). Additional vocalizations used in the Many-Standards control and their respective ABRs are shown in Extended Data Figure 1-1. d, Additional stimuli used for a cross-comparison with the vocalizations in a to evaluate the importance of the frequency-versus-time structure of the stimuli for deviance detection. The amplitude modulation of the communication AM call was produced by the animal itself and the call represents another example of a natural distress call of C. perspicillata. Both vocalization mimics are artificially generated. They resemble the natural vocalizations in their frequency range and duration but not in their temporal structure.
Figure 1-1
Additional stimuli used in the Many-Standards control and their respective ABRs. (a) Five additional social communication calls of C. perspicillata that were presented together with the stimuli in Fig. 1 in the Many-Standards control. (b) Grand averages of ABRs to the stimuli in (a), obtained during the Many-Standards control (n = 13 animals). Download Figure 1-1, TIF file.
Material and Methods
Animals
For the experiments of this study, 13 adult bats (7 males, 6 females) of the species C. perspicillata from the breeding colony of Goethe University Frankfurt were used. After being caught for the first time, all animals were held separately from the colony until the end of the study. Before every recording session, the animal was anesthetized by a mixture of ketamine (Ketavet 10%, Medistar; 7.5 mg per kg bodyweight) and xylazine (Rompun 2%, Bayer HealthCare; 16.5 mg per kg bodyweight) and the anesthesia was maintained by follow-up injections of the same mixture with reduced volume every 1–1.5 h, for up to 4 h total. A DC-powered heating pad that was attached to the animal holder was used to maintain the animal’s body temperature of 37°C. Two consecutive recording sessions in the same animal were at least 5 d apart. This study was approved by the Regierungspräsidium Darmstadt (permits: FR/1010 and FR/2007) and was performed in full compliance with current German laws.
Stimulation and recording procedure
Custom-written MATLAB (MathWorks) scripts were used for stimulation and data acquisition. The digital stimulus signal was D/A-converted by a 384 kHz Adi-2 Pro soundcard (RME Audio) before it was fed into a HiFi amplifier (Power Amplifier RB-1050, Rotel) and presented to the animal by a Fountek NeoPro 5i Ribbon Tweeter (Fountek Electronics). The speaker was positioned 15 cm away from the animal and pointed directly toward the left ear in a 45° azimuth angle relative to the head. To ensure a constant distance and angle between ear and speaker, the animal was head-fixed by a mouth-holder. All stimuli were natural vocalizations of C. perspicillata or vocalization mimics with durations between ∼2 and 10 ms (Fig 1, Extended Data Fig. 1-1). The social communication call that was used as target tone in Experiment 1 (Fig. 1a) is a distress call, a social vocalization that is emitted by the animal when under physical duress. Like all calls used in this study, it was recorded from a freely behaving bat. The echolocation and communication mimics are noise bursts covering very similar frequency ranges as their natural counterparts. They also resemble the vocalizations in their durations and rise/fall times, with only the temporal structure of the natural and artificial stimuli being fundamentally different from each other. All stimuli had an intensity of 60 dB SPL and were presented at a rate of 20 Hz, equivalent to a stimulus-onset asynchrony of 50 ms. In all stimulation sequences, only one exemplar of a respective vocalization was presented to the animal, serving as a representative example (shown in Fig. 1a,b; Extended Data Fig. 1-1).
The oddball paradigm that was used to study effects of deviance detection in the ABR consisted of two sequences of stimuli. In the first sequence, stimulus 1 was presented as standard (high probability, 90%) and stimulus 2 as deviant (low probability, 10%). The second sequence was presented consecutively and resembled the first one but with opposite roles of the stimuli, where now stimulus 1 was the deviant and stimulus 2 the standard (Fig. 1b). In total, a sequence contained 1,000 stimuli (900 standards, 100 deviants). To characterize the measured deviance detection effects in more detail, two control sequences were used. The first was the MS control (Schröger and Wolff, 1996), presenting the target stimuli (echolocation pulse and social communication call) in a pseudorandomly arranged sequence together with eight additional stimuli, all having a probability of occurrence of 10%. The other eight stimuli were the two vocalization mimics, an amplitude modulated (AM) communication (distress) call (Fig. 1d), and five other social communications of C. perspicillata that are related to different behaviors (Extended Data Fig. 1-1). The MS control is expected to generate responses that are unaffected by any modulatory effects of probability encoding (repetition suppression or deviant enhancement) since the stimuli are perceived neither as deviant nor standard. As a second control, the echolocation pulse and social communication call were presented in another sequence where their probability of occurrence was 50%, respectively. Like in the oddball paradigm, the sequences of both controls consisted of 1,000 stimuli each.
ABRs were differentially recorded by two electrodes—chlorinated silver wires (AG-10 T, diameter, 0.25 mm; uninsulated and chlorinated tip of 3 mm)—that were placed subcutaneously at the vertex of the animal’s skull and close to the bulla of the left ear. A ground electrode was clipped to the animal’s right thumb. This electrode arrangement has been used in former ABR recordings with bats (Burkard and Moss, 1994; Linnenschmidt and Wiegrebe, 2018; Hörpel and Firzlaff, 2020; Wetekam et al., 2020, 2022) and has been proven to result in large and reliable ABRs. The recorded responses were hardware filtered (0.1–3,000 Hz, 20 dB/decade roll-offs) and amplified by a factor of 20 k by a Dagan EX1 differential amplifier (Science Products) before they were A/D-converted by the soundcard and sent to the computer. Since the sampling rate of the soundcard (384 kHz) is much higher than what is needed to capture the relevant spectral content of the ABR (usually <2 kHz)—the raw ABR signal was down-sampled by a factor of 20 (resulting in a new sampling rate of 19.2 kHz). Instead of using only every 20th point of the input signal and rejecting the rest of the data, 20 consecutive datapoints were averaged to a single value. This way of down-sampling maintains more information of the original recording and usually results in a smoother down-sampled signal.
Data processing and statistical evaluation
All processing and statistical evaluation of the data was conducted in MATLAB. The recorded ABRs were bandpass filtered by a Butterworth filter (4th order) in two different ways, dependent on the analysis. For the broadband filtered responses, low- and high-cut frequencies of 0.1 and 2,500 Hz, respectively, were used, which did remove high-frequency noise from the signal but left the ABRs otherwise almost unchanged. On the other hand, the narrowband filter removed frequencies below 300 and above 2,500 Hz, abolishing all slow components of the response and allowing a more detailed inspection of the fast ABR waves i–iv. Before averaging, each trial was baseline corrected by calculating the mean voltage in a time window 1 ms prestimulus onset. This value was subtracted from the whole trial resulting in a prestimulus activity of 0 µV. Subsequently, the averaging procedure was restricted to those deviant responses that followed a standard response and, vice versa, those standard responses that preceded a deviant response. This method allows to use the same number of trials to calculate the deviant and standard average of each animal (here between 89 and 92 trials) and, at the same time, maximizes the effects of deviance detection in the responses (Wetekam et al., 2022). In the case of the 50% control, the same number of trials that was used to calculate the deviant and standard average for a given animal was used to randomly choose trials out of the 500 available responses to each stimulus. The MS average was calculated based on all responses to a given stimulus in the MS sequence (between 85 and 110 trials; mean difference to oddball responses, +2.4 trials). All responses were corrected for the sound-travelling delay caused by the distance between speaker and ear. In each graph, the time point of 0 ms represents the moment when the sound reached the bat’s ear.
To evaluate the response strength of each ABR, time windows were defined within which the response’s root mean square (RMS) value was calculated. This has been done successfully in the same species before (Wetekam et al., 2020, 2022). For all broadband filtered responses, this time window spanned from 0 to 10 ms, covering the whole ABR with all its fast and slow components. The detailed wave-by-wave analysis of the narrowband filtered responses was done using three consecutive time windows, each containing a different component of the ABR. Those windows had borders of 0–1 ms (wave i), 1 ms–2.2 ms (wave ii/iii) and 2.2–3.4 ms (wave iv) which are similar to previously reported ABR wave latencies of other bat species (Burkard and Moss, 1994; Linnenschmidt and Wiegrebe, 2018; Hörpel and Firzlaff, 2020).
To compare response strengths between conditions with each other, paired one-tailed t tests (deviant vs standard responses) and repeated-measure analyses of variance (ANOVAs; deviant vs standard vs control responses) with subsequent Bonferroni-corrected post hoc tests were used to evaluate differences between the calculated RMS values. Additionally, the effect size measure Cohen’s d was calculated for all significant comparisons which allow an estimation of strength of the measured deviance detection effects (Cohen, 1988). The data and code used for this study are freely available on G-Node (DOI: 10.12751/g-node.5eqq3r).
Results
Deviance detection in broadband filtered ABRs differs between echolocation and social communication sounds
In Experiment 1, an echolocation pulse and a social communication call were presented in the oddball paradigm to investigate differences of deviance detection between both vocalization types with the following results: ABRs to the echolocation pulse were significantly larger when the stimulus was perceived as a low-probability deviant (Fig. 2a, red ABR) than when it was a high-probability standard (blue ABR; ANOVA: F = 18.0, p = 1.7 * 10−5; standard-deviant: p = 1.2 * 10−4). This difference is present across the whole response; however, most prominently it appears in the last peak of the ABR, a slow wave that only becomes visible when the responses are broadband filtered between 0.1 and 2,500 Hz. This filtering method is different from the usual narrowband filters between ∼300–2,500 Hz that are used in many ABR studies and that will be discussed later. It has been proposed that deviance detection is driven by two mechanisms at the neural level: repetition suppression and deviant enhancement. To disentangle which mechanism underlies the neural responses, the MS control has been suggested (Schröger and Wolff, 1996; Fig. 1c, top). In this control, the target stimuli (Fig. 1a) are pseudorandomly presented together with multiple other stimuli (here, eight stimuli; Fig. 1d; Extended Data Fig. 1-1), which makes it impossible for the brain to detect regularities or deviations in the acoustic input. This results in responses that are unaffected by repetition suppression and deviant enhancement and thus can be used as baseline responses in comparison with the standard and deviant response. In this comparison, a reduction of response strength to the standard relative to the MS stimulus indicates repetition suppression of the standard response. On the other hand, a stronger deviant than MS response is evidence for deviant enhancement. That is, an increment in deviant response strength solely caused by the fact that the stimulus was unexpectedly perceived in a repetitive and not a variable acoustic context. The echolocation response that was recorded in the MS control was significantly smaller than the deviant response (ANOVA: F = 18.0, p = 1.7 * 10−5; MS control-deviant: p = 4.6 * 10−3) and not significantly different from the standard response (Fig. 2a, top). This observation shows that the neural mechanism driving deviance detection for echolocation is a deviant-related enhancement of the response (i.e., a prediction error response in the predictive coding framework) and not a repetition suppression effect on the standard response. Interestingly, the slow wave of the MS response has an earlier peak and offset latency compared with the deviant and standard ABR. This could indicate that additional neural mechanisms become active and modify the ABRs when the natural acoustic input becomes more complex, as occurs in the MS control compared with the oddball sequence. To further investigate deviance detection in the ABR, we used another common control paradigm, the so-called 50% control. Here, both target stimuli are presented in a sequence with equal probability of 50%. The analysis yielded a similar result to the MS control, that is the deviant response being significantly enlarged (ANOVA: F = 35.4, p = 6.9 * 10−8; 50% control-deviant: p = 5.2 * 10−5) and no difference between control and standard response (Fig. 2a, bottom). This suggests, in line with the MS control, that a deviant-related enhancement of the responses contributes more strongly to the overall effect than a repetition suppression of the standard response.
Deviance detection in broadband filtered ABRs differs between echolocation and social communication sounds (n = 13 animals). a, Grand averages of ABRs to an echolocation pulse presented as deviant (red) and standard (blue) as well as in the MS control (black, top) and the 50% control (magenta, bottom). The boxes framing the responses represent the time window taken for RMS calculation, covering the whole ABR response (0–10 ms poststimulus onset). The gray color of the boxes indicates a significant difference between deviant and standard response. Shaded areas around the ABRs depict the standard error of the mean. The inset on the right shows the RMS values calculated for each animal and condition as an estimation of response strength. b, As in a but the stimulus was a social communication call. The white color of the boxes framing the responses indicates that there was no significant difference between deviant and standard response.
As opposed to echolocation, the social communication sounds did not elicit deviance detection in this experiment (Fig. 2b). While deviant and standard responses were not significantly different from each other, the MS response was attenuated in comparison with the deviant response (Fig. 2b, top; ANOVA: F = 3.4, p = 4.9 * 10−2; MR control-deviant: p = 1.7 * 10−2). This is surprising since, as explained above, the MS control is expected to generate a baseline response that is affected by neither deviant enhancement nor repetition suppression and hence should be positioned between deviant and standard response. Likely, the attenuation of the MS communication response is the result of the same mechanisms that modified the timing of the MS echolocation response. At the same time, the 50% control was very similar to the deviant and standard response, confirming that probability encoding did not affect the ABR size to the social communication call (Fig. 2b, bottom).
For echolocation, deviance detection is measurable very early in narrowband filtered ABRs
To further characterize the effects of deviance detection on the echolocation response, we narrowband filtered the data (bandpass Butterworth, 300–2,500 Hz, 4th order) to analyze the fast ABR components in more detail (Fig. 3). ABR wave ii/iii as well as wave iv of the deviant response were significantly larger than the respective components of the standard response (wave ii/iii: t = 3.2, p = 8.1 * 10−3; wave iv: t = 2.9, p = 1.3 * 10−2). Given that wave ii and iii represent neural activity in the cochlear nucleus (CN) and superior olivary complex (SOC), respectively (Henry, 1979), this finding strongly supports the hypothesis that auditory probability encoding is happening already below the inferior colliculus (IC), as it has been suggested in former studies (Slabu et al., 2012; Wetekam et al., 2022).
For echolocation, deviance detection is measurable very early in narrowband filtered ABRs (n = 13 animals). Grand averages of ABRs to an echolocation pulse presented as deviant (red) and standard (blue), with a social communication call as context. The boxes framing the responses represent the time window taken for RMS calculation, covering the typical ABR peaks i, ii/iii, and iv. The color of the boxes indicates whether a significant difference between deviant and standard response could be measured (gray, yes; white, no). Shaded areas around the graphs depict the standard error of the mean. The inset on the right shows the RMS values calculated for each animal, condition, and time window as an estimation of response strength.
Behavioral meaning and complex sound features of a stimulus affect deviance detection in broadband filtered ABRs
The second experiment of this paper tackles the question of how low-level deviance detection is affected by individual stimulus parameters and possible behavioral meaning of the stimuli. Therefore, in addition to the previously used echolocation pulse and social communication call (Fig. 1a), an amplitude-modulated communication call (another distress vocalization of C. perspicillata) and two artificial vocalization-mimics that resembled the natural vocalizations in their frequency range and duration but not in their temporal structure (Fig. 1d) served as stimuli. The aim was to assess the relevance of the frequency-versus-time structure of a signal for producing deviance detection in broadband filtered ABRs. In addition, it was tested whether the AM property of a communication call influenced deviance detection, as AM appears in natural communication calls (Hechavarría et al., 2020; González-Palomares et al., 2021) and could bear additional meaning for the animal. To answer these questions, the five different stimuli were presented to the animals in all possible parings of the oddball paradigm.
When the echolocation pulse or the echolocation mimic served as target stimulus, significantly larger deviant than standard responses could be measured when any communication stimulus was the context (Fig. 4, Extended Data Fig. 4-1 for statistics; echoloc. in comm. context: t = 6.3, p = 3.9 * 10−5; echoloc. in comm. AM context: t = 4.7, p = 4.9 * 10−4; echoloc. in comm. mimic context: t = 6.1, p = 5.6 * 10−5; echoloc. mimic in comm. context: t = 7.6, p = 6.2 * 10−6; echoloc. mimic in comm. AM context: t = 3.4, p = 5.5 * 10−3; echoloc. mimic in comm. mimic context: t = 4.7, p = 5.5 * 10−4). Interestingly, deviance detection could also be recorded for the responses to the echolocation pulse when the echolocation mimic was the context (t = 2.3; p = 4.0 * 10−2), but not vice versa. This indicates that differences in auditory input beyond simple frequency deviations—for example, the frequency modulation (FM) of the echolocation pulse that is absent in the mimic—have a direct influence on subthalamic deviance detection in the bat brain. The fact that this effect is not present in the echolocation mimic responses when the echolocation pulse was the context supports the claim that the behavioral meaning of a stimulus plays a key role in low-level population-based deviance detection. Both natural communication calls—whether amplitude modulated or not—did not reveal significant deviance detection in any oddball combination except when presented with the echolocation mimic (comm. in echoloc. mimic context: t = 2.3, p = 4.0 * 10−2; comm. AM in echoloc. mimic context: t = 2.2, p = 4.8 * 10−2). This exception could be due to the very different physical properties of both call types where the artificial nature of the mimic increases the contrast even further. On the other hand, as in Experiment 1, the natural echolocation pulse as context did not cause deviance detection in the responses to either of the natural communication calls. In contrast, ABRs to the communication mimic did reveal strong deviance detection with significantly enlarged deviant responses when the AM communication call (t = 2.2; p = 4.8 * 10−2), the echolocation pulse (t = 3.2; p = 7.4 * 10−3) or the echolocation mimic (t = 4.8; p = 4.6 * 10−4) was the context. Only when the unmodulated communication call was the context, no significant difference between deviant and standard response could be measured. Together, these results indicate that the AM of the communication stimulus contributed to the differentiation between the true call and an artificial sound while it did not have a significant impact on the distinction between two different natural communication calls at subthalamic level.
Behavioral meaning and complex sound features of a stimulus affect deviance detection in broadband filtered ABRs (n = 13 animals). All possible parings of the oddball paradigm. Each column contains the recorded responses to the target stimulus of the oddball sequence while each row represents one context stimulus (the second stimulus of the oddball paradigm that served as context for the target stimulus). The stimuli tested were as follows: communication call (Comm.), AM communication call (Comm. AM), communication-mimic (Comm. Mimic), echolocation pulse (Echoloc.), and echolocation mimic (Echoloc. Mimic). Response plots like in Figure 2. The color of the boxes indicates whether a significant difference between deviant and standard response could be measured (gray, yes; white, no). If deviant and standard response differed significantly, Cohen’s d is provided as a measure of effect size (number in the gray boxes). For boxplots of the RMS values of each ABR, see Extended Data Figure 4-1.
Figure 4-1
RMS values for each stimulus combination and condition in the oddball paradigms of experiment 2. Download Figure 4-1, TIF file.
Discussion
In this study, different behaviorally relevant acoustic stimuli were used to characterize the extent and complexity of early deviance detection in the brainstem of bats. The most consistent deviance detection was obtained for echolocation stimuli, with effect sizes dependent on the acoustic context the echolocation pulse was perceived in. Clear deviance detection was also present for artificial stimuli that resembled the natural vocalizations in most of their physical properties except for the temporal structure. In contrast to this, natural social communication calls elicited only minor effects of deviance detection and only when presented together with an artificial stimulus.
The first experiment of this study demonstrates that deviance detection in the brainstem is strongly dependent on the stimulus, with the echolocation pulse eliciting significant effects of deviance detection while the social communication call did not. For the echolocation response, the strongest deviance detection was present in a late and slow ABR wave, only present in the broadband filtered recordings. This is in line with previous studies that investigated deviance detection in broadband filtered ABRs with pure tones (Duque et al., 2018; Wetekam et al., 2022) and confirms that this slow, most likely IC-generated (Land et al., 2016) wave plays a key role in ABR-based deviance detection. The MS control revealed that a deviant-related enhancement and not a repetition suppression of the standard response is the main effect causing deviance detection for the echolocation stimulus. Possibly, deviant stimuli cause the brainstem neurons to respond more synchronously than standard or MS stimuli, resulting in larger deviant ABR amplitudes. In line with this hypothesis, former studies have demonstrated the importance of synchronization and phase locking of brainstem neurons for speech (Abrams et al., 2006) and music (Musacchia et al., 2007; White-Schwoch et al., 2021) perception in humans. The fact that deviant enhancement and not repetition suppression is driving low-level deviance detection during the presentation of echolocation signals is interesting since previous studies have suggested repetition suppression to be the dominant mechanism causing deviance detection in subcortical nuclei (Parras et al., 2017; Carbajal and Malmierca, 2018). However, those studies used pure tones and recorded individual neurons instead of measuring vocalization-related summed potentials like we did here. It is possible that echolocation pulses evoke stronger deviant responses and less repetition suppression due to their high behavioral relevance compared with simple tone pips.
By using a narrowband filter, effects of deviance detection to echolocation in the CN and SOC could be detected. This is in line with a growing body of evidence that supports the idea that deviance detection is not firstly generated in the IC but is present along the whole auditory pathway (Slabu et al., 2012; Wetekam et al., 2022). In fact, effects of novelty detection have recently been described even for the cochlea, the first structure involved in auditory processing (Riecke et al., 2020; Otsuka et al., 2022). In these reports, the authors propose that the medial olivocochlear reflex is responsible for those effects by suppressing outer hair cell activity, mediated by feedback from the cortex. Since the ABRs presented here are averaged over many trials, it is possible that similar cortical feedback mechanisms are responsible for the very early effects seen in our ABR data. Previous studies could demonstrate a variety of efferent projections form the AC to subcortical structures, including the IC, SOC, CN, auditory nerve, and the cochlea (for reviews see Suga and Ma, 2003; Suga, 2008; Terreros and Delano, 2015; Elgueda and Delano, 2020). These corticofugal projections can have differential modulatory influences on neurons in subcortical structures, including suppression effects after repetitive stimulation (Perrot et al., 2006). In bats, corticofugal modulations also sharpen the responses of subcortical neurons to stimuli in the echolocation frequency band (Zhang and Suga, 2000; Xiao and Suga, 2002; Suga, 2008). This could be one of the reasons for the stronger effects of deviance detection when echolocation pulses were used as stimuli, compared with social communication calls. It can be hypothesized that regularity encoding in the lowest stations of the auditory pathway generally contributes to an efficient auditory processing and the saving of resources. Additionally, in the case of bats, it could be a prerequisite for the fast reaction times needed for echolocation-based navigation during flight.
In Experiment 2, different natural and artificial stimuli were presented in the oddball paradigm. The results demonstrate that deviance detection is sensitive not only to physical properties of the stimulus like carrier frequency, AM and FM, but also the behavioral meaning of the stimulus and the context it is perceived in. So far, modulatory effects of the behavioral meaning of a stimulus on the strength of deviance detection have been known for cortical areas (Yaron et al., 2020), but not for the brainstem. However, it is known that neurons in the bat IC are able to modify their response properties dependent on the spectral features of the acoustic input (Mittmann and Wenstrup, 1995; Leroy and Wenstrup, 2000) and that they respond to different types of vocalizations in a very selective manner (Klug et al., 2002; Pollak, 2013). This response selectivity is sensitive to fine differences in the temporal structure of natural vocalizations (Salles et al., 2020) and might be one of the fundamental mechanisms underlying deviance detection in the IC. The results of Experiment 2 also underpin the previous observation of echolocation pulses eliciting the strongest effects of deviance detection, especially compared with social communication calls. Evidence for differences in the cortical processing of novelty detection between echolocation and communication stimuli in C. perspicillata has been reported before (López-Jury et al., 2021, 2023) and is in line with the current brainstem data. A possible reason for this phenomenon is the fact that both natural communication calls used in this study are distress calls that the animal emits when it is under physical duress (Knörnschild et al., 2014; González-Palomares et al., 2021). Those distress calls might always elicit the strongest possible neural response since they are produced by other individuals and thus are always unexpected. On the contrary, echolocation pulses are self-generated vocalizations that are used to extract information about the environment the bat is moving in. It is crucial for the bat to rapidly extract the statistical parameters of the perceived pulse-echo sequences since they convey information that is relevant for orientation and navigation. This may explain the strong effects of echolocation-related deviance detection already in the lowest stations of the auditory pathway of bats. The deviance detection differences caused by echolocation and communication signals demonstrate that both types of stimuli are processed differently and potentially independently from each other. This observation links the current data with another phenomenon related to auditory processing, namely, auditory streaming. Auditory streaming suggests that animals can separate different categories of auditory input (e.g., echolocation and communication sequences) into different perceptual streams (Moss and Surlykke, 2001; Kanwal et al., 2003). Under this perspective, the asymmetry of deviance detection between echolocation and communication stimuli in the brainstem could be interpreted as an indicator that both signals are processed in separate streams, already very early in the ascending auditory pathway.
It can be hypothesized that the observed effects would be even larger in awake animals, as it has been shown that anesthesia can reduce the amount of stimulus-specific adaptation (SSA) of individual neurons (Cai et al., 2016; Parras et al., 2017). However, consistent with the data presented here, many former studies could show that deviance detection may be modulated by but does not require wakefulness or attention (Näätänen et al., 1978, 2007; Sculthorpe et al., 2008; Cai et al., 2016; Parras et al., 2017). As mentioned earlier, many bat species have a rich repertoire of communication calls, with the tested distress calls being only one example for these social vocalizations. It should be noted that other types of communication calls may elicit different subthalamic deviance detection responses. Follow-up studies investigating more combinations of social communication calls in the context of deviance detection will help to understand differences in the neural processing and meaning of different call types.
Conclusion
In this study, noninvasively recorded ABRs revealed that deviance detection responses to vocalizations that have different behavioral meanings for bats—navigation and communication—are processed in a complex and asymmetric way already at the earliest stations of the ascending auditory pathway. In fact, the results show that when considering the population response, subthalamic deviance detection is sensitive to physical (carrier frequency, FM and AM) as well as abstract stimulus features (behavioral meaning of a vocalization). By this, population-based subthalamic deviance detection showed a higher complexity than what has been reported before for cellular SSA of neurons in the same brain areas. Given the evolutionary preserved nature of the brainstem, it can be speculated that complex forms of auditory deviance detection can also be found in other mammalian species, including humans.
Footnotes
We thank Dr. Mirjam Knörnschild for providing us with some of the vocalizations of Carollia perspicillata that were used in this study. This work was funded by the Deutsche Forschungsgemeinschaft (KO 987/14-1).
The authors declare no competing financial interests.
- Correspondence should be addressed to Johannes Wetekam at wetekam{at}bio.uni-frankfurt.de or Manfred Kössl at koessl{at}bio.uni-frankfurt.de.
